Designing Effective Motor Imagery Protocols for Hybrid EEG-fNIRS Brain-Computer Interfaces in Clinical Research

Abstract

This article provides a comprehensive guide for researchers and biomedical professionals on designing and implementing motor imagery (MI) task protocols for hybrid electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) systems. It explores the foundational neuroscience behind MI, detailing how it activates cortical motor networks without physical movement. The content covers standardized methodological approaches for synchronized EEG-fNIRS data acquisition, participant instruction, and paradigm design for both upper and lower limbs. It further addresses critical troubleshooting aspects, such as mitigating participant non-responsiveness and managing physiological artifacts. Finally, the article evaluates validation frameworks and performance benchmarks for these multimodal protocols, highlighting their significant translational potential in neurorehabilitation, drug development, and the creation of precision rehabilitation systems for stroke and other neurological disorders.

The Neuroscience of Motor Imagery: Building a Theoretical Foundation for Hybrid EEG-fNIRS

Motor imagery (MI) is a cognitive process defined as the mental simulation of a movement without any overt motor output or muscle activation [1]. Within electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) research, MI is not a monolithic concept but comprises distinct psychological strategies, primarily kinesthetic motor imagery (KMI) and visual motor imagery (VMI) [2]. The precise definition and experimental isolation of these strategies are critical for designing robust brain-computer interface (BCI) protocols and interpreting neural data accurately. KMI involves the proprioceptive sensation of movement—the feeling of muscle contraction, joint forces, and body position [2]. In contrast, VMI is the visual representation of the movement, as if watching oneself perform an action from a first-person (internal) or third-person (external) perspective [2]. This article delineates these two forms of MI, provides validated protocols for their application in multimodal neuroimaging studies, and discusses their implications for therapeutic development and scientific research.

Theoretical Foundation and Neural Correlates

The theoretical underpinning of MI lies in the shared neural representations between actual movement execution and its mental simulation. It is established that MI activates brain regions largely overlapping with those involved in motor execution, including premotor and supplementary motor areas, the primary motor cortex, and somatosensory regions [1]. This overlapping activation forms the basis for its application in rehabilitation and BCI, as it can promote use-dependent cortical plasticity [1].

Crucially, while KMI and VMI share common neural substrates, they recruit distinct networks with different weights. KMI strongly engages the motor and somatosensory cortices, leveraging the dorsal processing stream, and is considered more effective for BCIs designed to mimic actual movement [2]. VMI, however, relies more heavily on the visual cortex and occipito-parietal regions, engaging the ventral visual stream [2]. A key study demonstrated that connectivity measures within the motor network, rather than local activity alone, could classify KMI and VMI with accuracies exceeding 98%, confirming their distinct neurophysiological signatures [2]. Specifically, degree centrality in the left primary somatosensory cortex (S1) in the alpha band was significantly higher during KMI than during VMI [2].

Table 1: Comparative Characteristics of Kinesthetic and Visual Motor Imageries

Feature	Kinesthetic Motor Imagery (KMI)	Visual Motor Imagery (VMI)
Core Definition	Mental simulation of the sensation of movement; proprioceptive and somatosensory imagination [2].	Mental visualization of the movement; creating a visual image of the action [2].
Primary Modality	Somatosensory, Proprioceptive	Visual
Dominant Neural Correlates	Motor cortex, Somatosensory cortex (S1), Premotor areas [2].	Visual cortex, Occipito-parietal regions [2].
Optimal Perspective	Internal perspective (first-person) [3].	Can be internal (first-person) or external (third-person).
Temporal Dynamics	Real-time, synchronous with actual movement duration [3].	Can be real-time or non-real-time.
BCI Suitability	Higher; more closely resembles motor execution networks [2] [4].	Lower; may introduce visual network "noise".

Experimental Protocols for EEG/fNIRS Research

Standardized protocols are essential for ensuring the fidelity of KMI and VMI tasks and the validity of the resulting neural data. The following protocols synthesize best practices from the literature, including the PETTLEP (Physical, Environment, Task, Timing, Learning, Emotion, Perspective) model [3].

Participant Screening and Familiarization

• Imagery Ability Assessment: Prior to the main experiment, screen participants for their imagery ability using standardized questionnaires like the Kinesthetic and Visual Imagery Questionnaire (KVIQ-10). This ensures participants can perform both types of imagery effectively. A minimum score (e.g., >2 on a 1-5 scale) is recommended for inclusion [2].
• Familiarization Session: Conduct a session before data acquisition to calibrate the participant's understanding and performance of MI. This is crucial for KMI.
  • For KMI Calibration: Use a dynamometer and a stress ball. Guide participants through:
    • Repeated maximal force exertions with the dynamometer.
    • Equivalent force applications using a stress ball.
    • Grip training at a standardized rhythm (e.g., one contraction per second).
  • This procedure reinforces the tactile and force-related sensations, enhancing the kinesthetic vividness of the subsequent MI [5].

Core Motor Imagery Experimental Protocol

The following workflow, adapted from established paradigms [5] [2], details a single trial for a left-hand/right-hand MI task suitable for both EEG and fNIRS.

Protocol Specifications:

• Participant Position: Participants should sit comfortably in a chair, with hands resting on thighs or a table. The physical position should mimic the real task environment as much as possible (PETTLEP principle) [3].
• Stimuli Presentation: Visual cues should be displayed on a monitor. Auditory cues (e.g., a 200 ms beep) signal the transition between phases [5].
• MI Task: During the 10-second execution phase, participants are instructed to imagine a grasping movement with the cued hand at a rate of one imagined grasp per second, synchronized with the pre-training rhythm [5].
• Task Blocks: A typical session includes at least two consecutive blocks, each containing 15 trials per hand (30 trials total). Sufficient rest (e.g., 1-2 minutes) should be provided between blocks to prevent fatigue [5].
• Instruction for KMI vs. VMI: The verbal instructions given to participants are the primary method for isolating the imagery type.
  • For KMI: "Imagine the sensation of grasping the ball with your [left/right] hand, feeling the muscle contraction and the pressure against your palm without actually moving." [2]
  • For VMI: "Visualize yourself grasping the ball with your [left/right] hand, as if you are watching a movie from your own eyes. Focus on the visual appearance of the movement." [2]

Table 2: Optimal Motor Imagery Training Parameters from Systematic Review

Parameter	Recommended Practice	Rationale
Session Duration	~17 minutes per session [3].	Balances sufficient data acquisition with user fatigue.
Frequency	~3 times per week [3].	Allows for consolidation between sessions.
Trials per Session	~34 trials [3].	Provides ample data while maintaining attention.
Total Intervention	~13 sessions (178 total minutes) [3].	Sufficient duration to induce measurable neuroplastic changes.
Instruction Mode	Acoustic, detailed, and standardized [3].	Ensures consistency across participants and trials.
Eyes During MI	Closed [3].	Reduces visual interference and enhances focus on internal simulation.

The Scientist's Toolkit: Research Reagent Solutions

This section details essential materials and tools required for setting up and executing a multimodal EEG-fNIRS MI study.

Table 3: Essential Research Reagents and Materials for EEG/fNIRS MI Studies

Item / Solution	Specification / Function	Example & Notes
Multimodal Neuroimaging System	Synchronized acquisition of electrophysiological (EEG) and hemodynamic (fNIRS) signals.	Custom-integrated systems using, e.g., a g.HIamp amplifier (EEG) and a NirScan system (fNIRS) [5].
Hybrid EEG-fNIRS Cap	Integrated cap with co-located electrodes and optodes for comprehensive cortical coverage.	Custom-designed cap (e.g., Model M, 54-58 cm) with 32 EEG electrodes and 32 fNIRS sources & 30 detectors (yielding 90 channels) [5].
Stimulus Presentation Software	Precisely timed delivery of visual and auditory cues, sending event markers to the acquisition systems.	E-Prime 3.0, PsychToolbox, Presentation [5].
Calibration Tools	Enhances kinesthetic vividness and standardizes the imagined movement.	Dynamometer (for measuring force) and stress ball (for tactile feedback) [5].
Imagery Ability Questionnaire	Screens participants for their innate capacity for visual and kinesthetic imagery.	Kinesthetic and Visual Imagery Questionnaire (KVIQ-10) [2].
Clinical Assessment Scales	For patient studies, quantifies motor impairment and functional independence.	Fugl-Meyer Assessment for Upper Extremities (FMA-UE), Modified Barthel Index (MBI) [5].

Data Analysis and Interpretation

Key Neural Features for Discrimination

The success of a protocol differentiating KMI from VMI relies on analyzing appropriate neural features.

