fMRI, EEG, and fNIRS: A Comparative Guide to Principles, Applications, and Optimization in Modern Brain Research

Abstract

This article provides a comprehensive overview of three cornerstone neuroimaging techniques—functional Magnetic Resonance Imaging (fMRI), Electroencephalography (EEG), and functional Near-Infrared Spectroscopy (fNIRS)—tailored for researchers and drug development professionals. It explores the foundational principles and physiological basis of each modality, contrasting their spatial and temporal resolution. The scope extends to methodological applications in clinical and cognitive neuroscience, including multimodal integration strategies. It further addresses critical troubleshooting, optimization approaches, and reproducibility challenges, culminating in a rigorous validation and comparative analysis of their strengths and limitations. This synthesis aims to serve as a strategic resource for selecting and applying these tools in biomedical research and therapeutic development.

Core Principles and Physiological Basis of fMRI, EEG, and fNIRS

Functional neuroimaging has revolutionized our understanding of the human brain by enabling non-invasive observation of brain activity in real time. Among the most prominent techniques in cognitive neuroscience and clinical research are functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS). While each modality offers unique insights into brain function, they all relate—directly or indirectly—to the fundamental physiological process of neurovascular coupling, which links neural activity to subsequent changes in cerebral blood flow and oxygenation. This whitepaper provides a comprehensive technical overview of these core neuroimaging modalities, with particular focus on their relationship to the hemodynamic response, their comparative strengths and limitations, and their application in contemporary neuroscience research and drug development.

Core Principles and Measurement Techniques

The Hemodynamic Response Function

The hemodynamic response function (HRF) represents the transfer function that links neural activity to the measured fMRI signal, effectively modeling the neurovascular coupling process [1]. When neurons become active, they trigger a complex physiological cascade that increases local cerebral blood flow to meet metabolic demands, typically occurring 2-6 seconds after the neural event [2] [3]. This delayed response follows a characteristic shape that can be mathematically modeled using a Gamma distribution, peaking approximately 4-6 seconds after stimulus onset before returning to baseline [3]. The HRF can be characterized by three primary parameters: response height (amplitude), time-to-peak (latency), and full-width at half-maximum (duration) [1].

The neurovascular coupling mechanism involves intricate relationships between neural activity, cerebral blood flow (CBF), cerebrovascular reactivity (CVR), and vasodilation/vasoconstriction of local blood vessels [1]. Neurometabolic modulators released by glutamatergic and GABAergic interneurons directly and indirectly modulate CBF, with higher glutamate concentration resulting in taller, quicker, and narrower HRFs, while higher GABA has opposite effects [1]. This physiological foundation forms the basis for interpreting signals across all hemodynamic-based neuroimaging modalities.

Functional Magnetic Resonance Imaging (fMRI)

Physical and Physiological Basis

fMRI relies on the Blood-Oxygen-Level Dependent (BOLD) contrast mechanism, which exploits the different magnetic properties of oxygenated hemoglobin (diamagnetic) and deoxygenated hemoglobin (paramagnetic) [2] [4]. When neuronal activity increases in a specific brain region, the subsequent metabolic demand triggers an increased blood flow to that region, disproportionately increasing oxygenated hemoglobin relative to oxygen consumption [4]. This alters the local magnetic properties, detectable using T2*-weighted sequences sensitive to magnetic field variations [4].

Data Acquisition and Experimental Design

Echo Planar Imaging (EPI) serves as the primary acquisition method for fMRI studies, allowing whole-brain acquisition every 2-3 seconds [4]. fMRI experiments typically employ either block designs or event-related designs. Block designs alternate between periods of task performance and rest (e.g., 20 seconds of finger tapping followed by 20 seconds of rest), repeated multiple times to enhance signal detection [4] [5]. Event-related designs present discrete, short-duration stimuli with variable inter-stimulus intervals, allowing analysis of individual hemodynamic responses to each stimulus [5]. The correlation between the acquired fMRI data and the expected hemodynamic response curve is quantified mathematically by the correlation coefficient, with voxels demonstrating strong correlations interpreted as active regions [4].

Functional Near-Infrared Spectroscopy (fNIRS)

Measurement Principle

fNIRS employs near-infrared light (650-1000 nm) to measure changes in cerebral hemoglobin concentrations [2] [6]. When light at specific wavelengths penetrates biological tissues, chromophores—particularly oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR)—exhibit characteristic absorption patterns [6]. By placing light sources and detectors on the scalp, fNIRS systems measure the attenuated light intensity, from which concentration changes of HbO and HbR can be computed using the Modified Beer-Lambert Law [6]. Continuous wave NIRS (CW-NIRS) represents the most extensively used approach in research and clinical settings due to its low cost and simplicity [6].

Physiological Basis and Relationship to fMRI

Like fMRI, fNIRS measures the hemodynamic response consequent to neural activity, with fNIRS measurements demonstrating similarity to the BOLD response obtained by fMRI [2] [6]. However, while fMRI's BOLD signal primarily reflects changes in deoxygenated hemoglobin, fNIRS separately quantifies both oxygenated and deoxygenated hemoglobin concentrations [2]. This provides complementary information about the hemodynamic response, with HbO typically increasing and HbR decreasing during neuronal activation [6].

Electroencephalography (EEG)

Neural Basis and Signal Generation

EEG measures the brain's electrical activity via electrodes placed on the scalp, detecting voltage changes resulting from synchronized firing of cortical neurons, primarily pyramidal cells [7] [6]. These post-synaptic potentials represent the summed activity of tens of thousands of synchronized pyramidal neurons within the cortex, whose dendritic trunks are coherently orientated parallel to each other and perpendicular to the cortical surface, enabling sufficient signal summation to propagate to the scalp [6]. EEG signals are typically divided into characteristic frequency bands: theta (4-7 Hz), alpha (8-14 Hz), beta (15-25 Hz), and gamma (>25 Hz), each associated with different brain states and functions [6].

Relationship to Hemodynamic Measures

While EEG directly measures neural electrical activity with millisecond temporal resolution, it bears an indirect relationship to the hemodynamic response through neurovascular coupling [6]. The synchronized electrical activity detected by EEG represents the initial neural event that subsequently triggers the hemodynamic response measured by fMRI and fNIRS, providing complementary information about different aspects of brain function with distinct temporal characteristics.

Comparative Analysis of Neuroimaging Modalities

Table 1: Technical comparison of primary neuroimaging modalities

Feature	fMRI	fNIRS	EEG
What it Measures	BOLD signal (deoxygenated hemoglobin)	HbO and HbR concentration changes	Electrical activity from cortical neurons
Spatial Resolution	High (millimeter-level)	Moderate (cortical surface only)	Low (centimeter-level)
Temporal Resolution	Low (seconds)	Moderate (1-2 seconds)	High (milliseconds)
Penetration Depth	Whole brain	Outer cortex (1-2.5 cm)	Cortical surface
Portability	Low (requires MRI scanner)	High (wearable systems available)	High (wireless systems available)
Motion Tolerance	Low (highly sensitive to movement)	Moderate (relatively robust to motion)	Low (susceptible to movement artifacts)
Subject Limitations	Metal implants, claustrophobia	Limited to cortical measurements	Few limitations
Cost	High (equipment and scanning)	Moderate	Low to moderate
Acoustic Noise	High (120+ dB)	Minimal	Minimal

Table 2: Experimental considerations for different research applications

Research Context	Recommended Modality	Rationale
Studying rapid cognitive processes (e.g., sensory perception, attention)	EEG	Millisecond temporal resolution ideal for capturing rapid neural dynamics [7]
Localizing cortical activation during sustained tasks	fNIRS	Good spatial resolution for surface cortical areas with greater movement tolerance [2] [7]
Mapping deep brain structures	fMRI	Whole-brain coverage enables imaging of subcortical regions [2]
Naturalistic settings (classrooms, sports, clinical environments)	fNIRS	Portability and robustness to movement artifacts [2] [7]
Presurgical mapping	fMRI	Gold standard for identifying eloquent cortex relative to pathological structures [4]
Multimodal brain investigation	EEG + fNIRS	Complementary electrical and hemodynamic information with good portability [6]
Studies with sensitive populations (infants, children, patients)	fNIRS	Quiet operation, tolerance to movement, and no physical restrictions [2]

Integrated Methodologies and Experimental Protocols

Simultaneous fNIRS-fMRI Acquisition

Several studies have successfully combined fNIRS and fMRI to gain complementary insights into brain activity patterns. In a study by Jalavandi et al., simultaneous fNIRS and fMRI measurements during motor tasks showed strong correlation between modalities, validating fNIRS as a reliable alternative for subjects unable to undergo fMRI scans [2]. Similarly, Huppert et al. performed simultaneous fNIRS and fMRI measurements during parametric median nerve stimulation, finding good correspondence between the modalities [2].

Experimental Protocol: Simultaneous fNIRS-fMRI for Motor Tasks

• Participant Preparation: Screen for MRI contraindications; familiarize participant with motor tasks.
• Equipment Setup: Position fMRI-compatible fNIRS optodes over motor cortex regions using the international 10-20 system for placement consistency.
• Task Design: Implement block design with alternating 20-second periods of motor execution (e.g., wrist movement) and rest, repeated 6-8 times.
• Synchronization: Use TTL pulses from fMRI scanner to synchronize fNIRS and fMRI data acquisition.
• Data Acquisition: Collect simultaneous BOLD fMRI signals and fNIRS hemoglobin concentration changes throughout task performance.
• Data Analysis: Perform separate preprocessing pipelines followed by correlation analysis between BOLD signals and HbO/HbR concentration changes.

Concurrent fNIRS-EEG Recordings

The integration of EEG and fNIRS offers numerous benefits by exploiting their complementary strengths—EEG provides superior temporal resolution while fNIRS offers better spatial resolution and noise robustness [6]. Three primary methodological approaches have emerged for concurrent fNIRS-EEG data analysis: EEG-informed fNIRS analyses, fNIRS-informed EEG analyses, and parallel fNIRS-EEG analyses [6].

Experimental Protocol: Concurrent fNIRS-EEG for Cognitive Tasks

• Hardware Configuration: Use integrated EEG-fNIRS caps with predefined compatible openings to avoid sensor interference.
• Sensor Placement: Apply international 10-20 system for consistent placement of both EEG electrodes and fNIRS optodes.
• Synchronization: Employ external hardware (TTL pulses) or shared acquisition software for temporal alignment of data streams.
• Task Paradigm: Implement event-related design with cognitive tasks (e.g., working memory, attention tasks) suitable for both electrical and hemodynamic response capture.
• Data Collection: Simultaneously record EEG electrical potentials and fNIRS hemoglobin concentration changes.
• Signal Processing: Apply separate preprocessing pipelines (filtering, artifact removal) followed by data fusion techniques such as joint Independent Component Analysis (jICA) or canonical correlation analysis (CCA).

HRF Characterization in Clinical Populations

The hemodynamic response function has demonstrated sensitivity to brain pathology and treatment response. In a study examining obsessive-compulsive disorder (OCD), HRF parameters (response height, time-to-peak, and full-width at half-maximum) were abnormal in OCD patients compared to healthy controls and normalized following cognitive-behavioral therapy [1]. Furthermore, pre-treatment HRF measures predicted treatment outcome with 86.4% accuracy using machine learning approaches [1].

Experimental Protocol: Resting-State HRF Characterization

• Participant Groups: Include patient population (e.g., OCD, depression) and matched healthy controls.
• Data Acquisition: Collect resting-state fMRI data over 8-10 minutes using standard EPI sequences.
• HRF Estimation: Perform deconvolution on pre-processed fMRI data to estimate voxel-level HRF parameters.
• Parameter Extraction: Calculate response height, time-to-peak, and full-width at half-maximum across regions of interest.
• Statistical Analysis: Compare HRF parameters between groups and correlate with clinical measures or treatment outcomes.

Visualization of Core Concepts

Neurovascular Coupling Pathway

Diagram 1: Neurovascular coupling pathway linking neural activity to hemodynamic responses

Experimental Workflow for Multimodal Neuroimaging

Diagram 2: Experimental workflow for concurrent EEG-fNIRS studies

Research Reagent Solutions and Essential Materials

Table 3: Essential research reagents and materials for neuroimaging studies

Item	Function/Purpose	Application Notes
fMRI Contrast Agents (Gadolinium)	Enhances tissue contrast in MRI images by altering magnetic properties of water molecules [8]	Administered via IV injection; facilitates detailed visualization of organs, blood vessels, and abnormalities [8]
EEG Electrode Gel	Ensures conductive connection between scalp and electrodes for optimal signal acquisition [7]	Reduces impedance; requires proper scalp preparation; various conductivity formulations available
fNIRS Optode Holders	Secures light sources and detectors in predetermined positions on scalp [6]	Ensures consistent source-detector distances; compatible with EEG caps for multimodal studies
Disinfecting Solutions	Cleans EEG electrodes and fNIRS optodes between uses	Prevents cross-contamination; maintains signal quality
Synchronization Hardware	Coordinates timing across multiple acquisition systems (TTL pulses, parallel ports) [6]	Essential for multimodal studies; ensures temporal alignment of data streams
Anatomical Localization Tools	Correlates functional data with anatomical structures (3D digitizers, MRI-compatible markers) [2]	Enables precise sensor placement and registration with structural images
Motion Stabilization Equipment	Minimizes head movement artifacts during data acquisition	Particularly important for EEG and fMRI; includes foam padding, chin rests, and bite bars

The integration of multiple neuroimaging modalities represents the future of comprehensive brain investigation, leveraging the complementary strengths of each technique to overcome individual limitations. fMRI continues to serve as the gold standard for spatial localization of brain activity, particularly for deep structures, while fNIRS offers an adaptable alternative for cortical mapping in naturalistic settings and with challenging populations. EEG remains unparalleled for capturing the temporal dynamics of neural processing. The hemodynamic response function provides a critical link between these modalities, reflecting the fundamental neurovascular coupling process that translates neural activity into measurable signals. As research continues to elucidate the relationships between the HRF, neural activity, and various disease states, these neuroimaging approaches will play increasingly important roles in both basic neuroscience and clinical applications, including pharmaceutical development and personalized treatment approaches. The ongoing development of integrated multimodal platforms and analytical approaches promises to further enhance our ability to non-invasively probe human brain function in health and disease.

Functional Magnetic Resonance Imaging (fMRI) has revolutionized non-invasive brain imaging since its development in the early 1990s, becoming a cornerstone of neurocognitive research and clinical practice [9]. The most common fMRI technique leverages the Blood-Oxygen-Level-Dependent (BOLD) contrast, which allows researchers and clinicians to visualize brain activity by detecting localized changes in blood oxygenation [10] [9]. This signal serves as an indirect marker of neural activity, capitalizing on the tight coupling between cerebral blood flow, energy demand, and neural firing [10]. The BOLD effect is fundamentally rooted in the magnetic properties of hemoglobin: oxygenated hemoglobin is diamagnetic while deoxygenated hemoglobin is paramagnetic [9]. This difference causes deoxygenated hemoglobin to act as an intrinsic contrast agent that produces local distortions in the magnetic field, leading to a detectable loss of T2* MRI signal [10] [9].

When neurons become active, a complex neurovascular response is triggered. The ensuing metabolic demand drives a regional increase in cerebral blood flow that overcompensates for the local oxygen consumption [10] [9]. This results in a localized decrease in deoxyhemoglobin concentration relative to the baseline, thereby reducing its paramagnetic effect and causing an increase in the T2* signal detected by MRI [9]. These signal changes, while subtle—typically ranging from about 2% on a 1.5 Tesla scanner to approximately 12% on a 7 Tesla system—can be reliably measured with appropriate statistical methods [9]. The temporal dynamics of this response are characterized by a predictable latency; the BOLD signal onset is typically delayed by ∼2 seconds after neural activity, peaks after 6–12 seconds, and often exhibits a prolonged post-stimulus undershoot before returning to baseline [10].

The Physiological Basis of the BOLD Signal

Neurovascular Coupling and Hemodynamic Response

The BOLD signal is an indirect measure of brain activity that depends on the intricate process of neurovascular coupling—the relationship between neural activity, metabolic demand, and subsequent hemodynamic changes [6] [11]. When a brain region becomes active, it consumes more oxygen and nutrients, triggering a complex biochemical signaling cascade that ultimately increases local cerebral blood flow (CBF) to meet this heightened demand [6]. The exact mechanisms remain under investigation but likely involve chemical mediators such as nitrous oxide and glutamate, with possible participation from astrocytes [9].

This hemodynamic response exhibits characteristic temporal properties that fundamentally constrain the temporal resolution of BOLD fMRI. The typical timeline includes:

• Onset delay: ∼2 seconds after neural activity begins [10]
• Time to peak: 6-12 seconds to reach maximum amplitude [10]
• Return to baseline: A similar ramp down, often followed by a prolonged post-stimulus undershoot [10]

The observed BOLD signal intensity enhancement reflects an increase in CBF that overcompensates for the increased oxygen consumption, resulting in an oversupply of oxygenated blood to active regions [10]. This phenomenon was first documented by Fox and Raichle (1986), who observed that the increase in cerebral blood flow exceeds the increase in cerebral metabolic rate of oxygen (CMRO2) during neural activation [10].

Biological Pathway of the BOLD Response

The following diagram illustrates the sequential physiological events that generate the measurable BOLD signal, from initial neural activation to the resulting MRI signal change.

Technical Implementation and Experimental Design

fMRI Acquisition and Pulse Sequences

BOLD fMRI detection requires specialized MRI pulse sequences sensitive to T2* variations. The most common approach involves using T2*-sensitive echo planar imaging (EPI) sequences, particularly gradient-echo (GRE) EPI, which can rapidly acquire whole-brain images [9]. Ogawa's initial discovery of BOLD contrast emerged from observations that gradient-echo pulse sequences produced signal dropouts from blood vessels due to susceptibility effects from deoxyhemoglobin, while spin-echo sequences did not [10]. Each individual scan through the brain typically takes between 333-3000 milliseconds depending on scanner model and pulse sequence parameters, enabling repeated sampling of the brain's hemodynamic state over time [9].

The sensitivity of BOLD fMRI is significantly influenced by magnetic field strength. Higher field strengths (e.g., 3T, 7T, and above) provide greater BOLD signal changes and improved spatial resolution [12] [9]. Recent advances in ultra-high field (7T) MRI have enabled more detailed investigations, including laminar fMRI that can differentiate activity across cortical layers with resolutions approaching 1 mm³ [12]. Innovative multi-contrast approaches now combine BOLD with complementary measures like cerebral blood flow (CBF) and cerebral blood volume (CBV) to provide a more comprehensive characterization of neurovascular responses [12].

Experimental Paradigms and Design

BOLD fMRI experiments employ standardized paradigms to elicit and measure brain activity:

Table 1: Common fMRI Experimental Paradigms

Paradigm Type	Design	Advantages	Limitations
Block Design	Stimuli presented in extended blocks (e.g., 20-30s) alternating with rest periods [9]	High statistical power for detecting activation [9]	Predictable presentation may induce habituation or strategy effects
Event-Related Design	Stimuli presented at random or pseudorandom intervals [9]	More naturalistic presentation; can model hemodynamic response to single events [9]	Lower statistical power than block designs
Mixed Designs	Combination of block and event-related elements	Benefits of both approaches; can separate sustained and transient activity	Increased analytic complexity

Clinical applications, particularly presurgical mapping, require special considerations. The Organization for Human Brain Mapping (OHBM) Clinical fMRI Working Group has established consensus recommendations for clinical language mapping, emphasizing task designs optimized for specific clinical objectives and modifications for patients with existing impairments [13]. For example, establishing language dominance often requires multiple language tasks targeting different components of the language system [13].

Analysis and Interpretation of BOLD Signals

Preprocessing and Statistical Analysis

fMRI data analysis involves multiple stages to transform raw MR images into interpretable statistical maps of brain activation:

• Preprocessing: Corrects for various confounding factors including patient head motion, slice timing differences, and image noise or artifacts [9]. In research settings, preprocessing often includes spatial normalization to align images to a common anatomical space or atlas [9].

• Statistical Analysis: Employs voxel-wise statistical comparisons of T2* signal between task and control conditions [9]. Common approaches include:

  • General linear model (GLM)
  • Correlation analysis
  • Voxel-wise t-tests

The final output is a statistical map showing voxels where stimulus-related activation exceeds a specified threshold (e.g., p-value, t-value, or z-score) [9]. For clinical applications, statistical thresholds often require individual customization due to inter-subject variability in cerebrovascular responsiveness, whereas research studies typically apply uniform thresholds across subjects with appropriate corrections for multiple comparisons [9].

Relationship to Underlying Neural Activity

A critical consideration in BOLD fMRI interpretation is understanding what specific aspects of neural activity the signal reflects. The BOLD signal appears to be more closely tied to local field potentials (LFPs) and input processing rather than spiking output [10]. Extracellular recordings suggest that the BOLD signal represents the weighted sum of all sinks and sources along multiple cells—essentially reflecting integrated synaptic activity rather than individual action potentials [10].

This relationship becomes particularly important when interpreting negative BOLD responses (NBR), which can reflect either reduced neuronal activity or heightened neuronal activity under certain conditions, depending on the complex interplay between CBF, CBV, and CMRO2 [12]. Advanced multi-contrast laminar fMRI at 7T has demonstrated distinct neurovascular and metabolic responses across cortical layers, suggesting potential feedback inhibition of neuronal activities in both superficial and deep cortical layers underlying negative BOLD signals [12].

Advanced Applications and Multimodal Integration

Clinical Applications of BOLD fMRI

BOLD fMRI has established significant clinical utility, particularly in presurgical planning:

Table 2: Primary Clinical Applications of BOLD fMRI

Application	Purpose	Clinical Utility
Presurgical Mapping	Identify eloquent cortex (motor, language, visual) near surgical targets [13] [9]	Define surgical risk, aid operative planning, expedite intraoperative mapping [9]
Language Lateralization	Determine hemisphere dominance for language functions [13]	Inform patient consent, decide whether to proceed with surgery [13]
Cerebrovascular Reactivity	Assess vascular reserve in patients with cerebrovascular disease [9]	Identify regions with impaired vasodilatory capacity [9]

When combined with diffusion tensor imaging (DTI) for visualizing white matter tracts, fMRI has been shown to reduce postoperative neurologic deficits [9]. However, it is crucial to recognize that fMRI provides statistical activation maps at arbitrary thresholds rather than direct anatomical representations, which must be considered when integrating them into surgical navigation systems [9].

Integration with EEG and fNIRS in Multimodal Studies

BOLD fMRI serves as a cornerstone in multimodal brain imaging approaches, often combined with electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) to overcome the inherent limitations of each individual technique:

• EEG-fMRI Integration: EEG provides millisecond-level temporal resolution of electrical brain activity, complementing fMRI's spatial precision [6]. Simultaneous EEG-fMRI recording presents technical challenges but enables direct correlation of electrical and hemodynamic brain events [6].

• fNIRS-fMRI Integration: fNIRS measures hemodynamic changes using near-infrared light, providing a portable alternative for measuring brain oxygenation [6] [11]. Simultaneous fNIRS-fMRI studies have demonstrated correlations between BOLD signals and hemoglobin concentration changes, validating fNIRS as a reliable hemodynamic measure [14] [11]. This integration is particularly valuable for studies in naturalistic settings where MRI scanners cannot be used [11].

The combination of these modalities capitalizes on their complementary strengths: fMRI's high spatial resolution, EEG's exceptional temporal resolution, and fNIRS's portability and suitability for long-term monitoring [6] [11]. This multimodal approach provides a more comprehensive understanding of brain function across different spatiotemporal scales and experimental contexts.

Research Toolkit: Essential Methods and Reagents

Table 3: Key Research Reagents and Solutions in BOLD fMRI Studies

Item	Function/Application	Technical Specifications
MRI Scanner	Image acquisition with T2* sensitivity	Clinical (1.5T, 3T) to research (7T, 11T) field strengths [12] [9]
EPI/GRE Pulse Sequence	T2*-weighted image acquisition	Enables rapid whole-brain imaging; sensitive to magnetic susceptibility [9]
Visual Presentation System	Stimulus delivery	MRI-compatible goggles, projectors, or LCD screens with precise timing [9]
Auditory Presentation System	Auditory stimulus delivery	MRI-compatible headphones or ear buds with artifact minimization [9]
Response Recording Device	Subject performance monitoring	MRI-compatible button boxes, joysticks, or eye-tracking systems [13]
Physiological Monitoring	Cardiorespiratory recording	Pulse oximetry, respiration monitoring for noise correction [9]
Analysis Software	Data processing and statistical mapping	FSL, SPM, AFNI; commercial clinical packages [9]

The BOLD signal has fundamentally transformed our ability to noninvasively investigate human brain function, providing a window into the complex interplay between neural activity, metabolism, and hemodynamics. While technically challenging and methodologically complex, BOLD fMRI continues to evolve through advancements in high-field imaging, sophisticated analysis techniques, and multimodal integration. Understanding its physiological basis, technical implementation, and analytical approaches is essential for proper interpretation and application across basic neuroscience and clinical domains. As technological innovations continue to emerge, BOLD fMRI remains an indispensable tool for unraveling the functional organization of the human brain.

Electroencephalography (EEG) is a non-invasive neurophysiological technique that measures the brain's spontaneous electrical activity from the scalp surface. First described by Hans Berger in 1929, EEG records the summation of synchronous postsynaptic potentials from billions of cortical neurons, primarily pyramidal cells [6] [15]. These electrical signals represent the macroscopic activity of the brain surface, requiring at least 6 square centimeters of synchronized cortical activity to be detectable by scalp electrodes [16]. The exquisite temporal resolution of EEG (on the millisecond scale) enables researchers to capture neural dynamics associated with rapid cognitive processes, making it an indispensable tool for studying brain function in both research and clinical settings [6] [17].

The electrical potentials measured by EEG originate mainly from excitatory and inhibitory postsynaptic currents generated by cortical pyramidal neurons. When many neurons within a narrow timeframe are activated, their current dipoles summate, producing measurable voltage fluctuations on the scalp [15]. The rhythmic oscillatory patterns of EEG reflect synchronized activity of neuronal circuits connecting different brain regions, with thalamic pacemaker neurons synchronizing cortical firing to generate characteristic rhythms [15]. This neurophysiological foundation enables EEG to serve as a direct measure of neural electrical activity, contrasting with hemodynamic-based techniques like fMRI and fNIRS that measure metabolic responses coupled to neural activity [6].

Core Electrophysiological Principles

From Neuronal Firing to Scalp Potentials

At the cellular level, the genesis of EEG signals begins with the electrical properties of individual neurons. Neuronal membranes maintain a resting potential of approximately -70mV through ion channels, most notably the sodium-potassium pump that exchanges three Na+ ions out for every two K+ ions into the cell [16]. When a neuron receives excitatory input, neurotransmitter binding leads to depolarization through the opening of Na+/K+ channels, causing Na+ influx that shifts the intracellular voltage toward positive values (e.g., from -70mV to -20mV) [16]. This depolarization creates a relatively negative voltage outside the cell, which EEG electrodes can detect when summed across thousands of synchronously activated neurons.

Critical to EEG generation is the specific architecture of cortical pyramidal neurons. Their dendritic trunks are coherently oriented, parallel with each other and perpendicular to the cortical surface, enabling sufficient summation and propagation of electrical signals to the scalp [6]. The arrangement creates what is effectively a dipole field, with charge separations that can be detected at a distance. When superficial cortical layers undergo excitatory postsynaptic potentials (EPSPs), the extracellular space near the scalp becomes negatively charged, producing negative deflections on EEG. Conversely, deep EPSPs or superficial inhibitory postsynaptic potentials (IPSPs) produce positive scalp deflections [16]. This relationship is crucial for accurate interpretation of EEG waveforms.

Table 1: Neural Events and Their EEG Correlates

Neural Event	Cortical Depth	Extracellular Potential	EEG Deflection
Excitatory (EPSP)	Superficial	Negative	Upward (Negative)
Excitatory (EPSP)	Deep	Positive	Downward (Positive)
Inhibitory (IPSP)	Superficial	Positive	Downward (Positive)
Inhibitory (IPSP)	Deep	Negative	Upward (Negative)

Signal Characteristics and Limitations

EEG signals possess distinctive characteristics that determine their utility and limitations in brain research. The voltage fluctuations measured on the scalp are typically in the range of 10-100 microvolts, requiring significant amplification for analysis [18]. Several factors constrain what EEG can detect: the electrical signals must pass through cerebrospinal fluid, skull, and scalp, which act as a low-pass filter attenuating high-frequency components and spatially blurring the source signals [6]. This biological filtering effect contributes to EEG's limited spatial resolution, as the electrical activity from a localized cortical region spreads before reaching scalp electrodes.

The orientation of neuronal dipoles significantly impacts their detectability on scalp EEG. Dipoles perpendicular to the cortical surface are well-detected, while those parallel or tangential to the scalp are poorly seen or missed completely [16]. This has important implications for localizing brain activity, particularly in regions with complex cortical folding patterns. For instance, activity originating in the interhemispheric fissure with transverse dipoles may appear to come from the contralateral hemisphere, creating potential false localization [16]. Understanding these fundamental principles is essential for proper experimental design and interpretation of EEG findings in research contexts.

EEG Signal Acquisition Methodology

Electrode Systems and Placement

Standardized electrode placement follows the International 10-20 system or its higher-resolution variants (10-10, 10-5 systems), which ensures consistent positioning relative to cranial landmarks and proportional coverage across head sizes [18] [19]. This system uses specific anatomical reference points (nasion, inion, preauricular points) to create a coordinate system for electrode placement, with letters indicating brain regions (F-frontal, C-central, P-parietal, O-occipital, T-temporal) and numbers indicating specific positions within those regions [19]. Modern high-density EEG systems can utilize 128, 256, or more electrodes to provide improved spatial sampling of brain electrical activity.

During acquisition, voltage differences between each electrode and a reference electrode are measured and amplified [6]. The choice of reference significantly influences the recorded signals, with common options including linked ears, averaged reference, or cephalic references. Proper electrode application requires careful skin preparation and conductive gel to maintain impedance below 5-10 kΩ, ensuring optimal signal quality [18]. Modern systems increasingly use active electrodes with built-in impedance conversion to improve signal quality and reduce environmental interference.

Technical Considerations and Equipment

EEG acquisition systems consist of electrodes, amplifiers, filters, and analog-to-digital converters. Amplifiers must have high common-mode rejection ratios (typically >100 dB) to reject noise that appears equally at all electrodes while amplifying the differential signals of interest [18]. Signal filtering is applied during acquisition, with typical bandpass settings of 0.1-100 Hz to capture relevant neural activity while excluding non-physiological frequencies. The sampling rate must be sufficiently high (usually 200-1000 Hz or higher) to avoid aliasing while capturing the frequency content of interest [20].

Equipment for EEG acquisition ranges from traditional laboratory-based systems to increasingly portable and wearable devices. Research-grade systems typically offer high channel counts (64-256 channels), precision timing, and extensive support for multimodal integration [18]. Portable systems such as the Emotiv EPOC (14 channels) and Muse (4 channels) provide greater flexibility for naturalistic experiments but with potentially reduced signal quality and spatial resolution [18]. The emergence of wearable EEG technology has enabled field studies and long-term monitoring outside traditional laboratory settings, expanding the methodological possibilities for brain research.

Table 2: EEG Acquisition Systems and Their Characteristics

System Type	Channel Count	Portability	Typical Applications	Key Considerations
Research-grade	64-256+	Low	Laboratory studies, clinical	High signal quality, precise timing
Clinical	19-32	Moderate	Diagnostic medicine	Standardized montages, medical safety
Portable/Wearable	4-32	High	Field studies, BCI, monitoring	Trade-offs between mobility and data quality

EEG Signal Processing and Analysis

Preprocessing and Denoising

Raw EEG signals are invariably contaminated with various artifacts and noise sources that must be addressed before meaningful analysis. Major artifact sources include ocular movements (EOG), muscle activity (EMG), cardiac signals (ECG), skin potentials, and environmental interference such as power line noise [18] [21]. Effective preprocessing pipelines typically include multiple stages: filtering (bandpass, notch), artifact detection, and correction or rejection of contaminated segments.

Advanced preprocessing techniques include Independent Component Analysis (ICA), which separates statistically independent sources from the recorded signals, enabling identification and removal of artifact-related components while preserving neural activity [18]. Other methods include regression-based approaches, blind source separation, and adaptive filtering. The specific choice of preprocessing methods depends on the experimental paradigm, artifact types, and subsequent analysis goals. Validation of preprocessing effectiveness is crucial, often involving both automated metrics and visual inspection to ensure meaningful neural signals are preserved while artifacts are adequately addressed.

Feature Extraction and Quantitative Analysis

Quantitative EEG (qEEG) analysis transforms raw waveforms into measurable features that characterize brain states and cognitive processes. The most fundamental analysis approach involves decomposing the EEG signal into characteristic frequency bands, each associated with different functional states [15]:

• Delta (0.1-4 Hz): Prominent in deep sleep, coma
• Theta (4-8 Hz): Seen in drowsiness, meditation
• Alpha (8-12 Hz): Characteristic of relaxed wakefulness with eyes closed
• Beta (12-30 Hz): Dominant during active thinking, focus
• Gamma (>30 Hz): Associated with cognitive engagement, sensory binding

Multiple mathematical approaches are available for feature extraction from EEG signals. The Fast Fourier Transform (FFT) calculates power spectral density to quantify the distribution of signal power across frequency bands [15] [21]. Wavelet Transform (WT) provides time-frequency representation with variable window sizes, offering superior analysis of non-stationary signals like EEG [21]. Eigenvector methods (Pisarenko, MUSIC, Minimum Norm) estimate signal frequency and power from noise-corrupted measurements [21]. Each method has distinct advantages and limitations, with selection depending on the specific analysis goals and signal characteristics.

Figure 1: EEG Signal Processing Pipeline

Experimental Protocols and Methodologies

Standard Experimental Paradigms

EEG research employs standardized protocols to investigate specific cognitive processes and brain functions. Event-Related Potentials (ERPs) are obtained by time-locking EEG segments to stimulus events and averaging across multiple trials to extract consistent neural responses embedded in background activity [18]. Common ERP components include the P300 (associated with attention and context updating), N400 (language processing), and MMN (deviance detection). These components provide precise temporal information about cognitive processes with millisecond resolution.

Resting-state EEG protocols record spontaneous brain activity during awake, relaxed states with eyes closed or open, providing measures of baseline brain dynamics and functional connectivity [19]. Quantitative features derived from resting-state EEG, such as spectral power ratios and network connectivity metrics, serve as biomarkers for various neurological and psychiatric conditions. Task-based EEG paradigms engage specific cognitive functions through carefully designed experimental tasks, enabling researchers to study the neural correlates of perception, attention, memory, decision-making, and other cognitive processes.

Clinical and Applied Protocols

In clinical neuroscience, specialized EEG protocols facilitate diagnosis and monitoring of neurological disorders. For epilepsy evaluation, prolonged EEG monitoring captures interictal and ictal activity, with automated detection algorithms employing measures of signal amplitude variation, pattern regularity, and frequency characteristics to identify seizure patterns [20]. In critical care settings, continuous EEG monitoring detects nonconvulsive seizures in comatose patients, utilizing quantitative trend analysis to simplify review of extended recordings [20].

Brain-Computer Interface (BCI) protocols establish real-time communication between brain activity and external devices, often using sensorimotor rhythms, P300 responses, or steady-state visually evoked potentials as control signals [18]. These protocols require specialized signal processing for real-time feature extraction and classification, with applications in assistive technology, neurorehabilitation, and human-computer interaction. Each protocol type demands specific experimental design considerations, including appropriate control conditions, trial structure, and timing parameters to ensure valid and interpretable results.

Comparative Framework with Other Neuroimaging Modalities

EEG versus fNIRS: Complementary Strengths

EEG and functional near-infrared spectroscopy (fNIRS) represent complementary approaches to non-invasive brain imaging, each with distinct strengths and limitations. While EEG measures direct electrical neural activity with millisecond temporal resolution, fNIRS detects hemodynamic responses (changes in oxygenated and deoxygenated hemoglobin) with slower temporal resolution (seconds) but better spatial localization for cortical areas [6] [17]. This fundamental difference in measured signals creates opportunities for multimodal integration that leverages the advantages of both techniques.

The spatial resolution of EEG is limited by the volume conduction of electrical signals through head tissues, typically localizing activity to regions of several square centimeters. fNIRS provides better spatial resolution for superficial cortical areas but is limited to measuring outer cortical layers (1-2.5 cm depth) [17]. fNIRS is more tolerant of movement artifacts than EEG, making it suitable for studies involving natural movements, children, or real-world environments [17]. The choice between modalities depends on the research question, with EEG preferred for studying rapid neural dynamics and fNIRS advantageous for investigating localized cortical activity during naturalistic tasks.

Table 3: Comparison of EEG and fNIRS Characteristics

Characteristic	EEG	fNIRS
What is Measured	Electrical activity of neurons	Hemodynamic response (blood oxygenation)
Signal Source	Postsynaptic potentials in cortical neurons	Changes in oxygenated/deoxygenated hemoglobin
Temporal Resolution	High (milliseconds)	Low (seconds)
Spatial Resolution	Low (centimeter-level)	Moderate (better than EEG)
Depth of Measurement	Cortical surface	Outer cortex (1-2.5 cm deep)
Sensitivity to Motion	High	Low
Portability	High	High
Best Use Cases	Fast cognitive tasks, ERP studies, seizure detection	Naturalistic studies, child development, motor rehabilitation

Multimodal Integration: EEG-fNIRS

Simultaneous EEG-fNIRS recording provides comprehensive assessment of brain function by capturing both electrical neural activity and hemodynamic responses [6] [19]. This multimodal approach enables investigation of neurovascular coupling - the relationship between neural activity and subsequent cerebral blood flow changes [6]. Integration approaches include EEG-informed fNIRS analysis, fNIRS-informed EEG analysis, and parallel analyses that combine information from both modalities [6].

Technical implementation of simultaneous EEG-fNIRS requires careful consideration of sensor placement compatibility, typically using integrated caps with predefined positions for both electrodes and optodes following the 10-20 system [17]. Hardware synchronization ensures temporal alignment of data streams, while specialized processing pipelines address modality-specific artifacts and combine features for enhanced classification of brain states [17] [19]. Multimodal integration has demonstrated particular utility in brain-computer interfaces, where combined electrical and hemodynamic features improve classification accuracy and information transfer rates compared to either modality alone.

Figure 2: EEG and fNIRS Complementary Strengths and Multimodal Integration

Essential Research Tools and Reagents

The Scientist's Toolkit for EEG research encompasses specialized equipment, software resources, and methodological components that enable comprehensive investigation of brain electrical activity. These tools facilitate signal acquisition, processing, analysis, and interpretation across diverse experimental contexts.

Table 4: Essential EEG Research Resources

Resource Category	Specific Examples	Function/Purpose
Acquisition Systems	Research-grade EEG systems (BrainAmp, Biosemi), Portable systems (Emotiv EPOC, Muse)	Signal recording with precise timing and amplification
Electrodes & Supplies	Active/passive electrodes, Electrode caps, Conductive gel, Abrasive preparations	Signal transduction from scalp to recording system
Analysis Software	EEGLAB, Brainstorm, MNE-Python, FieldTrip	Signal processing, visualization, and statistical analysis
Quantitative Tools	qEEGt Toolbox, VARETA source imaging	Normative comparison, source localization, spectral analysis
Experimental Platforms	Presentation, E-Prime, PsychToolbox	Stimulus presentation and experimental control
Reference Databases	Cuban Human Brain Mapping Project, CAMCAN	Normative comparisons, methodological validation

Advanced analytical tools like the qEEGt Toolbox enable quantitative EEG analysis integrated with the Montreal Neurological Institute neuroinformatics ecosystem, producing age-corrected normative statistical parametric maps of EEG source spectra [22]. This toolbox incorporates the Variable Resolution Electrical Tomography (VARETA) method for source imaging and provides z-spectra based on normative databases, facilitating comparison of individual subjects against population norms [22]. Such standardized processing pipelines enhance reproducibility and enable multi-site collaborations in EEG research.