• EEG Signatures: The most common feature is Event-Related Desynchronization (ERD) in the sensorimotor rhythm (mu and beta bands; ~8-30 Hz) over the contralateral sensorimotor cortex [4] [6]. KMI typically produces a stronger and more focused ERD in these rhythms compared to VMI.
• fNIRS Signatures: fNIRS measures changes in oxygenated (HbO) and deoxygenated (HbR) hemoglobin. A well-executed MI task should evoke an increase in HbO in the contralateral motor cortex [5] [6]. KMI is expected to elicit a more robust hemodynamic response in the primary sensorimotor cortex than VMI.
• Connectivity Measures: As demonstrated by Kim et al., functional connectivity measures (e.g., degree centrality) in the alpha and beta bands can be more effective than power analysis alone for classifying KMI and VMI, with accuracies over 98% [2].

Advanced Analysis: Transfer Learning

Given that Motor Execution (ME) shares neural mechanisms with MI, transfer learning is a promising approach to improve MI-BCI performance, especially for users struggling with pure MI. A recent study showed that a classification model trained on ME data could achieve statistically similar accuracy on MI tasks (65.93% for ME-trained model vs. 65.93% for within-task MI) [4]. Furthermore, combining ME and MI datasets for training significantly improved MI classification accuracy to 69.21%, demonstrating the practical utility of this approach for building more user-friendly BCIs [4].

The rigorous distinction between kinesthetic and visual motor imagery is a cornerstone of methodological validity in EEG and fNIRS research. Kinesthetic MI, characterized by the internal simulation of proprioceptive and somatosensory sensations, more effectively recruits the motor network crucial for rehabilitation and BCI applications. The protocols and tools outlined here provide a framework for researchers to design experiments that accurately isolate these cognitive states, thereby enhancing data quality, improving BCI classification performance, and ultimately contributing to the development of more effective neurorehabilitation therapies. Future work should focus on refining real-time classification of imagery type to provide adaptive feedback, further personalizing and optimizing training protocols for both healthy users and clinical populations.

A foundational principle in modern cognitive neuroscience is the concept of "functional equivalence," which proposes that the mental simulation of an action, known as motor imagery (MI), and its actual motor execution (ME) share overlapping neural substrates [7] [8]. This theory, supported by decades of neuroimaging research, suggests that imagining a movement activates a similar fronto-parietal network to performing it, albeit without overt motor output [8]. The core of this shared network includes the premotor cortex, parietal areas, and supplementary motor area (SMA). Notably, quantitative meta-analyses have revealed that MI and ME consistently co-activate subcortical structures like the cerebellum and putamen, a pattern distinct from action observation [8].

Understanding these shared correlates is crucial for developing non-invasive Brain-Computer Interfaces (BCIs) and neurorehabilitation protocols. The advent of mobile neuroimaging technologies, particularly electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS), has enabled the study of MI and ME during dynamic, real-world behaviors, providing new insights into their temporal and spatial neural dynamics [7] [9]. This application note details the experimental protocols and analytical frameworks for investigating these shared neural processes, with a focus on practical guidance for researchers in the field.

Table 1: Meta-Analysis Findings of Consistent Activations across Motor Execution (ME), Motor Imagery (MI), and Action Observation (AO).

Brain Region	ME	MI	AO	Notes
Premotor Cortex	✓	✓	✓	Core shared node of the action simulation network
Parietal Cortex	✓	✓	✓	Involved in motor planning and spatial integration
Supplementary Motor Area (SMA)	✓	✓	(~)	Strongly associated with MI and ME
Primary Motor Cortex (M1)	✓	(~)	(~)	More consistently and strongly activated in ME
Putamen	✓	✓		Subcortical structure, differentiation point for AO
Cerebellum	✓	✓		Subcortical structure, differentiation point for AO
Somatosensory Cortex	✓	(~)		Activated during ME, less consistent during MI

Table 2: Characteristic EEG Power Modulations during Motor Execution and Motor Imagery.

Frequency Band	Modulation Type	Associated Cognitive/Motor Process	Typical Topography
Alpha (8-12 Hz)	Decrease (ERD)	Activation of sensorimotor areas	Contralateral central (C3/C4)
Beta (13-30 Hz)	Decrease (ERD)	Movement preparation and execution	Contralateral central (C3/C4)
Beta (13-30 Hz)	Increase (ERS)	Post-movement reset, inhibition of motor processes	Contralateral central (C3/C4)

Experimental Protocols for MI and ME Research

Mobile EEG Protocol for Whole-Body Actions

This protocol is adapted from a study investigating the neural correlates of walking imagery and execution using mobile EEG [7].

• Objective: To compare neural oscillatory dynamics during the execution and imagery of a multi-step walking task.
• Participants: Healthy adults with no known neurological or motor disorders.
• Equipment: Mobile EEG system with a sufficient number of electrodes for coverage over sensorimotor cortices (e.g., 32-channel setup). A safe, clear walking path is required.
• Task Design:
  • Walking Execution (WE) Condition: Participants physically walk six steps along the path at a self-paced rhythm.
  • Walking Imagery (WI) Condition: While standing, participants vividly imagine walking the same number of steps, focusing on the kinesthetic sensation, without producing any overt movement.
  • Control Condition (e.g., Mental Counting): Participants count silently from one to six to control for general cognitive engagement unrelated to motor processes.
  • The conditions should be presented in a randomized or counterbalanced order across multiple trials.
• Data Analysis:
  • Preprocess EEG data (filtering, artifact removal).
  • Calculate time-frequency representations (TFRs) for each condition.
  • Extract event-related desynchronization (ERD) and synchronization (ERS) in the alpha and beta bands over sensorimotor electrodes.
  • Statistically compare the power modulations between WE, WI, and control conditions, focusing on the early and late phases of the action.

Hybrid EEG-fNIRS Protocol for Upper-Limb MI

This protocol outlines a standardized paradigm for collecting synchronized neural data, suitable for BCI and rehabilitation research [5] [10].

• Objective: To acquire high-quality, multimodal neural signals during left- and right-hand motor imagery.
• Participants: Can include both healthy controls and clinical populations (e.g., patients with intracerebral hemorrhage for specific datasets).
• Equipment:
  • Synchronized EEG and fNIRS systems.
  • A custom cap integrating EEG electrodes and fNIRS optodes over the sensorimotor cortex.
  • A display screen for presenting visual cues.
• Calibration Phase: To enhance MI vividness, include a grip strength calibration using a dynamometer or stress ball. This reinforces the tactile and temporal aspects of the movement [5].
• Task Design (Trial Structure):
  • Baseline Recording (2 min): Record signals with eyes closed and eyes open.
  • Visual Cue (2 s): A left- or right-pointing arrow indicates the required MI task.
  • Execution/MI Phase (10 s): Participants either execute or imagine a grasping movement with the cued hand at a rate of one per second. An auditory beep marks the start.
  • Inter-Trial Interval (15 s): A rest period with a blank screen [5].
  • A session should contain at least 30 trials (15 per hand), with multiple sessions and rest breaks to avoid fatigue.
• Data Acquisition:
  • Set EEG sampling rate to ≥256 Hz and fNIRS sampling rate to ~11 Hz.
  • Use event markers from stimulus presentation software (e.g., E-Prime) to synchronize both systems [5].

Invasive Recording Protocol for Effector Specificity

This advanced protocol is based on single-neuron recordings in humans and provides high-resolution data on motor coding [11].

• Objective: To investigate the representation of different body parts in the motor cortex (MC) and posterior parietal cortex (PPC).
• Participants: Typically a unique cohort (e.g., tetraplegic participants with implanted arrays as part of a clinical trial).
• Equipment: Utah microelectrode arrays implanted in the hand knob of the precentral gyrus (MC) and the superior parietal lobule (PPC).
• Task Design:
  • Participants attempt or imagine moving various effectors (e.g., eyes, head, shoulders, wrists, thumbs, legs, ankles).
  • A center-out paradigm is used where a visual cue specifies the effector and the direction of movement.
  • Neural activity is recorded during the movement attempt and during an inter-trial rest period.
• Data Analysis:
  • Spike-sort the recorded signals.
  • Analyze population and single-neuron tuning to different effectors.
  • Compare the strength and selectivity of representations between MC and PPC.

Signaling Pathways and Experimental Workflows

From Instruction to Cortical Activation: A Multimodal BCI Pathway

Hierarchical Organization of Motor Control

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Equipment and Software for Motor Imagery and Execution Research.

Item Name	Category	Specification/Model Example	Primary Function in Research
Mobile EEG System	Hardware	ActiCHamp (Brain Products), g.HIamp (g.tec)	Records electrical brain activity with high temporal resolution during dynamic movements.
fNIRS System	Hardware	NIRScout (NIRx), NirScan	Measures hemodynamic changes (HbO2/HbR) in the cortex, offering good spatial resolution.
Hybrid EEG-fNIRS Cap	Hardware	Custom-designed cap (e.g., EasyCap base)	Integrates electrodes and optodes for simultaneous multimodal data acquisition.
Stimulus Presentation Software	Software	E-Prime, PsychoPy	Prescribes experimental paradigms, displays cues, and sends synchronization triggers.
Common Spatial Pattern (CSP) Algorithm	Software/Algorithm	MATLAB, Python (scikit-learn)	Spatial filter that maximizes variance between two classes of EEG signals (e.g., left vs. right hand MI).
Deep Learning Decoders	Software/Algorithm	EEGNet, FBCNet, HA-FuseNet	Classifies MI tasks from raw or preprocessed EEG signals using neural networks.
g.tec HIamp Amplifier	Hardware	g.HIamp (g.tec)	High-quality signal amplification for EEG data acquisition in BCI applications.
Event-Related (De)Synchronization Analysis	Software/Algorithm	MATLAB FieldTrip, MNE-Python	Quantifies power decreases (ERD) and increases (ERS) in specific frequency bands related to an event.

Functional near-infrared spectroscopy (fNIRS) and electroencephalography (EEG) represent two pivotal non-invasive neuroimaging techniques for studying real-time brain activity. While fNIRS monitors cerebral hemodynamic responses by measuring changes in oxygenated and deoxygenated hemoglobin using near-infrared light, EEG detects electrical potentials generated by synchronized neuronal firing via electrodes placed on the scalp [12]. Their complementary characteristics—fNIRS offers improved spatial resolution and tolerance to movement artifacts, while EEG provides millisecond-scale temporal precision—make their integration particularly powerful for investigating complex brain functions [12] [13]. This integration is especially valuable in motor imagery (MI) research, where understanding both the rapid neuronal dynamics and their localized cortical activation patterns is essential for advancing brain-computer interfaces (BCIs) and neurorehabilitation strategies [14].

The fundamental synergy between these modalities stems from their measurement of different but related physiological processes. EEG captures direct neural electrical activity with high temporal resolution but limited spatial accuracy, whereas fNIRS measures the indirect hemodynamic response with better spatial localization but slower temporal evolution [12]. This complementary relationship enables researchers to overcome the limitations of either modality used independently, particularly for studying spatiotemporally complex processes like motor imagery, where both rapid signal changes and specific regional activations are of interest [15].

Fundamental Principles and Complementarity

Physiological Basis and Measurement Principles

Electroencephalography (EEG) measures the brain's electrical activity via electrodes placed on the scalp. These sensors detect voltage changes resulting from the synchronized firing of cortical neurons, primarily pyramidal cells. When these neurons fire synchronously, their postsynaptic potentials generate electrical fields that can be detected at the scalp surface [12]. One of EEG's greatest strengths is its exceptional temporal resolution—it captures neural dynamics on a millisecond scale, making it ideal for analyzing rapid cognitive processes like attention, sensory perception, and motor planning [12]. However, EEG's spatial resolution is limited due to the blurring and dispersion of electrical signals as they pass through the skull, scalp, and other tissues [15].

Functional Near-Infrared Spectroscopy (fNIRS) monitors cerebral hemodynamic responses by measuring how blood flow in the brain shifts in response to mental tasks or stimuli. It detects changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) using near-infrared light, typically at wavelengths between 650-925 nm, where biological tissues are relatively transparent while hemoglobin species exhibit distinct absorption spectra [16]. fNIRS offers better spatial resolution than EEG, particularly for surface cortical areas, as light scattering, though substantial, provides more localized information compared to electrical field dispersion [12]. However, fNIRS has slower temporal resolution (on the scale of seconds) as it reflects the indirect hemodynamic response to neural activity, which evolves over several seconds due to neurovascular coupling [12].

Spatiotemporal Resolution Comparison

Table 1: Comparative characteristics of EEG and fNIRS

Feature	EEG	fNIRS
What It Measures	Electrical activity of neurons	Hemodynamic response (blood oxygenation levels)
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, but limited to cortex)
Depth of Measurement	Cortical surface	Outer cortex (~1-2.5 cm deep)
Sensitivity to Motion	High - susceptible to movement artifacts	Low - more tolerant to subject movement
Portability	High - lightweight & wireless systems available	High - often used in mobile & wearable formats
Best Use Cases	Fast cognitive tasks, ERP studies, sleep research	Naturalistic studies, child development, motor rehab

Table 2: Signal characteristics and relationship to neural activity

Parameter	EEG	fNIRS
Direct/Indirect Neural Measure	Direct measurement of electrical activity	Indirect via neurovascular coupling
Primary Signal Origin	Synchronized pyramidal cell activity	Hemodynamic response to neural activity
Latency Following Neural Activation	Immediate (milliseconds)	Delayed (1-6 seconds)
Key Observable Phenomena in MI	Event-Related Desynchronization/Synchronization (ERD/ERS)	Hemodynamic response in motor cortex
Typical Motor Imagery Responses	μ-rhythm (8-13 Hz) suppression	Increased HbO in contralateral motor areas

Neurovascular Coupling: The Biological Bridge

The relationship between EEG and fNIRS signals is fundamentally governed by neurovascular coupling (NVC), the process where neuronal activity triggers subsequent hemodynamic changes [16]. During increased neural activity, there is an initial rise in oxygen consumption, followed by a compensatory increase in cerebral blood flow that overshoots metabolic demands. This results in a characteristic hemodynamic response function with a rapid increase in oxygenated hemoglobin (HbO) and a more modest decrease in deoxygenated hemoglobin (HbR) [16]. The coupling between electrical and hemodynamic activities enables the complementary use of EEG and fNIRS, with EEG capturing the initial neural activation and fNIRS reflecting the subsequent metabolic response.

Experimental Design and Protocols for Motor Imagery Research

Motor Imagery Paradigm Design

Well-designed motor imagery paradigms are crucial for eliciting robust and interpretable brain signals. A standardized MI experimental protocol typically includes:

Participant Preparation and Calibration: Before data acquisition, participants should undergo a grip strength calibration procedure to enhance MI vividness. This may include repeated maximal force exertions using a dynamometer, equivalent force applications using a stress ball, and grip training at a rate of one contraction per second [5]. This preparatory phase reinforces the tactile and force-related aspects of the grasping movement, standardizing its temporal rhythm and improving consistency across trials.

Trial Structure: A typical MI trial structure follows this sequence [5]:

• Visual cue presentation (2 s): A directional arrow (left/right) indicates the required MI task.
• Execution phase (10 s): A fixation cross appears while participants perform kinesthetic MI of the cued hand movement at approximately one imagined grasp per second.
• Inter-trial interval (15 s): A blank screen indicates the rest period to allow hemodynamic responses to return to baseline.

Session Organization: Each session should include at least 15 trials per condition (e.g., left-hand vs. right-hand MI), with multiple sessions conducted consecutively. Sufficient inter-session rest intervals (typically 5-10 minutes) should be implemented to mitigate fatigue effects and ensure data quality [5].

Simultaneous EEG-fNIRS Data Acquisition

Equipment Setup: For hybrid EEG-fNIRS acquisition, a 32-channel EEG configuration with a g.HIamp amplifier (sampling rate: 256 Hz) combined with a continuous-wave fNIRS system like NirScan (sampling rate: 11 Hz) provides effective multimodal recording [5]. A custom-designed hybrid EEG-fNIRS cap with pre-defined compatible openings ensures proper sensor placement without interference.

Sensor Placement: Both systems typically use the international 10-20 system for electrode/optode placement. The hybrid cap should strategically position 32 EEG electrodes for comprehensive cortical coverage while deploying fNIRS optodes (e.g., 32 sources and 30 detectors) in an optimized geometric matrix to create approximately 90 measurement channels through source-detector pairing at controlled separation distances (typically 3 cm) [5]. This configuration enables hemodynamic monitoring across prefrontal, motor, and association cortices simultaneously with electrophysiological recording.

Synchronization: Temporal synchronization between modalities is critical and can be achieved using event markers transmitted from experimental presentation software (e.g., E-Prime 3.0) that simultaneously trigger both recording systems during experimental paradigms [5].