EEG fundamentals rest upon the neurophysiological principles of synchronized neuronal activity that generates measurable electrical potentials on the scalp surface. The technique's exceptional temporal resolution provides direct insight into neural dynamics across diverse cognitive states and pathological conditions. While limitations in spatial resolution and sensitivity to deep sources persist, ongoing methodological advances in high-density recording, source localization, and multimodal integration continue to expand EEG's research applications.

Understanding the core principles of EEG signal generation, acquisition, and analysis is essential for leveraging this technology effectively in neuroscience research and clinical applications. The integration of EEG with complementary modalities like fNIRS creates powerful frameworks for comprehensive brain investigation, bridging gaps between electrical neural activity and hemodynamic responses. As neurotechnologies evolve, EEG maintains its fundamental role as a versatile, non-invasive window into human brain function, with particular utility for studying the temporal dynamics of cognitive processes that underlie perception, cognition, and behavior.

Functional Near-Infrared Spectroscopy (fNIRS) is a non-invasive, portable neuroimaging technique that measures cerebral hemodynamic activity by detecting changes in near-infrared light absorption by brain tissue [23] [24]. First reported by Jöbsis in 1977, fNIRS leverages the relative transparency of biological tissue to light in the 700-900 nm range, known as the "optical window," to infer neural activity indirectly via neurovascular coupling [25] [23] [6]. This article provides an in-depth technical examination of fNIRS fundamentals, detailing its physical basis, instrumentation, signal processing, and experimental protocols, contextualized within the broader landscape of brain research tools like fMRI and EEG.

Core Physical Principles of fNIRS

The Optical Window and Light-Tissue Interaction

fNIRS functionality relies on the optical properties of biological tissues and chromophores. Within the near-infrared spectrum (700-900 nm), skin, skull, and brain tissue scatter light but absorb it relatively weakly, allowing light to penetrate several centimeters and reach the cerebral cortex [23] [24]. The primary absorbers in this window are oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR), which have distinct and sensitive absorption spectra [23] [6]. Light emitted onto the scalp is either absorbed by these chromophores or scattered within the tissue. A detector placed a few centimeters away measures the back-scattered light, the attenuation of which depends on the absorption properties of the underlying tissue, which in turn reflects hemoglobin concentration changes [25] [24].

The Modified Beer-Lambert Law

The conversion of measured light intensity into hemodynamic changes is primarily achieved using the Modified Beer-Lambert Law (mBLL) for continuous-wave systems, the most common fNIRS type [23] [6]. The standard Beer-Lambert law is modified to account for significant light scattering in tissue. The fundamental equation is expressed as:

OD = log₁₀(I₀/I) = ϵ · [X] · l · DPF + G [23]

Where:

• OD is the optical density (unitless)
• I₀ and I are the input and detected light intensities, respectively
• ϵ is the extinction coefficient of the chromophore (HbO or HbR)
• [X] is the concentration of the chromophore
• l is the physical distance between source and detector
• DPF is the differential pathlength factor, accounting for increased photon pathlength due to scattering
• G is a geometry-dependent factor

To solve for the relative changes in HbO and HbR concentrations (Δ[HbO] and Δ[HbR]), measurements at a minimum of two wavelengths are required, forming a system of equations [23]:

(ΔODλ1)		(ϵλ1Hbd ϵλ1HbO2d)	(Δ[X]Hb)
(ΔODλ2)	=	(ϵλ2Hbd ϵλ2HbO2d)	(Δ[X]HbO2)

Here, d represents the total mean pathlength (l · DPF) [23]. The following diagram illustrates the complete photon migration and concentration calculation pathway.

fNIRS Instrumentation and System Types

fNIRS systems are categorized based on their light emission and detection techniques, each with distinct advantages and limitations. The key system types are detailed in the table below.

Table 1: Comparison of fNIRS System Types

System Type	Basic Principle	Key Measurements	Pathlength Knowledge	Relative Cost & Complexity	Primary Use/Advantage
Continuous Wave (CW) [23] [6]	Constant intensity light source	Light attenuation (ΔOD)	Estimated (uses DPF factor)	Low cost, simple	Most common; suitable for relative concentration changes
Frequency-Domain (FD) [23]	Amplitude-modulated light (~100 MHz)	Attenuation, phase shift, average pathlength	Directly measured	High cost, complex	Provides absolute concentrations of HbO and HbR
Time-Domain (TD) [23] [24]	Short light pulses (~picoseconds)	Photon time-of-flight	Directly measured	Highest cost, most complex	High depth resolution; can separate absorption and scattering

CW-fNIRS is the most prevalent system in research and clinical settings due to its simplicity, cost-effectiveness, and capacity for high-channel counts [23] [6]. The subsequent workflow diagram outlines the standard data acquisition and processing pipeline for a CW-fNIRS experiment.

Experimental Protocols and Methodologies

A Representative Experimental Protocol

A typical block-design fNIRS experiment involves participants performing a task in alternating blocks of activity and rest. The following methodology is adapted from a study investigating prefrontal cortex (PFC) activation during an auditory task with and without Active Noise Cancellation (ANC) technology [26] [27].

• Participants: 41 normal-hearing adults.
• Task: Auditory decision-making task performed in two conditions: ANC OFF and ANC ON.
• fNIRS Recording: Concentration changes of oxyhemoglobin (Δoxy-Hb) in the prefrontal cortex were measured using a multi-channel fNIRS system [26] [27].
• Additional Measures: Subjective listening effort was assessed using a Visual Analog Scale (VAS). Behavioral performance (accuracy and reaction time) was also collected [26].
• Procedure: The experiment consisted of multiple trials. Each trial began with a 30-second resting baseline, followed by the task period (self-paced, ~37-63 seconds), and concluded with another 30-second rest period [25] [26].

Key Research Reagent Solutions and Materials

The table below catalogs essential materials and their functions for a standard fNIRS experiment.

Table 2: Essential Materials and Reagents for fNIRS Research

Item	Function/Description	Example from Protocol
fNIRS Instrument	A system (typically CW) with sources and detectors to emit and record light.	32-channel CW-fNIRS system (TechEn CW6) [25].
fNIRS Cap/Probe	A headgear holding sources and detectors in predetermined locations.	Cap with sources and detectors placed according to the international 10-20 system [19] [28].
Short-Separation Detectors	Optional detectors placed ~8mm from a source to measure systemic physiological noise from the scalp for signal correction [23].	Used to separate brain signal from superficial scalp hemodynamics [23].
Anatomical Registration System	A 3D digitizer (e.g., FastSCAN stylus) to record the precise locations of optodes on the head for spatial mapping [25].	Used to register fNIRS probe position on each subject's head for inter-subject data registration [25].
Stimulus Presentation Software	Software to deliver controlled auditory, visual, or other stimuli to the participant.	Software presenting the auditory decision-making task [26]. Nintendo Wii Fit game for balance task [25].
Data Processing Software	Tools for converting raw signals, filtering, and statistical analysis (e.g., HOMER3, MNE-Python, NIRS Toolbox) [29] [23].	MNE-Python and Brainstorm software used for preprocessing optical density and converting to hemoglobin [19] [29].

Signal Preprocessing and Analysis

Raw fNIRS signals require extensive preprocessing to isolate the neural-related hemodynamic response. A standard pipeline, as implemented in tools like MNE-Python, includes the following steps [29]:

• Conversion to Optical Density: The raw light intensity signal is converted to optical density, which makes the signal linear with respect to changes in chromophore concentration [29].
• Quality Assessment: The Scalp Coupling Index (SCI) can be used to identify and exclude channels with poor contact between optodes and the scalp [19] [29].
• Conversion to Hemoglobin: The mBLL is applied to the optical density signals to compute relative concentration changes of HbO and HbR [29]. A common partial pathlength factor (ppf) of 0.1 is used [29].
• Filtering: A bandpass filter (e.g., 0.05 - 0.7 Hz) is applied to remove high-frequency noise (e.g., cardiac pulsation ~1 Hz) and very low-frequency drift [29].
• Epoching and Averaging: For task-based designs, the continuous data is segmented into epochs (e.g., -5 s to +15 s around stimulus onset). Epochs are then averaged across trials for each condition to improve the signal-to-noise ratio of the hemodynamic response [29].

fNIRS in the Broader Neuroimaging Landscape

fNIRS occupies a unique niche among non-invasive brain imaging techniques. The table below provides a comparative overview of its position relative to fMRI and EEG.

Table 3: fNIRS Compared to Other Primary Non-Invasive Brain Imaging Modalities

Feature	fNIRS	fMRI	EEG
What it Measures	Hemodynamic response (HbO/HbR) [6] [24]	Hemodynamic response (BOLD signal) [6]	Electrical activity from neurons [6] [28]
Temporal Resolution	Low (seconds) [6] [28]	Very Low (seconds) [6]	Very High (milliseconds) [6] [28]
Spatial Resolution	Moderate (surface cortex) [6] [28]	High (whole brain)	Low [6] [28]
Portability	High [6] [24]	Low (requires massive scanner)	High [6] [28]
Tolerance to Motion	Moderate/High [6] [28]	Low	Low [6] [28]
Measurement Depth	Superficial cortex (1-2 cm) [25] [28]	Whole brain	Primarily cortical surface [28]
Best Use Cases	Naturalistic studies, child development, clinical monitoring, mobility studies [24] [28]	Deep brain structures, high-precision anatomy	Rapid cognitive processes, event-related potentials, sleep studies [6] [28]

A significant trend is the multimodal integration of fNIRS with EEG, as the techniques are highly complementary [6] [28]. EEG provides direct, millisecond-level information on neural electrical activity, while fNIRS provides localized hemodynamic information with better spatial resolution and is less susceptible to motion artifacts [6]. This combination allows for a more comprehensive investigation of brain function and the relationship between electrical and hemodynamic activity (neurovascular coupling) [19] [6]. Integrated systems require careful synchronization of hardware and consideration of sensor placement compatibility, often using caps designed for both modalities [6] [28].

The human brain operates through two primary, interconnected physiological processes: rapid electrical neural signaling and a slower hemodynamic response that delivers energy. Modern neuroimaging techniques allow researchers to probe these processes non-invasively. Electroencephalography (EEG) directly measures electrical activity from populations of neurons, while functional Near-Infrared Spectroscopy (fNIRS) and functional Magnetic Resonance Imaging (fMRI) indirectly monitor neural activity by detecting associated changes in cerebral blood flow and oxygenation [6] [30]. Understanding the origins, relationships, and technical capabilities of these signals is fundamental to advancing brain research. This guide provides an in-depth technical comparison of these modalities, detailing their neurophysiological bases, measurement principles, and methodologies for integrated use.

Fundamental Neurophysiological Signals

The Origin and Measurement of Electrical Activity

Electroencephalography (EEG) captures the electrical fields generated by the synchronous firing of large groups of cortical pyramidal neurons. These cells are orientated perpendicularly to the cortical surface, allowing their post-synaptic potentials to summate effectively and propagate to the scalp [6]. The resulting voltage differences, typically in the microvolt range, are recorded via electrodes placed on the scalp [31].

EEG signals are categorized into rhythmic patterns based on their frequency, each associated with different brain states [31]:

• Delta (δ): 0.5-4 Hz, prominent during deep sleep.
• Theta (θ): 4-7 Hz, associated with drowsiness and cognitive processing.
• Alpha (α): 8-12 Hz, the dominant rhythm during awake, relaxed states with closed eyes.
• Beta (β): 13-30 Hz, linked to active, alert states and motor behavior.
• Gamma (γ): >30 Hz, involved in higher cognitive processing and sensory integration.

Infraslow oscillations (<0.5 Hz), though not part of conventional clinical EEG, have gained research interest for their role in cognitive tasks and long-range spatial coupling in the brain [31].

The Origin and Measurement of Hemodynamic Activity

Neural activity is metabolically expensive. To meet the increased demand for oxygen and glucose, a process called neurovascular coupling occurs. This involves a localized increase in cerebral blood flow (CBF) to active brain regions [6] [32]. Functional Near-Infrared Spectroscopy (fNIRS) leverages this principle by using near-infrared light (650-950 nm) to measure concentration changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) in the cortical tissue [6] [30]. The most common fNIRS systems, Continuous Wave (CW) systems, apply the Modified Beer-Lambert Law to attenuated light signals to compute these concentration changes [6] [32].

Functional Magnetic Resonance Imaging (fMRI) also measures the hemodynamic response, but it is primarily sensitive to the paramagnetic properties of deoxygenated hemoglobin. This is known as the Blood Oxygen Level Dependent (BOLD) signal. When neural activity increases, the local influx of oxygenated blood outweighs the oxygen consumption, leading to a relative decrease in deoxygenated hemoglobin and an increase in the BOLD signal [11].

Table 1: Core Principles of Electrical and Hemodynamic Signals

Feature	Electrical Activity (EEG)	Hemodynamic Response (fNIRS/fMRI)
Primary Origin	Synchronous post-synaptic potentials of cortical pyramidal neurons [6]	Neurovascular coupling; changes in cerebral blood flow and volume [6] [32]
Measured Signal	Voltage fluctuations on the scalp (microvolts) [31]	fNIRS: Concentration changes of HbO and HbR [6]fMRI: Blood Oxygen Level Dependent (BOLD) signal [11]
Temporal Relationship	Direct, instantaneous reflection of neural firing	Indirect, slow response lagging neural activity by 1-6 seconds [11]
Key Physiological Principle	Summation of ionic currents across neuronal membranes	Tight coupling between neuronal energy demand and vascular supply [6] [30]

Technical Comparison of Major Neuroimaging Modalities

EEG, fNIRS, and fMRI offer distinct windows into brain function due to their inherent technical capabilities and limitations. EEG provides millisecond-level temporal resolution, ideal for tracking the rapid dynamics of brain networks, but suffers from limited spatial resolution and difficulty in localizing deep sources due to the inverse problem [6]. fNIRS offers a balance, with better spatial resolution than EEG for cortical regions, good tolerance to motion artifacts, and high portability. However, it is confined to the cerebral cortex and has a lower temporal resolution than EEG, constrained by the slow hemodynamic response [6] [11]. fMRI stands out with its excellent spatial resolution (millimeter-level) and whole-brain coverage, including subcortical structures. Its primary drawbacks are poor temporal resolution (limited by the hemodynamic response), high cost, immobility, and sensitivity to motion artifacts [11].

Table 2: Technical Specifications of EEG, fNIRS, and fMRI

Characteristic	EEG	fNIRS	fMRI
Spatial Resolution	Poor (several centimeters) [6]	Moderate (1-3 cm) [11]	High (millimeter-level) [11]
Temporal Resolution	Excellent (milliseconds) [6]	Moderate (0.1 - 1 Hz) [6] [11]	Poor (0.3 - 2 Hz, limited by HRF) [11]
Penetration Depth	Whole brain (but surface-weighted)	Superficial cortex (1-3 cm) [11]	Whole brain (cortex & subcortex) [11]
Portability	High [6] [30]	High [30] [11]	Low (immobile scanner) [11]
Key Artifacts	Electrical noise (50/60 Hz), muscle activity, eye blinks [6]	Scalp blood flow, motion artifacts, hair [32] [11]	Motion, magnetic susceptibility, physiological noise (cardiac, respiratory) [11]
Measured Physiology	Direct neural electrical activity [6]	Hemodynamic (HbO, HbR concentration) [6]	Hemodynamic (BOLD signal) [11]

Methodologies for Integrated Experimental Protocols

Multimodal integration leverages the strengths of each technique to provide a more comprehensive picture of brain function. The rationale is rooted in the neurovascular coupling phenomenon, where electrical events trigger metabolic and vascular responses [6] [30].

Concurrent fNIRS-EEG Recordings

This combination is highly practical due to the portability of both systems and the absence of electromagnetic interference [6] [30]. The experimental workflow involves:

• Hardware Setup: Use integrated caps that house both EEG electrodes and fNIRS optodes. Ensure optode placement avoids blocking EEG electrodes and vice versa.
• Synchronization: Use a common trigger signal (e.g., TTL pulse) sent from the stimulus presentation computer to both the EEG and fNIRS acquisition systems to ensure temporal alignment of data streams.
• Data Acquisition: Simultaneously record EEG (raw voltage) and fNIRS (light intensity at multiple wavelengths) data throughout the experimental task.
• Data Processing and Analysis: Three primary analytical approaches are used [6]:
  • EEG-informed fNIRS analysis: Use features from the fast EEG signal (e.g., event-related potentials or power in specific bands) to model or constrain the slower fNIRS hemodynamic response.
  • fNIRS-informed EEG analysis: Use hemodynamic activation maps from fNIRS to guide source localization for EEG, improving spatial accuracy.
  • Parallel fNIRS and EEG analysis: Analyze each modality independently and fuse the results at the interpretation stage (e.g., correlating EEG band power with HbO concentration).

Combined fMRI-fNIRS Studies

This multimodal approach is often used to validate fNIRS signals against the gold-standard spatial resolution of fMRI or to extend fMRI findings to more naturalistic settings [11]. Methodologies include:

• Synchronous Acquisition: Conducting fNIRS and fMRI simultaneously inside the MRI scanner. This requires specialized, MRI-compatible fNIRS equipment that is non-magnetic and does not interfere with the magnetic field [11].
• Asynchronous Acquisition: Performing fNIRS and fMRI sessions separately on the same subjects using identical task paradigms. This allows for the use of standard fNIRS equipment in a natural setting after localizing the region of interest with fMRI.
• Data Fusion: Co-registering fNIRS channels to the subject's anatomical MRI. The high-resolution spatial maps from fMRI can be used to interpret the fNIRS data or to create a head model for advanced fNIRS analysis, such as diffuse optical tomography (DOT) [11].

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Key Materials and Equipment for Multimodal Brain Imaging Research

Item	Function & Description
Integrated EEG-fNIRS Cap	A flexible head cap with pre-configured layouts holding EEG electrodes and fNIRS optodes, enabling simultaneous data acquisition [6] [30].
MRI-Compatible fNIRS System	A specialized fNIRS device constructed from non-magnetic materials (e.g., fiber optics) for safe and artifact-free operation inside an MRI scanner [11].
Electrode Gel (Electrolyte)	Applied at the scalp-electrode interface to reduce impedance and ensure high-quality electrical signal acquisition for EEG [31].
fNIRS Optodes (Sources & Detectors)	Sources emit near-infrared light into the scalp, while detectors capture the light after it has traveled through brain tissue. The source-detector distance determines penetration depth and sensitivity [6] [30].
Trigger Interface Box	A hardware device that receives event markers from a stimulus computer and sends synchronized TTL pulses to all data acquisition systems, ensuring temporal alignment of data with the experimental paradigm [6].
Data Processing Pipelines (Software)	Computational tools and algorithms (e.g., for motion artifact correction, filtering, GLM analysis) are crucial for converting raw signals into interpretable neurophysiological data. Standardization is key for reproducibility [32] [33].

Signaling Pathways and Experimental Workflows

The following diagrams, created using Graphviz, illustrate the core neurophysiological pathway and a generalized workflow for a multimodal experiment.

Neurovascular Coupling Pathway

Neurovascular Coupling Links Neural Activity to Blood Flow

Multimodal fNIRS-EEG Experimental Workflow

Concurrent fNIRS-EEG Study Design

In brain research, the selection of a neuroimaging modality is fundamentally governed by a trilemma involving spatial resolution, temporal resolution, and penetration depth. This technical guide examines how functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) navigate these inherent trade-offs, each establishing a unique position within the research ecosystem. Understanding these core principles is essential for researchers and drug development professionals to select appropriate methodologies, interpret findings accurately, and design innovative experiments that leverage the strengths of each technique.

The quest to decode brain function relies on non-invasive technologies that can capture neural dynamics with varying degrees of spatial and temporal precision. While the ideal modality would offer millimeter spatial resolution, millisecond temporal resolution, and whole-brain coverage, current technological and physiological constraints make this impossible. Instead, each mainstream technique prioritizes certain dimensions at the expense of others, creating complementary profiles that can be leveraged through multimodal approaches [34] [2].

Technical Specifications and Comparative Analysis

Quantitative Comparison of Core Parameters

Table 1: Technical specifications of major neuroimaging modalities

Parameter	fMRI	EEG	fNIRS
Spatial Resolution	High (mm to sub-mm) [34]	Low (source localization challenges) [35]	Moderate (1-3 cm) [35] [2]
Temporal Resolution	Moderate (0.3-2 Hz, limited by hemodynamic response) [34]	Very High (millisecond range) [35] [36]	High (0.1-10 Hz) [35]
Penetration Depth	Full brain (cortical & subcortical) [34]	Superficial (scalp-level signals) [36]	Superficial cortical (2-3 cm) [37] [2]
Primary Signal Measured	Blood Oxygen Level Dependent (BOLD) [34]	Electrical potentials from neuronal firing [36]	Hemoglobin concentration changes (HbO, HbR) [38] [37]
Key Strength	Unparalleled spatial mapping of deep structures	Capturing rapid neural oscillations	Balance of portability, cost, and motion tolerance [35] [2]
Principal Limitation	Low temporal resolution, high cost, immobility	Poor spatial localization, sensitive to artifacts	Limited to cortical surfaces, cannot image subcortical areas [37] [34]

The Fundamental Trade-off Relationships

The quantitative specifications in Table 1 reveal the fundamental constraints of neuroimaging. The relationship between these core parameters can be visualized as follows:

Figure 1: The core trade-offs between spatial resolution, temporal resolution, and penetration depth in neuroimaging, with modality positioning.

Modality-Specific Technical Principles and Methodologies

Functional Magnetic Resonance Imaging (fMRI)

fMRI operates by detecting the Blood Oxygen Level Dependent (BOLD) signal, which exploits the different magnetic properties of oxygenated and deoxygenated hemoglobin. When neurons fire, they consume oxygen, triggering a complex hemodynamic response that ultimately delivers oxygenated blood in excess of demand. This changes the local ratio of oxygenated to deoxygenated hemoglobin, which in turn alters the magnetic properties of the tissue that fMRI can detect [34] [2]. This process typically lags behind neural activity by 4-6 seconds, fundamentally limiting the technique's temporal resolution [34].

Experimental Protocol: Typical fMRI Block Design

• Participant Preparation: Screen for metal implants; ensure compliance with MRI safety protocols.
• Stimulus Presentation: Use specialized MRI-compatible displays and response systems.
• Data Acquisition: Employ a series of alternating task and rest blocks (e.g., 30-second task blocks interspersed with 30-second rest blocks).
• Data Analysis: Preprocess data (motion correction, normalization), then perform statistical parametric mapping (e.g., SPM, FSL) to identify voxels with significant BOLD signal changes between conditions [34].

Electroencephalography (EEG)

EEG measures the electrical potentials generated by the summed postsynaptic activity of large, synchronously firing populations of pyramidal neurons in the cortex. These minute electrical signals are detected by electrodes placed on the scalp, amplified, and digitized. While EEG provides unparalleled temporal resolution for capturing neural oscillations, its spatial resolution is limited by the blurring effects of the skull and other tissues, as well as the challenge of solving the inverse problem to localize source activity [36].

Experimental Protocol: EEG for Attention State Classification

• System Setup: Apply a multi-channel EEG cap according to the 10-20 system; ensure impedances are below 5-10 kΩ.
• Paradigm Design: Implement a sustained attention task (e.g., oddball, continuous performance task) with randomized stimuli.
• Data Recording: Collect continuous EEG data at a high sampling rate (e.g., 500-1000 Hz).
• Feature Extraction: Segment data into epochs; extract features from time, frequency, and connectivity domains (e.g., power spectral density, phase-locking value, functional connectivity).
• Classification: Apply machine learning models (e.g., SVM) to classify mental attention states based on the extracted features [36].

Functional Near-Infrared Spectroscopy (fNIRS)

fNIRS leverages the relative transparency of biological tissues to near-infrared light (650-950 nm). Within this "optical window," light can penetrate the scalp and skull to reach the cerebral cortex. fNIRS measures changes in the concentration of oxygenated (HbO) and deoxygenated hemoglobin (HbR) by exploiting their distinct absorption spectra, typically using the modified Beer-Lambert law [37] [35]. The path of the light between source and detector is typically "banana-shaped," sampling a volume of cortical tissue [37].

Experimental Protocol: fNIRS with Riemannian Geometry Classification

• Optode Placement: Position sources and detectors over the region of interest (e.g., prefrontal cortex) with a 3-cm separation for adults.
• Signal Acquisition: Record changes in HbO and HbR concentrations at a sampling rate of ~10 Hz during cognitive tasks (e.g., motor imagery, verbal fluency).
• Covariance Matrix Estimation: Construct channel-wise kernel matrices that capture the spatial and temporal relationships between fNIRS signals, separately for HbO and HbR.
• Riemannian Classification: Apply classifiers such as the Riemannian Support Vector Machine or Tangent Space Logistic Regression to differentiate brain states based on the geometry of these covariance matrices in the manifold space [38].

Multimodal Integration: Overcoming Individual Limitations

Synergistic Approaches

No single modality provides a complete picture of brain function, leading to increased interest in multimodal integration. The combination of fMRI and fNIRS is particularly powerful: fMRI provides the "gold standard" for spatial localization—including of deep brain structures—while fNIRS validates these findings and adds superior temporal resolution in a portable format suitable for naturalistic settings [34] [2]. Similarly, EEG and fNIRS can be combined to capture both electrophysiological and hemodynamic aspects of neural activity with high temporal resolution and improved spatial specificity compared to EEG alone [39].

Experimental Protocol: Simultaneous EEG-fNIRS for Motor Imagery Classification

• Hardware Setup: Use integrated EEG-fNIRS systems with carefully co-registered electrodes and optodes.
• Data Acquisition: Record both EEG (electrical) and fNIRS (hemodynamic) signals simultaneously during a motor imagery task (e.g., imagining left vs. right hand movement).
• Feature Extraction: For EEG, extract spatiotemporal features using temporal convolution and attention mechanisms. For fNIRS, extract spatial patterns across channels and temporal dynamics of the hemodynamic response.
• Decision Fusion: Fuse the outputs from both modalities using evidence theory (e.g., Dempster-Shafer Theory) to model uncertainty and improve classification accuracy [39].

Table 2: Essential research reagents and materials for neuroimaging studies

Item	Function/Purpose	Example Application
fNIRS Optodes	Emit and detect near-infrared light through scalp and skull.	Measuring HbO/HbR concentration changes in prefrontal cortex during cognitive tasks [38] [40].
EEG Electrodes/Cap	Detect electrical potentials from scalp surface.	Recording event-related potentials (ERPs) during attention tasks [36].
Conductive Gel	Reduces impedance between scalp and EEG electrodes.	Ensuring high-quality signal acquisition in EEG studies [36].
3D Digitizer	Records precise spatial coordinates of fNIRS optodes/EEG electrodes.	Coregistering measurement locations with standard brain atlases [2].
MRI-Compatible Stimulus Presentation System	Presents visual/auditory stimuli within MRI environment.	Administering block-design paradigms during fMRI scanning [34].
Motion Stabilization Equipment	Minimizes head movement artifacts.	Improving data quality in fNIRS/EEG studies with moving subjects [35] [2].

Advanced Applications and Methodological Innovations

Riemannian Geometry in fNIRS Analysis

Traditional fNIRS analysis often relies on channel-based features such as mean activation or slope. A recent innovation applies Riemannian geometry to the classification of fNIRS signals. This approach utilizes covariance matrices derived from the temporal and spatial relationships between channels, effectively capturing the inherent duality of fNIRS signals (HbO and HbR). In one study, this method achieved a mean accuracy of 65% in an eight-choice classification of mental tasks, significantly outperforming traditional methods which reached only 42% accuracy [38]. This demonstrates how advanced analytical approaches can extract more information from existing data, partially overcoming inherent hardware limitations.

Deep Learning for Multimodal Data Fusion

Deep learning architectures are increasingly used to fuse data from multiple modalities, leveraging their complementary strengths. For instance, one framework uses Convolutional Neural Networks (CNNs) to process EEG data transformed into time-frequency images, while Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs) capture the temporal dynamics of fNIRS signals [39] [40]. The fusion of these heterogeneous data streams can occur at the feature level or decision level, with advanced methods employing evidence theory to manage uncertainty. Such approaches have achieved accuracies exceeding 83% in classifying motor imagery states, outperforming unimodal methods [39]. The workflow for such a multimodal classification pipeline is illustrated below:

Figure 2: A deep learning-based fusion pipeline for EEG and fNIRS signal classification.

The inherent trade-offs between spatial resolution, temporal resolution, and penetration depth continue to define the capabilities and applications of fMRI, EEG, and fNIRS in brain research. fMRI remains the gold standard for precise spatial mapping of the entire brain, EEG excels at capturing the rapid dynamics of neural electrical activity, and fNIRS offers a practical balance for studying cortical functions in naturalistic environments. Rather than seeking a single superior modality, the future of brain research lies in strategically combining these technologies, supported by advanced analytical methods like Riemannian geometry and deep learning. This multimodal approach, which leverages their complementary natures, provides the most promising path toward a comprehensive understanding of brain function in health and disease, ultimately accelerating discovery in basic neuroscience and therapeutic development.

Methodological Integration and Translational Applications in Research

Understanding the intricate functions of the human brain requires a multifaceted approach, leveraging neuroimaging techniques that capture complementary aspects of neural activity. Functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) have become cornerstone methodologies in cognitive neuroscience and clinical research. Each technique possesses unique strengths and limitations that define its optimal application scope, from highly controlled laboratory settings to ecologically valid naturalistic environments [11] [41]. This technical guide examines the core principles of these modalities, their synergistic potential, and detailed experimental protocols for deploying them across the research continuum. The transition toward naturalistic assessment represents a paradigm shift in neuroscience, enabling unprecedented investigation of brain function during dynamic, real-world behaviors [42].

Core Technical Principles and Comparative Analysis

Fundamental Biophysical Mechanisms

Functional Magnetic Resonance Imaging (fMRI) measures brain activity indirectly through the blood-oxygen-level-dependent (BOLD) contrast. This signal originates from the magnetic susceptibility differences between oxygenated (diamagnetic) and deoxygenated (paramagnetic) hemoglobin. When neural activity increases in a specific region, it triggers a complex hemodynamic response that typically peaks 4-6 seconds post-stimulus, delivering oxygenated blood in excess of metabolic demand. This results in a localized decrease in deoxygenated hemoglobin and a corresponding increase in the BOLD signal [41]. fMRI provides comprehensive whole-brain coverage, including deep subcortical structures such as the hippocampus and amygdala, with spatial resolution at the millimeter level [11] [34].

Electroencephalography (EEG) measures the electrical activity generated by synchronized neuronal firing, primarily from pyramidal cells in the cerebral cortex. Postsynaptic potentials create electrical dipoles that propagate through volume conduction to electrodes placed on the scalp. EEG captures these voltage fluctuations with exceptional temporal resolution in the millisecond range, enabling real-time tracking of neural dynamics during cognitive processing [43]. However, the spatial resolution is limited by the blurring and attenuation of electrical signals as they pass through the skull and scalp.

Functional Near-Infrared Spectroscopy (fNIRS) employs near-infrared light (650-950 nm) to measure hemodynamic changes associated with neural activity. Light photons at specific wavelengths are introduced at the scalp surface and undergo scattering and absorption as they propagate through biological tissues. Hemoglobin molecules are primary absorbers in this spectrum, with distinct absorption spectra for oxygenated (HbO) and deoxygenated (HbR) states. Using modified Beer-Lambert law calculations with appropriate differential pathlength factors, fNIRS quantifies relative concentration changes of HbO and HbR in superficial cortical layers [23]. The technique is sensitive to hemodynamic changes in the outer cortex, typically to depths of 1-3 centimeters [11] [34].

Technical Specification Comparative Analysis

Table 1: Comparative technical specifications of fMRI, EEG, and fNIRS

Parameter	fMRI	EEG	fNIRS
Spatial Resolution	High (1-3 mm) [11]	Low (Centimeter-level) [43]	Moderate (1-3 cm) [11]
Temporal Resolution	Low (0.3-2 Hz) - limited by hemodynamic response [11]	High (Milliseconds) [43]	Moderate (0.1-10 Hz) - limited by hemodynamic response [19]
Depth Penetration	Whole brain (cortical and subcortical) [11]	Cortical surface [43]	Superficial cortex (1-2.5 cm) [11] [43]
Portability	Low (Fixed scanner) [41]	High (Wearable systems available) [43]	High (Wearable systems available) [44] [42]
Tolerance to Motion	Low (Requires head immobilization) [41]	Moderate (Sensitive to motion artifacts) [43]	High (Relatively motion-tolerant) [41] [43]
Primary Signal Source	Hemodynamic (BOLD response) [41]	Electrical (Neural potentials) [43]	Hemodynamic (HbO/HbR concentration) [23]
Operational Requirements	High (Dedicated facility, magnetic shielding) [41]	Low-Moderate (Electrode application, shielded room preferred)	Moderate (Optode placement, ambient light control) [44]
Approximate Cost	Very High (>$1M) [41]	Low-Moderate [43]	Moderate-High [43]

Table 2: Application scope suitability across research environments

Research Context	Optimal Technique(s)	Key Considerations
Basic Laboratory Research	fMRI, EEG, fNIRS	Task control, signal quality, participant compliance [41]
Clinical Populations	fNIRS, EEG (fMRI when feasible)	Tolerance to motion, accessibility, patient comfort [44] [45]
Developmental Studies	fNIRS, EEG	Participant movement, ecological validity, comfort [11] [42]
Naturalistic Settings	fNIRS, EEG (Mobile systems)	Portability, motion tolerance, environmental interference [44] [42]
Social Interaction Research	fNIRS (Hyperscanning), EEG	Multi-person recording, ecological validity [11] [42]
Resting-State Connectivity	fMRI, fNIRS, EEG	Spatiotemporal requirements, participant state monitoring [19]
Precision Mental Health	fNIRS, fMRI (for baseline)	Individual-level mapping, longitudinal monitoring [44]

Signaling Pathways and Neural Correlates

The relationship between neural activity and the signals measured by each modality follows distinct biophysical pathways. Understanding these pathways is essential for interpreting neuroimaging data and designing appropriate experiments.

Diagram 1: Signaling pathways from neural activity to measured signals

EEG provides the most direct measure of neural activity, capturing electrical potentials generated primarily by pyramidal cell synchronization within cortical layers. These measurements reflect the integrated electrical activity of neuronal populations, with exquisite temporal precision that enables tracking of rapid cognitive processes such as sensory perception and attentional allocation [43].

In contrast, both fMRI and fNIRS measure hemodynamic responses that are indirectly coupled to neural activity through neurovascular coupling mechanisms. The process begins with neurotransmitter release and energy consumption at synapses, triggering metabolic demands that increase local oxygen consumption. This metabolic signal initiates a complex cascade of neurovascular coupling involving astrocytes, neurons, and vascular smooth muscle, ultimately leading to increased regional cerebral blood flow (CBF). The hemodynamic response function typically peaks 4-6 seconds after neural activity, creating the temporal lag characteristic of hemodynamic-based methods [11] [41].

fMRI specifically measures the BOLD contrast, which reflects the balance between oxygen delivery and consumption. Increased neural activity typically produces a decrease in deoxygenated hemoglobin concentration due to the disproportionate increase in oxygenated blood flow, resulting in the positive BOLD signal. fNIRS separately quantifies both oxygenated (HbO) and deoxygenated (HbR) hemoglobin concentration changes, providing a more comprehensive picture of the hemodynamic response [23]. The fNIRS HbR signal has been shown to correlate strongly with the fMRI BOLD signal, as both are sensitive to deoxygenated hemoglobin changes [11] [34].

Experimental Design and Protocol Specifications

Laboratory-Based Protocols

fMRI Block Design for Cognitive Task Localization

• Experimental Setup: Participants positioned supine in MRI scanner with head immobilization; visual stimuli presented via MRI-compatible display system; responses collected via MRI-compatible response devices [41].
• Task Structure: Alternating 30-second blocks of task condition (e.g., working memory N-back) and control condition (e.g., passive fixation); 5-6 cycles per run; total duration approximately 10 minutes.
• Data Acquisition Parameters: TR = 2000 ms, TE = 30 ms, flip angle = 75°, voxel size = 3×3×3 mm, 40 axial slices covering whole brain.
• Physiological Monitoring: Cardiac and respiratory monitoring throughout for noise regression in preprocessing.
• Quality Control: Real-time monitoring for excessive head motion (>2 mm translation, >2° rotation); repeated acquisition if motion exceeds threshold.

EEG Event-Related Potential (ERP) Protocol

• Experimental Setup: 64-channel EEG system following 10-20 international placement; electrode impedance maintained below 5 kΩ; electrically shielded room preferred.
• Stimulus Presentation: Rapid serial visual presentation with inter-stimulus interval jittered between 800-1200 ms; 100 trials per condition minimum.
• Data Acquisition: Sampling rate ≥1000 Hz; online filtering 0.1-100 Hz; common reference during acquisition.
• Task Paradigm: Oddball task with frequent standard stimuli (80%) and rare target stimuli (20%); participants respond to target stimuli via button press.
• Preprocessing Pipeline: Re-referencing to average reference; bandpass filtering 0.1-30 Hz; ocular artifact correction using ICA; epoch extraction -200 to 800 ms relative to stimulus onset; baseline correction; artifact rejection (±100 μV threshold).

fNIRS Protocol for Prefrontal Cortex Activation

• Experimental Setup: fNIRS optodes positioned over prefrontal regions according to 10-20 system (Fp1, Fp2, F7, F8, F3, F4); inter-optode distance 30 mm; inclusion of short-separation channels (8 mm) for superficial signal regression [44] [46].
• Task Paradigm: Continuous performance task with 5-minute duration; alternating 30-second task and 30-second rest periods.
• Data Acquisition: Dual-wavelength system (760 nm and 850 nm); sampling rate 10 Hz; signal quality assessment via scalp coupling index [19].
• Preprocessing Steps: Conversion to optical density; motion artifact detection and correction (e.g., wavelet-based methods); bandpass filtering 0.01-0.2 Hz; conversion to hemoglobin concentrations using modified Beer-Lambert law with appropriate differential pathlength factor [23].

Naturalistic Environment Protocols

Mobile fNIRS for Ecological Assessment

• System Configuration: Wireless, wearable fNIRS headset; tablet-based stimulus presentation; synchronized behavioral and brain activity recording [44].
• Paradigm Design: Real-world tasks such as walking navigation, social interaction, or driving simulation; protocol duration 15-45 minutes depending on research question.
• Motion Mitigation: Secure headband fitting; accelerometer integration for motion tracking; artifact-resistant analysis pipelines (e.g., robust regression, spline interpolation) [42].
• Data Quality Assurance: Signal-to-noise ratio monitoring throughout recording; ambient light control using opaque cap; verification of optode-scalp coupling post-session.

Hybrid EEG-fNIRS for Comprehensive Brain Mapping

• Hardware Integration: Simultaneous recording with integrated cap systems; careful placement to avoid optode-electrode interference; synchronized data acquisition via TTL pulses or Lab Streaming Layer [43] [19].
• Experimental Paradigm: Motor imagery tasks (e.g., left vs. right hand movement imagination) with 10-second epochs; resting-state recordings pre- and post-task.
• Data Fusion Approaches: Joint Independent Component Analysis (jICA) for identifying coupled electrical-hemodynamic components; canonical correlation analysis for multimodal connectivity assessment [19].