Diagram 1: Motor imagery experimental workflow

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Essential equipment and materials for hybrid EEG-fNIRS research

Item	Function/Purpose	Specification Guidelines
EEG Amplifier System	Records electrical brain activity	32+ channels; sampling rate ≥256 Hz; g.HIamp or equivalent
fNIRS System	Measures hemodynamic responses	Continuous-wave system; dual-wavelength; NirScan or equivalent
Hybrid EEG-fNIRS Cap	Simultaneous sensor placement	Custom design with 32 EEG electrodes + 62 fNIRS optodes (32 sources, 30 detectors)
Experimental Presentation Software	Paradigm delivery & synchronization	E-Prime, PsychToolbox, or equivalent with trigger capability
Grip Strength Measurement	MI task calibration	Dynamometer and stress ball for force calibration
Data Synchronization Unit	Temporal alignment of multimodal data	TTL pulses or parallel port synchronization between systems
Electrode Gel	EEG signal conduction	Standard electrolyte gel for optimal electrode-scalp interface
3D Digitizer	Sensor position registration	Precise localization of EEG electrodes & fNIRS optodes on scalp

Data Processing and Fusion Methodologies

Preprocessing Pipelines

EEG Preprocessing: Raw EEG data requires specific preprocessing steps before analysis [17]:

• Downsampling to 128 Hz to reduce computational load
• Band-pass filtering (8-25 Hz) to isolate μ-band and low-β band activity relevant to motor imagery
• Re-referencing to common average reference
• Artifact removal (ocular, cardiac, muscle) using independent component analysis (ICA) or regression techniques

fNIRS Preprocessing: fNIRS signals undergo different preprocessing [18]:

• Conversion of raw light intensity to optical density
• Filtering with a band-pass filter (0.01-0.2 Hz) to remove cardiac, respiratory, and very low-frequency drift components
• Motion artifact correction using wavelet-based or moving average approaches
• Conversion to hemoglobin concentration changes using the modified Beer-Lambert law

Multimodal Fusion Strategies

Three primary fusion approaches have been developed for integrating EEG and fNIRS data:

Early-Stage Fusion: This approach combines raw or minimally processed data from both modalities before feature extraction. A Y-shaped neural network architecture has demonstrated significant advantages for early fusion, with studies showing substantially higher performance compared to middle-stage and late-stage fusion (average accuracy of 76.21% in left-right hand MI classification) [17]. Early fusion preserves the maximum information content from both modalities but requires careful handling of their different temporal resolutions and physiological origins.

Middle-Stage (Feature-Level) Fusion: This method involves extracting features separately from each modality then combining them for classification. Common EEG features for MI include band power changes (especially μ-rhythm ERD/ERS), common spatial patterns (CSP), and time-frequency representations. Typical fNIRS features include mean HbO/HbR concentrations, slope, variance, and peak values during tasks [14]. The combined feature set is then used for classification with algorithms like LDA, SVM, or neural networks.

Late-Stage (Decision-Level) Fusion: In this approach, each modality is processed independently through complete pipelines including classification, with final decisions combined at the end via voting schemes or weighted averaging. While less complex than other fusion methods, late fusion may fail to capture important cross-modal interactions [19].

Diagram 2: Multimodal data fusion strategies

Advanced Fusion Frameworks

Recent advances in multimodal fusion include representation learning models that capture both shared and modality-specific information. The EEG-fNIRS Representation learning Model (EFRM) is pre-trained on large-scale unlabeled datasets (approximately 1250 hours of brain signals from 918 participants) and fine-tuned for specific tasks, demonstrating competitive performance even with minimal labeled data [20]. This approach leverages masked autoencoders for modality-specific feature extraction combined with contrastive learning to identify shared representations between simultaneously recorded EEG and fNIRS signals.

Applications in Motor Imagery and Neurorehabilitation

Clinical Applications in Stroke Rehabilitation

Motor imagery-based BCIs have emerged as a transformative approach for stroke rehabilitation, leveraging neuroplasticity to facilitate motor network reorganization through closed-loop feedback mechanisms [5]. The hybrid EEG-fNIRS approach is particularly valuable for intracerebral hemorrhage (ICH) rehabilitation, where conventional therapies often yield suboptimal results due to disrupted corticospinal pathways [5]. The HEFMI-ICH dataset, comprising recordings from 17 normal subjects and 20 ICH patients during standardized left-right hand motor imagery tasks, provides a valuable resource for developing and validating algorithms tailored to this population [5].

In clinical applications, fNIRS-EEG BCIs can provide real-time feedback during rehabilitation, helping patients with severe motor impairments to access their motor system and facilitate recovery across all stages of motor recovery [14]. Studies have demonstrated that active motor training through MI-BCIs enhances activity in the primary motor cortex, promoting functional recovery through neuroplastic mechanisms [14].

Performance Advantages of Hybrid Systems

Research consistently demonstrates the superiority of hybrid EEG-fNIRS systems over unimodal approaches:

Table 4: Performance comparison of unimodal vs. hybrid BCI systems in motor imagery classification

Modality	Average Classification Accuracy	Advantages	Limitations
EEG-only	~65% [17]	Millisecond temporal resolution; Direct neural activity measurement	Low spatial resolution; Sensitive to artifacts
fNIRS-only	~57% (HbO) [17]	Better spatial localization; Motion-tolerant	Slow hemodynamic response; Indirect neural measure
Hybrid EEG-fNIRS	76.21% [17]	Combines temporal & spatial advantages; Enhanced information content	Increased system complexity; Requires specialized fusion methods

The performance improvement of approximately 10-20% in hybrid systems demonstrates the significant advantage of combining complementary information from both modalities [5] [17]. This enhanced classification accuracy is particularly important for clinical applications, where reliable detection of motor intention is crucial for effective rehabilitation.

Implementation Protocols and Best Practices

Protocol for Hybrid System Setup

Step 1: Equipment Preparation

• Select compatible EEG and fNIRS systems with synchronization capabilities
• Use integrated caps with predefined openings for both modalities to ensure consistent placement
• Verify trigger connection between experimental presentation computer and both acquisition systems

Step 2: Participant Preparation

• Explain the motor imagery task and ensure participant understanding
• Conduct grip strength calibration using dynamometer and stress ball
• Measure head circumference and select appropriate cap size
• Prepare scalp surface (light abrasion if needed) for optimal EEG signal quality
• Apply electrode gel for EEG electrodes ensuring proper impedance (<10 kΩ)
• Verify fNIRS optode-scalp contact using signal quality metrics

Step 3: System Synchronization

• Implement temporal synchronization using TTL pulses or shared clock systems
• Verify synchronization with test triggers before data collection
• Record synchronization events for post-hoc verification

Data Quality Assurance Protocol

Real-Time Quality Metrics:

• EEG: Monitor impedance values, signal variance, and spectral characteristics
• fNIRS: Assess signal-to-noise ratio, detector saturation, and physiological signal presence

Post-Acquisition Quality Checks:

• Verify temporal alignment of EEG and fNIRS data using synchronization markers
• Check for motion artifacts in both modalities
• Ensure task-related responses are present (ERD/ERS in EEG, HbO changes in fNIRS)

Recommendations for Specific Research Scenarios

Controlled Laboratory Studies: For highly controlled environments with minimal movement, high-density EEG (64+ channels) combined with moderate-density fNIRS (30+ channels) provides optimal spatial sampling. Emphasis should be placed on precise sensor localization using 3D digitizers and co-registration with anatomical images when available [15].

Naturalistic and Clinical Settings: For studies involving patients or real-world environments, prioritize robustness over density. Use 16-32 EEG channels focused on motor areas combined with 15-20 fNIRS channels covering similar regions. Implement robust artifact removal algorithms and consider shorter experimental sessions to maintain participant engagement and data quality [14].

Longitudinal Studies: For research involving multiple sessions with the same participants, develop precise cap placement procedures using anatomical landmarks and measurement systems. Consider individual headcasts or 3D-printed caps for perfect reproducibility across sessions [18].

The integration of EEG and fNIRS technologies represents a powerful approach for motor imagery research and applications, particularly in the realm of neurorehabilitation. By combining EEG's millisecond temporal resolution with fNIRS's superior spatial localization, researchers can overcome the fundamental limitations of either modality used independently. The complementary nature of these signals—direct electrical activity and indirect hemodynamic responses—provides a more comprehensive picture of brain function during motor imagery tasks.

The protocols and methodologies outlined in this document provide a framework for implementing hybrid EEG-fNIRS systems in both research and clinical settings. As technology advances, particularly in the domains of wearable systems, real-time processing, and sophisticated fusion algorithms, the potential for these integrated approaches to transform motor rehabilitation and brain-computer interface applications continues to grow. The future of motor imagery research lies in leveraging the unique strengths of each modality while developing increasingly sophisticated methods for their integration, ultimately leading to more effective and personalized interventions for patients with motor impairments.