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential materials and their functions in multimodal brain imaging research

Category	Item	Specification/Function
fMRI Consumables	MRI-Compatible Response Devices	Fiber-optic or non-magnetic components for safe operation in high-field environment [41]
	Physiological Monitoring Equipment	Pulse oximeter, respiratory belt, for noise correction in BOLD signal [41]
	Head Stabilization Equipment	Foam padding, bite bars to minimize motion artifacts [41]
EEG Consumables	Electrolyte Gel/Solution	Reduces impedance at electrode-scalp interface (typically <5 kΩ target) [43]
	Abrasive Preparatory Gels	Mild skin preparation for improving electrode contact
	Disposable Electrodes	Ag/AgCl electrodes with chloride coating for stable potentials
fNIRS Consumables	Optical Probes/Sensors	Source-detector pairs with specific wavelength emission (650-950 nm) [23]
	Short-Separation Detectors	8 mm source-detector distance for measuring superficial signals [23]
	Optode Positioning Systems	Augmented reality guidance or physical templates for reproducible placement [44]
Data Acquisition & Analysis	Synchronization Hardware	TTL pulse generators, Lab Streaming Layer for multimodal timing [43] [42]
	Analysis Software Platforms	HOMER3, NIRS Toolbox, EEGLAB, SPM, FSL [23]
	Anatomical Registration Tools	AtlasViewer for fNIRS, Brainstorm for EEG/MRI coregistration [23]

Multimodal Integration and Experimental Workflows

Combining multiple neuroimaging modalities enables researchers to overcome the limitations of individual techniques, providing more comprehensive insights into brain function. The workflow for designing and implementing multimodal studies requires careful consideration of technical compatibility and experimental design.

Diagram 2: Multimodal neuroimaging experimental workflow

Synchronous Multimodal Acquisition

Synchronous fMRI-fNIRS recording provides complementary spatiotemporal information by combining fMRI's whole-brain coverage with fNIRS' superior temporal resolution and motion tolerance. This approach requires specialized MRI-compatible fNIRS equipment that minimizes electromagnetic interference and ensures participant safety [11] [34]. The protocol involves:

• Hardware Configuration: MRI-compatible fNIRS probes constructed with non-magnetic materials; fiber-optic cables extending outside scanner room; careful routing to avoid entanglement with MRI hardware.
• Synchronization: Trigger-based synchronization using TTL pulses from MRI console to fNIRS system; sampling clock alignment to account for differential drift.
• Quality Assurance: Assessment of fNIRS signal quality during fMRI acquisition; identification and exclusion of periods affected by MRI gradient artifacts; evaluation of potential heating risks.

Integrated EEG-fNIRS Platforms

Combining EEG and fNIRS capitalizes on their complementary strengths—EEG's millisecond temporal resolution for electrical dynamics and fNIRS' improved spatial localization for hemodynamic responses [43] [19]. Implementation considerations include:

• Hardware Integration: Specialized caps with integrated electrode and optode holders; careful arrangement to prevent physical interference; consideration of 10-20 system for consistent placement.
• Synchronization Methods: Hardware triggers via parallel port or TTL pulses; software-based synchronization using network time protocol; Lab Streaming Layer for unified data collection [43].
• Data Fusion Techniques: Joint independent component analysis (jICA) to identify coupled components across modalities; parallel ICA for maximizing spatial and temporal information; general linear modeling approaches incorporating both electrical and hemodynamic features.

Transitioning to Naturalistic Neuroscience

The emergence of wearable neuroimaging technologies, particularly mobile fNIRS and EEG systems, has enabled the new field of "Neuroscience of the Everyday World" (NEW) [42]. This paradigm shift moves brain imaging from constrained laboratory settings to ecologically valid environments where complex, natural behaviors can be studied.

Protocol Considerations for Naturalistic Settings:

• Participant Mobility: Secure but comfortable headgear; wireless data transmission; battery life sufficient for extended recording sessions (45+ minutes) [44].
• Environmental Controls: Ambient light management for fNIRS; electromagnetic interference minimization for EEG; consideration of temperature and humidity effects on signal quality.
• Behavioral Synchronization: Video recording for behavioral annotation; inertial measurement units for movement quantification; precise timing of stimulus presentation in real-world contexts.
• Data Quality Validation: Implementation of signal quality indices (e.g., scalp coupling index for fNIRS); motion artifact detection algorithms; verification of adequate signal-to-noise ratio in field conditions.

Applications of naturalistic neuroimaging include social interaction studies using hyperscanning (simultaneous recording from multiple individuals), sports performance monitoring, neurorehabilitation in real-world contexts, and investigation of neurological disorders during activities of daily living [45] [42]. These approaches provide unprecedented insights into brain function under ecologically valid conditions, bridging the gap between laboratory findings and real-world brain activity.

The integration of fMRI, EEG, and fNIRS represents a powerful framework for comprehensive brain investigation across the laboratory-to-naturalistic spectrum. Future developments will focus on enhanced multimodal integration, improved wearable technology, and advanced analytical approaches for complex naturalistic data [44] [42].

Critical challenges remain in standardization of acquisition protocols, development of robust analytical methods for naturalistic environments, and validation of findings across different populations and settings. The ongoing miniaturization of hardware, improvement in signal processing techniques, and development of novel experimental paradigms will continue to expand the application scopes of these complementary neuroimaging modalities [11] [42].

By strategically selecting and combining these techniques based on their specific strengths and application domains, researchers can design optimized studies that span the continuum from precise laboratory control to ecologically valid naturalistic assessment, ultimately advancing our understanding of brain function in health and disease.

Functional Magnetic Resonance Imaging (fMRI) has emerged as a predominant tool for examining neural activity in the human brain non-invasively. By measuring the Blood Oxygen Level-Dependent (BOLD) signal, fMRI provides an indirect correlate of neural activity with high spatial resolution, making it invaluable for cognitive neuroscience, clinical psychology, and surgical planning [47]. Unlike resting-state fMRI, which examines spontaneous signal synchronicity across brain regions without active tasks, task fMRI requires participants to engage in specific paradigms, enabling precise localization of brain regions involved in particular functions [47]. The strength of fMRI lies in its ability to map complex cognitive processes to specific neural substrates across deep brain structures, filling a critical niche that complements other neuroimaging modalities like EEG and fNIRS.

This technical guide examines the core principles, advanced analytical frameworks, and practical applications of fMRI, contextualizing its role within the broader neuroimaging toolkit. While electroencephalography (EEG) provides millisecond-level temporal resolution essential for capturing rapid neural dynamics, it suffers from limited spatial accuracy due to signal dispersion through the skull [48]. Conversely, functional near-infrared spectroscopy (fNIRS) measures hemodynamic responses similar to fMRI but is restricted to surface cortical areas, lacking fMRI's comprehensive whole-brain coverage including deeper structures [49] [48]. fMRI thus occupies a unique position in the neuroimaging spectrum, offering unparalleled spatial mapping capabilities throughout the entire brain, which is indispensable for investigating network connectivity and deep brain structures inaccessible to more superficial modalities.

Core Principles and Comparative Advantages

The Physiological Basis of fMRI

The fundamental principle underlying fMRI is neurovascular coupling—the relationship between neural activity and subsequent hemodynamic changes. When neurons become active, they trigger a complex physiological response that increases cerebral blood flow to the region, altering the relative concentrations of oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) [49]. These hemoglobin variants possess different magnetic properties: deoxygenated hemoglobin is paramagnetic and creates signal distortions, while oxygenated hemoglobin is diamagnetic. These differential magnetic properties form the basis of the BOLD contrast mechanism [49] [47]. The hemodynamic response function (HRF) characterizes the typical temporal evolution of this effect, with a gradual rise peaking approximately 5 seconds after neural activity begins, followed by a slower return to baseline and potential undershoot [49]. This inherent physiological delay represents a fundamental temporal constraint of the BOLD signal.

fMRI in the Neuroimaging Ecosystem

fTable: Comparison of Non-Invasive Neuroimaging Modalities

Feature	fMRI	EEG	fNIRS
What It Measures	BOLD signal (blood oxygenation)	Electrical activity from neurons	Hemodynamic response (HbO/HbR)
Spatial Resolution	High (millimeter-level)	Low (centimeter-level)	Moderate (surface cortex only)
Temporal Resolution	Moderate (seconds)	High (milliseconds)	Low (seconds)
Depth of Measurement	Whole brain	Cortical surface	Outer cortex (1–2.5 cm deep)
Tolerance to Motion	Low (requires head immobilization)	Moderate (susceptible to artifacts)	High (suitable for naturalistic settings)
Key Strengths	Whole-brain coverage including deep structures; excellent spatial resolution	Captures fast neural dynamics; direct neural signal	Portable; suitable for real-world environments
Primary Limitations	Indirect measure; expensive; noisy environment; temporal lag	Poor spatial resolution; sensitive to artifacts	Superficial measurement only; limited spatial coverage

This comparative analysis reveals the complementary nature of these modalities. EEG excels when research questions demand exquisite temporal precision, such as studying event-related potentials or rapid cognitive processes like sensory perception [48]. fNIRS finds its niche in studies requiring greater ecological validity, such as social interactions, developmental research with children, or investigations where participants must move freely [49] [48]. fMRI remains the modality of choice when comprehensive spatial mapping is paramount, particularly for investigating network connectivity across deep brain structures or localizing function for clinical applications like presurgical planning [47]. The choice between modalities should be guided by the specific research question, considering the trade-offs between spatial and temporal resolution, depth of coverage, and practical constraints of the experimental environment.

Advanced Analytical Frameworks

Preprocessing and Denoising Pipelines

fMRI data contain substantial noise from both structured (physiological) and unstructured (thermal) sources, with BOLD signal changes in cognitive tasks often representing less than 1% of the total signal variance [47]. Consequently, sophisticated preprocessing pipelines are essential to enhance the signal-to-noise ratio (SNR) before meaningful analysis can proceed. Spatial smoothing using Gaussian kernels has been a standard preprocessing step to improve SNR and mitigate anatomical variability across subjects in group-level analyses [47]. However, conventional isotropic smoothing compromises spatial specificity by introducing spatial blurring artifacts, where inactive voxels near active regions may be mistakenly identified as active [47].

Recent advances have introduced adaptive spatial smoothing techniques that tailor smoothing parameters for each voxel based on surrounding time series characteristics. Traditional methods like Canonical Correlation Analysis (CCA) determine optimal weights for neighboring voxels to maximize correlation with the task design matrix [47]. However, these approaches become computationally prohibitive when incorporating more neighboring voxels for higher-resolution data. Deep neural network (DNN) architectures now offer a sophisticated alternative by using multiple 3D convolutional layers to incorporate more neighboring voxels efficiently, learning data-driven spatial filters that adapt to various data characteristics without pre-specifying filter shapes or orientations [47]. This approach maintains spatial specificity while enhancing SNR, making it particularly valuable for applications requiring individual-level precision such as presurgical mapping.

For denoising, novel methods like DeepCor utilize deep generative models to disentangle and remove noise from fMRI data [50]. This contrastive autoencoder-based approach has demonstrated substantial improvements over established methods like CompCor, enhancing BOLD signal responses to stimuli by up to 215% in empirical tests [50]. These advanced denoising techniques are particularly crucial for functional connectivity analyses, where spurious correlations induced by motion and physiological noise can severely compromise results.

Statistical Modeling and Activation Detection

The General Linear Model (GLM) represents the cornerstone of task fMRI analysis, providing a statistical framework for detecting brain activation associated with experimental conditions [49] [47]. The GLM models the fMRI time series as a linear combination of explanatory variables (predictors derived from the experimental design) plus an error term:

y = Xβ + ε

Where y is the measured fMRI signal, X is the design matrix containing task predictors convolved with the hemodynamic response function, β represents the unknown coefficients to be estimated, and ε is the error term [47]. Model estimation accounts for temporal autocorrelation in the noise and often incorporates drift removal to eliminate slow signal fluctuations unrelated to neural activity.

For inference, statistic images are thresholded to identify significantly active voxels while controlling for multiple comparisons across thousands of voxels. Standard approaches include Familywise Error (FWE) correction and False Discovery Rate (FDR) [51]. Recent machine learning advances have further enhanced activation detection, with methods like self-supervised voxel shuffling frameworks that find optimal kernel mappings while mitigating overfitting without relying on spatial ground truth information [52]. Similarly, Class-Aware Hidden Markov Models provide simpler yet more accurate approaches for simultaneous functional connectivity estimation and classification compared to traditional two-step methods [52].

Diagram 1: fMRI Analysis Workflow. This flowchart illustrates the standard processing pipeline from raw data acquisition to final interpretation, highlighting key stages including preprocessing, statistical analysis, and output generation.

Network Connectivity Analysis

Beyond localized activation, fMRI enables comprehensive mapping of brain network connectivity through several analytical approaches. Seed-based correlation analysis examines functional connectivity by computing temporal correlations between a seed region's time series and all other brain voxels [52]. Independent Component Analysis (ICA) identifies spatially independent networks with synchronized temporal dynamics without requiring a priori seed selection [52]. More recently, multivariate pattern analysis and machine learning approaches have been employed to decode cognitive states and information representation from distributed activation patterns.

Advanced methods now enable real-time functional connectivity analysis, with recent software platforms demonstrating high-speed performance approaching that of conventional offline tools [52]. For clinical applications such as presurgical planning, automated resting-state fMRI analysis pipelines implement probabilistic functional atlases in seed-based correlation and ICA pipelines to map eloquent cortical areas near pathological lesions [52]. The emerging field of connectome fingerprinting further explores how individual differences in functional network architecture correlate with cognitive abilities, clinical symptoms, and genetic factors.

Innovative approaches are also advancing how we conceptualize connectivity. Studies investigating "brain-to-brain communication channels" have developed novel methods to identify both inter-brain network connections and their information content during social interactions [52]. Other methodological innovations include dynamic autocorrelation (dAC) measures to investigate time-dependent changes in neural timescales along hippocampal axes during resting-state [52], revealing how neural timescales adapt to environmental demands.

Experimental Protocols and Methodologies

Task fMRI Experimental Design

Effective fMRI research begins with meticulous experimental design that carefully balances cognitive paradigm selection with methodological constraints. The two primary design approaches are block designs and event-related designs. Block designs present sustained periods of a single condition (e.g., 30 seconds of face detection) alternating with control conditions (e.g., 30 seconds of house detection). This approach maximizes statistical power for detecting the hemodynamic response but sacrifices temporal precision [49]. Event-related designs present brief, discrete trials in randomized sequences, allowing analysis of individual trial responses and identification of rapid neural transitions, though with reduced statistical power [49].

Critical to any fMRI design is the selection of appropriate control conditions that isolate the cognitive process of interest. The fundamental principle is to compare two conditions that differ only in the specific cognitive component being studied while matching all other perceptual, motor, and cognitive processes [49]. For complex cognitive functions, parametric designs that vary task difficulty along a continuum or factorial designs that systematically manipulate multiple variables often provide more nuanced insights than simple categorical comparisons.

Diagram 2: Task fMRI Experimental Design Framework. This diagram outlines the key components and decision points in designing an effective fMRI experiment, including design selection, control conditions, and modeling approaches.

Data Acquisition Parameters

Optimal fMRI data acquisition requires careful parameter selection to balance spatial coverage, resolution, and temporal sampling. Key acquisition parameters include:

• Magnetic Field Strength: While 3.0 Tesla remains standard, 5.0 Tesla and higher systems provide improved signal-to-noise ratio and resolution [52].
• Repetition Time (TR): The time between successive volume acquisitions, typically 2-3 seconds for whole-brain coverage, representing a trade-off between temporal resolution, spatial coverage, and signal quality.
• Echo Time (TE): Optimized for BOLD contrast at the specific field strength (approximately 30ms at 3T).
• Voxel Size: Standard acquisitions use 3×3×3 mm³ voxels, with high-resolution protocols achieving 2mm isotropic or smaller for specialized applications [47].
• Slice Acquisition Order: Sequential or interleaved to minimize slice timing artifacts.

Recent technical advances have made sub-millimeter resolution fMRI increasingly feasible, allowing cognitive neuroscientists to detect subtle activity differences between anatomically close regions or subregions in complex tasks [47]. Multi-echo acquisition sequences and accelerated parallel imaging techniques further enhance data quality and temporal resolution.

Best Practices in Analysis and Reporting

Comprehensive reporting of methodological details is essential for reproducibility and proper interpretation of fMRI findings. Key reporting standards include [51]:

• Experimental Design: Number of blocks, trials, or experimental units per session and subject; duration of blocks or trials; task timing parameters.
• Image Acquisition Parameters: Magnet strength, pulse sequence, TE/TR, field of view, matrix size, voxel dimensions, slice orientation and order, number of sessions and volumes.
• Preprocessing Steps: Motion correction algorithms, slice timing correction, distortion correction, normalization methods, smoothing kernel size and type.
• Statistical Modeling: Software and version; hemodynamic response function specification; drift removal; autocorrelation modeling; contrast definitions.
• Inference Procedures: Whole-brain versus region of interest analysis; multiple comparison correction method; cluster-defining threshold if used.

Transparent results reporting should include unthresholded statistic maps to avoid misleading representations of activation patterns, along with time course plots for event-related analyses and scatter plots for correlation analyses [51].

Research Reagent Solutions

fTable: Essential Materials and Analytical Tools for fMRI Research

Item	Function/Purpose	Technical Specifications
High-Density MRI Scanner	Data acquisition with optimal BOLD contrast	3.0T standard; ultra-high field (7T+) for improved SNR
Multi-Channel Head Coils	Enhanced signal reception and parallel imaging capability	32-64 channels for optimal whole-brain coverage
Stimulus Presentation System	Precise delivery of experimental paradigms	MRI-compatible displays, response devices, and audio systems
Physiological Monitoring	Recording of cardiac and respiratory signals	Pulse oximeter, respiratory belt for noise correction
Analytical Software Platforms	Data preprocessing, statistical analysis, and visualization	SPM, FSL, AFNI, BrainVoyager, FreeSurfer
Spatial Smoothing Tools	Signal-to-noise ratio enhancement	Gaussian kernels (6-8mm FWHM); adaptive methods (CCA, DNN)
Deep Learning Denoising	Advanced noise removal from BOLD signal	DeepCor, 3DConv-lSTM, ME-DUNE architectures
Connectivity Analysis Tools	Mapping functional networks and interactions	Seed-based correlation, ICA, graph theory metrics
Brain Atlases	Anatomical reference and region definition	MNI, Talairach, AAL, Harvard-Oxford, Yeo networks

Emerging Computational Tools

The fMRI analytical landscape is rapidly evolving with the integration of sophisticated computational approaches:

• DeepCor: A contrastive autoencoder-based denoising method that outperforms CompCor by 215% in enhancing BOLD signal responses to stimuli [50].
• Adaptive Spatial Smoothing DNN: A deep neural network that builds on constrained canonical correlation analysis to incorporate more neighboring voxels for estimating optimal spatial smoothing without significantly increasing computational costs [47].
• Class-Aware Hidden Markov Models: Provide simpler and more accurate approaches for simultaneous functional connectivity estimation and classification compared to traditional two-step methods while maintaining model interpretability [52].
• Kolmogorov-Arnold Networks: Offer strong interpretability for fMRI task-condition classification, aligning with established literature on the functional roles of specific brain regions [52].
• VAVOOOM: A voxel and vicinity-based algorithm for noise reduction of fMRI datasets that improves temporal signal-to-noise ratio and facilitates accurate characterization of fMRI metrics in high-resolution data [52].

These tools represent the cutting edge of fMRI methodology, enabling more precise, reliable, and interpretable analyses that advance both basic neuroscience research and clinical applications.

The future of fMRI research points toward increasingly integrative, personalized, and computationally sophisticated approaches. Multimodal integration of fMRI with EEG, fNIRS, and other neuroimaging modalities provides complementary data streams that overcome the limitations of individual techniques [48]. Simultaneous EEG-fMRI recording, though technically challenging, offers particular promise for coupling the millisecond temporal resolution of EEG with the millimeter spatial precision of fMRI [48]. Personalized brain mapping represents another frontier, with studies demonstrating the consistency and stability of individualized cortical functional network parcellation at both 3.0T and emerging 5.0T MRI systems [52]. This approach lays the foundation for precision personalized medicine applications in neurology and psychiatry.

Ultra-high field strength scanners (7T and beyond) continue to push the boundaries of spatial resolution, enabling laminar and columnar-level fMRI that can discriminate activity across different cortical layers [47]. The development of real-time analysis platforms is making high-speed resting-state fMRI analysis feasible, approaching the performance and utility of conventional offline analysis toolboxes [52]. Finally, the integration of artificial intelligence and machine learning is revolutionizing every aspect of fMRI, from data acquisition and denoising to pattern classification and clinical prediction.

In conclusion, fMRI maintains a central position in the neuroimaging toolkit, offering unparalleled capabilities for mapping brain function across deep structures and distributed networks. When employed with rigorous experimental designs, appropriate analytical techniques, and comprehensive reporting standards, fMRI provides powerful insights into brain organization and function. As the field advances toward more integrated multimodal approaches and computationally sophisticated analyses, fMRI continues to evolve as an indispensable tool for both basic cognitive neuroscience and clinical applications.

In the landscape of modern neuroimaging, techniques are selected according to the specific temporal and spatial resolution requirements of the research or clinical application. Electroencephalography (EEG) occupies a unique and crucial niche within this spectrum, providing the highest temporal resolution of all non-invasive brain imaging methods, capable of tracking brain dynamics at the millisecond level [53]. This unparalleled speed is essential for capturing the rapid neural processes underlying perception, cognition, and action. In contrast, functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS) measure hemodynamic responses—changes in blood flow and oxygenation that are indirect and slow correlates of neural activity, occurring over seconds [53] [54]. While fMRI offers superior spatial resolution and fNIRS provides a portable compromise, EEG remains the gold standard for directly measuring the brain's electrical activity with exquisite temporal precision [54]. This whitepaper details the principles, methodologies, and applications of EEG, with a specific focus on Event-Related Potentials (ERPs), framing them within the broader context of multimodal brain research for scientists and drug development professionals.

Table 1: Comparison of Key Non-Invasive Neuroimaging Modalities

Feature	EEG	fMRI	fNIRS
What it Measures	Electrical activity from postsynaptic potentials [54]	Blood-oxygen-level-dependent (BOLD) signal [55]	Concentration changes in oxygenated & deoxygenated hemoglobin [53] [54]
Temporal Resolution	Milliseconds [54]	Seconds [54]	Seconds [54]
Spatial Resolution	Low (centimeter-level) [54]	High (millimeter-level) [53]	Moderate (cortical surface) [54]
Direct/Indirect Measure	Direct neural signal	Indirect hemodynamic response	Indirect hemodynamic response
Portability	High (wearable systems available) [54]	Low (requires fixed scanner)	High (wearable systems available) [54]
Primary Cost	Generally lower [54]	High	Generally higher than EEG [54]

Fundamentals of EEG and Event-Related Potentials (ERPs)

Electroencephalography (EEG) detects voltage fluctuations resulting from the ionic current within the neurons of the brain, specifically the summed postsynaptic potentials from large, synchronously firing groups of pyramidal cells [54]. When recorded from the scalp, these signals represent the brain's ongoing, spontaneous electrical activity.

Event-Related Potentials (ERPs) are a specific application of EEG. They are very small voltages generated in the brain structures in response to specific sensory, cognitive, or motor events [56]. The ERP technique involves time-locking the EEG recording to the onset of a stimulus and averaging the responses across many trials. This averaging process cancels out the random background brain activity and noise, leaving a stereotyped electrophysiological waveform that is directly related to the information processing of the event [57]. Mathematically, averaging N trials reduces the noise amplitude by a factor of 1/√N, allowing the underlying ERP signal to emerge [57].

ERPs are characterized by a series of positive (P) and negative (N) voltage deflections, called components. These components are typically named by their polarity and their latency in milliseconds (e.g., N100) or their ordinal position (e.g., P3) [57]. Early components (within ~100 ms) are generally considered "sensory" or "exogenous," as they are highly dependent on the physical parameters of the stimulus. Later components are termed "cognitive" or "endogenous," as they reflect the manner in which the subject evaluates the stimulus [56].

Figure 1: The Core Workflow for Extracting Event-Related Potentials (ERPs) from Raw EEG Data.

A Detailed Guide to Major ERP Components

ERP components serve as robust biomarkers for specific stages of neural processing, from basic sensory registration to higher-order cognitive evaluation. The following table summarizes the key characteristics and functional correlates of major ERP components.

Table 2: Characteristics and Functional Correlates of Major ERP Components

ERP Component	Typical Latency (ms)	Polarity	Functional Correlation
P50	40-75	Positive	Sensory gating; inhibition of response to redundant stimuli [56]
N100 (N1)	90-200	Negative	Orienting response; initial stimulus detection ("vertex potential") [56]
P200 (P2)	100-250	Positive	Early perceptual processing; may relate to sensation-seeking [56]
N200 (N2)	~200	Negative	Stimulus discrimination and classification; includes MMN (mismatch negativity) [56]
P300 (P3)	250-700	Positive	Context updating, attention allocation, and decision-making [56] [57]
N400	300-600	Negative	Semantic processing; violated semantic expectation [56]
P600	~600	Positive	Syntactic processing; reanalysis of grammatical structures [56]
CNV	Varies (S1-S2)	Negative	Expectancy and motor preparation ("contingent negative variation") [56]

Key Experimental Paradigms

Eliciting specific ERP components requires carefully designed experimental tasks:

• The Oddball Paradigm for P300: The P300, one of the most studied ERP components, is reliably elicited using the "oddball" paradigm. In this task, a series of stimuli are presented where one occurs infrequently. The subject is instructed to respond (e.g., with a button press) to this rare "target" stimulus and ignore the frequent "standard" stimuli. The P300 wave, a positive deflection peaking around 300-600 ms post-stimulus, is then observed in response to the target. Its amplitude is thought to reflect the allocation of attentional resources, while its latency indicates the speed of stimulus classification [56] [57].
• The Paired-Click Paradigm for P50 Sensory Gating: To assess pre-attentional sensory filtering mechanisms, the paired-click paradigm is used. Two identical auditory stimuli are presented in rapid succession (e.g., 500 ms apart). In a healthy brain, the P50 amplitude in response to the second stimulus (S2) is significantly smaller than that to the first (S1). The ratio of S2/S1 amplitudes measures sensory gating, a key inhibitory process that is often deficient in psychiatric disorders like schizophrenia [56].
• Sentence Reading Tasks for N400: The N400 component is commonly studied using language paradigms. It appears as a negative deflection when a subject reads a sentence that ends with a semantically incongruent word (e.g., "I take my coffee with cream and dog."). The amplitude of the N400 is inversely related to the semantic expectancy of the word within its context [56].

Technical Protocols and the Researcher's Toolkit

Conducting a rigorous ERP study demands meticulous attention to experimental design, data acquisition, and processing. Below is a detailed protocol and a list of essential research reagents and equipment.

Detailed Experimental Protocol for an ERP Study

• Participant Preparation: The participant's scalp is cleaned, and an electroconductive gel or saline solution is applied to each electrode to reduce impedance, typically to below 5-10 kΩ. The electrode cap is positioned according to the international 10-20 system or a higher-density variant [58] [19].
• Experimental Task Programming and Presentation: The stimulus paradigm (e.g., oddball task) is programmed using specialized software (e.g., PsychoPy [58]). Stimuli are presented with precise timing, and event markers (triggers) are sent simultaneously to the EEG recording system to mark the onset of each stimulus.
• EEG Data Acquisition: Data is recorded using a multi-channel EEG system (e.g., 32-channel [58]). Key acquisition parameters include:
  • Sampling Rate: ≥ 500 Hz (1000 Hz is common) to accurately capture millisecond-scale dynamics [19].
  • Hardware Filters: A high-pass filter (e.g., 0.1 Hz) and a low-pass filter (e.g., 100 Hz) are applied during recording to remove slow drifts and high-frequency noise.
  • Reference: Data is recorded against a reference electrode (e.g., Cz, linked mastoids).
• Data Preprocessing: This critical step removes artifacts from the continuous data.
  • Filtering: Digital band-pass filtering (e.g., 0.1-30 Hz) is applied to focus on the frequency range of ERPs.
  • Artifact Rejection: Automated or manual procedures are used to remove epochs contaminated by large artifacts such as eye blinks (electrooculogram, EOG), muscle activity (EMG), or head movements [57].
  • Epoching: The continuous EEG is segmented into epochs (e.g., -200 ms to +800 ms relative to stimulus onset).
  • Baseline Correction: The average voltage of the pre-stimulus period (e.g., -200 to 0 ms) is subtracted from the entire epoch.
• Averaging and Analysis: Artifact-free epochs for each condition are averaged to create the ERP waveforms. Component amplitude and latency are then quantified for statistical analysis.

Table 3: Essential Research Reagents and Equipment for EEG/ERP Research

Item / Solution	Function / Purpose
EEG Amplifier & Acquisition System	Amplifies microvolt-level brain signals and converts them to digital data for analysis [58].
Active/Passive Electrode Cap	Holds electrodes in standardized positions on the scalp (e.g., 10-20 system) [58] [19].
Electroconductive Gel / Saline Solution	Ensures a stable, low-impedance electrical connection between the scalp and the electrodes.
Abrasive Skin Prep Gel	Gently abrades the scalp to remove dead skin cells and reduce electrical impedance.
Stimulus Presentation Software (e.g., PsychoPy)	Presents visual, auditory, or other stimuli with precise timing and sends synchronization triggers [58].
ERP Analysis Software (e.g., EEGLAB, MNE, Brainstorm)	Provides a suite of tools for preprocessing, visualizing, and statistically analyzing EEG/ERP data [19].

Multimodal Integration: EEG with fNIRS and fMRI

No single neuroimaging modality provides a complete picture of brain function. The strength of EEG lies in its millisecond temporal resolution, but it suffers from limited spatial resolution due to the inverse problem—the difficulty in pinpointing the exact intracranial sources of scalp electrical signals [57]. Conversely, hemodynamic-based methods like fMRI and fNIRS offer superior spatial resolution but poor temporal resolution [53]. Consequently, integrating EEG with fNIRS or fMRI is a powerful approach to achieve concurrent high temporal and spatial resolution.

EEG-fNIRS Integration

This dual-modality approach is particularly synergistic because it combines a direct measure of neural electrical activity (EEG) with an indirect measure of its metabolic consequence (fNIRS). This allows researchers to probe the relationship between cortical electrical activity and hemodynamic/metabolic responses—a process known as neurovascular coupling [19]. Studies using simultaneous EEG-fNIRS have shown that structure-function coupling in the brain varies between electrical and hemodynamic networks, with differences most pronounced in higher-order association cortices [19].

Technical Considerations for Integration:

• Helmet Design: A significant challenge is the integration of EEG electrodes and fNIRS optodes into a single headgear. Solutions range from using flexible EEG caps with punctures for fNIRS fixtures to custom 3D-printed helmets that offer a more precise and stable fit for both sensor types [53].
• Synchronization: For precise temporal alignment, a unified processor can be used to acquire and process both EEG and fNIRS signals simultaneously. Alternatively, separate systems can be synchronized via triggers (e.g., TTL pulses) from a host computer, though this may offer less precision [53].
• Data Fusion: After separate preprocessing, data fusion techniques such as joint Independent Component Analysis (jICA) or machine learning models that combine features from both modalities can be employed to extract integrated information [53] [54].

Figure 2: A Framework for Simultaneous EEG-fNIRS Integration, Combining Temporal and Spatial Strengths.

Applications in Clinical Research and Drug Development

The objective and quantifiable nature of ERPs makes them valuable tools for understanding disease mechanisms and evaluating novel therapeutics, particularly in psychiatry and neurology.

Biomarkers in Psychiatric Disorders

ERP components show reliable abnormalities across several disorders, serving as potential endophenotypes—heritable, trait-like markers that reflect the genetic vulnerability to an illness [56] [59].

• Schizophrenia: One of the most robust findings is a reduction in P300 amplitude, which is thought to reflect deficits in attention and context updating. Patients also show diminished P50 suppression, indicating a sensory gating deficit that may underlie perceptual fragmentation, and reduced Mismatch Negativity (MMN), suggesting impaired pre-attentive auditory processing [56].
• Bipolar Affective Disorder and Depression: Psychotic bipolar disorder shares some P50 gating deficits with schizophrenia [56]. Major depression is often associated with a reduced P300 amplitude, particularly in patients with psychotic features or severe symptoms [56].
• Alcohol Dependence Syndrome: Acute ethanol intake suppresses P300 amplitude. Importantly, P300 abnormalities are also observed in abstinent individuals and their first-degree relatives, suggesting it may be a marker of neurobiological vulnerability [56].

De-risking Drug Development

In the pharmaceutical industry, EEG and ERPs are increasingly used as pharmacodynamic biomarkers to de-risk clinical trials [59]. They can objectively demonstrate that a drug has engaged its intended functional target in the brain.

• Phase 1 Trials: EEG can be used in early-stage trials to establish brain penetration, functional target engagement, and dose-response relationships. For example, in developing drugs for cognitive impairment associated with schizophrenia (CIAS), EEG/ERP signals can demonstrate pro-cognitive effects at doses lower than those causing adverse side effects, information that molecular imaging (e.g., PET) alone might miss [59].
• Patient Stratification: In later-phase trials, ERP biomarkers can be used to enrich study populations. By selecting patients who show specific neurophysiological deficits (e.g., reduced P300), researchers can create a more homogeneous cohort that is more likely to respond to a mechanism-driven treatment, thereby increasing the probability of trial success [59].

EEG and its derivative, ERPs, provide an indispensable window into the brain's millisecond-level dynamics, offering a direct measure of neural electrical activity that is complementary to hemodynamic methods like fMRI and fNIRS. The high temporal resolution of EEG is critical for dissecting the rapid cascade of sensory and cognitive processes. As part of a multimodal neuroimaging toolkit, EEG contributes to a more comprehensive understanding of brain structure-function relationships. For clinical researchers and drug developers, ERPs serve as robust, quantitative biomarkers for characterizing disease phenotypes and objectively evaluating the functional impact of novel therapeutics in the human brain. The continued integration of EEG with other modalities and its application within a precision psychiatry framework holds great promise for advancing both neuroscience and clinical care.

Functional neuroimaging has revolutionized our understanding of brain function, yet each modality presents distinct trade-offs between spatial resolution, temporal resolution, portability, and cost. While functional magnetic resonance imaging (fMRI) provides high spatial resolution and electroencephalography (EEG) offers millisecond temporal precision, functional near-infrared spectroscopy (fNIRS) has emerged as a particularly valuable tool for studying brain function in real-world settings and clinical populations. fNIRS operates on the principle of neurovascular coupling, the fundamental relationship between neuronal activity and subsequent hemodynamic responses [60]. As neurons become active, their increased energy demands trigger a complex vascular response that ultimately delivers oxygenated blood to support metabolic needs [60]. fNIRS captures this phenomenon by measuring changes in cerebral blood oxygenation, positioning itself as a portable complement to established modalities like fMRI and EEG within the neuroscientist's toolkit [61].

This technical guide explores the unique capabilities of fNIRS technology, with a specific focus on its application in motor rehabilitation for clinical populations such as stroke survivors. We examine the technical foundations of fNIRS, its integration with other modalities, detailed experimental protocols, and its growing role in guiding therapeutic interventions through neurofeedback.

Technical Foundations of fNIRS

Basic Principles and Mechanisms

fNIRS is a non-invasive optical imaging technique that measures hemodynamic changes in the cerebral cortex. The technology leverages the relative transparency of biological tissues to light in the near-infrared spectrum (700-900 nm), and the differential absorption properties of oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) in this range [60] [62]. When near-infrared light is emitted through the scalp and skull, it travels through brain tissue in a "banana-shaped" path before being detected by sensors placed several centimeters from the source [63] [62]. The attenuation of light between source and detector is used to calculate relative concentration changes in HbO and HbR based on the modified Beer-Lambert law [63].

The fundamental physiological process measured by fNIRS is the hemodynamic response to neuronal activity. When a specific brain region becomes active, it initially experiences a brief increase in oxygen consumption, leading to a slight rise in deoxygenated hemoglobin (the "initial dip") [60]. This is rapidly followed by a substantial increase in cerebral blood flow that overshoots metabolic demands, resulting in a characteristic increase in oxygenated hemoglobin and decrease in deoxygenated hemoglobin [63] [60]. This hemodynamic response unfolds over 2-6 seconds post-stimulus, defining the temporal resolution limits of fNIRS [64].

Comparative Analysis with Other Neuroimaging Modalities

Table 1: Comparison of Key Neuroimaging Modalities

Feature	fNIRS	EEG	fMRI
What it Measures	Hemodynamic response (HbO, HbR)	Electrical activity from neurons	Blood-oxygen-level-dependent (BOLD) signal
Temporal Resolution	Moderate (seconds)	High (milliseconds)	Slow (seconds)
Spatial Resolution	Moderate (cortical surface)	Low	High
Portability	High (wearable systems available)	High	Low (requires fixed scanner)
Motion Tolerance	Moderate to High	Low	Low
Cost	Moderate	Low	High
Best Use Cases	Naturalistic studies, rehabilitation, child development	Fast cognitive tasks, sleep research, seizure detection	Detailed brain mapping, deep brain structures

fNIRS occupies a unique position in the neuroimaging landscape, balancing reasonable spatial resolution with good portability and motion tolerance [64]. Unlike fMRI, which requires subjects to lie still in a confined scanner, fNIRS can be deployed in various settings, allowing study participants to move, speak, and engage in real-world activities while being monitored [61]. Compared to EEG, fNIRS offers superior spatial localization and is less susceptible to motion artifacts and electrical interference, though it cannot match EEG's millisecond-level temporal resolution [64]. This combination of characteristics makes fNIRS particularly suitable for studying complex, ecologically valid behaviors and for use with clinical populations who may have difficulty remaining still in traditional imaging environments [63] [61].

fNIRS in Motor Rehabilitation: Applications and Evidence

Stroke Motor Recovery

Stroke rehabilitation represents one of the most promising clinical applications for fNIRS technology. Motor impairment following stroke has profound consequences on functional independence and quality of life, creating substantial demand for effective rehabilitation services [65]. fNIRS enables researchers and clinicians to monitor cortical activation patterns during recovery, providing valuable insights into neuroplasticity and compensation mechanisms.

Studies utilizing fNIRS have revealed characteristic reorganization patterns in the motor cortex following stroke. During upper limb movements, stroke patients often demonstrate bilateral activation of motor cortices, in contrast to the predominantly contralateral activation observed in healthy individuals [63]. This shift toward ipsilateral involvement is thought to represent a compensatory mechanism where the unaffected hemisphere supports recovery of motor function [63]. Research by Sui et al. found that reorganization of the ipsilateral hemisphere was particularly important for motor recovery, validating the concept of healthy-side compensation during stroke rehabilitation [63].

The relationship between functional recovery and hemispheric activation patterns has been further elucidated by fNIRS studies. Delorme et al. reported a positive correlation between Fugl-Meyer assessment scores (a standardized measure of motor function) and lateralization indices in stroke patients, indicating that as motor function improved, cortical activation patterns transitioned from bilateral back toward the affected hemisphere [63]. This finding suggests that fNIRS can objectively track recovery progress and potentially guide therapeutic approaches.

Neurofeedback Applications

fNIRS has emerged as a powerful tool for neurofeedback-based rehabilitation approaches. Neurofeedback enables patients to consciously modulate their brain activity through real-time visual representation of their hemodynamic responses [66]. In motor rehabilitation, this typically involves patients engaging in motor imagery (mentally rehearsing physical movements) while receiving feedback about associated cortical activation [65] [66].

Shimadzu's NIRS neurorehabilitation system exemplifies this approach, presenting brain activity feedback in intuitively understandable formats such as progress bars, intensity scales, or circular scales to maintain patient engagement and motivation [66]. Clinical studies led by Mihara and colleagues have demonstrated that fNIRS-mediated neurofeedback enhances motor imagery-related cortical activation and improves gait and balance recovery in post-stroke patients [66]. This neurofeedback approach leverages the brain's natural capacity for neuroplasticity by providing patients with direct access to information about their functional networks, potentially leading to more efficient motor recovery than traditional rehabilitation methods alone [66].

Hybrid fNIRS-EEG Approaches

The integration of fNIRS with EEG creates a powerful multimodal platform that overcomes the inherent limitations of each individual modality. Hybrid fNIRS-EEG systems simultaneously measure electrical and hemodynamic activities in the brain, providing complementary information about neural processes [65] [19]. EEG contributes excellent temporal resolution for capturing rapid neural dynamics, while fNIRS provides better spatial localization of the underlying activated regions [65] [64].

This complementary relationship is particularly valuable for investigating structure-function relationships in brain networks. A 2024 study by using simultaneous EEG and fNIRS recordings found that fNIRS structure-function coupling resembled slower-frequency EEG coupling at rest, with variations across brain states and oscillations [19]. The study also revealed heterogeneous local relationships, with stronger coupling in sensory cortex and increased decoupling in association cortex, following the unimodal to transmodal gradient [19].

For motor rehabilitation, hybrid fNIRS-EEG brain-computer interfaces (BCIs) have shown particular promise. These systems can provide more robust control signals for assistive devices by leveraging the strengths of both modalities, and offer comprehensive feedback to patients and therapists throughout the rehabilitation process [65]. The combination of electrophysiological and hemodynamic information enhances classification accuracy for detecting motor intentions and monitoring recovery progress [65].

Experimental Protocols and Methodologies

Standardized Experimental Setup

Table 2: Key Research Reagents and Equipment for fNIRS Studies

Item	Function	Specifications/Considerations
fNIRS System	Measures hemodynamic responses	Continuous wave systems most common; 2+ wavelengths (760nm, 850nm)
Optodes	Light sources and detectors	Source-detector separation: 3-5cm (adults); determines penetration depth
EEG System	Measures electrical activity	30+ electrodes following 10-5 or 10-20 system; sampling rate ≥1000Hz
fNIRS-EEG Cap	Integrated sensor placement	Pre-defined compatible openings to avoid interference
Stimulus Presentation	Delivers experimental tasks	Software like PsychoPy; synchronized triggers
Motion Tracking	Monitors head movement	Accelerometers or camera-based systems
Signal Processing	Analyzes raw data	Motion correction, filtering, component analysis

Implementing rigorous fNIRS experiments requires careful attention to protocol design and technical parameters. A typical motor imagery study involves participants performing mental rehearsal of specific movements without physical execution, while fNIRS records associated hemodynamic changes in motor cortex regions [65]. The international 10-20 system for electrode placement provides a standardized framework for positioning fNIRS optodes, ensuring consistency across studies and subjects [64] [19].

Data acquisition parameters significantly impact signal quality. fNIRS systems typically employ two wavelengths (commonly 760nm and 850nm) to differentially capture HbO and HbR concentrations [19]. Source-detector separation distances of 3-5cm in adults allow sufficient penetration to cortical tissues while maintaining detectable light intensity [60]. Sampling rates for fNIRS generally range from 2-10Hz, sufficient to capture the relatively slow hemodynamic responses [63] [19].

For simultaneous fNIRS-EEG recordings, synchronization between systems is critical. This can be achieved through hardware triggers or shared clock systems, with careful attention to minimizing interference between modalities [64]. Integrated caps with pre-defined fNIRS-compatible openings help maintain proper optode and electrode positioning while avoiding physical interference [64].

Signal Processing and Data Analysis

fNIRS data processing involves multiple stages to extract meaningful hemodynamic responses from raw optical signals. Preprocessing typically includes:

• Optical density conversion of raw intensity measurements [19]
• Signal quality assessment using metrics like scalp-coupled index (SCI) to identify poor-quality channels [19]
• Motion artifact correction using algorithms like PCA-based approaches or global variance in temporal derivative (GVTD) metrics [19]
• Bandpass filtering (typically 0.02-0.08 Hz for resting-state studies) to remove physiological noise [19]
• Conversion to hemoglobin concentrations using the modified Beer-Lambert law [63]

Feature extraction for classification in BCI applications commonly includes temporal (mean, peak, variance, slope) and morphological (kurtosis, skewness) characteristics of HbO and HbR signals [65]. Machine learning classifiers such as Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and increasingly, deep learning approaches like Convolutional Neural Networks (CNN) are employed to decode motor intentions from these features [65].

For functional connectivity analysis, correlations between hemodynamic time courses from different brain regions are computed to investigate network properties and their alterations in clinical populations [63] [67].

Visualization of fNIRS Signaling Pathways and Experimental Workflows

Neurovascular Coupling Pathway

Diagram 1: Neurovascular coupling pathway measured by fNIRS

fNIRS Experimental Workflow

Diagram 2: fNIRS experimental workflow from setup to application

fNIRS has established itself as a valuable tool in the functional neuroimaging arsenal, particularly for studying motor rehabilitation in clinical populations. Its portability, tolerance to movement, and ability to provide continuous monitoring make it uniquely suited for investigating brain function in real-world contexts and with patients who cannot be easily studied using traditional neuroimaging methods [61]. The integration of fNIRS with other modalities, especially EEG, creates powerful multimodal platforms that overcome individual limitations and provide more comprehensive insights into brain dynamics [65] [19].

Future developments in fNIRS technology will likely focus on improving spatial resolution through high-density optode arrays, enhancing signal processing algorithms to better separate neural signals from physiological noise, and developing more sophisticated analytic approaches for understanding complex brain networks [60] [67]. As the technology continues to evolve, fNIRS is poised to play an increasingly important role in both basic neuroscience research and clinical applications, particularly in rehabilitation medicine where understanding brain plasticity and recovery processes is essential for developing more effective interventions.

For researchers and clinicians, fNIRS offers a practical means of bringing functional brain imaging out of the laboratory and into real-world settings, ultimately helping to bridge the gap between controlled experimental paradigms and the complex, dynamic nature of everyday brain function.

The quest to understand the intricate workings of the human brain necessitates neuroimaging technologies that can capture its complex spatiotemporal dynamics. While functional magnetic resonance imaging (fMRI) has emerged as the gold standard for non-invasive brain imaging due to its superior spatial resolution, its practical constraints limit ecological validity for many research paradigms [41]. Similarly, electroencephalography (EEG) provides millisecond temporal resolution but suffers from limited spatial accuracy [68]. In this landscape, the integration of functional near-infrared spectroscopy (fNIRS) and EEG has emerged as a powerful multimodal approach that capitalizes on their complementary strengths. This integration is particularly valuable for investigating complex neural systems such as the Action Observation Network (AON) and studying neurovascular coupling—the fundamental relationship between neuronal electrical activity and subsequent hemodynamic responses [69] [70].

This technical guide explores the theoretical foundations, methodological frameworks, and practical applications of combined fNIRS-EEG, with particular emphasis on its transformative potential for AON research. By simultaneously capturing electrical and hemodynamic aspects of brain function, this multimodal approach offers unprecedented insights into brain dynamics during ecologically valid paradigms, including motor execution, observation, and imagery tasks that are central to understanding the human mirroring system [71].

Technical Foundations: Principles of fMRI, EEG, and fNIRS

Functional Magnetic Resonance Imaging (fMRI)

fMRI measures brain activity indirectly through the blood oxygen level-dependent (BOLD) contrast, which exploits the different magnetic properties of oxygenated and deoxygenated hemoglobin [41]. When neurons become active, localized increases in cerebral blood flow deliver oxygenated blood that exceeds metabolic demands, resulting in a measurable signal change. fMRI provides excellent spatial resolution (typically millimeter-range) and whole-brain coverage, making it ideal for mapping neural networks. However, its practical limitations include sensitivity to motion artifacts, requirement for a supine position within a scanner bore, loud acoustic noise, and high operational costs [41]. These constraints particularly challenge the study of naturalistic motor behaviors and interactive paradigms, especially those involving the AON where participants need to observe and potentially execute actions [72].

Electroencephalography (EEG)

EEG measures the electrical activity generated by synchronized firing of neuronal populations, primarily pyramidal cells in the cerebral cortex [68]. Electrodes placed on the scalp detect voltage fluctuations resulting from postsynaptic potentials, providing direct measurement of neural electrical activity with millisecond temporal resolution. This exquisite temporal sensitivity makes EEG ideal for studying rapid cognitive processes, event-related potentials, and neural oscillations. However, EEG's spatial resolution is limited by the blurring and distortion of electrical signals as they pass through the skull and scalp, a phenomenon known as the volume conduction problem [68]. This limitation is particularly problematic for localizing activity within precise cortical structures, such as those comprising the AON.

Functional Near-Infrared Spectroscopy (fNIRS)

fNIRS is an optical neuroimaging technique that measures cortical hemodynamic responses by exploiting the differential absorption properties of oxygenated (HbO) and deoxygenated hemoglobin (HbR) in the near-infrared spectrum (700-900 nm) [69]. Based on the modified Beer-Lambert law, fNIRS quantifies changes in hemoglobin concentration associated with neural activity, measuring similar hemodynamic responses to fMRI but with greater portability and motion tolerance [41]. fNIRS typically penetrates 1-2 cm into the cortical surface, providing better spatial resolution than EEG but more limited than fMRI [68]. Its key advantages include relative immunity to motion artifacts, quiet operation, and portability, enabling studies in naturalistic settings and with populations challenging to scan with fMRI (e.g., infants, clinical populations) [41].

Table 1: Comparative Analysis of Neuroimaging Modalities

Feature	fMRI	EEG	fNIRS
What It Measures	BOLD signal (blood oxygenation)	Electrical activity	Hemodynamic response (HbO/HbR)
Temporal Resolution	Low (seconds)	High (milliseconds)	Low (seconds)
Spatial Resolution	High (millimeter-range)	Low (centimeter-range)	Moderate (surface cortex)
Depth Penetration	Whole brain	Cortical surface	Outer cortex (1-2.5 cm)
Portability	Low (fixed scanner)	High (wearable systems)	High (wearable systems)
Motion Tolerance	Low	Moderate	High
Key Strengths	Whole-brain coverage, high spatial resolution	Excellent temporal resolution, direct neural measurement	Good balance of spatial resolution and ecological validity

The fNIRS-EEG Integration: Technical Synergies

The combination of fNIRS and EEG creates a powerful multimodal approach that captures complementary aspects of brain function—electrical activity and hemodynamic response—within a single experimental framework [70]. This integration is particularly valuable for investigating neurovascular coupling, the fundamental relationship between neuronal electrical activity and subsequent hemodynamic changes that underlies BOLD fMRI signals [41].

From a technical perspective, fNIRS and EEG are highly compatible modalities. Both are non-invasive, portable, and relatively tolerant of movement compared to fMRI, enabling studies in ecologically valid settings [68]. They can be recorded simultaneously using integrated systems or carefully synchronized separate equipment. Modern solutions include high-density EEG caps with pre-defined openings for fNIRS optodes or specialized holders that allow both sensor types to be co-located on the scalp [70].

The complementary nature of fNIRS and EEG data provides a more complete picture of brain activity than either modality alone. While EEG captures rapid neural dynamics with millisecond precision, fNIRS provides better spatial localization of the underlying cortical generators [68]. This synergy is particularly advantageous for studying complex cognitive-motor processes that involve both rapid neural dynamics and sustained cortical activation patterns, such as those engaged during action observation, execution, and imagery [71].

Studying the Action Observation Network (AON)

The AON: Neural Correlates and Functions

The Action Observation Network (AON) is a system of brain regions engaged both when performing actions and when observing others perform similar actions [72]. Originally conceptualized as a "mirror neuron system" in non-human primate research, the human AON is thought to facilitate action understanding, imitation, and social cognition by mapping observed actions onto the observer's own motor representations [72]. Key cortical regions comprising the AON include the ventral premotor cortex (PMC), inferior parietal lobule (IPL), superior temporal sulcus (STS), and portions of the primary motor cortex (M1) [71].

The AON is typically studied through paradigms comparing brain activity during action execution, observation, and imagery. According to the Simulation Hypothesis, these three conditions share overlapping neural mechanisms, making them particularly relevant for motor learning and rehabilitation [71]. However, technical limitations of individual neuroimaging modalities have historically constrained our ability to fully characterize AON dynamics during ecologically valid paradigms.

Methodological Limitations of Unimodal Approaches

EEG Studies of AON: EEG research has primarily investigated the AON through mu rhythm (8-12 Hz) desynchronization at central scalp sites during both action execution and observation [72]. While this approach provides excellent temporal resolution for tracking the rapid dynamics of AON engagement, it suffers from limited spatial precision and confounding signals from attention-related alpha rhythms that show similar desynchronization patterns [72]. The poor spatial resolution of EEG makes it difficult to distinguish activity from specific AON regions, particularly those in deeper cortical structures.

fMRI Studies of AON: fMRI provides excellent spatial localization of AON regions but is severely limited by its sensitivity to motion artifacts [72]. This constraint has led most fMRI studies of the AON to focus exclusively on action observation rather than execution, or to employ highly constrained motor tasks that lack ecological validity [72]. The typical supine position within the scanner bore also prevents naturalistic social interactions and live action observation paradigms.

fNIRS as an Optimal Tool for AON Research: fNIRS addresses many limitations of both EEG and fMRI for AON research. Its tolerance for movement artifacts enables studies involving actual motor execution and naturalistic observation conditions [73]. Compared to EEG, fNIRS provides superior spatial localization of cortical AON regions, while its portability allows for face-to-face interaction paradigms that are impossible in the fMRI environment [71]. Several fNIRS studies have successfully investigated lateralization patterns in the AON during both action execution and observation, demonstrating the technique's sensitivity to key functional properties of this network [73].

Experimental Protocols and Methodologies

Simultaneous fNIRS-EEG Acquisition for AON Studies

A growing body of research demonstrates the effectiveness of simultaneous fNIRS-EEG recordings for studying the AON. A 2023 study by researchers from the National Institutes of Health and University of Maryland provides a exemplary protocol for investigating motor execution, observation, and imagery using a multimodal approach [71].

Participant Preparation and Equipment Setup:

• Participants are fitted with an integrated fNIRS-EEG cap containing 128 EEG electrodes and 24 fNIRS channels (8 sources, 10 detectors) covering sensorimotor and parietal cortices [71].
• fNIRS optodes are embedded within the elastic EEG cap, with inter-optode spacing maintained at approximately 2.88 ± 0.13 cm [71].
• A 24-channel continuous-wave fNIRS system (e.g., Hitachi ETG-4100) measures changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentration at two wavelengths (695 nm and 830 nm) with a sampling rate of 10 Hz [71].
• EEG data is collected simultaneously from the integrated electrodes, typically at higher sampling rates (e.g., 512 Hz or higher) to capture rapid neural dynamics [74].
• After the experiment, each participant's fNIRS optodes are digitized in reference to anatomical landmarks (nasion, inion, preauricular points) using a 3D magnetic space digitizer to account for individual differences in head size and cap placement [71].

Experimental Paradigm: The protocol employs a live-action paradigm where participants sit face-to-face with an experimenter across a table, enabling ecologically valid social interaction during the tasks [71]:

• Motor Execution (ME): Participants physically perform actions (e.g., grasping, lifting, and moving a cup) with their right hand in response to an audio cue ("Your turn").

• Motor Observation (MO): Participants observe the experimenter performing the same actions in response to an audio cue ("My turn").

• Motor Imagery (MI): Participants mentally simulate performing the action without any overt movement in response to an audio cue ("Imagine your turn").

Each condition typically involves multiple trials (e.g., 5 blocks of 8 trials each) with randomized presentation to avoid order effects. Trial durations of 15 seconds allow for robust hemodynamic response capture while maintaining participant engagement [71].

fNIRS-Specific Lateralization Protocol for AON

A specialized fNIRS protocol for investigating lateralization in the AON involves measuring brain activity over both left and right motor cortices during self-action and observation conditions [73]:

Participant Characteristics:

• Sample size of approximately 40 participants to ensure adequate statistical power for lateralization comparisons [73].
• Inclusion of handedness assessments using the Edinburgh Handedness Inventory to account for individual differences in motor lateralization [73].
• Measurement of manual dexterity using the Purdue Pegboard Task to quantify performance differences between hands [73].

Self-Action Task Protocol: Participants complete five fine motor tasks with both left and right hands:

• Putting plastic pennies in a bank
• Moving pom poms between bowls with tweezers
• Placing pegs in a pegboard
• Stringing large wooden beads on a rod
• Turning large wooden bolts on a screw [73]

Each task is performed for 15-second trials, with five blocks of eight trials for each hand, totaling 40 trials per hand across the five tasks [73].

Observation Task Protocol: Participants observe a research assistant performing the same five fine motor tasks with both left and right hands while counting repetitions to maintain attention. The same trial structure as the self-action condition ensures comparability between conditions [73].

Table 2: Key Experimental Parameters from Representative fNIRS-EEG AON Studies

Parameter	fNIRS-EEG AON Study [71]	fNIRS Lateralization Study [73]
Participants	21 healthy adults	41 right-handed undergraduates
fNIRS Channels	24	Bilateral motor cortex coverage
EEG Channels	128	Not applicable
Tasks	Motor execution, observation, imagery	Self-action and observation with left/right hands
Trial Duration	Not specified	15 seconds
Key Findings	Consistent activation in left inferior parietal lobe across conditions	Lateralization during self-action but not observation
Analysis Method	Structured sparse multiset CCA	Hemodynamic response lateralization indices

Data Analysis Approaches

Preprocessing Pipelines

fNIRS Data Preprocessing: Raw fNIRS data undergoes multiple preprocessing steps to enhance signal quality:

• Signal Quality Assessment and Channel Rejection: Channels with poor signal-to-noise ratio are identified and excluded from analysis [75].
• Motion Artifact Correction: Algorithms such as wavelet-based methods or moving standard deviation approaches detect and correct motion-induced signal distortions [75].
• Filtering and Drift Removal: Bandpass filtering (typically 0.01-0.2 Hz) removes physiological noise from cardiac cycles (~1 Hz), respiration (~0.3 Hz), and very low-frequency drifts [75].
• Hemoglobin Concentration Calculation: The modified Beer-Lambert law converts optical density changes to relative concentrations of oxygenated (HbO) and deoxygenated hemoglobin (HbR) [75].

EEG Data Preprocessing: EEG data preprocessing typically includes:

• Filtering: Bandpass filtering (e.g., 0.1-40 Hz) to remove slow drifts and high-frequency noise.
• Artifact Removal: Identification and removal of artifacts from eye blinks, muscle activity, and cardiac signals using independent component analysis (ICA) or regression-based approaches.
• Epoching: Segmenting data into trials time-locked to experimental events.
• Spectral Analysis: Computation of power in frequency bands of interest (e.g., mu rhythm: 8-13 Hz) for event-related desynchronization (ERD) analysis [72].

Multimodal Data Fusion Techniques

The true power of simultaneous fNIRS-EEG recording emerges through data fusion techniques that integrate information from both modalities:

Structured Sparse Multiset Canonical Correlation Analysis (ssmCCA): This advanced fusion method identifies components that maximize correlations between fNIRS hemodynamic responses and EEG spectral features [71]. ssmCCA incorporates structured sparsity constraints to select variables that are consistent across modalities while maintaining interpretability, effectively pinpointing brain regions where both electrical and hemodynamic responses show consistent task-related activation patterns [71].

Joint Independent Component Analysis (jICA): jICA decomposes combined fNIRS-EEG datasets into spatially independent components that share similar temporal profiles, identifying networks that co-activate across both modalities [68].

Canonical Correlation Analysis (CCA): CCA finds linear combinations of fNIRS and EEG features that maximize their correlation, revealing underlying relationships between hemodynamic and electrical brain responses [68].

These fusion approaches have demonstrated particular utility for AON research, consistently identifying activation in the left inferior parietal lobe, supramarginal gyrus, and post-central gyrus during motor execution, observation, and imagery—regions that show less consistent activation in unimodal analyses [71].

Diagram 1: fNIRS-EEG Data Analysis Workflow. This flowchart illustrates the comprehensive pipeline from data acquisition through multimodal fusion for AON characterization.

Table 3: Research Reagent Solutions for fNIRS-EEG AON Studies

Item	Function	Example Specifications
Integrated fNIRS-EEG System	Simultaneous acquisition of electrical and hemodynamic signals	24 fNIRS channels, 128 EEG electrodes, synchronization capability [71]
3D Magnetic Digitizer	Coregistration of optode/electrode positions with anatomical landmarks	Fastrak Polhemus system for digitizing nasion, inion, preauricular points [71]
Structured Sparse Multiset CCA	Advanced data fusion algorithm	MATLAB/Python implementation for identifying correlated components across modalities [71]
fNIRS Motion Correction Algorithms	Artifact removal from movement	Wavelet-based methods, moving standard deviation approaches [75]
EEG Artifact Removal Tools	Ocular and muscle artifact reduction	Independent Component Analysis (ICA), signal regression techniques [72]
International 10-20 System Cap	Standardized sensor placement	Integrated cap with predefined openings for fNIRS optodes and EEG electrodes [68]
Hemodynamic Response Modeling	Analysis of fNIRS response dynamics	General Linear Model (GLM) with canonical HRF, block averaging approaches [75]
EEG Spectral Analysis Tools	Computation of frequency-specific power	Time-frequency analysis, event-related desynchronization/synchronization (ERD/ERS) [72]

Future Directions and Clinical Applications

The integration of fNIRS and EEG holds particular promise for advancing clinical neuroscience and rehabilitation. In brain-computer interfaces (BCIs), combining fNIRS with EEG has been shown to improve classification accuracy for motor execution and imagery tasks compared to single-modality approaches [69]. This enhancement is particularly valuable for developing more robust communication and control systems for individuals with severe motor disabilities.

In clinical populations, fNIRS-EEG integration offers new avenues for monitoring rehabilitation progress in conditions such as stroke, Parkinson's disease, and epilepsy [70]. The ability to simultaneously track electrical and hemodynamic correlates of neural function provides a more comprehensive assessment of recovery processes than either modality alone. Furthermore, the portability of these systems enables monitoring in naturalistic settings, including during actual rehabilitation sessions rather than in isolated scanning environments.

Future technical developments will likely focus on improving the integration of fNIRS and EEG systems, enhancing data fusion algorithms, and developing more sophisticated computational models of neurovascular coupling. The growing availability of commercial integrated systems is making this multimodal approach increasingly accessible to researchers [70]. As these technologies continue to evolve, simultaneous fNIRS-EEG recording is poised to become a standard methodology for cognitive neuroscience research, particularly for studying complex brain networks like the AON in ecologically valid contexts.

Diagram 2: Neurovascular Coupling in AON Studies. This diagram illustrates the relationship between neuronal activity, hemodynamic response, and their measurement through EEG and fNIRS during AON paradigms.

The multimodal integration of fNIRS and EEG represents a significant advancement in neuroimaging methodology, offering unique insights into brain function by combining complementary information from electrical and hemodynamic signals. For the study of the Action Observation Network and neurovascular coupling, this approach provides unprecedented opportunities to investigate brain dynamics during ecologically valid paradigms involving motor execution, observation, and imagery. The continued development of integrated hardware systems and sophisticated data fusion algorithms will further enhance the utility of this multimodal approach, solidifying its role as a cornerstone methodology in cognitive neuroscience and clinical research.

The pursuit of understanding the human brain's intricate functions necessitates neuroimaging technologies that can capture its dynamic activity with high fidelity. Among the plethora of available techniques, functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS) have emerged as particularly powerful tools, each with distinct advantages and limitations. Framed within a broader thesis on the basic principles of fMRI, EEG, and fNIRS for brain research, this guide examines the synergistic potential of combined fMRI-fNIRS approaches. The integration of these modalities is driven by the recognition that no single imaging method can fully capture the spatiotemporal complexity of neural processes [34] [41]. Whereas fMRI provides high spatial resolution for deep brain structures, fNIRS offers superior temporal resolution and portability for naturalistic settings [34]. This technical review explores how their combined use validates cortical activation patterns and enhances spatiotemporal mapping, thereby advancing both fundamental neuroscience and clinical applications in drug development and neurological disorders.

Fundamental Principles of fMRI and fNIRS

Functional Magnetic Resonance Imaging (fMRI)

fMRI has been a cornerstone of neuroimaging since its inception in the early 1990s, valued particularly for its ability to visualize brain activity with high spatial resolution [34] [41]. The technique relies on the blood oxygen level dependent (BOLD) contrast, which exploits the different magnetic properties of oxygenated and deoxygenated hemoglobin [41]. When neurons become active, localized increases in cerebral blood flow deliver oxygenated blood, creating a measurable signal change detectable by MRI scanners [34]. This enables researchers to localize brain regions involved in specific cognitive, sensory, and motor tasks with millimeter-level precision, covering both cortical and subcortical structures including the hippocampus, amygdala, and thalamus [34]. Despite these strengths, fMRI suffers from several limitations: constrained temporal resolution (typically 0.33-2 Hz) due to the slow hemodynamic response, high sensitivity to motion artifacts, requirement for expensive, immobile equipment, and practical constraints that limit ecological validity of tasks performed in the scanner environment [34] [41].

Functional Near-Infrared Spectroscopy (fNIRS)

fNIRS operates on fundamentally different principles, utilizing near-infrared light (650-950 nm) to measure changes in hemoglobin concentrations in cortical tissue [34] [41]. By measuring the differential absorption of light by oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR), fNIRS provides an indirect measure of neural activity with millisecond-level temporal precision [34]. Key advantages include significantly greater tolerance for movement artifacts, portability for bedside monitoring or field studies, lower cost, and compatibility with populations often excluded from fMRI studies (e.g., infants, individuals with implants) [34] [41]. These benefits come with trade-offs: fNIRS typically offers lower spatial resolution (1-3 cm), limited penetration depth that restricts measurement to superficial cortical regions, and susceptibility to confounding signals from extracerebral tissues such as scalp blood flow [34].

Table 1: Technical Comparison of fMRI and fNIRS

Feature	fMRI	fNIRS
Spatial Resolution	Millimeter-level (1-3 mm)	Centimeter-level (1-3 cm)
Temporal Resolution	Limited by hemodynamic response (0.33-2 Hz)	Superior (millisecond precision)
Depth Penetration	Whole-brain (cortical & subcortical)	Superficial cortical regions only (2-3 cm)
Portability	Immobile equipment	Portable to wearable systems available
Motion Tolerance	Highly sensitive	Relatively tolerant
Primary Signal	BOLD contrast	HbO and HbR concentration changes
Key Advantage	High spatial resolution, whole-brain coverage	Ecological validity, patient accessibility

Methodological Synergy and Spatial Correspondence

Complementary Strengths for Enhanced Mapping

The integration of fMRI and fNIRS creates a powerful methodological synergy that leverages their complementary strengths. This combination enables robust spatiotemporal mapping of neural activity by aligning spatially detailed fMRI maps with temporally dynamic fNIRS signals [34]. The fundamental relationship between their measured signals stems from their shared basis in neurovascular coupling - the mechanism linking neural activity to subsequent hemodynamic changes [76]. Both techniques measure aspects of the same underlying physiological processes, with the fMRI BOLD signal being most closely theoretically linked to changes in deoxygenated hemoglobin (HbR) concentration, which fNIRS can directly measure [76]. This theoretical relationship provides the foundation for cross-modal validation and signal interpretation.

Empirical Evidence for Spatial Correspondence

Substantial empirical evidence demonstrates strong spatial correspondence between fMRI and fNIRS measurements. A 2024 study with 22 healthy adults performing motor and visual tasks found fNIRS overlapped with up to 68% of fMRI activation areas in group-level analyses, with an average overlap of 47.25% in within-subject analyses [77]. This indicates that fNIRS reliably detects task-related activity in regions identified by fMRI, particularly in superficial cortical areas. Research focusing on motor tasks has further validated this correspondence. A 2023 study investigating motor execution and imagery found that subject-specific fNIRS signals could identify corresponding motor activation clusters in fMRI data, with significant peak activation overlapping individually-defined primary and premotor cortices [76]. Notably, no statistically significant differences were observed in spatial correspondence between HbO, HbR, and total hemoglobin measures for motor tasks [76].

Table 2: Quantitative Measures of Spatial Correspondence Between fNIRS and fMRI

Study Reference	Task Paradigm	Spatial Overlap Metric	Key Finding
NeuroImage 2024 [77]	Motor & Visual	Group-level: up to 68% Within-subject: 47.25% average	Promising clinical utility for superficial cortical assessment
Scientific Reports 2023 [76]	Motor Execution & Imagery	Significant peak activation overlap in M1/PMC	Both HbO and HbR provide usable spatial information
Scientific Reports 2022 [78]	Motor Execution & Imagery	Topographical similarity (Spearman correlation)	SMA activation reliably measured with fNIRS

Experimental Protocols and Methodologies

Motor Task Paradigms

Motor tasks represent one of the most validated experimental paradigms for fMRI-fNIRS integration. A standardized approach involves block-designed experiments comparing activation during motor execution and motor imagery against baseline periods [76] [78]. A typical protocol involves:

• Participants: Healthy adults with no neurological history, typically 10-20 participants per study [76] [78]
• Task Design: Alternating blocks of motor action, motor imagery, and baseline (e.g., 30-second blocks, repeated 4-9 times) [76]
• Motor Tasks: Bilateral finger tapping sequences (e.g., 1-2-1-4-3-4 finger sequence) for execution conditions, and kinesthetic imagination of the same sequence without overt movement for imagery conditions [76]
• fMRI Parameters: 3T scanner, T2*-weighted echo-planar imaging, TR=1500ms, TE=30ms, 3mm isotropic voxels [76]
• fNIRS Setup: Continuous-wave systems with 16 sources, 15 detectors (30mm optode distance), sampling at 5.08Hz, covering bilateral motor areas [76]

This protocol enables direct comparison of hemodynamic responses across modalities and validation of fNIRS against the gold standard of fMRI for localizing motor cortex activation.

Naturalistic Task Validation

Beyond constrained laboratory tasks, methodological approaches have been developed to validate fNIRS against fMRI during more ecologically valid tasks. One innovative protocol modified the dance video game "Dance Dance Revolution" for compatibility with both fMRI and fNIRS environments [79]. The methodology involves:

• Task Modification: Creating identical but simplified versions of naturalistic tasks that can be performed in both environments
• Block Design: Alternating 30-second game periods with 30-second rest periods
• fMRI Adaptation: Using foot buttons for limited movement (left/right toe presses) within scanner constraints
• fNIRS Testing: Full-body movement version in naturalistic setting
• ROI Analysis: Using fMRI-defined regions of interest (e.g., superior and middle temporal gyrus) to guide fNIRS analysis

This approach demonstrates that task-related increases in oxyhemoglobin measured by fNIRS in naturalistic settings correspond to activation patterns observed in constrained fMRI environments [79].

Supplementary Motor Area (SMA) Activation Protocol

Targeting specific brain regions for validation requires specialized protocols. For investigating SMA activation during motor imagery - particularly relevant for neurofeedback applications - studies have employed:

• Participants: Older adults (typically 55+ years) to enhance clinical relevance [78]
• Tasks: Motor execution of hand movements, motor imagery of hand movements, and motor imagery of whole-body movements [78]
• fNIRS Configuration: Continuous-wave systems with optodes positioned over SMA based on individual anatomy or standard positioning systems [78]
• Signal Analysis: Comparison of both HbO and HbR concentration changes, with emphasis on HbR for its potentially greater specificity in motor imagery [78]
• Validation: Consecutive or simultaneous fMRI sessions coregistered to fNIRS channels using individual anatomical scans

This protocol has demonstrated that SMA activation can be reliably measured with fNIRS, with spatial patterns showing significant topographical similarity to fMRI activation maps [78].

Implementation Frameworks and Integration Modes

Synchronous versus Asynchronous Acquisition

The integration of fMRI and fNIRS can be implemented through different temporal frameworks, each with distinct advantages and applications. Synchronous acquisition involves simultaneous data collection from both modalities, enabling direct temporal correlation of signals and precise investigation of neurovascular coupling dynamics [34]. This approach requires specialized hardware solutions to mitigate electromagnetic interference between systems and careful synchronization of data streams. In contrast, asynchronous acquisition involves separate data collection sessions, typically used for spatial registration and validation purposes [34] [76]. This approach is methodologically simpler and allows optimization of each modality's recording environment, though it precludes direct temporal comparison of signals.

Data Fusion and Analytical Approaches

Combining information from fMRI and fNIRS requires sophisticated data fusion techniques that operate at multiple levels:

• Spatial Coregistration: Individual fNIRS channels are mapped to corresponding cortical locations using digitized optode positions coregistered to structural MRI scans [76] [78]
• Signal Processing: Generalized Linear Models (GLM) are commonly applied to both fMRI and fNIRS data to estimate task-related responses, enabling direct comparison of activation patterns [76]
• Multivariate Pattern Analysis: Advanced techniques like structured sparse multiset Canonical Correlation Analysis (ssmCCA) can fuse electrical and hemodynamic responses to identify brain regions consistently detected by multiple modalities [71]
• Image Reconstruction: High-density fNIRS arrays enable tomographic reconstruction of activation maps, which can be quantitatively compared to fMRI statistical parametric maps [77]

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential Materials for Combined fMRI-fNIRS Research

Item	Function/Purpose	Technical Specifications
fNIRS System	Measures cortical hemodynamics via near-infrared light	Continuous-wave (CW) or time-resolved (TR) systems; 650-950nm wavelengths; 16+ sources/detectors [76] [78]
MRI-Compatible fNIRS Probes	Enables simultaneous acquisition without electromagnetic interference	Carbon-fiber electrodes; fiber-optic bundles; MRI-safe materials [34]
3D Digitization System	Coregisters fNIRS optode positions with anatomical MRI	Magnetic or optical digitizer; records nasion, inion, preauricular landmarks [76] [71]
Multimodal Analysis Software	Processes and integrates fMRI and fNIRS data	Homer3, BrainVoyager, SPM, AtlasViewer, fOLD toolbox [76] [78]
Custom Headgear	Secure, reproducible positioning of fNIRS optodes	3D-printed or thermoplastic helmets; integrated EEG-fNIRS configurations [80]
Short-Distance Detectors	Measures and removes superficial physiological noise	8mm source-detector separation; records extracerebral signals [76]
Physiological Monitors	Records confounding physiological signals	Pulse oximeter, blood pressure monitor, respiration belt [41]

Applications and Clinical Translation

Brain-Computer Interfaces and Communication Systems

The combined use of fMRI and fNIRS has significantly advanced brain-computer interface development, particularly for communication systems designed for completely paralyzed or "locked-in" patients [81]. These systems typically employ:

• Mental Imagery Tasks: Tennis imagery versus spatial navigation imagery to encode "yes/no" responses [81]
• Spatiotemporal Encoding: Combining different mental tasks with distinct timing windows to enhance classification accuracy
• fMRI-Guided fNIRS: Using fMRI localization to optimize fNIRS probe placement for target regions
• Signal Processing: General Linear Model analysis or multivariate pattern analysis for single-trial classification

Studies have demonstrated that healthy participants can achieve binary communication accuracies up to 100% using such systems, with fNIRS providing the mobility needed for bedside implementation [81].

Clinical Populations and Neurorehabilitation

The combined approach holds particular promise for clinical applications:

• Stroke Rehabilitation: fNIRS neurofeedback training targeting motor areas localized with fMRI [34] [78]
• Parkinson's Disease: SMA upregulation training using fNIRS, with fMRI validation of target engagement [78]
• Neurodevelopmental Disorders: Mapping atypical brain function in autism spectrum disorder using fMRI, with fNIRS enabling longitudinal monitoring [34]
• Aging and Neurodegenerative Diseases: Assessing cognitive decline with fNIRS in naturalistic settings, informed by fMRI localization [41]

Naturalistic Neuroscience and Drug Development

The combination enables novel research approaches with significant implications for drug development:

• Therapeutic Target Identification: Using fMRI to precisely localize pathological brain activity patterns
• Treatment Response Monitoring: Employing fNIRS for repeated measures of target engagement in clinical trials
• Cognitive Drug Effects: Assessing pharmaceutical impacts on brain function in more ecologically valid settings
• Biomarker Development: Establishing fNIRS-based biomarkers validated against fMRI gold standard

The integration of fMRI and fNIRS represents a powerful multimodal approach that successfully addresses fundamental limitations of each individual technique. Through systematic validation studies, researchers have established strong spatial correspondence between these modalities, particularly in superficial cortical regions, with fNIRS demonstrating up to 68% overlap with fMRI activation maps [77]. The complementary nature of fMRI's high spatial resolution and fNIRS's temporal precision and portability enables enhanced spatiotemporal mapping of brain function across diverse populations and settings [34]. As methodological standards continue to evolve and hardware innovations address current limitations, combined fMRI-fNIRS approaches are poised to significantly advance both basic neuroscience and clinical applications, from neurorehabilitation to drug development. This integration represents a pragmatic solution to the enduring challenge of capturing the brain's dynamic activity with both spatial and temporal precision, moving the field closer to comprehensive measurements of brain function in health and disease.

Addressing Technical Challenges and Optimizing Data Quality

Non-invasive neuroimaging techniques, such as functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS), have become cornerstones of modern brain research. Their value in elucidating neural mechanisms and informing drug development is immense. However, the data acquired from these modalities are invariably contaminated by various artifacts, which can obscure the true neural signals and lead to erroneous interpretations. A comprehensive understanding of these artifacts—categorized broadly as motion, physiological noise, and environmental interference—is therefore a fundamental prerequisite for robust scientific discovery. This guide provides an in-depth technical overview of these common artifacts, detailing their origins, characteristics, and, crucially, evidence-based methodologies for their mitigation, providing researchers and scientists with the tools to ensure data integrity.

Artifacts in Functional Magnetic Resonance Imaging (fMRI)

fMRI measures brain activity indirectly through the Blood Oxygen Level Dependent (BOLD) signal, which is sensitive to changes in blood oxygenation. This measurement is highly susceptible to noise. Motion artifacts are a predominant challenge, as even minute head movements (on the order of millimeters) can cause signal changes that mimic or mask true BOLD activation [11]. These artifacts are particularly problematic because they are often correlated with the task paradigm. Physiological noise arises from cardiac pulsation (∼1 Hz) and respiration (∼0.2-0.3 Hz), which cause rhythmic movement of the brain and changes in blood flow and oxygenation. These processes introduce temporal autocorrelation and spatial structure into the noise. Furthermore, scanner drift, a low-frequency noise component, can arise from hardware instability over the course of a long scan [11].

Experimental Protocols and Mitigation Strategies

A multi-pronged approach is essential for mitigating fMRI artifacts. The following protocols are considered standard practice:

• Prospective Motion Correction: Real-time tracking of head position using external cameras can be used to adjust the imaging sequence dynamically, minimizing the impact of motion during data acquisition.
• Retrospective Image Registration: As a foundational post-processing step, individual volume images are realigned to a reference volume (e.g., the first or the mean) using rigid-body transformation. This corrects for head motion between scans.
• Physiological Noise Modeling: Recording cardiac and respiratory cycles during the fMRI scan using pulse oximeters and breathing belts is crucial. These recordings are used to create Regressors of No Interest (e.g., using RETROICOR or similar techniques) that are included in the general linear model (GLM) to remove the variance associated with these physiological processes.
• Advanced Acquisition Sequences: Sequences like multiband imaging can reduce motion sensitivity by acquiring multiple slices simultaneously, shortening the repetition time (TR).