Neurovascular coupling (NVC) is the fundamental biological process that creates a tight temporal and regional linkage between neuronal activity and subsequent changes in cerebral blood flow (CBF). This mechanism ensures that active brain regions promptly receive the necessary oxygen and glucose to meet metabolic demands, forming the physiological basis for functional brain imaging techniques that rely on hemodynamic responses, such as functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS) [21] [22]. The brain, while representing only 2% of body weight, consumes over 20% of the body's oxygen and glucose at rest, with almost all adenosine triphosphate production dependent on oxidative glucose metabolism. This high energy demand, coupled with limited intracellular energy storage capacity, makes precise neurovascular coupling essential for normal brain function [21].

The neurovascular unit comprises three major components: vascular smooth muscle cells, neurons, and astrocyte glial cells, which work in concert to regulate local blood flow [21]. When neurons become activated, they initiate a complex signaling cascade that ultimately leads to the dilation of parenchymal arterioles, significantly increasing local blood flow. This process, known as functional hyperemia, typically results in a 4-fold greater increase in CBF relative to the increase in ATP consumption, representing a substantial overcompensation that ensures adequate oxygen supply to active neural tissue [21]. This oversupply of oxygenated blood to activated brain regions provides the foundation for blood-oxygen-level-dependent (BOLD) contrast mechanisms used in various neuroimaging modalities [21].

Understanding neurovascular coupling is particularly crucial in the context of motor imagery (MI) research, where researchers aim to decode imagined movements through their associated neural and hemodynamic signatures. Motor imagery shares similar neural mechanisms with actual movement execution (ME) and movement observation (MO), activating comparable motor regions despite the absence of physical movement [4] [23]. This parallel activation pattern makes NVC particularly relevant for developing brain-computer interfaces (BCIs) that can accurately interpret intention from brain activity alone.

Physiological Mechanisms and Signaling Pathways

Anatomical Structure of the Cerebral Vasculature

The cerebral vasculature is uniquely organized to support precise regulation of blood flow. The internal carotid and vertebral arteries serve as the primary conduit vessels, delivering blood to the brain, with the internal carotid arteries transmitting approximately 70% of total CBF in healthy individuals [21]. These large arteries branch extensively, forming the circle of Willis—a redundant anastomotic safety net designed to maintain perfusion if one or more extracranial cerebral arteries become disrupted. Pial arteries wrap the brain surface within the pia mater before penetrating into the cortex, where they become parenchymal arteries completely enveloped by astrocytic end-feet [21].

A critical distinction exists in how different segments of the cerebrovascular tree are regulated. Extracranial arteries, large cerebral arteries, and pial arteries on the brain surface are richly innervated by "extrinsic" perivascular neurons from both sympathetic and parasympathetic origins. In contrast, parenchymal arterioles are primarily regulated by "intrinsic" factors within the brain parenchyma, mediated directly by neuronal activation and astrocytic modulation [21]. This differential innervation means that parenchymal arteries exhibit greater basal tone and respond differently to neurotransmitters compared to their upstream counterparts, allowing for precise local control of blood flow that directly matches neuronal activity.

Cellular Signaling Mechanisms

The process of neurovascular coupling involves sophisticated communication between neurons, astrocytes, and vascular smooth muscle cells. When neurons fire action potentials and engage in synaptic transmission, they release neurotransmitters including glutamate, which initiates the neurovascular coupling response [21]. The signaling mechanism unfolds through several parallel pathways:

• Neuronal Nitric Oxide Pathway: Glutamate released from presynaptic terminals activates N-methyl-D-aspartate (NMDA) receptors on postsynaptic neurons. This activation stimulates neuronal nitric oxide synthase (nNOS), producing nitric oxide (NO) that diffuses to nearby parenchymal arterioles acting as a potent vasodilator [21].
• Astrocyte-Mediated Pathways: Astrocytes respond to synaptic glutamate through metabotropic glutamate receptors, activating a cascading pathway involving the production of arachidonic acid. This intermediate then produces both epoxyeicosatrienoic acid (EET) and prostaglandins, which serve to dilate parenchymal arterioles [21].
• Metabolic Feedback Signals: When neuronal ATP levels decrease, adenosine is released directly from neurons to act as a vasodilator, providing a metabolic feedback mechanism that regulates blood flow based on energy demand [21].
• GABA Interneuron Integration: GABAergic interneurons play a crucial integrating role between glutamatergic pyramidal cells and microvessels. When these interneurons fire, they release various vasoactive substances including NO, acetylcholine, neuropeptide-Y, and vasoactive intestinal peptide (VIP), creating complex, fine-tuned regulation of local cerebral blood flow [21].

The following diagram illustrates the coordinated signaling pathways between neural activity and vascular response in neurovascular coupling:

Figure 1: Neurovascular Coupling Signaling Pathways. This diagram illustrates the coordinated cellular mechanisms linking neuronal activity to vascular responses. Key pathways include astrocyte-mediated production of vasoactive compounds, neuronal nitric oxide release, and metabolic feedback signals.

Quantitative Benchmarks in Motor Imagery Research

Performance Metrics Across Modalities

Motor imagery BCI research has established quantitative benchmarks that demonstrate both the capabilities and limitations of current technologies. The classification accuracy of motor imagery tasks varies significantly based on the modality used, the specific tasks performed, and individual user characteristics. The following table summarizes key performance metrics from recent motor imagery studies:

Table 1: Classification Accuracy Across Motor Tasks and Modalities

Task Type	Modality	Classification Accuracy	Notes	Source
Motor Execution (ME)	EEG	67.05%	Within-task classification	[4]
Motor Imagery (MI)	EEG	65.93%	Within-task classification	[4]
Motor Observation (MO)	EEG	73.16%	Within-task classification	[4]
ME to MI Transfer	EEG	65.93%	ME-trained model tested on MI data	[4]
MO to MI Transfer	EEG	60.82%	MO-trained model tested on MI data	[4]
ME + MI Combined	EEG	69.21%	Outperformed within-task accuracy	[4]
Public Datasets (Average)	EEG	66.53%	Meta-analysis of 861 sessions	[23]
High-Quality Dataset (2-class)	EEG	85.32%	Using EEGNet algorithm	[24]
High-Quality Dataset (3-class)	EEG	76.90%	Using DeepConvNet algorithm	[24]

The data reveals several important patterns. First, motor observation tasks generally yield higher classification accuracy compared to motor execution and imagery, likely due to more consistent neural patterns across participants [4]. Second, transfer learning approaches show promise—models trained on motor execution data can achieve statistically similar accuracy on motor imagery tasks compared to within-task training, suggesting shared neural mechanisms [4]. Third, combining datasets from different task types (e.g., ME and MI) can improve classification performance beyond within-task approaches, highlighting the value of multi-task paradigms for enhancing BCI robustness [4].

BCI Performance Distribution and User Variability

A significant challenge in motor imagery BCI research is the substantial variability in user performance. The phenomenon of "BCI illiteracy" or "BCI poor performers" refers to individuals who struggle to achieve sufficient control proficiency for effective BCI use [23]. Meta-analyses of public datasets indicate that approximately 36.27% of users fall into the poor performer category, with classification accuracy below 70% [23]. This distribution highlights the critical need for improved protocols and signal processing techniques that can accommodate user variability.

Research demonstrates that transfer learning approaches particularly benefit poor performers. Among low performers with within-task accuracy of 70% or less, 90% showed improvement when using models trained on motor execution data, and 76.2% improved with models trained on motor observation data [4]. This suggests that alternative training paradigms may help overcome the challenges faced by BCI poor performers.

Experimental Protocols for Motor Imagery Research

Comprehensive Motor Imagery Protocol

Well-designed experimental protocols are essential for collecting high-quality, reproducible data in motor imagery research. Based on best practices identified across multiple studies, the following protocol provides a framework for conducting motor imagery experiments with concurrent EEG and fNIRS recording:

Table 2: Standardized Motor Imagery Experimental Protocol

Protocol Component	Specifications	Duration	Purpose
Participant Preparation	EEG cap placement, fNIRS probe arrangement, signal quality check	30-45 minutes	Ensure proper equipment setup and signal quality
Pre-experiment Baseline	Eye-open rest, eye-closed rest	60 seconds each	Establish baseline brain activity
Trial Structure	Fixation cross, visual/auditory cue, motor imagery period, rest	7.5 seconds total	Standardized trial sequence
Cue Presentation	Visual (arrow, text, or video) + auditory instruction	1.5 seconds	Clear task instruction
Motor Imagery Period	Kinaesthetic motor imagery of specified movement	4 seconds	Core experimental condition
Inter-trial Interval	Fixation cross display	2 seconds	Neural response return to baseline
Tasks	Left hand grasping, right hand grasping, foot hooking	Counterbalanced	Multiple movement representations
Trials per Session	40 trials per class (2-class) or 60 trials (3-class)	~35-48 minutes	Sufficient data for classification
Session Schedule	3 sessions on different days	Multi-day	Account for inter-session variability

This protocol incorporates evidence-based recommendations from motor imagery research. The timing parameters align with findings that typical motor imagery trials include pre-rest (2.38s), imagination ready (1.64s), imagination (4.26s), and post-rest (3.38s) periods, with imagination periods ranging from 1-10 seconds across studies [23]. The recommendation for multiple sessions across different days addresses the significant inter-session variability in EEG signals and helps develop more robust classification models [24].