The following diagram illustrates a typical fMRI data processing workflow incorporating these key artifact correction steps.

Artifacts in Electroencephalography (EEG)

EEG records the brain's electrical activity at the scalp with millisecond temporal resolution, making it highly susceptible to both physiological and non-physiological artifacts. Motion artifacts result from changes in electrode-skin contact due to head or body movement, often manifesting as high-amplitude, low-frequency shifts or spikes in the signal. Cable movement can generate triboelectric noise due to friction within the cable components [82].

Physiological artifacts are generated by the subject's own body and include:

• Ocular Artifacts: High-amplitude signals from eye blinks and eye movements.
• Cardiac Artifacts: The electrical signal from the heart (electrocardiogram, ECG) that can be picked up by EEG electrodes.
• Muscle Artifacts: High-frequency noise from the tension of cranial and neck muscles (electromyogram, EMG).

Environmental interference is dominated by mains interference (50/60 Hz noise from power lines), which can be exacerbated by poor electrode impedance or improper grounding [82]. An electrode pop, caused by a sudden change in impedance at a single electrode (e.g., from a loose lead), appears as a sudden, large voltage shift [82].

Table 1: Common Non-Physiological EEG Artifacts and Mitigation Strategies

Artifact Type	Characteristics in Signal	Prevention and Correction Methods
Mains Interference	Monotonous waves at 50/60 Hz [82].	Use of active shielding in cables [82]; notch filtering in post-processing [82].
Electrode Pop	Sudden, large shift in DC offset or signal out of range [82].	Secure electrode attachment; check impedances before/during recording [82].
Cable Movement	Sudden, high-amplitude changes [82].	Use low-noise cables; secure cables to reduce sway; active shielding [82].

Experimental Protocols for EEG Artifacts

A rigorous experimental setup is the first line of defense against EEG artifacts.

• Preparation Protocol: Abrade the skin to reduce impedance at each electrode site to below 5-10 kΩ. Use conductive paste and secure electrodes with caps or tape to prevent movement. Ensure the reference electrode has a particularly stable and low-impedance connection.
• Hardware Selection: Utilize amplifiers with active shielding to minimize mains interference and cable movement artifacts [82]. Employ systems that allow for continuous impedance monitoring during the experiment.
• Post-Processing Algorithms: Independent Component Analysis (ICA) is a widely used and effective method for isolating and removing stereotypical artifacts like blinks, eye movements, and cardiac signals. Other methods include regression-based techniques and adaptive filtering.

Artifacts in Functional Near-Infrared Spectroscopy (fNIRS)

fNIRS measures hemodynamic changes by detecting near-infrared light passed through the scalp and skull. Its signals are a mixture of cortical brain activity and various noise sources. Motion artifacts are a significant challenge, particularly in mobile or clinical populations. They occur when movement causes a decoupling between the optodes and the scalp, resulting in signal spikes or baseline shifts [83]. These artifacts can be high-frequency spikes or low-frequency shifts that are temporally correlated with the task, making them particularly difficult to separate from the true hemodynamic response [83].

Physiological noises are a major confound in fNIRS as they are also of hemodynamic origin. These include:

• Cardiac oscillations (∼1 Hz)
• Respiratory waves (∼0.2-0.3 Hz)
• Mayer waves (very-low-frequency oscillations ∼0.1 Hz) associated with blood pressure regulation [84].

Environmental interference is less prominent than in EEG but can include stray light entering the system, which can be prevented by ensuring optodes are properly attached and shielded.

Quantitative Comparison of fNIRS Motion Artifact Correction Techniques

Several algorithms have been developed to correct for motion artifacts in fNIRS. Their performance can be quantified using metrics like the change in Signal-to-Noise Ratio (ΔSNR) and the percentage reduction in motion artifacts (η). Recent research demonstrates the efficacy of advanced methods.

Table 2: Performance of fNIRS Motion Artifact Correction Techniques

Correction Method	Description	Reported Performance (Average)	Key Findings
Wavelet Packet Decomposition (WPD) [85]	Single-stage decomposition using wavelet packets (e.g., db1, db2).	ΔSNR: 16.11 dB; η: 26.40% [85]	A robust standalone method for motion artifact reduction.
WPD with Canonical Correlation Analysis (WPD-CCA) [85]	Two-stage method: WPD followed by CCA for enhanced denoising.	ΔSNR: 16.55 dB; η: 41.40% [85]	Superior to single-stage WPD, providing significantly greater artifact reduction (η increased by 56.82%) [85].
Adaptive Filtering with Short-Separation Channels [84]	Uses a separate, short-distance channel (<1 cm) to measure and subtract superficial noise.	N/A	Effectively reduces physiological and superficial noises; suitable for both offline and online processing [84].

Experimental Protocols for fNIRS Artifacts

The following methodologies are critical for obtaining clean fNIRS data.

• Short-Separation Channel Protocol: Incorporate optode pairs with a small source-detector distance (e.g., 0.8 cm) into the probe design. These channels are predominantly sensitive to systemic physiological noise in the scalp. This signal can be used as a nuisance regressor in a General Linear Model (GLM) or in an adaptive filter to clean the signals from standard long-separation channels (∼3 cm) [84].
• Motion Artifact Correction Pipeline: Implement a processing pipeline that includes:
  • Identification: Detect motion artifacts automatically based on signal slope, amplitude, or standard deviation thresholds.
  • Correction: Apply a robust correction algorithm such as WPD-CCA [85] or a moving average spline interpolation method [83]. Evidence from real functional data suggests wavelet-based methods are highly effective [83].
  • Band-Pass Filtering: Apply a band-pass filter (e.g., 0.01 - 0.2 Hz) to remove slow drifts and high-frequency cardiac noise.
• Physiological Noise Modeling: For high-quality data, model the physiological noises explicitly as a sum of sinusoidal functions (for cardiac, respiratory, and Mayer waves) within a recursive least-squares (RLS) or Kalman filtering framework [84].

The logical relationship between the sources of fNIRS artifacts and the corresponding correction pathways is synthesized in the diagram below.

The Scientist's Toolkit: Essential Research Reagents and Materials

The following table details key materials and solutions used in fNIRS and EEG research to prevent and manage artifacts.

Table 3: Key Research Reagents and Materials for Artifact Management

Item	Function in Research	Application Context
Active Shielding Cables	Minimizes capacitive coupling from power lines and reduces triboelectric noise from cable movement [82].	EEG, fNIRS
Low-Impedance Electrode Paste/Gel	Ensures stable electrical connection between EEG electrode and skin, minimizing electrode pop and baseline drift [82].	EEG
Short-Separation fNIRS Optodes	Optodes placed 0.5-1.0 cm apart to selectively measure hemodynamic changes in the scalp, used as a regressor to remove superficial noise from standard channels [84].	fNIRS
Conductive EEG Caps	Holds electrodes in stable positions according to international systems (10-20, 10-10), ensuring consistent geometry and reducing motion artifacts.	EEG
Physiological Monitoring Kit	Includes pulse oximeter (for cardiac cycle) and respiratory belt. Provides essential data for modeling and removing physiological noise.	fMRI, fNIRS
Recursive Least-Squares (RLS) Estimator Algorithm	An adaptive filtering method used to dynamically estimate and remove physiological and superficial noises from fNIRS data in offline or online modes [84].	fNIRS (Data Analysis)

Functional near-infrared spectroscopy (fNIRS) has emerged as a prominent neuroimaging tool that noninvasively measures brain activity by detecting hemodynamic responses associated with neural activation. Unlike fMRI, which requires bulky, expensive equipment and restricts natural movement, fNIRS offers a portable, cost-effective alternative that enables brain monitoring in real-world environments and diverse populations [86] [87]. Similarly, while EEG provides excellent temporal resolution, it suffers from limited spatial resolution and difficulty in localizing neural activity sources [53]. Despite these advantages, fNIRS faces a fundamental challenge: optimizing spatial specificity to accurately pinpoint and reliably measure activity from specific brain regions, particularly across repeated sessions [86] [46].

This technical guide addresses how high-density (HD) array configurations and advanced source localization techniques can significantly enhance fNIRS spatial specificity. Spatial specificity refers to the technique's ability to accurately localize hemodynamic changes to their precise cortical origins, which is crucial for distinguishing adjacent functional areas and tracking changes over time in both basic research and clinical applications such as drug development [86] [88]. Traditional sparse fNIRS arrays with 30 mm channel spacing suffer from limited spatial resolution, poor depth sensitivity, reduced sensitivity to focal activations, and inconsistent measurement reliability across sessions [88]. These limitations can be substantially mitigated through optimized HD array designs and computational approaches that leverage anatomical information.

High-Density fNIRS Arrays: Principles and Quantitative Advantages

Fundamental Principles of HD-fNIRS

High-density fNIRS arrays are characterized by smaller inter-optode spacing (typically ≤15 mm) and multiple, overlapping source-detector distances, creating a rich sampling network across the scalp surface. This configuration enables tomographic image reconstruction through Diffuse Optical Tomography (DOT), which significantly improves spatial resolution and depth discrimination compared to traditional sparse arrays that typically feature 30 mm spacing with non-overlapping channels [88] [89]. The core principle involves using multiple source-detector distances with varying sensitivity profiles to different tissue depths, allowing mathematical inversion techniques to reconstruct three-dimensional images of cortical activation [89].

In HD-DOT, each measurement samples a mixture of responses from various depths, with larger source-detector distances exhibiting greater sensitivity to deeper tissues. The dense grid of optodes provides a comprehensive range of such distances, enabling tomographic reconstruction that effectively "unmixes" these signals to localize activity within the brain [89]. This approach stands in stark contrast to traditional sparse fNIRS, which assumes that each channel primarily reflects activity from a specific region directly beneath the optodes—an oversimplification that leads to significant spatial inaccuracy and signal contamination from superficial tissues.

Quantitative Performance Comparison

Table 1: Performance Comparison Between Sparse and High-Density fNIRS Arrays

Performance Metric	Sparse Arrays (30mm spacing)	High-Density Arrays (13mm spacing)	Ultra-High-Density Arrays (6.5mm spacing)
Spatial Resolution	20-30 mm	13-16 mm	5-7 mm (simulated)
Localization Error	High	Moderate	30-50% reduction vs. HD [89]
Signal-to-Noise Ratio	Standard	Improved	1.4-2.0x improvement vs. HD [89]
Depth Sensitivity	Limited	Good	Excellent
Inter-subject Consistency	Low to moderate	Improved	Highest
Typical Source-Detector Distances	Single distance (~30 mm)	Multiple distances (13-40 mm)	Multiple distances (6.5-40 mm)

Table 2: Impact of Array Density on Experimental Outcomes in Visual and Motor Tasks

Experimental Factor	Sparse Array Performance	High-Density Array Performance	Evidence Source
Detection of Incongruent Stroop Task	Suitable for detection	Suitable for detection and localization	[88]
Detection of Congruent Stroop Task	Poor detection	Robust detection and localization	[88]
Reproducibility Across Sessions	Lower reproducibility due to sensitivity to optode placement shifts	Higher reproducibility despite placement variations	[46]
Image Space Localization	Poor localization quality	Superior localization and sensitivity	[88]
Oxyhemoglobin (HbO) Detection	More reproducible than deoxyhemoglobin (HbR)	Improved detection for both HbO and HbR	[46]

The transition from sparse to high-density arrays represents a substantial advancement in fNIRS technology. Research demonstrates that HD arrays outperform sparse layouts particularly in detecting and localizing brain activity during lower cognitive load tasks, which often present challenges for traditional fNIRS configurations [88]. Furthermore, HD configurations with multiple source-detector distances enable more effective separation of cerebral signals from superficial artifacts through superficial signal regression techniques, significantly enhancing the brain specificity of the measurements [88].

Source Localization and Anatomical Coregistration

Source Localization Methodologies

Source localization in fNIRS refers to computational approaches that estimate the precise anatomical origins of measured hemodynamic signals. This process typically involves combining HD-fNIRS data with anatomical information to constrain and refine the interpretation of cortical activation patterns. Unlike EEG source localization, which must solve an ill-posed inverse problem, fNIRS source localization benefits from more favorable mathematical properties but requires accurate modeling of light propagation through heterogeneous tissues [46] [19].

The fundamental approach involves constructing a forward model that describes how light propagates through head tissues, typically using the diffusion approximation of the radiative transfer equation. This model, often represented as a Jacobian matrix (sensitivity matrix), quantifies how changes in absorption within each brain voxel affect the measured signals at each source-detector pair [89]. The inverse solution then estimates the spatial distribution of hemodynamic changes that best explain the measured data, often using regularized reconstruction algorithms to mitigate noise amplification [89].

Recent research has demonstrated that source localization significantly improves the reliability of fNIRS for capturing brain activity compared to traditional channel-based analyses. One study investigating reproducibility across multiple sessions found that "source localization improves the reliability of fNIRS to capture brain activity" [46]. This enhancement occurs because source localization projects measurements onto anatomical space, reducing the confounding effects of inter-session optode placement variations that typically plague traditional channel-based fNIRS analyses.

Anatomical Coregistration Protocols

Accurate source localization requires precise coregistration of fNIRS optodes with anatomical references. The following protocol outlines the standard methodology for achieving this crucial registration:

• Optode Position Digitization: Using 3D digitization systems (e.g., electromagnetic or optical trackers), record the precise spatial coordinates of each fNIRS source and detector relative to anatomical landmarks (nasion, inion, preauricular points) and reference points on the scalp [46] [19].

• Head Surface Mapping: Capture additional points across the scalp surface to create a dense representation of head morphology, improving registration accuracy.

• Atlas Registration: Coregister the digitized optode positions with standardized head templates (e.g., ICBM152) or individual MRI scans when available. This typically involves affine or nonlinear transformations to align the digitized head surface with the template surface [19].

• Channel Projection: Project the measurement channels onto the cortical surface through the sensitivity profile of each source-detector pair, typically using pre-calculated sensitivity distributions based on Monte Carlo simulations or finite-element models of light transport [19] [89].

• Volume Registration: For group-level analyses, normalize individual brains to a standard atlas space (e.g., MNI space) to enable cross-participant comparisons and group statistical inferences [19].

This anatomical coregistration process is fundamental for enhancing spatial specificity, as it enables researchers to precisely relate measured hemodynamic changes to specific cortical regions, significantly improving the interpretability and reproducibility of fNIRS findings [46].

Implementation Protocols and Methodologies

HD-fNIRS Experimental Implementation Protocol

Implementing high-density fNIRS requires careful attention to both technical and practical considerations. The following protocol outlines key steps for successful HD-fNIRS experimentation:

Probe Design and Manufacturing:

• Customized Helmets: Utilize 3D printing or thermoplastic materials to create customized helmets that ensure consistent optode placement and adequate scalp coupling across participants [53]. These approaches help maintain consistent source-detector distances and pressure, improving data quality.
• Optode Arrangement: Configure sources and detectors in overlapping, multidistance patterns with typical nearest-neighbor distances of 10-15 mm for HD arrays, or as low as 6.5 mm for ultra-high-density systems [88] [89].
• Source-Detector Distances: Include multiple distance ranges, typically from 10-45 mm, to sample different tissue depths. Short-separation channels (<15 mm) are essential for measuring and regressing out superficial physiological artifacts [88].

Signal Quality Optimization:

• Scalp Coupling Assessment: Implement real-time quality metrics such as the scalp-coupling index (SCI) to identify poor-quality channels during setup [19]. Exclude channels with SCI <0.7 from analysis.
• Motion Artifact Mitigation: Utilize accelerometer-based motion detection, signal quality indices, and algorithmic corrections (e.g., PCA-based, wavelet-based) to identify and compensate for movement artifacts [86] [19].
• Physiological Noise Reduction: Apply superficial signal regression using short-separation channels to remove systemic physiological contaminants from cardiac, respiratory, and blood pressure oscillations [19].

Data Acquisition Parameters:

• Sampling Rate: Acquire data at sufficient sampling rates (typically ≥10 Hz) to adequately capture hemodynamic responses and physiological noise components [19].
• Light Intensity Management: Optimize source power and detector gain to maintain signals within dynamic range while maximizing signal-to-noise ratio, particularly critical for the smallest source-detector distances in UHD arrays [89].

Image Reconstruction and Analysis Workflow

Table 3: HD-fNIRS Data Processing Pipeline

Processing Stage	Key Operations	Common Algorithms/Tools
Preprocessing	Signal quality assessment, motion artifact correction, physiological noise removal, bandpass filtering (0.01-0.5 Hz)	SCI, GVTD, PCA/ICA, wavelet filtering, superficial signal regression [19]
Forward Modeling	Light transport modeling, sensitivity profile calculation, head model creation	Diffusion approximation, finite element method, Monte Carlo simulations [89]
Image Reconstruction	Solving inverse problem to reconstruct 3D hemodynamic images	Regularized linear inversion, Tikhonov regularization, model-based reconstruction [89]
Statistical Analysis	General linear modeling, group-level inference, functional connectivity	GLM, random effects analysis, correlation/coherence analysis [88] [19]

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 4: Essential Research Materials for HD-fNIRS Implementation

Category	Item	Specification/Function	Application Notes
Optoelectronics	Laser Diodes or LEDs	Wavelengths: 760 nm & 850 nm (±10 nm)	Targets differential absorption of HbO and HbR [90] [91]
	Detectors (SPADs or APDs)	High sensitivity, 140 dB dynamic range	Critical for detecting weak signals in UHD arrays [89]
Probe Design	3D-Printed Helmets	Custom-fit to individual head morphology	Ensures consistent optode placement and coupling [53]
	Thermoplastic Sheets	Moldable at ~60°C, rigid when cool	Alternative for custom helmet creation [53]
	Optode Holders	Flexible yet stable attachment	Maintains consistent pressure and positioning
Data Acquisition	Digitization System	Electromagnetic or optical tracking	Records 3D optode positions for coregistration [46] [19]
	DAQ Hardware	High dynamic range (>120 dB), sampling rate ≥10 Hz	Preserves signal fidelity across intensity ranges [89]
Computational Tools	Head Models	Multi-layered (scalp, skull, CSF, GM, WM)	Realistic light propagation modeling [89]
	Atlas Templates	ICBM152, MNI, or Desikan-Killiany atlas	Standardized spatial reference for group analysis [19]
Software	Reconstruction Algorithms	Regularized inverse solutions (Tikhonov)	Solves ill-posed inverse problem [89]
	Analysis Platforms	MNE, Brainstorm, Homer2, NIRS-KIT	Open-source tools for fNIRS processing [19]

The integration of high-density array designs and sophisticated source localization techniques represents a transformative advancement in fNIRS technology, substantially bridging the spatial resolution gap with fMRI while maintaining fNIRS's inherent advantages of portability, cost-effectiveness, and tolerance of movement. These methodological refinements enable more precise mapping of brain function, enhanced reproducibility across longitudinal studies, and improved discrimination of adjacent functional areas—capabilities particularly valuable for drug development research requiring sensitive detection of neuromodulatory effects.

While implementation challenges remain, including increased computational demands and more complex experimental setups, the demonstrated improvements in spatial specificity, signal quality, and measurement reliability firmly establish HD-fNIRS as a powerful neuroimaging modality. As these technologies continue to evolve, they promise to further expand the applications of fNIRS in both basic neuroscience and clinical research contexts, potentially enabling increasingly sophisticated investigations of brain function in naturalistic environments and diverse participant populations.

Functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) have revolutionized our ability to non-invasively study human brain function. Each technique offers unique advantages: fMRI provides high spatial resolution, EEG delivers excellent temporal resolution, and fNIRS offers a portable compromise with relative motion tolerance [11] [41]. However, the signals measured by each of these modalities are profoundly susceptible to various artifacts that can obscure genuine neural activity and compromise data interpretation. The pursuit of improved signal quality through advanced preprocessing and real-time artifact removal therefore represents a cornerstone of reliable brain research.

Artifacts introduce noise that can significantly reduce the signal-to-noise ratio (SNR) of neuroimaging data [92]. In EEG, where signals are typically measured in microvolts, contamination can be particularly problematic [93]. For fNIRS, motion artifacts (MAs) have been identified as a major challenge that can deteriorate measurement quality [92]. Similarly, fMRI is sensitive to motion artifacts and physiological noise, which can be exacerbated by the scanner environment [11] [41]. The impact of these artifacts extends beyond basic research—in clinical applications and drug development, compromised signal quality can lead to misinterpretation of brain function and potentially affect diagnostic or therapeutic decisions. A recent large-scale reproducibility initiative (the fNIRS Reproducibility Study Hub) highlighted that how researchers handle poor-quality data significantly affects their results, underscoring the critical importance of standardized preprocessing approaches [33].

Characterizing Artifacts Across Neuroimaging Modalities

Physiological Artifacts

Physiological artifacts originate from the subject's own bodily functions and represent a pervasive challenge across all three neuroimaging modalities.

Table 1: Physiological Artifacts in Neuroimaging

Artifact Type	Origin	Effect on Signals	Primary Modalities Affected
Ocular Artifacts	Eye blinks and movements	Sharp, high-amplitude deflections; low-frequency contamination	EEG (frontal), fNIRS (prefrontal)
Muscle Artifacts	Muscle contractions (jaw, neck, face)	High-frequency noise; broadband EMG interference	EEG, fNIRS
Cardiac Artifacts	Heartbeat and pulsatile blood flow	Rhythmic waveforms at heart rate; pulse artifacts	EEG, fNIRS, fMRI
Respiration	Chest movements and breathing cycles	Slow baseline drifts; potential motion effects	EEG, fNIRS, fMRI
Perspiration	Sweat gland activity	Slow potential shifts; impedance changes	EEG, fNIRS

In EEG recordings, ocular artifacts generate significant contamination, particularly over frontal electrodes. The eye behaves as an electric dipole due to the charge difference between the cornea and retina, and when this dipole moves during blinks or saccades, it creates a field disturbance measurable on the scalp. These EOG artifacts typically reach 100-200 µV, often an order of magnitude larger than EEG signals [93]. Muscle artifacts (EMG) present another major challenge, generating broadband noise that overlaps with and can obscure important EEG rhythms in the beta (13-30 Hz) and gamma (>30 Hz) ranges [94] [93].

For fNIRS, systemic physiological processes represent a major confound. Cardiac activity introduces pulsatile artifacts, while respiration can cause low-frequency oscillations in the signals. These physiological noises are particularly challenging because they overlap with the hemodynamic responses that fNIRS aims to measure [95]. Recent approaches such as Systemic Physiology Augmented fNIRS (SPA-fNIRS) have emphasized the importance of concurrently measuring systemic physiological parameters to better distinguish cerebral from non-cerebral signals [95].

Non-Physiological Artifacts

Non-physiological artifacts stem from external sources, equipment issues, or subject motion unrelated to physiological processes.

Table 2: Non-Physiological Artifacts in Neuroimaging

Artifact Type	Origin	Effect on Signals	Primary Modalities Affected
Motion Artifacts	Head/body movement; cable displacement	Sudden signal shifts; baseline instability	fNIRS, EEG, fMRI
Electrode/Sensor Issues	Poor contact; impedance changes	Transient spikes; signal drift	EEG, fNIRS
Instrument Noise	Power line interference; amplifier noise	50/60 Hz line noise; high-frequency components	EEG, fNIRS
Environment Artifacts	Electromagnetic interference; scanner artifacts	Signal distortion; additional noise sources	All modalities
Scanner Artifacts	Magnetic field inhomogeneities; gradient noise	Image distortion; signal loss	fMRI

Motion artifacts present a particularly significant challenge for fNIRS and EEG. In fNIRS, motion artifacts arise from imperfect contact between optodes and the scalp, including displacement, non-orthogonal contact, and oscillation of the optodes [92]. These artifacts can be caused by various movements including head nodding, shaking, tilting, facial muscle movement, and even jaw movements during talking, eating, or drinking [92]. The direct impact is a significant deterioration in the signal-to-noise ratio, which can compromise data quality and interpretation [92].

In EEG, cable movement and electrode "pops" caused by sudden impedance changes create transient spikes and signal distortions that can mimic pathological activity or normal brain rhythms [93]. fMRI is especially susceptible to motion artifacts from head movement, which can cause image misregistration, spin history effects, and signal dropouts [11] [41]. The fMRI environment also introduces unique artifacts including gradient noise from switching magnetic fields and physiological noise from cardiac and respiratory cycles that interact with the magnetic field [41].

Figure 1: Classification of common artifacts in neuroimaging signals

Advanced Preprocessing Pipelines and Methodologies

EEG Artifact Removal Techniques

EEG artifact removal has evolved from simple regression-based approaches to sophisticated blind source separation methods. The traditional regression method operates under the assumption that each EEG channel represents the cumulative sum of pure EEG data and a proportion of artifact [94]. This approach defines amplitude relationships between reference channels (like EOG) and EEG channels through transmission factors, then subtracts the estimated artifacts from the EEG signal according to the equation:

EEGcor = EEGraw − γF(HEOG) − δF(VEOG)

where γ and δ depend on the transmission coefficient between EOG and EEG channels, and HEOG and VEOG represent recordings from horizontal and vertical EOG channels respectively [94]. However, this method is affected by bidirectional interference, where ocular potentials contaminate EEG data while EEG data simultaneously contaminates ocular recordings [94].

Blind Source Separation (BSS) approaches, particularly Independent Component Analysis (ICA), have emerged as the most commonly used algorithms for EEG artifact removal [94]. ICA operates by decomposing the multichannel EEG signal into statistically independent components, many of which often represent artifacts from ocular, muscular, or cardiac sources. These artifactual components can be visually identified or automatically detected based on their spatial, temporal, and spectral characteristics before reconstructing the signal without them [94] [93]. Modern implementations often combine ICA with deep learning approaches such as CNN-LSTM models to detect and isolate artifacts at scale [93].

fNIRS Motion Artifact Correction

fNIRS research has developed specialized approaches to handle motion artifacts, which represent one of the most significant challenges for this modality. These approaches can be broadly categorized into hardware-based and algorithmic solutions.

Hardware-based solutions often incorporate additional sensors to detect motion. Accelerometer-based methods include Adaptive Filtering, Active Noise Cancellation (ANC), Accelerometer-Based Motion Artifact Removal (ABAMAR), and Blind Source Separation, Accelerometer-based Artifact Rejection and Detection (BLISSA2RD) [92]. These methods use motion information from accelerometers to identify and remove motion-contaminated segments or to adaptively filter artifacts. The introduction of accelerometers improves the feasibility of real-time rejection of motion artifacts [92].

Algorithmic solutions include:

• Moving Average Filters: Simple temporal filtering approaches that can reduce high-frequency motion components but may also remove neural signals of interest [92].
• Wavelet-Based Methods: These decompose the fNIRS signal into different frequency components, allowing identification and removal of motion-related components in the wavelet domain before reconstructing the signal [94].
• Correlation-Based Signal Improvement (CBSI): This method exploits the negative correlation between oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentrations during genuine neural activity, unlike motion artifacts which typically affect both similarly [92].
• Targeted Principal Component Analysis (tPCA): Identifies and removes motion artifacts by detecting outliers in the principal component space [92].

Recent comparative studies have shown that the performance of different motion correction methods can vary depending on the type and intensity of motion artifacts, with no single method universally superior across all scenarios [92].

fMRI Noise Reduction Strategies

fMRI preprocessing pipelines incorporate multiple stages to address various noise sources. Key steps include:

• Slice-timing correction: Compensates for differences in acquisition time between different slices of the brain.
• Realignment: Corrects for head motion by aligning all volumes to a reference image.
• Spatial normalization: Transforms individual brains into a standard space to enable group-level analysis.
• Spatial smoothing: Improves signal-to-noise ratio by averaging signals across neighboring voxels.
• Temporal filtering: Removes low-frequency drifts and high-frequency noise outside the expected hemodynamic response range.

Additionally, advanced techniques such as RETROICOR (Retrospective Image Correction) address physiological noise by modeling and removing cardiac and respiratory-related fluctuations using concurrently recorded physiological data [41]. CompCor (Component-Based Noise Correction) identifies noise components from high-variance areas such as white matter and cerebrospinal fluid and removes these from the signal [41].

Figure 2: Comparative preprocessing pipelines for EEG, fNIRS, and fMRI

Real-Time Artifact Removal Implementations

Requirements for Real-Time Processing

Real-time artifact removal imposes specific constraints that differ from offline processing. The primary requirements include:

• Computational efficiency: Algorithms must process data within strict time constraints to enable immediate feedback.
• Causal implementation: Methods can only use current and past data points, not future information.
• Robustness: Algorithms must handle variable data quality without extensive parameter tuning.
• Minimal signal distortion: Preservation of neural signals is crucial while removing artifacts.

For brain-computer interfaces (BCIs) and neurofeedback applications, these requirements become particularly stringent, as delays in processing can significantly impact system performance and user experience [92].

Real-Time Approaches by Modality

EEG: Real-time artifact removal often employs adaptive filtering techniques that continuously update filter parameters based on incoming data. Reference-based methods use dedicated EOG, ECG, or EMG channels to adaptively subtract artifacts from EEG signals [94] [93]. For systems without reference channels, real-time ICA implementations have been developed, though these require sufficient computational resources to perform source separation quickly [93].

fNIRS: Real-time motion artifact correction has been advanced through accelerometer-based methods such as ABAMAR and ABMARA, which use motion information from accelerometers to identify and correct artifacts as they occur [92]. These methods can be implemented with minimal computational overhead, making them suitable for real-time applications. Simple moving average filters and correlation-based approaches also lend themselves well to real-time implementation due to their computational efficiency [92].

fMRI: Real-time processing has gained traction for neurofeedback applications, where subjects learn to self-regulate brain activity based on immediate feedback. Realignment and slice-timing correction are commonly implemented in real-time, while more computationally intensive processes like ICA are typically reserved for offline analysis due to their higher computational demands [41].

Table 3: Performance Comparison of Artifact Removal Techniques

Method	Computation Load	Suitable for Real-Time	Key Advantages	Key Limitations
Regression	Low	Yes	Simple implementation; minimal computation	Requires reference channels; bidirectional contamination
ICA	High	Limited	Comprehensive artifact removal; no reference needed	Computationally intensive; requires multichannel data
Adaptive Filtering	Medium	Yes	Continuously adapts to changing signals	Requires reference channels
Wavelet Denoising	Medium	Yes	Effective for transient artifacts	Parameter selection critical; may remove neural signals
tPCA	Medium	Limited	Effective for motion artifacts	May remove neural signals; requires parameter tuning
Accelerometer-Based	Low	Yes	Direct motion measurement; computationally efficient	Requires additional hardware

Experimental Protocols for Artifact Removal Validation

Protocol for Evaluating fNIRS Motion Artifact Correction

A comprehensive protocol for evaluating motion artifact correction methods should include controlled introduction of artifacts and quantitative performance metrics [92]:

• Participant Preparation: Secure fNIRS optodes according to standard positioning systems (e.g., international 10-20 system). For hardware-based methods, ensure proper placement of accelerometers or other motion sensors.

• Data Acquisition:

  • Record resting-state data for baseline signal characteristics.
  • Instruct participants to perform standardized movements (head nodding, shaking, tilting) at predetermined intervals.
  • Include tasks that induce subtle motion artifacts (e.g., verbal fluency tasks, jaw movements).

• Validation Metrics:

  • Noise Suppression: Evaluate reduction in motion artifact power.
  • Signal Distortion: Assess preservation of hemodynamic responses during task conditions.
  • Temporal Characteristics: Measure processing delays for real-time applications.

This protocol enables direct comparison of different correction methods and facilitates optimization of parameters for specific experimental conditions [92].

Protocol for Multimodal Integration Validation

With the growing interest in combining multiple neuroimaging modalities, validation protocols must assess how artifact removal in one modality affects integrated data:

• Synchronous Data Acquisition: Implement simultaneous recording of fMRI and EEG, or fNIRS and EEG, with careful timing synchronization [96].

• Cross-Modality Artifact Assessment:

  • Evaluate how EEG artifact removal affects fMRI-EEG integration.
  • Assess how fNIRS motion correction impacts concurrent EEG analysis.
  • For fMRI-EEG, specifically address ballistocardiographic artifacts caused by cardiac pulsatility in the magnetic field [41].

• Performance Benchmarks:

  • Quantify integration accuracy using standardized tasks with known neural correlates.
  • Compare results against ground-truth measurements where possible (e.g., intracranial recordings in clinical populations).

Advanced algorithms for multimodal integration have demonstrated average accuracy of 90.2% (±5.0%) in recent systematic assessments, though significant methodological heterogeneity remains a challenge [96].

Table 4: Research Reagent Solutions for Artifact Management

Tool/Category	Specific Examples	Function/Purpose	Compatible Modalities
Reference Sensors	EOG, ECG, EMG electrodes	Provide reference signals for regression-based artifact removal	EEG, fNIRS
Motion Tracking	Accelerometers, gyroscopes, IMUs	Detect motion for hardware-based artifact correction	fNIRS, EEG
Analysis Toolboxes	EEGLAB, NIRS Brain AnalyzIR, SPM, FSL	Provide implemented algorithms for artifact removal	EEG, fNIRS, fMRI
Quality Metrics	SNR calculations, Pearson correlation, RMSE	Quantify effectiveness of artifact removal	All modalities
Physiological Monitors	Pulse oximeters, respiratory belts, GSR	Monitor systemic physiology for augmented analysis	fNIRS, fMRI
Hardware Solutions	MRI-compatible EEG systems, fiber-optic fNIRS	Enable simultaneous multimodal acquisition	EEG-fMRI, fNIRS-fMRI

Advanced preprocessing and real-time artifact removal are essential components of modern brain research using fMRI, EEG, and fNIRS. While each modality faces unique challenges, common principles emerge across techniques, including the value of multimodal reference signals, the importance of balancing noise removal with signal preservation, and the need for standardized validation approaches.

Future developments will likely focus on several key areas. First, machine learning and deep learning approaches show promise for automated, adaptive artifact removal that can learn from specific data characteristics [96]. Second, the field is moving toward better standardization of preprocessing pipelines to enhance reproducibility, as evidenced by initiatives like the fNIRS Reproducibility Study Hub [33]. Third, hardware improvements continue to reduce susceptibility to artifacts at the acquisition stage.

For researchers and drug development professionals, implementing robust artifact management strategies is not merely a technical consideration but a fundamental requirement for generating reliable, interpretable results. As the field advances, the integration of sophisticated artifact removal with multimodal data fusion promises to unlock new insights into brain function in increasingly naturalistic settings.

In the evolving landscape of brain research, the quest for reproducible findings represents a fundamental pillar of scientific integrity. As neuroimaging techniques like functional Near-Infrared Spectroscopy (fNIRS) gain prominence for their portability and applicability in naturalistic settings, understanding the factors that affect reproducibility becomes paramount [33] [97]. The complexity of modern analysis pipelines, while offering powerful analytical flexibility, introduces substantial variability in how researchers process, interpret, and derive conclusions from the same neural data [33]. This technical guide examines the impact of analysis pipelines and researcher expertise on reproducibility, framing this discussion within the broader context of multimodal brain research involving fNIRS, fMRI, and EEG.

The reproducibility crisis has touched numerous scientific fields, and functional neuroimaging is no exception [33]. Initiatives like the fNIRS Reproducibility Study Hub (FRESH) have emerged to systematically quantify how analytical choices and researcher experience affect outcomes [33] [98]. Their findings reveal that while nearly 80% of research teams agree on group-level findings for strongly supported hypotheses, individual-level analyses show considerably lower agreement [33] [97]. This variability stems not from technical limitations alone but from the intricate interplay between methodological choices and researcher expertise—a relationship that this guide will explore in depth.

The FRESH Initiative: A Community-Wide Reproducibility Assessment

Experimental Design and Methodology

The FRESH initiative represents one of the most comprehensive efforts to date to quantify variability in fNIRS analysis. In this landmark study, 38 research teams worldwide independently analyzed the same two fNIRS datasets using their preferred methodologies [33] [98]. The experimental design was structured to mirror real-world research conditions while maintaining scientific rigor:

• Dataset Provision: Participating teams received identical fNIRS datasets with standardized descriptions of the experimental design and research questions.
• Analysis Independence: Teams applied their own analysis pipelines without collaboration or discussion of approaches with other participants.
• Hypothesis Testing: Researchers addressed predefined hypotheses to enable direct comparison of outcomes across teams.
• Metadata Collection: The study collected comprehensive metadata on researchers' fNIRS experience, analysis confidence, and specific methodological choices [33] [98].

This innovative methodology allowed the FRESH consortium to disentangle the effects of analysis decisions from other potential confounding factors, providing unprecedented insights into the sources of variability in neuroimaging research.

Key Quantitative Findings on Reproducibility

The FRESH study yielded critical quantitative insights into the reproducibility of fNIRS research, with findings that likely extend to other neuroimaging modalities. The results revealed several key patterns that directly inform our understanding of how analysis pipelines and expertise impact reproducibility.

Table 1: Key Reproducibility Findings from the FRESH Community Study

Metric	Finding	Implication
Group-Level Agreement	Nearly 80% of teams agreed on group-level results for strongly literature-supported hypotheses [33]	High consensus possible when biological effects are strong
Individual-Level Agreement	Lower agreement on individual-level analyses [33]	Greater challenge in single-subject applications
Researcher Experience	Teams with higher self-reported analysis confidence (correlated with fNIRS experience) showed greater agreement [33]	Expertise development improves reproducibility
Data Quality Impact	Individual-level agreement improved with better data quality [33]	Quality control is foundational to reproducibility

These findings underscore that reproducibility is not a binary outcome but exists on a spectrum influenced by multiple interacting factors. The stronger agreement at the group level aligns with established practices in neuroimaging, where averaging across participants reduces individual variability and noise. The correlation between researcher experience and agreement highlights the often-overlooked human dimension in analytical reproducibility.

Primary Factors Affecting Reproducibility

The FRESH study identified three primary categories of analytical variability that significantly impact reproducibility in fNIRS research. Understanding these sources of variation is essential for developing strategies to enhance reliability.

Table 2: Primary Sources of Variability in fNIRS Analysis Pipelines

Variability Category	Specific Examples	Impact Level
Handling of Poor-Quality Data	Approaches to motion artifact correction, signal quality thresholds, channel exclusion criteria [33]	High
Response Modeling	Hemodynamic response function selection, physiological noise modeling, temporal derivative inclusion [33]	High
Statistical Analysis	Multiple comparison correction methods, statistical thresholds, cluster-size estimation techniques [33]	High
Preprocessing Choices	Filter types and cutoffs, physiological noise removal methods, short-channel regression application [46] [99]	Medium-High

The handling of poor-quality data emerged as a particularly influential factor, with different approaches to artifact correction and signal quality assessment leading to divergent conclusions [33]. Similarly, choices in response modeling—such as how the hemodynamic response is characterized and fitted—introduced significant variability, especially for individual-level analyses. These findings highlight that analytical flexibility, while valuable for exploring different aspects of neural signals, can substantially impact research outcomes and interpretations.

The Role of Researcher Expertise

Beyond specific methodological choices, the FRESH investigation revealed that researcher expertise plays a crucial role in reproducibility. Teams with more experience in fNIRS analysis and higher self-reported confidence produced results that showed greater agreement with the consensus [33] [97]. This expertise effect manifested in several ways:

• More appropriate pipeline selection for specific research questions and data characteristics
• Better judgment in quality control decisions and artifact management
• More informed parameter selection based on deeper understanding of the method's principles
• More comprehensive reporting of analytical choices and their justification

The correlation between expertise and reproducibility underscores the importance of both technical standardization and comprehensive training in neuroimaging methods. It suggests that as the field matures and researchers gain more experience, reproducibility may naturally improve—but that this process can be accelerated through explicit guidance and knowledge sharing.

Experimental Protocols for Assessing Reproducibility

Test-Retest fNIRS Protocol

Establishing robust experimental protocols is essential for systematic assessment of reproducibility. The following test-retest protocol has been validated for evaluating within-subject reproducibility in fNIRS studies [46] [99]:

• Participant Preparation: Screen participants for inclusion criteria (e.g., TMS compatibility if used for localization). Record demographic and physiological characteristics including hair properties that may affect signal quality [99].
• Optode Placement: Utilize neuronavigation techniques (e.g., TMS-guided localization) for precise targeting of brain regions. Mark optode positions on the scalp for consistent placement across sessions [46] [99]. Digitize optode positions for each session to quantify placement shifts [46].
• Session Structure: Conduct multiple sessions on separate days (e.g., 1-13 day intervals). Each session should include identical protocols with consistent timing and conditions [99].
• Task Design: Implement block-design protocols with appropriate rest periods. For motor tasks, use self-paced or metronome-guided movements at consistent frequencies (e.g., 1Hz grasping). Include sufficient trials (e.g., 30 trials per condition) to ensure statistical power [99].
• Data Acquisition: Use consistent device parameters across sessions. Collect at appropriate sampling frequencies (e.g., 8.98Hz or higher). Include short-separation channels for enhanced signal processing [99].
• Quality Assessment: Calculate signal-to-noise ratios for each channel. Monitor physiological parameters (e.g., heart rate, Mayer waves) that may affect signal quality [99].

Reproducibility Quantification Methods

To objectively assess reproducibility, researchers should employ multiple quantitative measures:

• Intraclass Correlation Coefficient (ICC): Calculate ICC values for both single subjects (ICCsingle) and groups to quantify consistency across sessions [99].
• Spatial Overlap Analysis: Quantify the spatial consistency of activation patterns across sessions using dice coefficients or similar overlap metrics [46].
• Amplitude Consistency: Compare effect sizes (e.g., beta values) and peak amplitudes of hemodynamic responses across sessions [46] [99].
• Classification Accuracy: For paradigm-based studies, compute classification accuracies for task conditions across sessions as a measure of functional reproducibility [99].

These protocols and metrics provide a standardized approach for assessing reproducibility, enabling more direct comparisons across studies and laboratories.

Visualization: Analytical Variability and Reproducibility Framework

The following diagram illustrates the key factors affecting reproducibility in fNIRS research and their interrelationships, as revealed by current research:

Analytical Variability and Reproducibility Framework - This visualization maps the complex relationships between analysis pipeline choices, researcher expertise, and reproducibility outcomes in fNIRS research, highlighting the factors that most significantly impact reliability.

Standardization Initiatives and Best Practices

Community-Led Standardization Efforts

In response to reproducibility challenges, the fNIRS community has mobilized to develop comprehensive standardization resources. The Society for functional Near Infrared Spectroscopy (SfNIRS) has established a Standardization and Open Science Committee dedicated to enhancing reproducibility through several key initiatives [100]:

• Best Practices Guidelines: Development and maintenance of evidence-based guidelines for hardware characterization, experimental design, data acquisition, analysis, and reporting [100].
• Data Format Standards: Establishment and promotion of the Shared Near-Infrared File Format (SNIRF) and integration with the Brain Imaging Data Structure (BIDS) standard to ensure interoperability and facilitate data sharing [100].
• Terminology Harmonization: Creation of a community-sourced fNIRS glossary to establish consistent terminology across the field [97] [100].
• Open Science Promotion: Encouragement of data, code, and protocol sharing through frameworks like the OpenfNIRS ecosystem [97] [100].

These initiatives represent a proactive approach to addressing reproducibility challenges before they become entrenched in the literature—a strategy from which other neuroimaging modalities might benefit.

Technical Recommendations for Enhanced Reproducibility

Based on current evidence, researchers can adopt several technical practices to enhance the reproducibility of their fNIRS studies:

• Implement Short-Channel Regression: Utilize short-separation channels (<8mm for adults) to regress out confounding physiological signals from scalp circulation, which has been shown to improve single-subject reproducibility (increasing ICCsingle from 0.64 to 0.81 in one study) [99].
• Adopt Source Localization Techniques: Move beyond channel-based analysis to anatomical source reconstruction using digitized optode positions and individual or template head models, which improves spatial accuracy and reproducibility [46].
• Prioritize Oxyhemoglobin (HbO) Signals: Focus on HbO measurements, which have demonstrated higher reproducibility across sessions compared to deoxyhemoglobin (HbR) signals [46].
• Standardize Quality Metrics: Apply consistent signal quality thresholds (e.g., raw signal-to-noise ratio threshold at 40dB) to identify measurements with sufficient quality for analysis [99].
• Document Analysis Pipelines Completely: Utilize checklists from best practices guidelines to ensure comprehensive reporting of all analytical choices [33] [100].

Table 3: Essential Research Reagents and Resources for Reproducible fNIRS Research

Resource Category	Specific Examples	Primary Function
Standardized Data Formats	SNIRF file format, BIDS-fNIRS extension [100]	Ensure data interoperability and facilitate sharing
Analysis Software & Toolboxes	Homer2, Homer3, NIRS-KIT, MNE-NIRS, openfnirs.org resources [100] [98]	Provide validated processing pipelines
Quality Assessment Tools	Signal-to-noise ratio calculators, motion artifact detection algorithms, physiological noise monitors [99]	Standardize quality control procedures
Probe Placement Guides	TMS-guided neuronavigation, photogrammetry systems, 10-20 system alignment [46] [99]	Improve spatial consistency across sessions
Experimental Paradigms	FRESH test datasets, validated block designs, event-related designs [49] [98]	Provide benchmarking and validation resources
Reporting Guidelines	Best practices checklist, methodological reporting standards [33] [100]	Ensure comprehensive methodology documentation

This toolkit represents the foundational resources that support reproducible fNIRS research. By leveraging these community-developed resources, researchers can enhance the reliability and comparability of their findings while contributing to the overall robustness of the field.

The reproducibility of fNIRS research—and neuroimaging more broadly—is neither predetermined nor elusive. Rather, it is actively shaped by the methodological choices researchers make and the expertise they bring to their analytical practices. The evidence from community-wide initiatives like FRESH demonstrates that while analytical variability presents significant challenges, consistent patterns of agreement emerge when researchers employ rigorous methods and when biological signals are strong [33].

The path forward requires a dual commitment: to the continued development and adoption of standardized practices while preserving the analytical flexibility that drives methodological innovation. By embracing open science principles, leveraging community-developed resources, and prioritizing comprehensive training, the neuroimaging community can enhance the reproducibility of fNIRS research without stifling its creative potential. As fNIRS continues to expand the boundaries of where brain activity can be measured—from laboratory settings to real-world environments—this commitment to reproducibility will ensure that the findings we generate are both ecologically valid and scientifically robust.

Understanding the intricate functions of the human brain requires multimodal neuroimaging approaches that integrate complementary technologies. Functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) each provide unique insights into brain structure and function, yet no single modality can comprehensively capture the multifaceted nature of neural activity [11]. This technical guide examines core hardware optimization challenges in multimodal integration, focusing on probe design, spatial co-registration, and custom headgear solutions essential for effective simultaneous data acquisition. Framed within the broader principles of fMRI, EEG, and fNIRS for brain research, this review addresses persistent technical barriers and innovative solutions driving the field forward.

The fundamental challenge in multimodal neuroimaging lies in leveraging the distinct advantages of each technique while mitigating their individual limitations. fMRI provides high spatial resolution for visualizing deep brain structures but requires expensive, immobile equipment and is sensitive to motion artifacts [11]. fNIRS offers superior temporal resolution, portability, and tolerance for movement but is confined to superficial cortical regions with lower spatial resolution [11]. EEG delivers millisecond-level temporal precision for capturing rapid neural dynamics but suffers from poor spatial localization [53]. Combining these modalities creates powerful synergies but introduces significant technical complexities in hardware integration, spatial alignment, and experimental design.

Fundamental Principles and Comparative Strengths of Neuroimaging Modalities

Core Biophysical Principles

Each neuroimaging modality measures distinct physiological phenomena related to neural activity. fMRI detects Blood Oxygen Level Dependent (BOLD) signals, reflecting changes in blood flow and oxygenation associated with neural firing. This technique provides high spatial resolution (millimeter-level) whole-brain coverage but has limited temporal resolution (0.33-2 Hz) due to the slow hemodynamic response [11].

fNIRS utilizes near-infrared light (650-950 nm) to measure changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) concentrations in the superficial cortex. Similar to fMRI, it measures hemodynamic responses but offers higher temporal resolution (often millisecond-level) and greater portability [11]. The technique relies on the modified Beer-Lambert law to calculate chromophore concentrations from light attenuation measurements [101].

EEG records electrical potentials generated by synchronous synaptic activity of cortical neurons, primarily from pyramidal cells. It provides exquisite temporal resolution (millisecond precision) but limited spatial resolution due to the blurring effects of the skull and scalp [53] [102].

Comparative Analysis of Technical Specifications

Table 1: Technical comparison of major neuroimaging modalities

Parameter	fMRI	fNIRS	EEG
Spatial Resolution	High (1-3 mm)	Moderate (1-3 cm)	Low (Several cm)
Temporal Resolution	Low (0.3-2 Hz)	Moderate (~0.1-10 Hz)	High (≥1000 Hz)
Depth Penetration	Whole brain	Superficial cortex (2-3 cm)	Superficial cortex
Portability	Low	High	High
Motion Tolerance	Low	Moderate	High
Cost	High	Moderate	Low-Moderate

The complementary nature of these modalities is visually represented in the spatial-temporal resolution trade-offs. EEG provides millisecond temporal precision but poor spatial localization, while fNIRS offers better spatial resolution than EEG with good temporal characteristics, and fMRI delivers the highest spatial resolution across the entire brain with slower temporal response [53] [102].

Figure 1: Resolution characteristics and measurement targets of EEG, fNIRS, and fMRI

Probe Design Considerations for Multimodal Integration

fNIRS Probe Configurations: Sparse vs. High-Density Arrays

Probe design fundamentally influences signal quality, spatial specificity, and depth sensitivity in fNIRS. Traditional sparse arrays employ a non-overlapping grid pattern with 30mm channel spacing, providing basic cortical coverage but limited spatial resolution and sensitivity [103]. These configurations suffer from poor inter-subject consistency and an inability to differentiate adjacent functional regions.

High-density (HD) arrays with overlapping, multi-distance channels represent a significant advancement in probe design. HD configurations improve spatial resolution, depth sensitivity, and signal-to-noise ratio by employing multiple source-detector distances ranging from 10-45mm [103]. This overlapping design enables better localization of brain activity through three-dimensional image reconstruction in Diffuse Optical Tomography (DOT).

Table 2: Comparison of sparse versus high-density fNIRS arrays

Characteristic	Sparse Arrays	High-Density Arrays
Channel Layout	Non-overlapping grid	Overlapping, multi-distance
Typical Spacing	30 mm	10-45 mm (multiple distances)
Spatial Resolution	Low (1-3 cm)	High (<1 cm)
Depth Sensitivity	Limited	Improved with multiple distances
Localization Accuracy	Moderate	High
Setup Complexity	Low	High
Setup Time	Shorter	Longer
Cost	Lower	Higher

Comparative studies demonstrate the superior performance of HD arrays. In Word-Color Stroop tasks, HD arrays outperformed sparse configurations in detecting and localizing brain activity, particularly during lower cognitive load tasks where sparse arrays showed limited sensitivity [103]. HD arrays also demonstrated improved inter-subject consistency in localization, a critical factor for group-level analyses.

Optode Design and Scalp Coupling

Effective light transmission and detection requires optimal optode-scalp coupling. Standard fNIRS optodes utilize various tip designs tailored to specific applications and populations. Blunt tip optodes are ideal for neonatal and infant measurements, while dual-tip designs provide enhanced comfort and stability for extended recordings [104]. Low-profile optodes are essential for concurrent fMRI or TMS studies where electromagnetic compatibility is crucial [104].

Proper optode pressure and alignment is critical for signal quality. Insufficient contact pressure creates signal dropout, while excessive pressure causes participant discomfort and potential skin irritation. Customized headgear solutions with adjustable mounting systems help maintain consistent pressure across all optodes throughout experimental sessions.

Spatial Co-registration Methodologies

The Co-registration Imperative

Spatial co-registration transforms fNIRS channel positions from scalp-based measurements to corresponding cortical locations, enabling meaningful anatomical interpretation of findings. Without proper co-registration, fNIRS data remains limited to relative channel positions without reference to underlying neuroanatomy [105]. This process is particularly crucial for multimodal integration, where aligning fNIRS data with structural MRI or EEG electrode positions enables direct comparison across modalities.

Individual-Level Co-registration with Structural MRI

The most accurate co-registration approach uses subject-specific structural MRI. Vitamin E capsules or other MRI-visible markers are placed on key fNIRS optodes during MRI acquisition, creating visible reference points on structural images [101]. These markers enable precise mapping between scalp positions and cortical structures.

The balloon-inflation algorithm has emerged as an efficient automatic projection method that accurately represents underlying cortical anatomy [101]. This algorithm projects fNIRS channel locations from the scalp to the cortical surface along paths that approximate the normal to the cortical sheet, improving upon manual methods that were time-consuming and error-prone.

Following MRI acquisition and marker identification, processing pipelines typically involve:

• Segmentation of structural MRI to identify brain tissue and cortical surface
• Co-registration of digitized optode positions with scalp surface
• Projection of fNIRS channels to cortical surface using mathematical algorithms
• Transformation to standard stereotaxic space (MNI or Talairach) for group-level analysis

Probabilistic and Virtual Registration Methods

When individual MRI is unavailable, probabilistic registration methods provide a practical alternative. These techniques utilize reference MRI databases to establish statistical relationships between scalp landmarks (based on the 10-20 system) and cortical structures [105]. The probabilistic approach enables MRI-free co-registration for standalone fNIRS systems, greatly enhancing accessibility, though with reduced individual accuracy.

Virtual registration methods have been developed that require only three or four scalp landmarks, making them particularly suitable for field studies or clinical settings where MRI access is limited [101]. These methods are implemented in various fNIRS software packages including HomER2, fNIRS_SPM, and POTATo.

Figure 2: fNIRS-MRI co-registration workflow for individual-level analysis

Custom Headgear Solutions for Multimodal Integration

Integrated Cap Design Principles

Effective multimodal headgear must address competing spatial demands while maintaining signal quality and participant comfort. Key design considerations include:

• Spatial Allocation: EEG electrodes and fNIRS optodes compete for scalp real estate, particularly in regions of interest [102]. Strategic placement is essential, with priority given to modality-specific hotspots based on research questions.
• Fabric Characteristics: Dark-colored cap fabric (preferably black) reduces unwanted optical reflection and improves fNIRS signal quality [102]. The material should provide sufficient tension to maintain optode contact while ensuring participant comfort during extended recordings.
• Holder Compatibility: Cap slits must physically accommodate both EEG electrode holders and fNIRS optode fixtures, requiring caps with high slit density (128-160 slits) for flexible montages [102].

Advanced Headgear Fabrication Techniques

Several innovative approaches address the limitations of standard elastic caps:

3D-Printed Custom Helmets: Subject-specific helmets created via 3D scanning and printing offer optimal fit and precise sensor placement [53]. While production costs are higher, these solutions provide unparalleled stability and reproducibility for longitudinal studies.

Thermoplastic Custom Helmets: Composite polymer cryogenic thermoplastic sheets provide a cost-effective alternative to 3D printing [53]. These materials soften at approximately 60°C, allowing molding to individual head shapes, then retaining stability when cooled.

Integrated Modular Systems: Commercial solutions like the NIRx NIRScaps system offer flexible integration of fNIRS with both active and passive EEG electrodes [104]. These systems provide standardized components for reliable and reproducible montages across participants.

Population-Specific Adaptations

Headgear design requires specialization for different populations. Infant sleep research presents unique challenges, requiring comfortable systems that maintain coupling during natural movement without disrupting sleep [106]. Success rates improved from 45% to 90% when transitioning from non-wearable to wearable wireless NIRS-EEG systems in infant studies, highlighting the importance of age-appropriate design [106].

Older adult populations present different challenges, including age-related brain atrophy and cerebrovascular changes that affect signal quality [101]. Co-registration approaches must account for these morphological differences when projecting channels to cortical surfaces.

Experimental Protocols and Validation Methodologies

Synchronization Approaches for Multimodal Recordings

Precise temporal alignment is crucial for correlating neural events across modalities. Two primary synchronization methods are employed:

Hardware Trigger Sharing: Direct electrical connection between systems shares event markers with minimal latency. This approach provides the most precise synchronization but requires compatible trigger interfaces across devices [102].

Lab Streaming Layer (LSL) Protocol: An open-source system for unified collection of measurement time series across multiple devices [102]. LSL provides software-based synchronization with sub-millisecond precision and greater flexibility for integrating heterogeneous systems.

Experimental Paradigms for Multimodal Validation

Motor execution, observation, and imagery paradigms effectively demonstrate multimodal integration. A published protocol recorded simultaneous fNIRS-EEG during cup grasping tasks, revealing complementary activation patterns [71]. The fNIRS component identified hemodynamic responses in left angular gyrus and right supramarginal gyrus, while EEG detected electrical activity in bilateral central, right frontal, and parietal regions [71].

Structured sparse multiset Canonical Correlation Analysis (ssmCCA) has been employed to fuse fNIRS and EEG data, consistently identifying activation in the left inferior parietal lobe, superior marginal gyrus, and post-central gyrus across conditions [71]. This multimodal fusion approach validates findings across modalities and provides a more comprehensive picture of neural activity.

Cognitive tasks like the Word-Color Stroop paradigm demonstrate the sensitivity advantages of high-density fNIRS arrays. HD configurations successfully detected activation during both congruent and incongruent conditions, while sparse arrays only reliably detected activation during the more demanding incongruent condition [103].

Signal Quality Assessment and Artifact Management

Rigorous quality assessment is essential for multimodal studies. Quantitative Toolbox for NIRS (QT-NIRS) provides standardized metrics for evaluating signal quality, including good time window percentage and channel inclusion rates [106]. In infant sleep studies, wearable systems achieved 50% channel inclusion rates while significantly improving sleep duration and study success rates [106].

Motion artifacts present particular challenges that affect modalities differently. EEG is susceptible to muscle and movement artifacts, while fNIRS signals can be contaminated by scalp blood flow changes. Short-separation channels in fNIRS arrays enable regression of superficial contaminants, improving specificity to cerebral hemodynamics [103].

The Scientist's Toolkit: Essential Research Reagents

Table 3: Essential hardware and software solutions for multimodal neuroimaging

Tool Category	Specific Examples	Function & Application
fNIRS Systems	Hitachi ETG-4000, NIRx NIRScout	Hemodynamic response measurement with various probe configurations
EEG Systems	Brain Products BrainAMP, Electrical Geodesics	Electrical potential recording with high temporal resolution
Integrated Caps	actiCAP (Easycap), NIRx NIRScaps	Physical platform for co-locating EEG electrodes and fNIRS optodes
Spatial Digitizers	Fastrak (Polhemus)	3D coordinate measurement for optode/electrode positioning
Co-registration Software	HomER2, fNIRS_SPM, ArrayDesigner	Spatial mapping of channels to cortical locations
Synchronization Solutions	Lab Streaming Layer (LSL)	Temporal alignment of multimodal data streams
HD-fNIRS Arrays	GowerLabs Lumo, Custom DOT	High-density mapping with overlapping channels
Specialized Optodes	Dual-tip, Low-profile, Blunt-tip	Population-specific and application-specific light coupling

Hardware optimization for multimodal neuroimaging requires careful consideration of probe design, spatial co-registration, and integrated headgear solutions. The complementary nature of fMRI, EEG, and fNIRS provides powerful opportunities for comprehensive brain mapping, but realizing this potential demands technical excellence across multiple domains.

Future advancements will likely focus on improved miniaturization of components, enhanced electromagnetic compatibility for simultaneous fMRI-fNIRS recordings, and more sophisticated co-registration algorithms that account for individual anatomical variations. Standardization of montages, registration protocols, and data formats will be crucial for facilitating data sharing and comparison across research sites.

The development of increasingly wearable, robust multimodal systems opens new possibilities for studying brain function in naturalistic environments and clinical populations. As these technologies mature, they will continue to transform our understanding of brain function in health and disease, ultimately enhancing diagnostic and therapeutic strategies in neuroscience and clinical practice.

The advancement of brain research using neuroimaging technologies like functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) is fundamentally reliant on the consistency and reproducibility of the data produced. However, the field currently grapples with a "combinatorial explosion" of data-processing methodologies, where vast and sometimes arbitrary choices in analysis workflows can lead to drastically different conclusions from the same dataset [107]. This variability poses a significant challenge for comparing results across studies, validating biomarkers, and translating research findings into clinical practice, such as in drug development [107] [108].

The absence of standardized guidelines introduces heterogeneity in study design and reporting, complicating the assessment of a biomarker's reliability, validity, and ultimate utility [108]. Recognizing this, the neuroscientific community is increasingly focusing on systematic evaluation and the development of reporting standards. This article provides an in-depth technical guide to the current state of standardization efforts for fMRI, EEG, and fNIRS data processing and reporting, offering detailed methodologies and resources to help researchers navigate the path toward consensus.

The Standardization Landscape: Challenges and Initiatives

The push for standardization is driven by the need to enhance the reliability, reproducibility, and clinical applicability of neuroimaging findings. The core challenge lies in the multitude of methodological choices available at every stage of data processing, from the definition of network nodes in a functional connectome to the correction of physiological artifacts [107].

A landmark systematic evaluation of fMRI data-processing pipelines revealed the profound impact of these choices. Researchers assessed 768 distinct pipelines for constructing functional brain networks from resting-state fMRI data and found that the majority of pipelines failed at least one critical criterion for reliability and sensitivity [107]. This underscores a critical point: an uninformed choice of pipeline can produce results that are not only misleading but replicably so. Despite this variability, the same study identified a subset of optimal pipelines that consistently satisfied all evaluation criteria, demonstrating that achieving robust and generalizable results is feasible with informed, standardized practices [107].

Parallel efforts are underway for EEG. An international working group under the International Federation of Clinical Neurophysiology (IFCN) has developed the GREENBEAN (Guidelines for Reporting EEG/Neurophysiology Biomarker Evaluation for Application to Neurology and Neuropsychiatry) checklist [108]. This initiative classifies EEG biomarker validation studies into distinct phases, mirroring the framework for therapeutic studies, and provides a detailed checklist of methodological factors that must be reported to ensure transparency and allow for replication [108].

Table 1: Key Standardization Initiatives in Neuroimaging

Initiative/Aspect	Modality	Core Focus	Key Outcome/Recommendation
Pipeline Evaluation [107]	fMRI	Functional connectomics	Identified optimal, reliable end-to-end pipelines for network construction from rs-fMRI.
GREENBEAN Checklist [108]	EEG	Biomarker validation	Provides a reporting framework and checklist for EEG biomarker studies (Phases 1-4).
Multimodal Integration [109]	EEG-fNIRS	Experimental platform & protocol	Proposes a standardized protocol for evaluating multimodal NF, including open-source software.
Carbon Footprint Assessment [110]	fMRI	Environmental sustainability	Benchmarks carbon emissions of preprocessing software, informing eco-conscious pipeline choices.

Beyond analytical consistency, newer considerations like the environmental sustainability of data processing are emerging. A 2025 study quantified the carbon footprint of fMRI data preprocessing and statistical analysis across three common software packages (FSL, SPM, and fMRIPrep), finding that emissions from fMRIPrep were 23-30 times larger than those from SPM and FSL, albeit with slightly superior statistical sensitivity in some cases [110]. This highlights a potential trade-off between data quality and computational cost, urging researchers to make conscious choices about when and how to use computationally expensive tools.

Modality-Specific Standardization Efforts

Functional Magnetic Resonance Imaging (fMRI)

The standardization of fMRI processing, particularly for functional connectomics, requires meticulous attention to each step of the pipeline. A systematic framework for evaluation has identified key choice points and their impact on the resulting network topology [107].

Key Steps in fMRI Network Construction:

• Global Signal Regression (GSR): A controversial preprocessing step that significantly influences downstream results. Evaluations must be conducted separately for data processed with and without GSR [107].
• Node Definition (Parcellation): Nodes can be defined using discrete anatomical or functional parcellations (e.g., with ~100, 200, or 300-400 regions) or continuous, overlapping maps from Independent Components Analysis (ICA) [107].
• Edge Definition: The statistical relationship between node time-series can be defined using Pearson correlation or mutual information [107].
• Edge Filtering: A critical step to retain meaningful connections. Methods include imposing a fixed density (e.g., 5%, 10%), a minimum weight threshold (e.g., 0.3, 0.5), or data-driven methods like Efficiency Cost Optimisation (ECO) [107].
• Network Type: The final network can be analyzed as a binary (connections present/absent) or weighted (connections have strength) graph [107].

The evaluation of pipelines should be based on multiple criteria, including minimizing motion confounds and spurious test-retest discrepancies while maintaining sensitivity to inter-subject differences and experimental effects [107]. The portrait divergence (PDiv) measure, which captures dissimilarity between networks across all scales of organization, is a valuable tool for this comprehensive assessment [107].

Electroencephalography (EEG)

Standardization in EEG is crucial for the development and validation of biomarkers. The GREENBEAN guidelines address this by defining four phases of validation [108]:

• Phase 1-2: Preliminary studies that do not constitute formal validation.
• Phase 3: Studies designed to provide compelling evidence of a biomarker's validity.
• Phase 4: Studies assessing clinical utility and generalizability in real-world settings.

The guidelines emphasize transparent reporting of technical standards and experimental design to mitigate bias and enhance the interpretation of results. This includes detailed reporting on participant characteristics, recording parameters, data preprocessing steps, feature extraction, and statistical analysis [108].

Functional Near-Infrared Spectroscopy (fNIRS)

While fNIRS is prized for its portability and applicability in ecological settings, its literature, particularly in fields like occupational workload assessment, suffers from small sample sizes and a lack of standardized signal processing methods [111]. A systematic review highlighted that only about 41% of fNIRS studies on occupational workload applied corrections for systemic and extracerebral artifacts, which are essential for data quality [111]. Furthermore, inconsistent optode placement across studies hinders the comparability of findings. Future efforts must focus on harmonizing signal-processing workflows and standardizing sensor layouts to improve the validity and cross-study integration of fNIRS research [111].

Multimodal Integration: EEG-fNIRS

The integration of complementary modalities like EEG and fNIRS is a promising strategy to overcome the limitations of single techniques. However, this introduces additional complexity in standardization, requiring seamless coordination and synchronization of distinct systems [109]. A 2025 study presented a fully operational experimental platform for multimodal EEG-fNIRS neurofeedback, which includes a custom cap integrating both sensor types and open-source software for real-time signal processing [109]. Making such platforms and their source code publicly available is a concrete step toward standardizing multimodal research, allowing other groups to replicate and build upon established methods.

A Framework for Standardized Experimental Protocols

Protocol: Evaluating fMRI Processing Pipelines

This protocol is based on the systematic evaluation of functional connectomics pipelines [107].

1. Objective: To identify an optimal fMRI data-processing pipeline that produces reliable and biologically relevant functional brain networks. 2. Experimental Design:

• Utilize a test-retest dataset with repeated scans from the same individuals across short (minutes), medium (weeks), and long-term (months) delays.
• Include an experimental manipulation (e.g., pharmacological intervention) to test sensitivity to change. 3. Data Acquisition:
• Collect resting-state fMRI data. An independent dataset like the UCLA Consortium for Neuropsychiatric Phenomics LA5c Study can be used for validation [107]. 4. Data Processing:
• Systematically vary the following parameters to generate a large number of pipelines (e.g., 768):
  • With and without Global Signal Regression (GSR).
  • Brain parcellation scheme and number of nodes.
  • Edge definition (Pearson correlation, mutual information).
  • Edge filtering method and level (e.g., density-based, threshold-based, ECO).
  • Network type (binary, weighted). 5. Outcome Measures & Analysis:
• Test-Retest Reliability: Use portrait divergence (PDiv) to quantify topological dissimilarity between networks from the same individual across repeated scans. Lower PDiv indicates higher reliability.
• Sensitivity: Assess the pipeline's ability to detect:
  • Inter-subject differences.
  • The effect of the experimental manipulation (e.g., propofol anesthesia).
• Generalizability: Validate the performance of top-performing pipelines on an independent dataset acquired with different parameters (e.g., HCP dataset).

Table 2: Essential Research Reagents and Computational Tools

Item/Tool Name	Function/Application	Context & Consideration
fMRIPrep [110]	Robust, standardized fMRI data preprocessing.	Higher statistical sensitivity but has a significantly larger carbon footprint (30x FSL).
FSL FEAT [110]	fMRI preprocessing and first-level analysis.	Lower carbon footprint; performance comparable to fMRIPrep in some regions.
SPM [110]	Statistical analysis of brain imaging data.	Lower carbon footprint; performance varies by brain region.
EEG-fNIRS Integrated Cap [109]	Simultaneous acquisition of electrical and hemodynamic brain activity.	Custom hardware for multimodal studies; requires co-registration of sensors.
GREENBEAN Checklist [108]	Reporting standard for EEG biomarker validation studies.	Ensures transparent reporting of technical and methodological details.
OpenNeuro Datasets [110]	Public repository for sharing raw neuroimaging data.	Facilitates replication and use of standardized datasets for pipeline testing.

Protocol: Multimodal EEG-fNIRS Neurofeedback

This protocol is derived from a study evaluating the effects of multimodal neurofeedback during motor imagery [109].

1. Objective: To assess the benefits of combining EEG and fNIRS signals for neurofeedback (NF) in the context of motor imagery (MI), compared to unimodal NF. 2. Participants: Thirty right-handed healthy volunteers. 3. Experimental Design:

• A randomized controlled study where each participant undergoes three NF conditions in a random order: EEG-only, fNIRS-only, and combined EEG-fNIRS. 4. Data Acquisition:
• Hardware: Use a custom cap that integrates a 32-channel EEG system and a continuous-wave fNIRS system with 16 detectors and 16 sources.
• Sensor Placement: Position EEG electrodes and fNIRS channels over the sensorimotor cortices according to the 10-10 international system. 5. Task:
• Participants perform left-hand motor imagery.
• A visual feedback interface (e.g., a ball moving along a one-dimensional gauge) is presented in real-time, its movement controlled by the NF score derived from brain activity in the right primary motor cortex. 6. Data Processing & NF Calculation:
• EEG Feature: Event-Related Desynchronization (ERD) in the sensorimotor rhythm.
• fNIRS Features: Changes in oxygenated (HbO) and deoxygenated (HbR) hemoglobin concentrations.
• NF Score: In the combined condition, calculate a single NF score that integrates the standardized features from both EEG and fNIRS signals in real-time. 7. Outcome Measures:
• Primary: The NF score and its correlation with the neuroimaging modality.
• Secondary: Participants' subjective feeling of NF control, motor imagery vividness, and performance.

Diagram 1: EEG-fNIRS multimodal neurofeedback workflow.

The Scientist's Toolkit: Reagents and Materials

Implementing standardized protocols requires a set of well-defined tools and materials. The table below details key resources for the featured experiments.

Table 3: The Scientist's Toolkit for Standardized Neuroimaging Research

Category	Item	Specifications / Context of Use
Software & Platforms	fMRI Pipeline Tools	FSL, SPM, fMRIPrep; choice involves trade-offs between statistical sensitivity and computational cost [110].
	EEG-fNIRS Platform	Custom software for real-time multimodal signal processing and NF calculation; source code available via git repository [109].
Reporting Aids	GREENBEAN Checklist	Guideline for reporting EEG biomarker studies to ensure methodological transparency and combat bias [108].
Hardware	EEG System	32-channel system (e.g., ActiCHamp) for electrical brain activity recording [109].
	fNIRS System	Continuous-wave system with 16 detectors & 16 sources (e.g., NIRScout XP) for hemodynamic activity recording [109].
	Integrated Cap	Custom cap (e.g., EasyCap) holding both EEG electrodes and fNIRS optodes over sensorimotor areas [109].
Datasets	Test-Retest fMRI Data	Publicly available datasets (e.g., UCLA CNP LA5c on OpenNeuro) for pipeline validation and reliability testing [107] [110].

Diagram 2: Evaluating fMRI pipelines with a multi-criteria framework.

The journey toward comprehensive standardization in fMRI, EEG, and fNIRS data processing and reporting is well underway. The field is moving from recognizing the problem of methodological heterogeneity to actively developing and testing solutions. These include systematic evaluations of processing pipelines, the creation of reporting guidelines like GREENBEAN, the development of open-source platforms for multimodal integration, and even the consideration of environmental sustainability. For researchers and drug development professionals, adhering to these emerging best practices is no longer optional but fundamental to producing trustworthy, replicable, and clinically translatable neuroscience research. By consciously selecting validated pipelines, transparently reporting methods, and adopting shared tools, the community can solidify the foundation upon which our understanding of the human brain is built.

Validation, Comparative Analysis, and Strategic Modality Selection

The quest to decode human brain function relies on sophisticated neuroimaging technologies, each with distinct strengths and limitations. Functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) have emerged as fundamental tools in cognitive neuroscience and clinical research. fMRI provides high spatial resolution for deep brain structures but lacks temporal precision and portability. EEG offers millisecond-level temporal resolution but suffers from limited spatial accuracy. fNIRS strikes a balance with moderate spatiotemporal capabilities, portability, and motion robustness [34] [30]. This technical guide explores how simultaneous multimodal integration of fMRI-fNIRS and EEG-fNIRS addresses the inherent limitations of individual modalities, enabling cross-validation and enriching our understanding of neural activity through complementary data perspectives. Such integration is particularly valuable for brain-computer interfaces (BCIs), neurological disorder diagnosis, and therapeutic monitoring [34] [112].

The fundamental motivation for multimodal recording lies in the complementary nature of the signals: EEG captures direct, rapid electrical neural activity, while fMRI and fNIRS measure hemodynamic responses that are indirect correlates of neural activity with different temporal characteristics [30]. By combining these modalities, researchers can achieve a more comprehensive picture of brain function, validating findings across measurement techniques and leveraging the strengths of each to overcome their individual limitations.

Technical Comparative Analysis of Neuroimaging Modalities

Table 1: Fundamental Characteristics of Core Neuroimaging Modalities

Feature	fMRI	EEG	fNIRS
Spatial Resolution	High (millimeter-level) [34]	Low (~2 cm) [113]	Moderate (1-3 cm) [34]
Temporal Resolution	Low (0.3-2 Hz, limited by hemodynamics) [34]	Very High (milliseconds) [113] [114]	Moderate (~0.1 Hz, limited by hemodynamics) [34]
Penetration Depth	Whole brain (cortical and subcortical) [34]	Cortical surface [113]	Superficial cortex (2-3 cm depth) [34]
Measured Signal	BOLD (primarily deoxygenated hemoglobin) [34]	Electrical potential differences [30]	Oxygenated & deoxygenated hemoglobin concentration changes [30]
Portability & Cost	Low (immobile, high cost) [113] [34]	High (portable, low cost) [30]	High (portable, low cost) [34] [30]
Motion Robustness	Low (highly sensitive) [34]	Moderate (sensitive to artifacts) [115]	High (relatively robust) [34] [115]

Table 2: Multimodal Integration Approaches and Their Applications

Integration Type	Primary Purpose	Advantages	Example Applications
fMRI-fNIRS	Spatial validation and naturalistic studies [34]	Leverages fMRI's spatial precision to validate fNIRS signals; combines deep brain coverage with cortical specificity [34] [116]	SMA activation validation [116], neurovascular coupling studies [34]
EEG-fNIRS	Spatiotemporal completeness for BCIs [114] [112] [115]	Complementary temporal (EEG) and spatial (fNIRS) features; enhanced classification accuracy [112] [115]	Motor imagery classification [114] [39], emotion recognition [114], mental workload assessment [112]
Feature-Level Fusion	Data integration before classification [112]	Preserves raw information; can optimize complementarity and minimize redundancy [112]	Mutual information-based feature selection [112], hybrid BCI systems [115]
Decision-Level Fusion	Integration after initial classification [112]	Leverages specialized classifiers for each modality; reduces computational complexity [112]	Motor imagery tasks [112], mental arithmetic [114]

Experimental Protocols and Methodologies

Simultaneous fMRI-fNIRS Recording Protocols

3.1.1 Motor Execution and Imagery Paradigm A validated protocol for cross-validating supplementary motor area (SMA) activation involves having participants perform executed and imagined motor tasks. In a study with older participants, researchers collected separate fMRI and continuous-wave fNIRS sessions involving executed hand movements, motor imagery (MI) of hand movements, and MI of whole-body movements. Individual anatomical MRI data served three critical functions: defining regions of interest for fMRI analysis, extracting the fMRI BOLD response from cortical regions corresponding to fNIRS channels, and selecting optimal fNIRS channels based on individual anatomy rather than standard brain templates. The results demonstrated that fNIRS could reliably detect SMA activation during both motor execution and imagery, with deoxygenated hemoglobin (Δ[HbR]) proving to be the more specific signal for motor imagery tasks, particularly for whole-body movements [116].

3.1.2 Hardware Integration Considerations Simultaneous fMRI-fNIRS recording requires specialized hardware configurations to ensure safety and data quality. MRI-compatible fNIRS modules must utilize sufficiently long optical fibers and MRI-compatible optodes to function within the high-magnetic-field environment without causing electromagnetic interference. Digital trigger inputs are essential for synchronizing fNIRS and fMRI data acquisition, enabling precise temporal alignment of the recorded signals for subsequent analysis [117].

Simultaneous EEG-fNIRS Recording Protocols

3.2.1 Motor Imagery and Mental Arithmetic Tasks A common paradigm for EEG-fNIRS research involves classifying motor imagery tasks. Participants are typically seated comfortably while performing mental rehearsals of specific actions, such as hand squeezing imagery. A standardized trial structure includes: (1) instruction period (2 seconds), (2) task period (10 seconds for imagery), and (3) rest period (15 seconds). Visual cues (e.g., left or right arrows) indicate which hand to imagine moving. The simultaneous recording setup typically employs integrated caps containing both EEG electrodes and fNIRS optodes, with data synchronization achieved through hardware triggers [114] [71].

3.2.2 Signal Processing and Feature Extraction Pipeline

• EEG Preprocessing: Signals are re-referenced, bandpass filtered (0.5-50 Hz), down-sampled to 200 Hz, and subjected to independent component analysis (ICA) for ocular artifact removal. Baseline correction is applied using a reference period (-3 to 0s before task onset) [114].
• fNIRS Processing: Raw intensity signals are converted to oxygenated (HbO) and deoxygenated (HbR) hemoglobin concentration changes using the modified Beer-Lambert law. Motion artifacts are corrected using wavelet-based or principal component analysis methods [116].
• Feature Extraction: For EEG, band power features in sensorimotor rhythms (8-30 Hz) are commonly extracted. For fNIRS, mean HbO and HbR concentrations during task periods versus baseline are used. Temporal features like the initial dip (0-2s) can be particularly informative for improving BCI response times [115].

Signaling Pathways and Neural Correlates

The neurophysiological basis for multimodal integration lies in the relationship between electrical neural activity and subsequent hemodynamic responses. This relationship, known as neurovascular coupling, forms the foundation for correlating EEG signals with fNIRS and fMRI measurements.

Electrical to Hemodynamic Transition: When neurons become active, they trigger a complex cascade of metabolic and vascular events. EEG captures the immediate electrical activity—primarily postsynaptic potentials from pyramidal cells—with millisecond precision. This neural activity increases local energy demands, leading to a hemodynamic response that typically peaks 4-6 seconds after the electrical event. The hemodynamic response involves increased blood flow to active regions, altering the relative concentrations of oxygenated and deoxygenated hemoglobin [30]. fNIRS directly measures these hemoglobin concentration changes, while fMRI's BOLD signal primarily reflects deoxygenated hemoglobin levels [34] [116].