Task-to-Task Transfer Learning Protocol

Recent research demonstrates that transfer learning between different motor tasks can enhance BCI performance and reduce calibration time. The following specialized protocol enables researchers to implement task-to-task transfer learning approaches:

• Motor Execution (ME) Data Collection: Participants perform actual physical movements (e.g., left/right hand grasping) following visual cues. Record EEG and/or fNIRS during 4-second movement execution periods, with adequate rest between trials [4].

• Motor Imagery (MI) Data Collection: Using the same experimental setup and participants, collect motor imagery data where participants imagine performing the same movements without physical execution [4].

• Model Training and Transfer:

  • Train classification models exclusively on ME data
  • Validate model performance on held-out ME data
  • Test trained models directly on MI data without retraining
  • Fine-tune models with limited MI data (e.g., 50-shot learning) to adapt to individual users [4]

• Performance Comparison: Compare transfer learning performance against within-task models trained and tested on MI data alone [4].

This approach leverages the shared neural mechanisms between motor execution and motor imagery, which exhibit similar event-related desynchronization in the alpha rhythm, albeit with slight differences in temporal patterns [4]. The protocol is particularly valuable for addressing the challenge of BCI poor performers, as transfer learning has been shown to benefit this population significantly.

Multimodal Integration and Analysis Approaches

Concurrent fNIRS-EEG Experimental Design

Integrating fNIRS and EEG provides complementary information about brain activity by capturing both hemodynamic responses (fNIRS) and electrical activity (EEG) simultaneously. This multimodal approach leverages the high temporal resolution of EEG (millisecond range) with the superior spatial resolution and noise resistance of fNIRS [13] [22]. The experimental workflow for concurrent fNIRS-EEG studies involves several critical stages:

Figure 2: Concurrent fNIRS-EEG Experimental Workflow. This diagram outlines the key stages in multimodal brain imaging studies, from experimental design through data interpretation, highlighting the three primary analysis approaches for integrated data.

Multimodal Data Analysis Methods

The analysis of concurrent fNIRS-EEG data can be approached through three primary methodologies, each with distinct advantages for specific research questions:

Table 3: Multimodal fNIRS-EEG Analysis Approaches

Analysis Method	Description	Implementation	Best Use Cases
Parallel Analysis	fNIRS and EEG data are analyzed independently, with results compared or correlated post-hoc	Separate processing pipelines for fNIRS (GLM, Hb concentration changes) and EEG (time-frequency analysis, ERP), then compare spatial and temporal activation patterns	Initial exploratory studies, validation of findings across modalities, clinical applications where established single-modality methods are preferred
fNIRS-Informed EEG Analysis	fNIRS activation maps guide EEG source localization to improve spatial accuracy	Use fNIRS-derived activation areas as spatial constraints for EEG inverse problem solutions, applying anatomical priors based on hemodynamic responses	Studying neural generators of EEG components, pinpointing sources of pathological activity in epilepsy, linking hemodynamic responses to electrical source dynamics
EEG-Informed fNIRS Analysis	EEG features create regressors that model the temporal characteristics of hemodynamic responses	Extract EEG features (ERP amplitudes, band power changes) to create customized predictors for General Linear Model (GLM) analysis of fNIRS data	Improving detection of task-evoked hemodynamic responses, studying neurovascular coupling dynamics, enhancing BCI classification accuracy

The asymmetric approaches (informed analyses) particularly leverage the neurovascular coupling relationship to enhance the strengths of each modality. fNIRS-informed EEG analysis addresses the volume conduction problem in EEG, where electrical activity recorded at each scalp electrode reflects blurred contributions from multiple neural sources [25]. By using fNIRS activation maps as spatial constraints, researchers can achieve more accurate source localization of EEG signals [22] [25].

Conversely, EEG-informed fNIRS analysis tackles the challenge of modeling the hemodynamic response function in fNIRS data. Traditional GLM approaches typically use canonical hemodynamic response functions convolved with boxcar functions representing stimulus timing. However, EEG features such as event-related potential components or frequency band power changes may better represent the precise timing and intensity of neural activity, creating more accurate models of the expected hemodynamic response [25].

The Scientist's Toolkit: Essential Research Solutions

Successful investigation of neurovascular coupling in motor imagery research requires specialized equipment and analytical tools. The following table catalogs essential research solutions and their applications:

Table 4: Essential Research Tools for Neurovascular Coupling Studies

Tool Category	Specific Solutions	Function	Application in NVC Research
fNIRS Hardware	NIRx NIRScout, Artinis systems	Measures concentration changes in oxygenated and deoxygenated hemoglobin	Tracking hemodynamic responses during motor imagery tasks
EEG Hardware	BrainAMP, Neuracle wireless systems	Records electrical brain activity with high temporal resolution	Capturing event-related desynchronization during motor imagery
Integrated Systems	NIRx combined fNIRS-EEG caps	Enables simultaneous multimodal data acquisition	Studying neurovascular coupling dynamics during motor tasks
Analysis Software	NIRS Toolbox, HOMER2/3, nirsLAB	Processes fNIRS signals, removes artifacts, calculates Hb concentrations	Analyzing hemodynamic responses to motor imagery
EEG Analysis Tools	EEGLAB, BCILAB, MNE-Python	Processes EEG signals, extracts features, classifies patterns	Detecting motor imagery-related ERD/ERS patterns
Multimodal Analysis	EEG-informed fNIRS GLM, fNIRS-constrained EEG source localization	Integrates data from both modalities for enhanced analysis	Investigating coupling between electrical and hemodynamic activity
Real-Time Processing	Turbo-Satori, BCILAB	Enables neurofeedback and brain-computer interface applications	Motor imagery BCI systems for rehabilitation and control
Stimulus Presentation	PsychToolbox, Presentation, E-Prime	Controls experimental paradigms and timing	Precisely timed motor imagery cue presentation

The selection of appropriate tools depends on specific research goals and constraints. For studies focusing primarily on hemodynamic responses with minimal motion artifacts, fNIRS may be sufficient alone. For research requiring precise temporal resolution of neural dynamics, EEG provides superior capabilities. However, for comprehensive investigation of neurovascular coupling mechanisms, integrated systems that simultaneously capture both electrical and hemodynamic aspects of brain activity offer the most complete picture [13] [22] [25].

Advanced analysis platforms like the NIRS Toolbox provide comprehensive functionality for fNIRS data processing, including signal processing, visualization, statistical analysis, and multimodal integration [26]. Similarly, HOMER2 and HOMER3 represent widely adopted open-source solutions for fNIRS analysis, offering flexible processing pipelines and integration capabilities with EEG data [26]. For real-time applications such as neurofeedback or BCI, specialized software like Turbo-Satori provides optimized performance for closed-loop paradigms [26].

The integration of these tools within a coherent experimental framework enables researchers to address fundamental questions about neurovascular coupling while advancing practical applications in brain-computer interfaces, neurorehabilitation, and cognitive neuroscience.

Motor imagery (MI)-based brain-computer interfaces (BCIs) represent a transformative technology for neurorehabilitation and assistive device control, particularly for patients with motor impairments resulting from conditions such as intracerebral hemorrhage (ICH) [5] [27]. Traditional unimodal BCI systems, relying solely on electroencephalography (EEG) or functional near-infrared spectroscopy (fNIRS), face fundamental limitations that restrict their clinical application and real-world performance [28]. Hybrid EEG-fNIRS systems synergistically combine the millisecond temporal resolution of EEG with the superior spatial localization and noise resistance of fNIRS, creating a more robust and accurate platform for decoding motor intentions [29] [27]. This integration is particularly crucial for addressing the neurophysiological heterogeneity in patient populations, where neurovascular uncoupling and altered neural signatures can impede reliable classification [5]. The resulting enhanced classification accuracy and robustness demonstrated by hybrid systems marks a significant advancement toward clinically viable BCI technologies for personalized rehabilitation.