Temporal Discrepancies and Complementarity: The fundamental timing difference between electrical events (measured by EEG) and hemodynamic responses (measured by fNIRS/fMRI) is not a limitation but rather an opportunity for comprehensive brain mapping. EEG provides exquisite temporal detail about when neural events occur, while fNIRS and fMRI offer better spatial localization of where these events take place. Simultaneous recording allows researchers to study the precise temporal relationship between these different manifestations of neural activity, providing insights into neurovascular coupling efficiency and its potential alterations in neurological disorders [30] [71].

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential Equipment and Solutions for Multimodal Recordings

Item	Function/Purpose	Technical Specifications	Example Applications
MRI-Compatible fNIRS Module	Simultaneous fMRI-fNIRS recording [117]	Long optical fibers (>5m), non-magnetic optodes, electromagnetic interference shielding	SMA validation studies [116], neurovascular coupling research [34]
Integrated EEG-fNIRS Caps	Co-registered multimodal data acquisition [114] [71]	Embedded optodes and electrodes in elastic cap, standardized positioning according to 10-5/10-20 system	Motor imagery BCI [114] [39], action observation studies [71]
Continuous Wave fNIRS Systems	Hemodynamic response measurement [30]	Multiple wavelengths (695±830 nm), source-detector distances (1.5-3 cm), sampling rate ≥10 Hz	Portable BCI applications [115], clinical population monitoring [116]
High-Density EEG Systems	Electrical neural activity recording [114]	≥16 channels, sampling rate ≥500 Hz, high input impedance, referential recording	Temporal dynamics analysis [71], artifact identification [114]
3D Magnetic Space Digitizer	Precise optode/electrode localization [71]	Spatial accuracy <1 cm, compatibility with anatomical landmarks (nasion, preauricular)	Individualized channel selection [116], accurate spatial registration [71]
Data Synchronization Unit	Temporal alignment of multimodal data [117]	Digital trigger inputs/outputs, sub-millisecond precision, compatible with all modalities	Temporal correlation analysis [30], event-related potential studies [71]

Advanced Data Fusion and Analysis Approaches

Mutual Information-Based Feature Selection

A sophisticated framework for EEG-fNIRS fusion employs mutual information criteria to optimize feature selection by balancing complementarity, redundancy, and relevance. This approach involves extracting multiple spectral and temporal features from both modalities and then selecting an optimized subset that maximizes information about the target variable (e.g., task condition or pathology) while minimizing redundant information between features. Studies applying this method to classify amyotrophic lateral sclerosis (ALS) patients and controls during visuo-mental tasks demonstrated considerably improved hybrid classification performance compared to individual modalities or conventional fusion without feature selection [112].

Cross-Modal Transfer Learning

Innovative transfer learning approaches enable knowledge transfer between EEG and fNIRS modalities. The R-CSP-E method regularizes Common Spatial Pattern (CSP) analysis by incorporating information from EEG signals when computing fNIRS spatial filters. This approach uses Independent Component Analysis (ICA) to establish correspondence between the sources of the two signals, then modifies the CSP algorithm to leverage EEG information for improved fNIRS feature extraction. Experimental results show this cross-modal transfer can increase classification accuracy by up to 5% for motor imagery tasks [114].

Deep Learning with Uncertainty Modeling

Recent advances in deep learning facilitate end-to-end fusion of EEG and fNIRS signals. For EEG data, dual-scale temporal convolution and depthwise separable convolution extract spatiotemporal features, while hybrid attention mechanisms enhance sensitivity to salient neural patterns. For fNIRS, spatial convolution across channels explores regional activation differences, and parallel temporal convolution combined with gated recurrent units (GRUs) captures hemodynamic dynamics. Decision fusion incorporates Dirichlet distribution parameter estimation to model uncertainty, with Dempster-Shafer Theory enabling evidence-based integration of both modalities, achieving 83.26% accuracy in motor imagery classification [39].

Simultaneous multimodal recordings of fMRI-fNIRS and EEG-fNIRS represent a powerful paradigm in cognitive neuroscience and clinical research. By leveraging the complementary strengths of each modality, researchers can overcome fundamental limitations of individual techniques, enabling more comprehensive brain mapping and robust validation of findings. The integration approaches outlined in this guide—from hardware configurations and experimental paradigms to advanced data fusion algorithms—provide a framework for designing rigorous multimodal studies. As these methodologies continue to evolve, they promise to deepen our understanding of brain function and enhance the efficacy of clinical applications in neurology and psychiatry.

Understanding the fundamental trade-off between spatial localization and temporal fidelity is paramount when selecting a neuroimaging technique for brain research. Non-invasive methods like functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and functional near-infrared spectroscopy (fNIRS) provide complementary windows into brain activity, each with distinct strengths and limitations. This whitepaper provides an in-depth technical comparison of these core modalities, framing their performance within the context of spatial accuracy and temporal resolution. Aimed at researchers and drug development professionals, this guide synthesizes current evidence and methodologies to inform experimental design and tool selection in cognitive neuroscience and clinical research.

Core Technical Specifications and Performance Metrics

The following table summarizes the fundamental performance characteristics of fMRI, EEG, and fNIRS, which are defined by the physiological signals they measure.

Table 1: Fundamental Characteristics of Non-Invasive Neuroimaging Modalities

Feature	fMRI	EEG	fNIRS
Primary Signal Measured	Blood-Oxygen-Level-Dependent (BOLD) response [118]	Electrical potentials from synchronized neuronal firing [119]	Changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) [120] [119]
Spatial Resolution	High (Millimeter-level) [118]	Low (Centimeter-level) [119]	Moderate (Better than EEG, limited to cortical surface) [119]
Temporal Resolution	Low (Seconds) [118]	High (Millisecond-level) [119]	Moderate (Seconds) [120]
Depth of Measurement	Whole-brain [86]	Cortical surface [119]	Outer cortex (1–2.5 cm deep) [119]
Key Strength	Excellent spatial localization of deep and subcortical structures [118]	Exquisite timing for tracking rapid neural dynamics [119]	Good balance of spatial specificity and motion tolerance for cortical areas [120] [86]
Primary Limitation	Slow, indirect hemodynamic response; expensive, non-portable [118]	Poor spatial localization; highly susceptible to motion and electrical artifacts [120] [119]	Limited to superficial cortical regions; physiologically delayed signal [120] [119]

Quantitative Performance Data in Cognitive and Clinical Tasks

Performance varies significantly based on the experimental paradigm and population. The table below provides representative quantitative data from recent studies.

Table 2: Representative Performance Metrics in Applied Research Contexts

Modality / Paradigm	Task / Context	Key Performance Metrics	Noted Challenges
EEG - Motor Imagery (MI)	BCI for prosthetic control [120]	High temporal resolution allows for rapid decoding of movement intent; classification accuracy is sufficient for BCI control.	High susceptibility to electrical noise and motion artifacts [120]; low spatial resolution [119].
EEG - Inner Speech Decoding	Classification of covertly imagined words [121]	Spectro-temporal Transformer model achieved 82.4% accuracy in classifying 8 imagined words [121].	Significant inter-subject variability; requires extensive training and validation [121].
fNIRS - Motor Imagery (MI)	BCI for prosthetic control [120]	Improved robustness to electrical and motion noise compared to EEG [120].	Inherent physiological delay (~2-6 seconds) limits speed of control [120] [119].
fNIRS - Clinical Populations	Chronic pain assessment in Alzheimer's disease [122]	Detected significant correlations between HbO and apathy/depression scores in prefrontal and somatosensory cortices [122].	Provides localized cortical biomarkers for neuropsychiatric symptoms [122].
Hybrid EEG-fNIRS	Motor Imagery & BCIs [120]	Enhances performance by combining EEG's temporal precision with fNIRS's spatial specificity [120].	Increased system complexity and data fusion challenges [120].

Detailed Experimental Protocols

EEG for Inner Speech Decoding

This protocol is adapted from the inner speech classification study using a spectro-temporal Transformer model [121].

• Objective: To decode covertly imagined words from non-invasive EEG signals in a cross-subject validation framework.
• Participant Preparation: Healthy, right-handed adults are fitted with a high-density EEG cap (e.g., 73-channel BioSemi system) following the standard 10-20 placement system. Scalp preparation may be necessary to reduce electrode impedance [121].
• Paradigm: Participants perform structured inner speech trials. Each trial involves the visual presentation of a target word from a set of eight (e.g., social words: "child," "wife"; number words: "four," "six"). Participants are instructed to covertly imagine saying the word when cued. Each word is presented in multiple randomized trials (e.g., 40 trials per word) [121].
• Data Acquisition: EEG data is continuously recorded at a high sampling rate (e.g., 2048 Hz) and stored for offline analysis [121].
• Preprocessing:
  • Bandpass Filtering: Raw data is filtered between 0.1 Hz and 50 Hz to remove slow drifts and high-frequency noise [121].
  • Epoching: Continuous data is segmented into epochs time-locked to the stimulus onset.
  • Artifact Rejection: Epochs containing high-amplitude artifacts (e.g., exceeding ±300 μV) are automatically or manually rejected [121].
• Feature Extraction & Classification: Processed epochs are used to train a deep learning model, such as a spectro-temporal Transformer. This model uses wavelet-based time-frequency features and self-attention mechanisms to classify the imagined word from the EEG pattern. Performance is rigorously evaluated using leave-one-subject-out (LOSO) cross-validation to assess generalizability [121].

fNIRS for Clinical Biomarker Assessment

This protocol is based on studies investigating fNIRS-derived biomarkers in individuals with Alzheimer's disease and related dementias (ADRD) [122].

• Objective: To correlate cortical hemodynamic responses with neuropsychiatric symptoms (e.g., apathy, depression) and pain in individuals with ADRD.
• Participant Preparation: Participants (e.g., diagnosed with mild-moderate ADRD) are fitted with an fNIRS headpiece. Optodes are positioned over brain regions of interest (ROIs) such as the prefrontal cortex (PFC) and primary motor and somatosensory (S1) cortices, bilaterally. Placement can be guided by the international 10-20 system [122] [119].
• Paradigm:
  • Baseline Resting State: Hemodynamic signals are recorded during a period of rest to establish a baseline.
  • Stimulation Task: Participants undergo a sub-threshold thermal pain stimulation protocol. Alternatively, cognitive tasks (e.g., verbal fluency) can be used to evoke cortical responses [122].
  • Clinical Assessments: Participants or their caregivers complete standardized self-report measures of pain, depression, and apathy [122].
• Data Acquisition: fNIRS records changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentrations at a typical sampling rate of ~10 Hz [86].
• Preprocessing & Analysis:
  • Signal Processing: Raw optical signals are converted to HbO and HbR concentration changes using the modified Beer-Lambert law.
  • Artifact Correction: Algorithms are applied to correct for motion artifacts and physiological noise (e.g., from heart rate and respiration) [86].
  • Statistical Modeling: Hemodynamic response features (e.g., mean HbO amplitude during stimulation) are extracted. Linear mixed-effects models are then used to test for correlations between these fNIRS features and the clinical assessment scores, often stratifying by factors like cognitive function level [122].

Visualization of Modality Trade-offs and Integration

The following diagram illustrates the core trade-off between spatial and temporal resolution in neuroimaging and the conceptual framework for multimodal integration.

Neuroimaging Trade offs and Fusion

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Key Materials and Solutions for Neuroimaging Experiments

Item / Solution	Function / Application	Technical Notes
High-Density EEG Cap	Holds electrodes in standardized positions (e.g., 10-20 system) for consistent signal acquisition across subjects.	Integrated caps with fNIRS-compatible openings are available for multimodal studies [119].
Electrode Conductive Gel/Paste	Improves electrical conductivity between the scalp and EEG electrodes, reducing impedance and signal noise.	Materials like NeuroPrep gel or Ten20 paste are commonly used, though they require post-session cleaning [120].
fNIRS Optodes	Emit and detect near-infrared light through the scalp to measure hemodynamic changes.	Include specific wavelength sources (e.g., ~760 nm & ~850 nm) and detectors. Placement is critical for spatial specificity [86].
Anatomic Landmark Digitizer	Records the 3D spatial coordinates of EEG/fNIRS sensors and cranial landmarks (nasion, inion).	Enables co-registration of sensor locations with individual or standard brain anatomy (e.g., MRI templates) for improved spatial accuracy [86].
Stimulus Presentation Software	Precisely controls the timing and delivery of experimental tasks (visual, auditory) and records event markers.	Critical for event-related designs; must synchronize with neuroimaging data acquisition hardware [121].
Biophysical Forward Model	A computational model that predicts the signals measured by sensors (EEG/MEG/fNIRS) from underlying neural source activity.	Derived from anatomical (MRI) data; used in source localization and multimodal fusion models [118].

Within the field of non-invasive neuroimaging, functional Magnetic Resonance Imaging (fMRI), Electroencephalography (EEG), and functional Near-Infrared Spectroscopy (fNIRS) have emerged as cornerstone technologies for investigating brain function. Each technique capitalizes on a distinct biological principle to measure correlates of neural activity, resulting in a unique profile of strengths and limitations. This technical guide provides a comparative analysis of fMRI, EEG, and fNIRS, focusing on the critical parameters of cost, portability, participant suitability, and clinical utility. The objective is to furnish researchers, scientists, and drug development professionals with a structured overview to inform experimental design, technology selection, and investment in brain research initiatives.

Core Principles and Physiological Basis

The fundamental difference between these modalities lies in the physiological phenomena they capture. fMRI measures the Blood-Oxygen-Level-Dependent (BOLD) signal, which is an indirect correlate of neural activity based on the magnetic properties of oxygenated versus deoxygenated hemoglobin [11] [2]. In contrast, EEG directly measures the brain's electrical activity, specifically the post-synaptic potentials of pyramidal neurons, which are detected as voltage fluctuations on the scalp [123] [102]. fNIRS, like fMRI, is an indirect measure of neural activity through hemodynamics. It uses near-infrared light to measure concentration changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) in the cortical tissue [11] [2].

The following diagram summarizes the basic signaling pathways for each modality.

Figure 1: Signaling Pathways of Neuroimaging Modalities. EEG measures direct electrical activity, while fMRI and fNIRS measure the slower hemodynamic response to neural activity.

Comparative Advantages Analysis

The distinct physiological bases of EEG, fMRI, and fNIRS translate into divergent technical profiles, which are summarized in Table 1 below.

Table 1: Comparative analysis of non-invasive neuroimaging modalities

Feature	EEG	fMRI	fNIRS
Spatial Resolution	Low (~10 mm) [124]	High (millimeter level) [11]	Moderate (~1-3 cm) [11]
Temporal Resolution	High (millisecond level) [125]	Low (0.33-2 Hz, limited by HRF) [11]	Moderate (~1 s) [124]
Portability	High (fully portable systems available)	None (requires immobile scanner) [11]	High (fully portable, wearable systems) [2]
Participant Suitability	High (robust, suitable for diverse populations)	Low (motion-sensitive, loud, excludes metallic implants) [2]	High (motion-tolerant, quiet, safe for implants) [2]
Approximate Cost	Low to Moderate	Very High (expensive equipment and upkeep) [2]	Moderate (often a one-time investment) [2]
Depth Penetration	Superficial and deep sources (with poor localization)	Whole brain (cortical and subcortical) [11]	Superficial cortical only (up to ~2 cm) [125] [2]

Key Trade-offs and Synergies

• Spatial vs. Temporal Resolution: The primary trade-off is between EEG's millisecond-level temporal precision and fMRI's millimeter-level spatial accuracy. fNIRS occupies a middle ground, offering better spatial resolution than EEG and better temporal resolution than fMRI [11] [102].
• Ecological Validity: fMRI requires a highly controlled, restrictive environment. In contrast, the portability and motion-tolerance of EEG and fNIRS enable brain imaging in naturalistic settings and during active behaviors, greatly enhancing ecological validity [11] [2].
• Complementary Nature: Due to their complementary strengths, combining these modalities is a powerful strategy. Simultaneous EEG-fNIRS is particularly promising as the signals do not interfere, and together they provide a more comprehensive picture of brain dynamics [102].

Experimental Design and Protocol Integration

Designing experiments for multimodal neuroimaging requires careful consideration of the temporal and physical constraints of each technique. A block design is often optimal for fNIRS and fMRI to capture the slow hemodynamic response, while event-related markers are essential for EEG to analyze trial-by-temporal dynamics [102].

The following workflow outlines a protocol for a simultaneous EEG-fNIRS experiment, a common multimodal approach.

Figure 2: Simultaneous EEG-fNIRS Experimental Workflow. Key steps include montage planning, synchronized data acquisition, and joint data analysis.

The Scientist's Toolkit: Essential Research Reagents and Materials

Successful execution of neuroimaging studies, particularly multimodal ones, relies on a suite of specialized equipment and software.

Table 2: Essential materials and solutions for neuroimaging research

Item	Function	Example Use-Case
High-Density EEG Cap	Holds electrodes in standardized positions (e.g., 10-20 system) for recording electrical activity.	Essential for all EEG studies; a 128-slit cap is recommended for hybrid EEG-fNIRS setups to accommodate both sensors [102].
fNIRS Optodes	Emit and detect near-infrared light through the scalp and skull to measure hemodynamic changes.	Placed on the cap over the region of interest (e.g., prefrontal cortex for cognitive tasks) [124] [125].
Hybrid Cap Solution	A single cap designed with sufficient slits/holes to host both EEG electrodes and fNIRS optodes.	Critical for simultaneous EEG-fNIRS recordings, minimizing setup complexity and ensuring stable sensor placement [102].
Synchronization Interface	Hardware or software (e.g., Lab Streaming Layer - LSL) to align data streams from different devices with precise timing.	Mandatory for multimodal studies to ensure event markers and data samples are aligned for integrated analysis [102].
Montage Design Software	Tools (e.g., MATLAB-based ArrayDesigner) to plan the optimal layout of sensors on the scalp.	Used during study planning to define the EEG-fNIRS montage based on the targeted brain regions [102].
Linear Discriminant Analysis	A classification algorithm used to decode brain states or commands from extracted signal features.	Commonly used in Brain-Computer Interface (BCI) studies using EEG, fNIRS, or hybrid systems to translate brain signals into commands [124] [123].

Clinical and Research Applications

The choice of neuroimaging modality is heavily influenced by the specific clinical or research application.

• fMRI remains the gold standard for pre-surgical mapping and for studies requiring precise spatial localization of deep brain structures [11] [2].
• EEG is indispensable for diagnosing epilepsy, monitoring sleep stages, and for any research requiring precise temporal tracking of brain network dynamics [123].
• fNIRS shows significant promise for bedside monitoring of cerebral oxygenation in clinical settings, long-term studies of neurodevelopment in infants, and rehabilitation therapy where patient movement is necessary [11] [2].
• Hybrid BCI Systems: Combining EEG and fNIRS is an active research area for developing more robust brain-computer interfaces. For example, EEG can control left/right commands via motor imagery, while fNIRS from the prefrontal cortex can generate forward/backward commands via mental arithmetic, thereby increasing the number of control signals without sacrificing accuracy [124] [123].

fMRI, EEG, and fNIRS each offer a unique window into brain function. There is no single "best" technology; rather, the optimal choice is dictated by the specific research question, participant population, and practical constraints. fMRI provides unparalleled spatial detail for hypothesis testing about localized brain function, EEG excels at capturing the rapid dynamics of brain networks, and fNIRS offers a flexible tool for studying the brain in real-world environments. The future of neuroimaging lies in the intelligent integration of these complementary modalities, leveraging their synergistic potential to achieve a more holistic and powerful understanding of the human brain in health and disease.

The integration of multimodal neuroimaging techniques is revolutionizing our understanding of brain function and recovery mechanisms following neurological injury. Within this landscape, the combination of quantitative electroencephalography (qEEG) and functional near-infrared spectroscopy (fNIRS) presents a particularly powerful synergy for investigating stroke motor recovery. This case study explores the correlation between qEEG parameters and fNIRS hemodynamic responses within the context of post-stroke rehabilitation, providing a technical guide for researchers and clinicians aiming to implement these technologies. The complementary nature of these modalities—qEEG offering millisecond-level temporal resolution of electrical brain activity and fNIRS providing vascular-based spatial localization of cortical activation—enables a more comprehensive assessment of neuroplastic changes during recovery [126] [11]. This integrated approach is particularly valuable for evaluating the neural mechanisms underlying innovative rehabilitation interventions, such as robot-assisted gait training and repetitive transcranial magnetic stimulation (rTMS), which promote functional recovery through neuroplastic reorganization [127] [128].

Theoretical Foundations and Principles

Basic Principles of qEEG, fNIRS, and fMRI

Understanding the correlated use of qEEG and fNIRS requires foundational knowledge of these modalities and their relationship with the gold-standard functional magnetic resonance imaging (fMRI).

• qEEG represents an advanced evolution of traditional EEG, employing complex mathematical algorithms to transform raw electrical signals into quantifiable metrics. It involves recording digital EEG signals that are processed and analyzed to extract features such as spectral power in specific frequency bands (delta, theta, alpha, beta), coherence, and network properties [126] [129]. These quantitative parameters provide insight into the oscillatory dynamics and functional connectivity of neural networks, with changes in these metrics reflecting pathophysiological states or recovery processes in conditions such as stroke.

• fNIRS is a non-invasive optical neuroimaging technique that measures cortical hemodynamic responses by detecting near-infrared light attenuation through cerebral tissues. It primarily monitors concentration changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) in the superficial layers of the cortex, providing an indirect measure of neural activity based on neurovascular coupling [11] [130]. Its portability, resistance to motion artifacts, and flexibility for use in naturalistic settings make it particularly suitable for rehabilitation environments and longitudinal monitoring.

• fMRI, while not the focus of this case study, provides essential context as the benchmark for spatial resolution in functional neuroimaging. It measures brain activity by detecting Blood Oxygen Level Dependent (BOLD) signals, offering high spatial resolution (millimeter-level) and whole-brain coverage, including subcortical structures [11]. However, its low temporal resolution (limited by the hemodynamic response lag), high cost, and restriction to controlled laboratory settings present limitations for dynamic rehabilitation assessment.

The Synergistic Value of Multimodal Integration

The combination of qEEG and fNIRS capitalizes on their complementary strengths to overcome the limitations of either modality used in isolation. While fNIRS provides superior spatial localization compared to qEEG, its resolution remains lower than fMRI and is restricted to cortical regions [11]. Conversely, qEEG offers unparalleled temporal resolution to capture rapid neural dynamics but suffers from limited spatial specificity and sensitivity to artifacts [131]. Together, they enable robust spatiotemporal mapping of neural activity, with fNIRS anchoring the electrical signatures from qEEG to specific cortical regions.

This synergy is further enhanced by their shared basis in hemodynamic responses and neurovascular coupling. Advanced integration methods, such as the EEG-fNIRS multilayer brain network analysis, quantify neurovascular coupling strength by estimating the subject-specific hemodynamic response function, which then informs the construction of interlayer edges in a comprehensive brain network model [127]. This approach has been shown to significantly improve the prediction of rehabilitation outcomes compared to unimodal methods [127].

Quantitative Parameters and Correlative Evidence

Key qEEG Parameters in Stroke Motor Recovery

qEEG provides several quantitative metrics that serve as sensitive indicators of brain state and recovery progression following stroke. The table below summarizes the most clinically relevant parameters in the context of motor rehabilitation.

Table 1: Key qEEG Parameters for Assessing Stroke Motor Recovery

Parameter	Description	Clinical Significance in Stroke	Reference
Delta/Alpha Ratio (DAR)	Ratio of power in delta (0.3-3.5 Hz) to alpha (8-13 Hz) bands	Increased values indicate greater pathological slow-wave activity and correlate with more severe motor impairment; decreases with recovery.	[126] [128]
Power Ratio Index (PRI)	Ratio of (delta + theta) to (alpha + beta) power	Serves as a composite measure of abnormal slowing; reduction signifies improved motor function recovery.	[128]
Brain Symmetry Index (BSI)	Measure of interhemispheric power asymmetry	Elevated values reflect asymmetry due to stroke lesion; normalization correlates with motor improvement.	[126]
Mean Alpha Frequency	Dominant rhythm in the alpha band	Reduction indicates cerebral dysfunction; normalization associated with better outcomes.	[126]
Coherence	Synchronization of oscillations between different brain regions	Altered inter- and intra-hemispheric connectivity, particularly in theta band, relates to motor network reorganization.	[126]

Key fNIRS Parameters and Their Correlation with qEEG

fNIRS measures hemodynamic changes that reflect the metabolic demands of neural activity, providing a spatially contextualized correlate to qEEG parameters.

Table 2: Key fNIRS Hemodynamic Parameters and Correlations with qEEG

Parameter	Description	Correlation with qEEG/Clinical Significance	Reference
Oxy-Hemoglobin (HbO)	Concentration of oxygenated hemoglobin	Primary indicator of cortical activation; increases in premotor and SMA correlate with improved DAR/PRI.	[128] [132]
Deoxy-Hemoglobin (HbR)	Concentration of deoxygenated hemoglobin	Typically decreases with activation; specific patterns in temporal and DLPFC are predictive of upper limb recovery.	[132] [133]
Total Hemoglobin (HbT)	Sum of HbO and HbR	Indicator of total blood volume change; connectivity patterns involving PMC/PSMC are predictive of motor outcomes.	[132]
Hemodynamic Response Function (HRF)	Model of the hemodynamic response to neural activity	Used to quantify neurovascular coupling strength, linking EEG-derived electrical activity to fNIRS-measured hemodynamics.	[127]
Functional Connectivity Edges	Statistical dependencies between time series of different brain regions	Number and strength of connections, especially involving DLPFC, Temporal, and PSMC, predict motor function recovery.	[127] [132]

Empirical evidence demonstrates significant correlations between these multimodal parameters. A randomized controlled study on robot-assisted gait training found that reduced qEEG parameters (DAR and PRI) in the premotor and primary motor cortices were associated with increased fNIRS-measured HbO activation in these same regions following intervention [128]. Furthermore, these neurophysiological improvements were significantly correlated with clinical scores on the Fugl-Meyer Assessment for lower limbs [128]. Another study on upper limb recovery established that a combination of fNIRS-based functional connectivity features could predict motor outcomes with high accuracy (AUC = 0.971 in the training dataset) [132].

Experimental Protocols and Methodologies

Multimodal Data Acquisition Setup

A standardized experimental setup is crucial for obtaining high-quality, synchronized qEEG-fNIRS data. The following protocol synthesizes methodologies from recent clinical studies [131] [128].

• Equipment and Montage: Use a commercially available hybrid EEG-fNIRS cap or custom-designed cap integrating both modalities. A typical configuration may include 32 EEG electrodes positioned according to the international 10-20 system and 32 fNIRS sources with 30 detectors to create approximately 90 measurement channels [131]. Optodes should provide coverage over key motor regions: primary motor cortex (M1), premotor cortex (PMC), supplementary motor area (SMA), and dorsolateral prefrontal cortex (DLPFC).

• Synchronization: Employ a common trigger system (e.g., E-Prime software) to send simultaneous event markers to both the EEG and fNIRS acquisition systems, ensuring temporal alignment of data streams [131]. Sampling rates should be set appropriately for each modality (e.g., 256 Hz for EEG, 11 Hz for fNIRS) [131].

• Paradigm Design: Implement a block-design motor imagery (MI) or motor execution (ME) task. A typical trial structure includes: (1) a visual cue (2-3 s) indicating the required movement (e.g., left-hand or right-hand grasping), (2) an execution/imagery phase (10 s), and (3) a rest interval (15-20 s) [131] [128]. Multiple trials (e.g., 30 per session) should be collected across at least two sessions.

Data Processing and Analysis Workflow

Processing synchronized qEEG-fNIRS data requires parallel pipelines that converge for integrated analysis.

• qEEG Preprocessing:

  • Filtering: Apply band-pass (e.g., 0.5-45 Hz) and notch (50/60 Hz) filters.
  • Artifact Removal: Identify and correct for ocular, cardiac, and motion artifacts using algorithms like Independent Component Analysis (ICA).
  • Spectral Analysis: Compute power spectral density for standard frequency bands (delta, theta, alpha, beta) via Fast Fourier Transform (FFT) or Wavelet analysis.
  • Parameter Calculation: Extract DAR, PRI, BSI, and coherence metrics from the cleaned spectra [126] [128].

• fNIRS Preprocessing:

  • Conversion: Convert raw light intensity signals to optical density and then to HbO and HbR concentrations using the modified Beer-Lambert law [132].
  • Artifact Correction: Use algorithms (e.g., in HOMER2 toolbox) for motion artifact detection and correction (spline interpolation method) [132].
  • Filtering: Apply a band-pass filter (e.g., 0.01-0.1 Hz) to remove physiological noise (cardiac, respiratory) and drift [132].
  • Hemodynamic Analysis: Calculate beta values (β) reflecting task-related cortical activation and determine functional connectivity between regions of interest (ROIs) [132] [128].

• Multimodal Integration and Correlation:

  • Temporal Alignment: Ensure qEEG and fNIRS data streams are aligned based on synchronization markers.
  • Neurovascular Coupling: Quantify the relationship between EEG-derived electrical activity (e.g., power in specific bands) and the fNIRS-measured hemodynamic response (HRF) [127].
  • Network Analysis: Construct multilayer brain networks where EEG and fNIRS data form separate layers interconnected through neurovascular coupling edges; calculate network metrics like the multiplex clustering coefficient [127].
  • Statistical Correlation: Perform correlation analysis (e.g., Pearson's R) between qEEG parameters (DAR, PRI) and fNIRS parameters (HbO activation, connectivity strength) across subjects and sessions, and correlate these neurophysiological changes with clinical motor scores (e.g., FMA-UE) [127] [128].

The Scientist's Toolkit: Essential Research Reagents and Materials

Successful implementation of a correlated qEEG-fNIRS study requires specific hardware, software, and assessment tools.

Table 3: Essential Materials for qEEG-fNIRS Stroke Research

Category	Item	Specification / Example	Primary Function
Hardware	Hybrid EEG-fNIRS Cap	Custom design with 32 EEG electrodes & 62 fNIRS optodes (32 sources, 30 detectors) [131]	Integrated signal acquisition from the scalp.
Signal Acquisition	EEG Amplifier	g.HIamp amplifier (g.tec) [131]	Amplification and digitization of electrical signals.
	fNIRS System	NirScan (Danyang Huichuang) [131] or LIGHTNIRS [128]	Emission and detection of near-infrared light.
Software	EEG Processing Toolbox	EEGLAB, HOMER2 (for fNIRS) [132]	Preprocessing, artifact removal, and spectral analysis.
	Statistical Software	SPSS, R Language [132]	Statistical analysis and model building (e.g., LASSO regression).
Experimental Control	Stimulus Presentation	E-Prime 3.0 [131]	Precise control and synchronization of task paradigms.
Clinical Assessment	Motor Function Scales	Fugl-Meyer Assessment (FMA-UE/FMA-LE) [131] [128]	Quantitative clinical evaluation of motor impairment.
	Activities of Daily Living	Modified Barthel Index (MBI) [131]	Assessment of functional independence.

Signaling Pathways and Neurophysiological Workflow

The correlation between qEEG and fNIRS signals is fundamentally grounded in the neurovascular coupling pathway, which links neuronal electrical activity to subsequent hemodynamic changes.

This neurovascular pathway explains the temporal relationship between the modalities: qEEG captures the immediate electrical activity of neuronal assemblies, while fNIRS detects the slower hemodynamic response that follows 4-6 seconds later [11]. In stroke, this coupling can be disrupted—a condition known as neurovascular uncoupling—making the combined assessment particularly valuable for identifying true neural activation versus compromised vascular response [127]. Successful rehabilitation interventions, such as rTMS combined with motor training, have been shown to improve neurovascular coupling levels, which is reflected in stronger correlations between qEEG and fNIRS metrics and is associated with better motor outcomes [127].

The correlated use of qEEG and fNIRS provides a powerful, multimodal framework for investigating the complex neurophysiological processes underlying motor recovery after stroke. This case study has detailed the key parameters, experimental protocols, and analytical workflows required to implement this integrated approach. The synergy between qEEG's temporal precision and fNIRS's spatial localization offers unprecedented insights into neuroplastic reorganization, enabling researchers to move beyond traditional clinical scales and develop objective, biologically grounded biomarkers of recovery. As hardware compatibility and data fusion techniques continue to advance, this multimodal paradigm holds significant promise for personalizing rehabilitation strategies, predicting individual patient outcomes, and accelerating the development of novel therapeutic interventions for stroke survivors.

The pursuit of robust neurophysiological biomarkers is paramount for advancing the prognosis and treatment of neurological disorders. This technical guide provides an in-depth examination of two prominent approaches in non-invasive brain imaging: quantitative Electroencephalography (qEEG) parameters, specifically the Power Ratio Index (PRI) and Brain Symmetry Index (BSI), and functional Near-Infrared Spectroscopy (fNIRS) measures, namely oxygenated and deoxygenated hemoglobin (HbO/HbR). Framed within the broader principles of fMRI, EEG, and fNIRS for brain research, this review synthesizes current evidence on the application of these biomarkers, detailing their physiological basis, methodologies for derivation, and prognostic utility, particularly in post-stroke motor recovery. The integration of these multimodal biomarkers, which capture complementary electrophysiological and hemodynamic facets of brain activity, presents a powerful toolkit for researchers and drug development professionals aiming to objectify patient stratification and evaluate therapeutic efficacy.

Modern brain research rests on a triad of core non-invasive neuroimaging modalities: functional Magnetic Resonance Imaging (fMRI), Electroencephalography (EEG), and functional Near-Infrared Spectroscopy (fNIRS). Each offers a unique window into brain function with complementary strengths and limitations.

• fMRI is considered the gold standard for mapping brain activity with high spatial resolution (millimeter range), providing whole-brain coverage of cortical and subcortical structures. It measures the Blood-Oxygen-Level-Dependent (BOLD) signal, an indirect correlate of neural activity based on neurovascular coupling [41] [11]. However, its low temporal resolution (∼0.3-2 Hz), high cost, immobility, and sensitivity to motion artifacts limit its use for repeated measurements and ecological validity [134] [11].

• EEG records the brain's electrical activity directly from the scalp with millisecond temporal resolution, capturing the rapid dynamics of neural communication. It is cost-effective, portable, and allows for long-term monitoring. Its primary limitation is its poor spatial resolution and difficulty in localizing the source of neural activity deep within the brain [134] [53].

• fNIRS occupies a middle ground, measuring hemodynamic responses (changes in HbO and HbR concentrations) associated with neural activity. While its spatial resolution (1-3 cm) is superior to EEG but inferior to fMRI, and its penetration is limited to the cortical surface, it offers an excellent balance of portability, tolerance to motion, and relatively good temporal resolution (often up to 10 Hz) [41] [11]. This makes it ideal for studying brain function in naturalistic settings and across diverse populations [134] [53].

The following diagram illustrates the spatiotemporal landscape of these key neuroimaging modalities.

Figure 1: Spatiotemporal Resolution of Neuroimaging Modalities. This diagram situates EEG, fNIRS, and fMRI within the landscape of brain imaging techniques, highlighting their relative strengths in temporal and spatial resolution. EEG excels in tracking fast neural dynamics, fNIRS offers a balance for superficial cortical monitoring, and fMRI provides detailed spatial maps of deep and superficial brain structures. MEG (Magnetoencephalography) and ECoG (Electrocorticography) are included for context [53] [41].

Quantitative EEG (qEEG) Biomarkers: PRI and BSI

Quantitative EEG involves the computerized processing of raw EEG data to extract objective metrics that reflect the brain's functional state. The Power Ratio Index (PRI) and Brain Symmetry Index (BSI) are two such parameters with significant prognostic value, particularly in stroke.

Power Ratio Index (PRI)

• Physiological Basis: PRI capitalizes on the shift in spectral power distribution following brain injury. Healthy, active brains exhibit dominant fast-wave activity (alpha, beta). Brain lesions, such as those from stroke, lead to neuronal dysfunction and a relative increase in slow-wave activity (delta, theta) [134].
• Calculation and Definition: PRI is defined as the ratio of the combined power in the slow-wave frequency bands (Delta: 0.5–4 Hz, Theta: 4–7 Hz) to the combined power in the fast-wave bands (Alpha: 8–12 Hz, Beta: 13–30 Hz) [134]. A higher PRI indicates a greater predominance of pathological slow waves.
• Prognostic Utility: An elevated PRI has long been recognized as a marker of poor functional outcome after stroke [134]. Studies have shown a significant inverse correlation between PRI and established clinical scales such as the Fugl-Meyer Assessment (FMA) for motor function [134]. The related Delta/Alpha Ratio (DAR) has also been inversely correlated with the Modified Rankin Scale (mRS) and cognitive scores (MMSE, MoCA) [134].

Brain Symmetry Index (BSI)

• Physiological Basis: The BSI quantifies the breakdown in normal interhemispheric balance caused by unilateral brain lesions. A healthy brain displays symmetrical electrical activity between hemispheres. A stroke creates a focus of abnormal, often slowed, activity, leading to measurable asymmetry [134].
• Calculation and Definition: The BSI is typically calculated as the mean absolute difference in power spectra between homologous hemispheres across the 1–25 Hz frequency range. Values range from 0 (perfect symmetry) to 1 (maximal asymmetry) [134].
• Prognostic Utility: A higher BSI, indicating greater asymmetry, is consistently associated with worse neurological status (e.g., higher NIHSS scores) and poorer motor recovery at follow-up periods ranging from 2 to 6 months [134]. Its correlation with upper extremity motor function (FMA-UE) is notably strong, though the link to lower extremity function may be less significant [134].

Table 1: Quantitative EEG (qEEG) Biomarkers for Prognosis

Biomarker	Physiological Basis	Calculation	Interpretation	Prognostic Correlation
Power Ratio Index (PRI)	Shift from fast-wave to slow-wave activity post-injury [134]	(PowerDelta + PowerTheta) / (PowerAlpha + PowerBeta)	Higher value = greater slow-wave activity, worse outcome [134]	Inverse correlation with FMA, mRS; predicts poor motor recovery [134]
Brain Symmetry Index (BSI)	Breakdown of interhemispheric balance due to unilateral lesion [134]	Mean absolute difference in power spectra between hemispheres (1-25 Hz) [134]	0 = perfect symmetry; 1 = maximal asymmetry [134]	Positive correlation with NIHSS; negative predictor of motor function recovery [134]

fNIRS Hemodynamic Biomarkers: HbO and HbR

fNIRS is an optical technique that measures hemodynamic changes secondary to neural activity, based on the principle of neurovascular coupling.

• Physiological Basis and Relationship to fMRI: When a brain region becomes active, it triggers a localized hemodynamic response. This leads to an increase in cerebral blood flow that delivers oxygenated blood, resulting in a rise in HbO and a subsequent decrease in HbR [41]. This is the same physiological phenomenon measured by the fMRI BOLD signal, which is inversely correlated with HbR [41]. fNIRS provides a more direct measure of these chromophores.
• Measurement Principle: fNIRS uses near-infrared light (650-950 nm) transmitted through the scalp and skull. HbO and HbR have distinct absorption spectra in this range. By measuring light attenuation at multiple wavelengths, the Modified Beer-Lambert Law allows for the calculation of concentration changes in both HbO and HbR [41] [75].
• Prognostic Utility: In stroke, fNIRS can reveal disruptions in the functional motor network. It is used to assess cortical reorganization during rehabilitation [134]. The hemodynamic response, particularly the task-evoked HbO signal, can serve as a marker for the recovery of neurovascular coupling and the integrity of the cortical motor network. Its portability allows for monitoring during real-world rehabilitation exercises, providing insights into brain function during recovery [134] [11].

The signaling pathway from neural activity to the measurable fNIRS signal is outlined below.

Figure 2: From Neural Activity to Hemodynamic Signals. This diagram depicts the neurovascular coupling pathway. Neural activity triggers a localized hemodynamic response, increasing cerebral blood flow and altering HbO/HbR concentrations. These changes are directly measured by fNIRS and are also the source of the fMRI BOLD signal, which is inversely related to HbR [134] [41].

Table 2: fNIRS Hemodynamic Biomarkers for Prognosis

Biomarker	Physiological Basis	Measurement Technique	Interpretation	Prognostic Utility
Oxygenated Hemoglobin (HbO)	Increase in regional cerebral blood flow and blood volume due to neurovascular coupling [41]	Continuous-wave fNIRS; application of Modified Beer-Lambert Law to light attenuation at 2+ wavelengths (e.g., 695, 830 nm) [75] [71]	Increase indicates area of neural activation [41] [71]	Marker of functional network integrity and neurovascular health; tracks reorganisation in motor cortex post-stroke [134] [11]
Deoxygenated Hemoglobin (HbR)	Regional oxygen consumption during neural activity [41]	Derived simultaneously with HbO from multi-wavelength fNIRS data [75] [71]	Decrease indicates area of neural activation (inverse of BOLD fMRI) [41]	Provides complementary information to HbO; its dynamics can refine assessment of hemodynamic response function [41]

Experimental Protocols and Methodologies

Protocol for Acquiring EEG Biomarkers (PRI/BSI) in Stroke

A standardized protocol for deriving PRI and BSI in a prognostic study involves several key stages [134] [135]:

• Participant Setup: Record EEG using a cap with electrodes placed according to the international 10–10 system. A sampling rate of ≥256 Hz and impedance kept below 5 kΩ is recommended.
• Data Acquisition: Record resting-state EEG for a sufficient duration (e.g., 5-10 minutes) with eyes closed to minimize artifacts. For task-based paradigms, a motor execution/observation/imagery protocol can be used.
• Preprocessing: Apply band-pass filtering (e.g., 0.5–45 Hz). Identify and remove artifacts (eye blinks, muscle activity) manually or using automated algorithms (e.g., ICA). Select artifact-free epochs for analysis.
• Power Spectral Density (PSD) Calculation: Compute the PSD for each EEG channel and frequency band (Delta, Theta, Alpha, Beta) using methods like Fast Fourier Transform (FFT) or Welch's method.
• Biomarker Calculation:
  • PRI: Sum the PSD for Delta and Theta bands; sum the PSD for Alpha and Beta bands. Divide the slow-wave sum by the fast-wave sum.
  • BSI: For each frequency band, calculate the difference in PSD between homologous left and right hemisphere channels. The BSI is the average of the absolute values of these differences across the 1–25 Hz range.

Protocol for fNIRS Hemodynamic Recording in Motor Tasks

A typical fNIRS protocol for assessing motor function includes [75] [71]:

• System Setup and Probe Placement: Use a continuous-wave fNIRS system with wavelengths (e.g., 695 nm and 830 nm). Place sources and detectors over the targeted brain regions (e.g., sensorimotor cortex, parietal cortex) with a source-detector separation of ~2.5-3.0 cm to ensure cortical sensitivity. Use a digitizer to record optode locations relative to cranial landmarks.
• Experimental Paradigm: Employ a block design. For example:
  • Resting State: 30-second baseline.
  • Task Blocks (e.g., Motor Execution): 20-second periods where the participant performs a motor task (e.g., grasping and moving a cup with the affected hand), repeated 5-10 times.
  • Interleave task blocks with 20-30 second rest periods.
• Data Preprocessing:
  • Convert raw light intensity to optical density.
  • Apply signal quality metrics and reject channels with poor signal-to-noise ratio.
  • Detect and correct for motion artifacts using algorithms (e.g., spline interpolation, wavelet-based methods).
  • Apply a band-pass filter (e.g., 0.01–0.2 Hz) to remove physiological noise (cardiac, respiratory) and drift.
  • Use the Modified Beer-Lambert Law to convert optical density changes to concentration changes of HbO and HbR.
• Data Analysis: For block designs, use a General Linear Model (GLM) with a canonical hemodynamic response function to model the expected HbO/HbR response to task blocks. Alternatively, block averaging can be used to visualize the mean hemodynamic response across all trials.

The workflow for a combined fNIRS-EEG experiment, which allows for the simultaneous capture of both biomarker types, is summarized below.

Figure 3: Workflow for a Combined fNIRS-EEG Experiment. This diagram outlines the key stages in a multimodal neuroimaging study, from setting up an integrated cap and acquiring synchronized data, to preprocessing modality-specific signals and finally extracting and fusing the biomarkers (PRI/BSI and HbO/HbR) for a comprehensive analysis [53] [71].

The Scientist's Toolkit: Essential Research Reagents and Materials

Successful acquisition and analysis of EEG and fNIRS biomarkers require a suite of specialized hardware and software tools.

Table 3: Essential Research Tools for EEG/fNIRS Biomarker Studies

Category	Item	Specification / Example	Critical Function
Core Hardware	EEG System	Amplifier with >64 channels, sampling rate ≥256 Hz, high-input impedance	Captures electrical potentials from the scalp with high temporal fidelity [134] [135]
	fNIRS System	Continuous-wave system with ≥2 wavelengths (e.g., 695 & 830 nm), sample rate ≥10 Hz	Measures light attenuation to derive hemodynamic changes [75] [71]
	Integrated Cap	Elastic cap with embedded EEG electrodes and fNIRS optodes; allows for precise co-registration	Enables simultaneous multimodal data acquisition from the same scalp locations [53] [71]
Acquisition Support	3D Digitizer	Electromagnetic or optical system (e.g., Polhemus Fastrak)	Records precise 3D locations of optodes/electrodes for accurate spatial registration to brain anatomy [71]
	Stimulation Software	Presentation or PsychoPy	Precisely controls and records the timing of experimental paradigms and triggers for data synchronization
Data Processing & Analysis	EEG Analysis Suite	EEGLAB, BrainVision Analyzer, or custom scripts in Python/MATLAB	Performs preprocessing, artifact removal, and calculation of spectral power and connectivity metrics (PRI, BSI) [134] [135]
	fNIRS Analysis Suite	Homer2, Homer3, NIRS-KIT, or custom scripts	Processes raw fNIRS signals through conversion, filtering, and statistical analysis to extract HbO/HbR responses [75]
	Multimodal Fusion Tools	Custom implementations of methods like ssmCCA	Fuses EEG and fNIRS data to identify neural activity consistently detected by both modalities [71] [136]

Integrated Analysis: The Multimodal Advantage

The combination of EEG and fNIRS is more than just simultaneous recording; it is a synergistic approach that provides a more holistic view of brain function by coupling electrophysiology and hemodynamics [134] [53].

• Methodologies for Integration: Integration can be achieved through unified hardware processors for precise synchronization or post-hoc alignment of data from separate systems [53]. Analytically, fusion methods like structured sparse multiset Canonical Correlation Analysis (ssmCCA) are used to identify latent variables that represent shared variance between the electrical (EEG) and hemodynamic (fNIRS) datasets, pinpointing brain regions where both signals consistently indicate activation [71].
• Revealing New Insights: Multimodal studies have demonstrated that activations detected by unimodal EEG and fNIRS do not always fully overlap, likely due to their different sensitivity profiles [71]. However, fusion analysis can consistently identify core regions of shared activation, such as the left inferior parietal lobe during motor tasks, providing stronger evidence for the engagement of specific neural networks like the Action Observation Network (AON) [71].
• Prognostic and Clinical Value: The combined use of PRI/BSI and HbO/HbR can offer a more accurate prediction of motor recovery after stroke than either modality alone [134]. The integrated features related to neurovascular coupling may serve as superior biomarkers for assessing the efficacy of rehabilitation interventions and for tailoring personalized treatment plans [134] [136].

The identification of sensitive and reliable biomarkers is a cornerstone of modern translational neuroscience. Quantitative EEG parameters (PRI and BSI) and fNIRS hemodynamic measures (HbO and HbR) offer powerful, complementary, and practical tools for prognostic assessment in neurological disorders. While qEEG provides direct insight into the electrical consequences of brain injury, fNIRS maps the accompanying hemodynamic changes. The integration of these modalities creates a multimodal framework that overcomes the limitations of individual techniques, yielding a more comprehensive and physiologically grounded understanding of brain (dys)function and recovery mechanisms. For researchers and drug development professionals, this combined biomarker approach holds immense promise for objectifying diagnosis, stratifying patients, monitoring disease progression, and rigorously evaluating the impact of novel therapeutics.

In contemporary neuroscience and neuropharmacology, functional Magnetic Resonance Imaging (fMRI), Electroencephalography (EEG), and functional Near-Infrared Spectroscopy (fNIRS) constitute a foundational trio of non-invasive technologies for investigating human brain function. Each modality captures distinct facets of neural activity through different biophysical principles, resulting in complementary profiles of strengths and limitations. fMRI measures brain activity indirectly by detecting blood-oxygen-level-dependent (BOLD) signals associated with cerebral blood flow and oxygenation [11]. EEG records electrical potentials generated by synchronized neuronal firing directly from the scalp surface with millisecond precision [137]. fNIRS employs near-infrared light to measure hemodynamic responses by quantifying changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentrations within cortical brain regions [35]. Understanding these fundamental measurement principles is critical for selecting the optimal technology for specific research objectives in basic neuroscience and applied drug development.

The increasing complexity of brain research demands a strategic approach to neurotechnology selection. No single modality can comprehensively capture the brain's multifaceted operations, which span multiple temporal and spatial scales. This whitepaper establishes an evidence-based framework for matching imaging modality strengths to specific research goals, experimental paradigms, and participant populations. By synthesizing current technical capabilities and methodological considerations, we provide researchers and drug development professionals with a structured decision-making tool to optimize study design, enhance data quality, and maximize the validity of neuroscientific findings.

Technical Specifications and Comparative Analysis

Quantitative Comparison of Core Modalities

Table 1: Comprehensive technical specification comparison across fMRI, fNIRS, and EEG

Parameter	fMRI	fNIRS	EEG
Spatial Resolution	High (mm to sub-mm) [11]	Moderate (1-3 cm) [35]	Low (source localization challenges) [35]
Temporal Resolution	Moderate (seconds) [11]	High (0.1-10 Hz) [35]	Very High (millisecond range) [35]
Penetration Depth	Whole brain (cortical & subcortical) [11]	Superficial cortex (1-2.5 cm) [35]	Cortical surface [137]
Portability	Low (fixed scanner) [2]	High (wearable systems) [2]	High (wearable systems) [137]
Motion Tolerance	Low (requires complete stillness) [2]	High (resistant to motion artifacts) [35]	Moderate (susceptible to artifacts) [137]
Participant Limitations	Claustrophobia, metal implants [2]	Minimal; suitable for all populations [2]	Minimal; suitable for all populations [137]
Environmental Constraints	Magnetic shielding required [11]	Minimal; naturalistic settings possible [97]	Shielded rooms preferred for high quality [137]
Operational Costs	Very High [2]	Moderate [2]	Low to Moderate [137]
Primary Signal	Hemodynamic (BOLD response) [11]	Hemodynamic (HbO/HbR concentration) [35]	Electrophysiological (neuronal potentials) [137]
Best Application Examples	Detailed spatial mapping of deep brain structures, network connectivity [11]	Naturalistic studies, clinical populations, longitudinal monitoring [97]	Event-related potentials, brain-computer interfaces, seizure detection [137]

Operational Characteristics and Practical Implementation

Table 2: Operational characteristics and practical implementation considerations

Consideration	fMRI	fNIRS	EEG
Setup Time	Extensive (30-60 minutes)	Moderate (10-20 minutes)	Moderate (15-30 minutes, including gel)
Operator Expertise	Advanced technical training required	Moderate technical knowledge	Moderate technical knowledge
Participant Preparation	Metal screening, safety briefing	Hair/skin preparation minimal	Scalp abrasion, conductive gel application
Data Acquisition Complexity	High (sequence optimization required)	Moderate (source-detector positioning)	Moderate (impedance checking)
Real-time Feedback Capability	Limited (processing delays)	Good (minimal delay)	Excellent (instantaneous)
Acoustic Noise	Significant (requires hearing protection) [2]	Silent	Silent
Subject Comfort	Low (confinement, noise) [2]	High (lightweight, silent) [97]	Moderate (cap pressure, gel)

Modality-Specific Experimental Protocols

fMRI Protocol for Pharmacological Challenge Studies

Objective: To map whole-brain neural activity changes in response to psychoactive compounds with high spatial precision.

Materials and Equipment:

• 3T or higher MRI scanner with standard head coil
• Compatible automated injection system for contrast agents or study compounds
• MR-compatible physiologic monitoring system (cardiac, respiratory)
• Presentation system for visual/auditory stimuli
• Response recording devices (fiber-optic button boxes)

Procedure:

• Participant Screening: Conduct comprehensive metal screening and contraindication assessment. Verify participant eligibility for pharmacological intervention.
• Scanner Preparation: Calibrate scanner, ensure compatibility with injection system, and position participant with comfortable but firm head immobilization.
• Structural Acquisition: Collect high-resolution T1-weighted anatomical images (e.g., MPRAGE sequence) for subsequent functional data co-registration.
• Baseline Functional Scan: Acquire BOLD fMRI data during resting state or task performance to establish pre-intervention baseline.
• Compound Administration: Administer study compound or placebo via calibrated infusion pump according to randomized, double-blind protocol.
• Post-Intervention Scanning: Conduct repeated functional acquisitions at predetermined timepoints to capture hemodynamic response to compound.
• Data Quality Assurance: Monitor head motion in real-time with automated tools; reacquire sequences if motion exceeds predetermined threshold (e.g., >2mm translation).

Analysis Considerations: Utilize statistical parametric mapping (SPM) or FMRIB Software Library (FSL) for preprocessing (realignment, normalization, smoothing) and general linear model (GLM) analysis. Account for hemodynamic response function delay when modeling drug effects [11].

Simultaneous EEG-fNIRS Protocol for Cognitive Motivation Studies

Objective: To investigate the temporal dynamics and spatial localization of neural processes during motivated decision-making tasks.

Materials and Equipment:

• Integrated EEG-fNIRS system with synchronized data acquisition
• fNIRS cap with sources and detectors positioned over prefrontal and parietal cortices
• EEG cap with active or passive electrodes following international 10-20 system
• Stimulus presentation computer with precise timing capability
• Behavioral response recording interface

Procedure:

• System Setup: Position combined EEG-fNIRS cap ensuring proper optode and electrode contact. For EEG, achieve electrode impedances <10 kΩ. For fNIRS, verify signal quality from all channels.
• Signal Quality Assessment: Collect brief resting-state data to verify both EEG and fNIRS signal quality before experimental tasks.
• Experimental Paradigm: Implement a validated cognitive task such as the visual decision-making paradigm where participants view scenes and decide whether to remember them [138].
• Data Synchronization: Implement hardware or software synchronization between EEG, fNIRS, and stimulus presentation systems using TTL pulses or network synchronization protocols.
• Simultaneous Data Acquisition: Record continuous EEG and fNIRS data throughout task performance, including appropriate baseline periods.
• Behavioral Data Collection: Record participant responses and reaction times with millisecond precision.
• Post-Task Procedures: Document sensor locations using 3D digitization for accurate spatial registration.

Analysis Considerations:

• EEG Analysis: Apply preprocessing steps (filtering, artifact removal, independent component analysis) followed by event-related potential (ERP) analysis focused on components of interest (e.g., P300, FRN) [138].
• fNIRS Analysis: Process raw intensity signals, convert to optical density, then to HbO and HbR concentrations using modified Beer-Lambert law. Apply hemodynamic response function and general linear modeling [138].
• Data Integration: Temporally align EEG ERP components with fNIRS hemodynamic responses to investigate neurovascular coupling dynamics during decision-making processes.

fNIRS Protocol for Naturalistic Cognitive Assessment

Objective: To measure prefrontal cortex hemodynamic responses during ecologically valid tasks in naturalistic settings.

Materials and Equipment:

• Portable, wearable fNIRS system with wireless data transmission capability
• Flexible fNIRS probe sets targeting dorsolateral and medial prefrontal cortex regions
• Portable computer or tablet for task administration
• Optional motion capture system for movement synchronization

Procedure:

• Probe Placement: Position fNIRS sources and detectors over prefrontal regions using anatomical landmarks (nasion, inion) or EEG 10-20 system reference (Fp1, Fp2, F7, F8, etc.).
• Signal Quality Verification: Check signal-to-noise ratio for all channels and adjust probe contact as needed.
• Baseline Recording: Collect 5-10 minutes of resting-state data with eyes open or closed condition.
• Task Administration: Implement naturalistic paradigms such as walking while performing cognitive tasks, interactive gaming environments, or real-world problem-solving activities [97].
• Data Recording: Collect continuous fNIRS data throughout task performance with event markers synchronized to task conditions.
• Environmental Monitoring: Document environmental conditions that may influence physiological signals (temperature, ambient light).
• Data Export: Transfer wireless data to secure storage for offline analysis.

Analysis Considerations: Utilize motion artifact correction algorithms (e.g., wavelet-based, accelerometer-based), bandpass filtering to separate physiological noise (cardiac, respiratory), and general linear model analysis with appropriate regressors for task conditions. For game-based learning studies, focus on HbO changes in prefrontal cortex as indicator of cognitive load [139].

Research Reagent Solutions and Essential Materials

Table 3: Essential research materials and their applications in neuroimaging studies

Material/Equipment	Primary Function	Application Examples	Technical Notes
fNIRS Optodes	Emit and detect near-infrared light through scalp and skull	Measuring cortical hemodynamics during cognitive tasks [35]	Source-detector distance (typically 3cm) determines penetration depth; flexible arrays improve comfort
EEG Electrodes	Record electrical potentials from scalp surface	Event-related potential studies, brain-computer interfaces [137]	Ag/AgCl electrodes provide optimal signal; impedance should be maintained below 10 kΩ
Conductive Gel/Electrolyte	Facilitate electrical signal transmission between scalp and electrodes	Improving signal quality in EEG recordings [137]	Saline-based solutions suitable for short studies; gel electrolytes provide better stability for longer recordings
3D Digitizer	Precisely map sensor locations on head surface	Co-registration of fNIRS/EEG data with anatomical MRI [2]	Essential for accurate spatial localization and group-level analysis
MRI-Compatible Stimulus Presentation System	Deliver visual/auditory stimuli within scanner environment	fMRI studies of sensory processing and cognitive tasks [11]	Must be MR-compatible (non-magnetic) and project via waveguide or fiber optic connection
Motion Tracking System	Monitor and quantify head movement	Motion artifact correction in fNIRS and EEG [33]	Accelerometer-based systems particularly valuable for naturalistic studies
Pharmacological Injection System	Precisely administer compounds during scanning	Drug challenge studies in fMRI [11]	Must be MR-compatible with precise flow rate control; typically placed outside scanner room
Signal Processing Software (e.g., Homer2, SPM, EEGLAB)	Preprocess, analyze, and visualize neuroimaging data	All modalities require specialized processing pipelines [33]	Open-source platforms promote reproducibility; standardized pipelines enhance comparability

Integrated Framework for Modality Selection

Decision Matrix for Modality Selection

Table 4: Evidence-based modality selection guide for common research scenarios

Research Objective	Recommended Primary Modality	Complementary Modality	Rationale	Key Methodological Considerations
Mapping deep brain structures	fMRI [11]	-	Superior spatial resolution for subcortical regions	Requires participant compliance with stillness; contraindicated with metal implants
Studying rapid neural dynamics (<100ms)	EEG [137]	fNIRS	Millisecond temporal resolution for electrical events	Limited spatial precision; sensitive to muscle artifacts
Naturalistic studies with movement	fNIRS [97]	EEG	High motion tolerance and portability	Superficial cortical coverage only; limited penetration depth
Clinical populations (children, elderly)	fNIRS [2]	EEG	High tolerance; minimal participant burden	Appropriate for cortical measurements only
Neurovascular coupling mechanisms	Simultaneous EEG-fNIRS [138]	-	Direct correlation of electrical and hemodynamic responses	Technical challenges in hardware integration and data synchronization
Resting-state network analysis	fMRI [11]	fNIRS	Whole-brain coverage for connectivity mapping	Requires advanced analytical approaches (ICA, graph theory)
Brain-computer interface applications	EEG [137]	fNIRS	Real-time signal processing capabilities	Balance between speed (EEG) and spatial specificity (fNIRS) needed
Pharmacological intervention studies	fMRI [11]	EEG/fNIRS	Whole-brain assessment of drug effects	Challenges in controlling for non-neural vascular effects

Advanced Applications and Multimodal Integration

The integration of multiple neuroimaging modalities represents a powerful approach to overcome the limitations of individual techniques. Simultaneous EEG-fNIRS recording, for instance, combines excellent temporal resolution from EEG with improved spatial localization from fNIRS, providing a more comprehensive picture of brain function [138]. This integrated approach is particularly valuable for studying complex cognitive processes where both rapid neural dynamics and their metabolic consequences are of interest.

In motor imagery classification for brain-computer interfaces, deep learning approaches that fuse EEG and fNIRS signals have demonstrated superior performance compared to unimodal systems, achieving accuracy improvements of up to 3.78% over state-of-the-art methods [39]. The complementary nature of these signals—electrical activity from EEG and hemodynamic responses from fNIRS—provides a richer feature set for decoding neural states.

For drug development applications, combined fMRI-fNIRS approaches enable both precise spatial localization of drug effects and longitudinal monitoring of treatment response. This is particularly relevant for neurological and psychiatric disorders such as schizophrenia, where fNIRS has emerged as a valuable tool for exploring neural mechanisms and monitoring treatment efficacy [35].

The strategic selection of neuroimaging modalities should be driven by specific research questions rather than technological availability alone. fMRI remains indispensable for studies requiring detailed spatial mapping of deep brain structures, while EEG provides unparalleled temporal resolution for capturing rapid neural dynamics. fNIRS offers a unique balance of portability, motion tolerance, and moderate spatial resolution that makes it particularly valuable for naturalistic studies and clinical populations.

As the field advances, methodological standardization becomes increasingly important for reproducibility. Recent initiatives like the fNIRS Reproducibility Study Hub (FRESH) have demonstrated that while experienced researchers generally converge on similar findings using different analysis pipelines, clearer methodological standards are needed to enhance reliability across the field [33]. Similarly, efforts to develop comprehensive glossaries and best practice guidelines contribute to more consistent implementation and reporting across studies [97].

For drug development professionals, this evidence-based framework supports informed decision-making in neurotechnology selection throughout the drug development pipeline—from early target engagement studies to late-phase treatment monitoring. By aligning modality strengths with specific research objectives, neuroscientists and clinical researchers can optimize study design, enhance data quality, and accelerate discoveries in brain science and therapeutic development.

Conclusion

fMRI, EEG, and fNIRS are not mutually exclusive but are powerful complementary tools in the neuroscience arsenal. fMRI remains unparalleled for detailed anatomical mapping of deep brain structures, while EEG excels at capturing the brain's rapid electrical dynamics. fNIRS emerges as a highly versatile modality, bridging the gap with its portability, robustness to motion, and applicability in diverse populations and real-world settings. The future of brain research lies in strategic multimodal integration, leveraging combined fNIRS-EEG or fMRI-fNIRS to overcome individual limitations and gain a more holistic understanding of brain function. For biomedical and clinical research, this implies more precise diagnostic biomarkers, enhanced monitoring of therapeutic interventions, and the development of sophisticated brain-computer interfaces. Overcoming challenges related to standardization, reproducibility, and data fusion will be pivotal in translating these technological advances into tangible clinical outcomes.

References

• 1. FMRI hemodynamic response function (HRF) as a novel ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7865013/]
• 2. Comparison of fNIRS with other neuroimaging modalities [https://www.artinis.com/blogpost-all/comparison-fnirs-vs-fmri]
• 3. Chapter 3: The Hemodynamic Response Function (HRF) [https://andysbrainbook.readthedocs.io/en/latest/fMRI_Short_Course/Statistics/03_Stats_HRF_Overview.html]
• 4. Functional Neuroimaging: Fundamental Principles and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4757157/]
• 5. Study design in fMRI: Basic principles [https://www.sciencedirect.com/science/article/pii/S0278262605001752]
• 6. Concurrent fNIRS and EEG for Brain Function Investigation [https://pmc.ncbi.nlm.nih.gov/articles/PMC9371171/]
• 7. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOoq89aTG-y98t4H0-QNRiLT_xlUOhjCduUhBrkOevhKrej0qPZoi]
• 8. What You Need to Know About MRI With Contrast [https://americanhealthimaging.com/mri-with-contrast/]
• 9. What is BOLD fMRI? [https://www.asfnr.org/what-is-bold-fmri]
• 10. The Underpinnings of the BOLD Functional Magnetic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6741096/]
• 11. Applications and advances of combined fMRI-fNIRs ... [https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2025.1542075/full]
• 12. Laminar multi-contrast fMRI at 7T allows differentiation of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12254000/]
• 13. Consensus recommendations for clinical functional MRI ... [https://apertureneuro.org/article/128149-consensus-recommendations-for-clinical-functional-mri-applied-to-language-mapping]
• 14. A Quantitative Comparison of Simultaneous BOLD fMRI ... [https://www.sciencedirect.com/science/article/pii/S1053811902912279]
• 15. EEG: Basic Principles and Overview [https://www.openanesthesia.org/keywords/eeg-basic-principles-and-overview/]
• 16. Basic EEG Electrophysiology [https://www.learningeeg.com/basic-eeg-electrophysiology]
• 17. vs fNIRS : EEG Non-Invasive Brain Comparing ... - iMotions Imaging [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/]
• 18. Electroencephalography Signal Processing [https://pmc.ncbi.nlm.nih.gov/articles/PMC10385593/]
• 19. Comparing structure–function relationships in brain ... [https://www.nature.com/articles/s41598-024-79817-x]
• 20. Quantitative EEG Analysis for Automated Detection ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4342615/]
• 21. Methods of EEG Signal Features Extraction Using Linear ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4045570/]
• 22. A Quantitative EEG Toolbox for the MNI Neuroinformatics ... [https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2020.00033/full]
• 23. Functional near-infrared spectroscopy [https://en.wikipedia.org/wiki/Functional_near-infrared_spectroscopy]
• 24. Functional near Infrared Spectroscopy - an overview [https://www.sciencedirect.com/topics/materials-science/functional-near-infrared-spectroscopy]
• 25. Functional Near-infrared Spectroscopy (fNIRS) of Brain ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC3294084/]
• 26. Investigating the neurophysiological effects of active noise ... [https://www.sciencedirect.com/science/article/abs/pii/S0378595525002254]
• 27. Investigating the neurophysiological effects of active noise ... [https://pubmed.ncbi.nlm.nih.gov/40882583/]
• 28. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOoqNk3pQ5Js-b0yLiA4S-Wb8msxXlkrqDeMaI9HwWNiLkRbn0HjL]
• 29. Preprocessing functional near-infrared spectroscopy ( fNIRS ) data [https://mne.tools/stable/auto_tutorials/preprocessing/70_fnirs_processing.html]
• 30. Simultaneous functional near-infrared spectroscopy and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5566595/]
• 31. EEG Normal Waveforms - StatPearls - NCBI Bookshelf [https://www.ncbi.nlm.nih.gov/books/NBK539805/]
• 32. Data Processing in Functional Near-Infrared Spectroscopy ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8151801/]
• 33. fNIRS reproducibility varies with data quality, analysis ... [https://www.nature.com/articles/s42003-025-08412-1]
• 34. Applications and advances of combined fMRI-fNIRs ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11958174/]
• 35. Functional Near-Infrared Spectroscopy in Schizophrenia ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12413766/]
• 36. Hybrid EEG Feature Learning Method for Cross-Session ... [https://www.mdpi.com/2076-3425/15/8/805]
• 37. Functional near-infrared spectroscopy for the assessment ... [https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2025.1524806/full]
• 38. Riemannian geometry boosts functional near-infrared ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12523035/]
• 39. EEG–fNIRS signal integration for motor imagery ... [https://www.sciencedirect.com/science/article/pii/S2772528625000299]
• 40. Real-World fNIRS-Based Brain-Computer Interfaces [https://arxiv.org/html/2505.10536v1]
• 41. Functional Magnetic Resonance Imaging and Functional Near ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5563305/]
• 42. Towards Neuroscience of the Everyday World (NEW) using ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7943029/]
• 43. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOor1YB0jtdHG0gKoc145VMn3r-0zhZMRmJU3z25_tXB03IjgWZhT]
• 44. Wearable fNIRS platform for dense sampling and precision ... [https://www.nature.com/articles/s41746-025-01690-3]
• 45. Unfolding the real-time neural mechanisms in addiction [https://www.sciencedirect.com/science/article/pii/S2772392522000438]
• 46. Reproducibility of fNIRS within Subject for Visual and Motor ... [https://www.sciencedirect.com/science/article/pii/S1053811925004951]
• 47. A deep neural network for adaptive spatial smoothing of ... [https://www.frontiersin.org/journals/neuroimaging/articles/10.3389/fnimg.2025.1554769/full]
• 48. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOop2Nk-fEeD4Wpsnk04AIt7nasry0nZx809r1PgAIsM5HOk8qr-3]
• 49. How to design real-world functional near-infrared ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12618016/]
• 50. DeepCor: denoising fMRI data with contrastive autoencoders [https://www.nature.com/articles/s41592-025-02967-x]
• 51. Guidelines for Presenting Neuroimaging Analyses [https://www.nisox.org/research/GuideFor/]
• 52. ISMRM25 - fMRI Analysis: Software, Pipelines & Methods [https://www.ismrm.org/25/program-files/D-123.htm]
• 53. A Cross-Disciplinary Review of the fNIRS-EEG Dual-Modality ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11506513/]
• 54. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOorLTcKZD6ezrFvAgeooByJZ4ytGMTuPAuvFc-ass8qFpyyUph5b]
• 55. Visualization of the Dynamic Brain Activation Pattern ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9688554/]
• 56. Event-related potential: An overview - PMC - PubMed Central [https://pmc.ncbi.nlm.nih.gov/articles/PMC3016705/]
• 57. Event-related potential [https://en.wikipedia.org/wiki/Event-related_potential]
• 58. EEG dataset for energy data visualizations [https://www.sciencedirect.com/science/article/pii/S2352340923009654]
• 59. Opportunities for use of neuroimaging in de-risking drug ... [https://www.nature.com/articles/s41386-024-01970-8]
• 60. Functional Near-Infrared Spectroscopy and Its Clinical ... [https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2020.00724/full]
• 61. Functional near-infrared spectroscopy reveals brain activity ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9231602/]
• 62. A Narrative Review on Clinical Applications of fNIRS - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC7573058/]
• 63. Applications of Functional Near-Infrared Spectroscopy (fNIRS ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11188690/]
• 64. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOopT6nlDoFUeZYhOuwxrZmVXeZB67mS2quIu-yWG78NZKuWMJLxJ]
• 65. fNIRS-EEG BCIs for Motor Rehabilitation: A Review - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC10740927/]
• 66. NIRS Neurorehabilitation | Research & Development [https://www.shimadzu.com/research_and_development/technology_branding/neurotech/rehabilitation.html]
• 67. A review on functional near-infrared spectroscopy and ... [https://www.sciencedirect.com/science/article/pii/S2590093521000084]
• 68. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOorhPm2O9m1Jp-SSyS8n48UDonv6Sv2TlYZkEkk_iB7wkTFnl3Af]
• 69. Multi-Modal Integration of EEG-fNIRS for Brain-Computer ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5651279/]
• 70. Multimodal Measurement: Integration of EEG & fNIRS - TMSi [https://www.tmsi.artinis.com/blog/eeg-fnirs-integration]
• 71. Simultaneous multimodal fNIRS-EEG recordings reveal ... [https://www.nature.com/articles/s41598-023-31609-5]
• 72. Characterizing the Action-Observation Network Through ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7930074/]
• 73. An fNIRS Study of Brain Lateralization During Observation ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8825352/]
• 74. Brain activity characteristics of RGB stimulus: an EEG study [https://www.nature.com/articles/s41598-023-46450-z]
• 75. Best practices for fNIRS publications - PMC - PubMed Central [https://pmc.ncbi.nlm.nih.gov/articles/PMC7793571/]
• 76. Multimodal assessment of the spatial correspondence ... [https://www.nature.com/articles/s41598-023-29123-9]
• 77. Spatial correspondence of cortical activity measured with ... [https://www.sciencedirect.com/science/article/pii/S1053811924000648]
• 78. -based fMRI of continuous-wave validation of supplementary... fNIRS [https://www.nature.com/articles/s41598-022-06519-7?error=cookies_not_supported&code=e4aca4e6-a008-4e06-8136-452e5227f48c]
• 79. of fMRI Measurements During a Naturalistic Task Validation fNIRS [https://www.jove.com/t/52116/fmri-validation-of-fnirs-measurements-during-a-naturalistic-task]
• 80. A Cross-Disciplinary Review of the fNIRS-EEG Dual ... [https://www.mdpi.com/2076-3425/14/10/1022]
• 81. Brain-Based Binary Communication Using Spatiotemporal ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2020.00113/full]
• 82. Common non-physiological (external) EEG artifacts - TMSi [https://www.tmsi.artinis.com/blog/common-non-physiological-external-eeg-artifacts]
• 83. Motion artifacts in functional near-infrared spectroscopy [https://pmc.ncbi.nlm.nih.gov/articles/PMC3762942/]
• 84. Adaptive filtering of physiological noises in fNIRS data [https://biomedical-engineering-online.biomedcentral.com/articles/10.1186/s12938-018-0613-2]
• 85. Motion Artifacts Correction from Single-Channel EEG and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9102309/]
• 86. Optimizing spatial specificity and signal quality in fNIRS [https://www.frontiersin.org/journals/neuroergonomics/articles/10.3389/fnrgo.2024.1286586/full]
• 87. Optimizing spatial specificity and signal quality in fNIRS [https://pmc.ncbi.nlm.nih.gov/articles/PMC11188482/]
• 88. High-density multidistance fNIRS enhances detection of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12412631/]
• 89. Ultra high density imaging arrays in diffuse optical ... [https://www.nature.com/articles/s41598-025-85858-7]
• 90. Functional near Infrared Spectroscopy - an overview [https://www.sciencedirect.com/topics/neuroscience/functional-near-infrared-spectroscopy]
• 91. Functional near-infrared spectroscopy as a tool to assess ... [https://sajce.co.za/index.php/sajce/article/view/1138/2042]
• 92. Motion artifacts removal and evaluation techniques for ... [https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2022.878750/full]
• 93. EEG Artifacts: Types, Detection, and Removal Techniques ... [https://www.bitbrain.com/blog/eeg-artifacts]
• 94. Removal of Artifacts from EEG Signals: A Review - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC6427454/]
• 95. Color-dependent changes in humans during a verbal ... [https://www.nature.com/articles/s41598-021-88059-0]
• 96. Systematic Review of Advanced Algorithms for Brain Mapping ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12650610/]
• 97. Mapping Out Brain Studies, Beyond the Lab [https://www.bu.edu/neurophotonics/2025/10/28/mapping-out-brain-studies-beyond-the-lab/]
• 98. FRESH: fNIRS REproducibility Study Hub [https://openfnirs.org/data/fresh/]
• 99. Characterizing reproducibility of cerebral hemodynamic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8901194/]
• 100. Standardization and Open Science Committee [https://fnirs.org/about-us/committees/standardization-committee/]
• 101. Spatial co-registration of functional near-infrared spectroscopy ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5581202/]
• 102. EEG-fNIRS Cookbook [https://www.brainproducts.com/support-resources/eeg-fnirs-cookbook/]
• 103. High-Density Multi-Distance fNIRS Enhances Detection of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11952576/]
• 104. NIRS Caps & Probes | NIRScaps fNIRS data collection [https://nirx.net/nirscap]
• 105. Spatial registration for functional near-infrared ... [https://www.sciencedirect.com/science/article/pii/S1053811913007805]
• 106. Developing customized NIRS-EEG for infant sleep research [https://pmc.ncbi.nlm.nih.gov/articles/PMC10519625/]
• 107. Systematic evaluation of fMRI data-processing pipelines for ... [https://www.nature.com/articles/s41467-024-48781-5]
• 108. The GREENBEAN checklist for reporting studies evaluating ... [https://www.sciencedirect.com/science/article/pii/S1388245725006297]
• 109. Evaluating the effects of multimodal EEG-fNIRS ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12425317/]
• 110. Comparing the carbon footprint of fMRI data processing and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12319921/]
• 111. The application of fNIRS in studies on occupational workload [https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2025.1560605/full]
• 112. Multimodal fusion of EEG-fNIRS: a mutual information-based ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7984774/]
• 113. Simultaneous EEG and fNIRS recordings for semantic ... [https://www.nature.com/articles/s41597-025-04967-0]
• 114. Cross-Modal Transfer Learning From EEG to Functional Near ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9021696/]
• 115. Enhancing Performance of a Hybrid EEG-fNIRS System ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2017.00462/full]
• 116. fMRI-based validation of continuous-wave fNIRS ... [https://www.nature.com/articles/s41598-022-06519-7]
• 117. fNIRS - fMRI | Concurrent NIRS MRI Recording ... [https://nirx.net/fnirs-fmri]
• 118. Estimating Brain Activity with High Spatial and Temporal ... [https://arxiv.org/html/2510.09415v1]
• 119. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOora0HxM3cwucBT2Ctl08iw-VrOXDS6N0Sm8usKiETmwgT2KjdiE]
• 120. Exploring Imagined Movement for Brain–Computer Interface ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12468679/]
• 121. Deep learning for inner speech recognition [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2025.1668935/full]
• 122. Functional near-infrared spectroscopy (fNIRS) detects brain ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12489781/]
• 123. Feature Extraction and Classification Methods for Hybrid ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6032997/]
• 124. Decoding of four movement directions using hybrid NIRS ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4009438/]
• 125. Simultaneous EEG and fNIRS recordings for semantic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11993746/]
• 126. The Role of Quantitative EEG in the Diagnosis of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7175442/]
• 127. EEG-fNIRS multilayer brain network analysis revealed ... [https://pubmed.ncbi.nlm.nih.gov/40531644/]
• 128. Neural mechanisms underlying the improvement of gait ... [https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01656-2]
• 129. Quantitative Electroencephalography - an overview [https://www.sciencedirect.com/topics/biochemistry-genetics-and-molecular-biology/quantitative-electroencephalography]
• 130. Advances in Functional Near-Infrared Spectroscopy [https://pubmed.ncbi.nlm.nih.gov/40796273/]
• 131. HEFMI-ICH: a hybrid EEG-fNIRS motor imagery dataset for ... [https://www.nature.com/articles/s41597-025-06100-7]
• 132. Predicting upper limb motor dysfunction after ischemic stroke [https://pmc.ncbi.nlm.nih.gov/articles/PMC12150403/]
• 133. Predictive Nomogram for Post-Stroke Motor Dysfunction ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2025.1526455/full]
• 134. Current implications of EEG and fNIRS as functional ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11629311/]
• 135. Functional connectivity in EEG: a multiclass classification ... [https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2025.1550581/full]
• 136. Investigating the interaction between EEG and fNIRS [https://www.sciencedirect.com/science/article/pii/S1877750324002096]
• 137. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOopHFg90XN6oGR0j_o7MaoYJWi_a_JpIycxkrmjvkKIpdwubYWPA]
• 138. Simultaneous EEG-fNIRS study of visual cognitive processing [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0325017]
• 139. A Scoping Review of Functional Near-Infrared ... [https://arxiv.org/html/2411.02650v2]