Performance Comparison of BCI Modalities

Quantitative comparisons across studies consistently demonstrate the performance superiority of hybrid EEG-fNIRS systems over unimodal approaches. The complementary nature of electrical and hemodynamic signals enables more precise discrimination of motor imagery tasks.

Table 1: Classification Accuracy of Different BCI Modalities and Fusion Strategies

Modality	Fusion Method	Dataset/Context	Reported Accuracy	Reference
Hybrid EEG-fNIRS	Deep Learning + Evidence Theory	TU-Berlin-A Dataset	83.26%	[29]
Hybrid EEG-fNIRS	Transfer Learning	Cross-Subject Generalization	82.30%	[27]
Hybrid EEG-fNIRS	Transfer Learning	Second Public Dataset	87.24%	[27]
EEG-only	Deep Learning (EEGNet)	Real-time Robotic Finger Control (2-finger)	80.56%	[30]
EEG-only	Deep Learning (EEGNet)	Real-time Robotic Finger Control (3-finger)	60.61%	[30]
Hybrid EEG-fNIRS	Feature-level Fusion	ICH Patients vs. Normal Subjects	~5-10% improvement vs. unimodal	[5]

Table 2: Comparative Characteristics of EEG and fNIRS Modalities

Characteristic	EEG	fNIRS
Measured Signal	Electrical cortical activity	Hemodynamic response (Oxy-Hb, Deoxy-Hb)
Temporal Resolution	Millisecond-level (High)	~1 second (Lower)
Spatial Resolution	Low (Limited by volume conduction)	High (5–10 mm)
Susceptibility to Noise	High (Electrical, motion artifacts)	Robust to electrical noise
Principal Advantage	Real-time control capability	Superior spatial localization

Detailed Experimental Protocols

Protocol 1: Upper-Limb Motor Imagery for Intracerebral Hemorrhage Rehabilitation

This protocol is designed for acquiring synchronized EEG-fNIRS data from both healthy subjects and ICH patients, forming the basis for developing robust cross-subject classification models [5] [27].

• Participant Preparation and Clinical Assessment: Recruit 17 normal subjects and 20 ICH patients. For patients, conduct clinical assessments using the Fugl-Meyer Assessment for Upper Extremities (FMA-UE), Modified Barthel Index (MBI), and modified Rankin Scale (mRS) to quantify motor impairment and functional status [5]. Obtain written informed consent approved by the institutional ethics committee.
• Motor Imagery Familiarization: Enhance MI vividness through a grip strength calibration procedure. This involves participants performing repeated 5 kg maximal force exertions using a dynamometer, equivalent force applications with a stress ball, and grip training at one contraction per second to reinforce tactile and temporal aspects of the movement [5].
• Data Acquisition Setup: Use a custom-designed hybrid cap integrating a 32-channel EEG configuration (e.g., g.HIamp amplifier) and a continuous-wave fNIRS system (e.g., NirScan) with 32 sources and 30 detectors, creating 90 measurement channels. Set sampling rates to 256 Hz for EEG and 11 Hz for fNIRS. Ensure temporal synchronization using event markers from presentation software like E-Prime 3.0 [5].
• Experimental Paradigm: Seat participants 25 cm from a monitor. The paradigm consists of at least two sessions, each containing 15 trials per hand.
  • Visual Cue (2 s): A yellow directional arrow (left/right) is displayed.
  • Execution Phase (10 s): A central yellow fixation cross appears with an auditory cue. Participants perform kinesthetic MI of a grasping movement with the cued hand at 1 Hz.
  • Inter-trial Interval (15 s): A blank screen is shown for rest [5].
• Data Preprocessing and Analysis: Preprocess EEG signals for noise and artifacts. For fNIRS, apply a bandpass filter (e.g., 0.01–0.3 Hz) to remove physiological noise and convert light intensity changes to oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) concentrations using the Modified Beer-Lambert Law [31].

Protocol 2: Real-time Robotic Hand Control at Individual Finger Level

This protocol enables real-time decoding of individual finger movements for precise robotic hand control, advancing BCI dexterity [30].

• Participant Setup: Recruit able-bodied, experienced BCI users. Apply a high-density EEG cap following the 10-20 system.
• Offline Model Training: In an initial session, participants perform executed or imagined movements of individual fingers (thumb, index, pinky) in a cued paradigm. Record EEG data to train a subject-specific base decoder (e.g., EEGNet-8.2) [30].
• Online Real-time Control Sessions: Conduct two online sessions where participants perform the same MI tasks.
  • Real-time Feedback: Provide visual feedback (target finger color change) and physical feedback via a robotic hand that moves the decoded finger in real time.
  • Model Fine-tuning: After the first half of each online session, use the collected data to fine-tune the base model, creating a session-specific adapted decoder to combat inter-session variability [30].
• Performance Evaluation: Assess online task performance using majority voting accuracy across trials. Evaluate binary (e.g., thumb vs. pinky) and ternary (e.g., thumb vs. index vs. pinky) classification tasks, calculating precision and recall for each class [30].

Protocol 3: Lower-Limb Motor Imagery for Exoskeleton Control

This protocol focuses on acquiring fNIRS data for lower-limb MI tasks, which is critical for developing rehabilitation protocols for gait restoration [31].

• Participant and Setup: Position 21 healthy participants and one amputee in a 45-degree Fowler's position to enhance cerebral blood flow. Use an fNIRS system (e.g., NIRSport2) with 8 sources and 8 detectors (20 channels) placed over the motor cortex, following the EEG 10-20 system. Secure IMU sensors to the ankle to monitor for unintended physical movements [31].
• Task Paradigm: The protocol is structured in blocks for different joint movements (ankle and knee).
  • Initial Rest (45 s): Hemodynamic stabilization.
  • MI Task (5 s): e.g., Ankle plantarflexion, guided by on-screen cues.
  • Rest (10 s): Return to baseline.
  • This cycle repeats for three trials per task type (ankle dorsiflexion/plantarflexion, knee flexion/extension) for right leg, left leg, and both legs [31].
• Data Processing: Preprocess fNIRS signals using a fourth-order Butterworth bandpass filter (0.01–0.3 Hz) and Z-transform normalization before converting to HbO/HbR concentrations [31].

Technical Workflow of a Hybrid EEG-fNIRS BCI System

The following diagram illustrates the integrated signal processing and decision fusion pathway that characterizes advanced hybrid systems.

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Key Materials and Equipment for Hybrid EEG-fNIRS Research

Item	Specification/Example	Primary Function
Hybrid EEG-fNIRS Cap	Custom design with 32 EEG electrodes, 32 fNIRS sources, 30 detectors [5]	Simultaneous acquisition of electrophysiological and hemodynamic signals from the scalp.
EEG Amplifier	g.HIamp (g.tec) [5]	Amplifies microvolt-level electrical signals from the brain with high fidelity.
fNIRS System	NirScan (Danyang Huichuang) [5] or NIRSport2 (NIRx) [31]	Emits near-infrared light and detects attenuated light to measure hemodynamic changes.
Synchronization Software	E-Prime 3.0 (Psychology Software Tools) [5]	Presents paradigms and sends simultaneous event markers to both EEG and fNIRS systems.
fNIRS Processing Algorithm	Modified Beer-Lambert Law [31]	Converts raw light intensity data into concentration changes of oxyhemoglobin and deoxyhemoglobin.
Clinical Assessment Scales	Fugl-Meyer Assessment (FMA-UE), Modified Barthel Index (MBI) [5] [27]	Quantifies motor function and independence in patients for correlation with neural data.
Deep Learning Decoder	EEGNet [30], Custom CNN-GRU architectures [29]	Automatically extracts discriminative features from raw or preprocessed signals for classification.
Transfer Learning Framework	Wasserstein metric-driven source domain selection [27]	Improves model generalization from normal subjects to patient populations.

From Theory to Practice: Implementing Standardized EEG-fNIRS MI Protocols

The design of experimental paradigms, specifically the temporal structuring of cue, execution, and rest intervals, is a critical determinant of data quality in motor imagery (MI) research using electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS). Well-designed paradigms effectively capture the neural correlates of motor imagery while minimizing artifacts and participant fatigue. This protocol outlines evidence-based guidelines for structuring these temporal components within motor imagery tasks, framed specifically for multimodal EEG-fNIRS research. The synchronization of these modalities capitalizes on their complementary strengths: EEG's millisecond temporal resolution for capturing rapid neuronal activation patterns during MI tasks, and fNIRS's superior spatial localization for tracking slower hemodynamic changes associated with cortical reorganization [5].

Motor imagery shares neural mechanisms with actual movements, with both imaginary and actual movements demonstrating event-related desynchronization in the mu rhythm (8-12 Hz) of the motor cortex [32]. The "functional equivalence" hypothesis suggests that MI activates similar neural regions to actual movement execution, including the primary motor cortex, supplementary motor area, and premotor cortex, making it particularly valuable for rehabilitation applications [31]. However, approximately 20% of BCI users face challenges with MI-based paradigms, often attributed to suboptimal experimental designs that fail to account for individual differences in cognitive processing and fatigue factors [32].

Theoretical Foundations of Interval Timing

The temporal parameters of MI paradigms must accommodate the distinct physiological signals captured by EEG and fNIRS. EEG records electrical neural activity with millisecond resolution, while fNIRS measures hemodynamic responses characterized by the slow rise-time of the hemodynamic response function (HRF), which typically peaks approximately 5 seconds after neural activity begins [33]. This fundamental difference necessitates careful consideration of interval timing to effectively capture signals from both modalities.

The cue interval serves to orient attention and prepare for the upcoming motor imagery task. The execution interval must be sufficiently long to allow for the development of detectable hemodynamic changes while capturing event-related desynchronization/synchronization in EEG signals. The rest interval is critical for allowing hemodynamic responses to return to baseline levels before subsequent trials, thereby preventing carryover effects that can contaminate data [31].

Cognitive motor processes engaged during MI include motor planning, kinesthetic visualization, and working memory components. The Simulation Hypothesis suggests that motor execution, observation, and imagery rely on a shared neural network (the Action Observation Network), though activation levels differ between these conditions [34]. Paradigm design must therefore facilitate the cognitive processes involved in mental simulation of movements without actual motor execution.

Quantitative Parameters for Interval Structuring

Based on analysis of current literature, optimal timing parameters for MI paradigms vary based on research objectives, participant populations, and modality requirements. The table below summarizes evidence-based interval timing from recent studies:

Table 1: Experimentally Validated Timing Parameters for MI Paradigms

Study & Population	Modality	Cue Interval	Execution Interval	Rest Interval	Total Trial Duration
HEFMI-ICH (ICH patients & healthy) [5]	Hybrid EEG-fNIRS	2 s	10 s	15 s	27 s
Lower-limb MI (Healthy) [31]	fNIRS	5 s (instruction)	5 s	10 s	20 s
Acute Stroke Patients [35]	EEG	4 s (instruction + preparation)	4 s	4 s	12 s
ME/MO/MI Comparison [34]	Hybrid EEG-fNIRS	Audio cue (1-2 s)	Self-paced (≈5 s)	Variable	Variable
Multimodal Neurofeedback [36]	Hybrid EEG-fNIRS	Visual cue (2-3 s)	8-10 s	10-15 s	20-28 s

A meta-analysis of public motor imagery/execution datasets revealed that trial lengths typically range from 2.5 to 29 seconds with a mean of 9.8 seconds [32]. Typically, each trial consists of multiple sections: pre-rest (mean 2.38 s), imagination ready (1.64 s), imagination (4.26 s, ranging from 1 to 10 s), and post-rest (3.38 s) [32]. These parameters provide a framework for designing paradigms while emphasizing the need for flexibility based on specific research requirements.

The execution interval is particularly critical as it must accommodate different temporal response characteristics between modalities. For fNIRS, studies have demonstrated that a minimum of 5-10 seconds execution time is necessary to capture the hemodynamic response, with the HRF typically peaking 5-7 seconds after task initiation [33] [31]. For EEG, shorter execution periods of 4-5 seconds can effectively capture event-related desynchronization, though consistency across trials is essential for signal averaging [32] [35].

Table 2: Considerations for Interval Timing Based on Research Factors

Research Factor	Impact on Interval Timing	Recommended Adjustment
Patient Population	ICH and stroke patients may exhibit slower cognitive processing and fatigue more rapidly [5]	Extend rest intervals by 20-40%; consider shorter execution intervals with more trials
Lower-limb vs Upper-limb MI	Lower-limb imagery may require different cognitive engagement [31]	Pilot studies recommended to optimize timing for specific tasks
Multimodal vs Unimodal	Hybrid systems require accommodating both hemodynamic and electrophysiological responses [5]	Execution intervals of 8-10 seconds optimal for capturing both signal types
Participant Training	Naïve participants may require longer cue and execution intervals initially [5]	Implement practice sessions; consider adaptive paradigms that adjust timing based on performance

Standardized Experimental Protocol

Participant Preparation and Calibration

Effective participant preparation is essential for quality data collection. For the HEFMI-ICH dataset, researchers implemented a grip strength calibration procedure before data acquisition to enhance MI vividness [5]. This protocol involves:

• Equipment Setup: Position participants 25-80 cm from the visual display monitor in an ergonomic seated position [5] [35]. For fNIRS, ensure proper optode-scalp contact, particularly challenging in participants with thick hair [31].

• Kinesthetic Calibration: Conduct a 5-10 minute preparatory phase using a dynamometer and stress ball comprising:

  • Repeated 5 kg maximal force exertions (or voluntary maximum efforts)
  • Equivalent force applications using a stress ball
  • Grip training at a rate of one contraction per second [5]

• Task Instruction: Clearly explain the concept of motor imagery, emphasizing that it involves "issuing a command to the hand without any overt movement" rather than simply visualizing the action [5]. Differentiate MI from motor execution and motor observation, as these engage shared but distinct neural mechanisms [34].

• Baseline Recording: Collect 1-minute eyes-closed followed by 1-minute eyes-open baseline recordings demarcated by an auditory cue (200 ms beep) [5].

Trial Structure Implementation

The following workflow details the implementation of a standardized trial structure optimized for hybrid EEG-fNIRS recordings:

Diagram 1: Motor Imagery Trial Structure

Session Configuration

A typical experimental session should comprise:

• Session Duration: Total session time of approximately 20-30 minutes, including preparation [35].
• Trial Count: Minimum of 30 trials per session (15 left/right hand MI trials each) with at least two consecutive sessions [5].
• Break Intervals: Implement sufficient inter-session rest intervals (3-5 minutes) to mitigate fatigue effects and ensure data quality [5].
• Counterbalancing: Alternate between left and right hand imagery trials in pseudorandom order to avoid sequence effects.
• Trial Timing: The HEFMI-ICH paradigm demonstrates effective timing: visual cue presentation (2 s), execution phase (10 s), and inter-trial interval (15 s) [5].

Multimodal Integration Considerations

Hybrid EEG-fNIRS systems require temporal synchronization between modalities. The HEFMI-ICH dataset utilized event markers transmitted from E-Prime 3.0 to simultaneously trigger both recording systems during experimental paradigms [5]. This synchronization is crucial for correlating electrical activity (EEG) with hemodynamic responses (fNIRS).

The neural correlates captured by each modality during properly timed intervals include:

Diagram 2: Multimodal Signal Acquisition

Structured sparse multiset Canonical Correlation Analysis (ssmCCA) has been successfully employed to fuse fNIRS and EEG data, pinpointing brain regions consistently detected by both modalities [34]. This integration enhances classification accuracy by 5-10% compared to unimodal systems [5].

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Equipment for Hybrid EEG-fNIRS Motor Imagery Research

Equipment Category	Specific Examples	Function in Research	Technical Specifications
EEG Acquisition System	g.HIamp amplifier (g.tec); ZhenTec NT1 [5] [35]	Records electrical neural activity with high temporal resolution	32-channel configuration; 256-500 Hz sampling rate; Common-mode rejection ratio: 120 dB [5] [35]
fNIRS Acquisition System	NirScan (Danyang Huichuang); NIRSport2 (NIRx) [5] [31]	Measures hemodynamic responses via near-infrared light	760 nm & 850 nm wavelengths; 10-11 Hz sampling rate; 3 cm source-detector spacing [5] [31]
Stimulus Presentation Software	E-Prime 3.0; PsychoPy [5] [31]	Controls experimental paradigm and sends synchronization triggers	Precision timing for visual/auditory cue presentation; Marker transmission to recording systems
Custom Integrated Caps	Model M hybrid cap (54-58 cm) [5]	Simultaneous EEG electrode and fNIRS optode placement	32 EEG electrodes; 32 optical sources; 30 photodetectors; 90 fNIRS measurement channels
Motion Monitoring	Inertial Measurement Unit (IMU) sensors [31]	Detects unintended physical movements during MI	Differentiates MI from motor execution; Ensures task compliance
Data Processing Platforms	EEGLAB; MNE-Python; Satori 2.0; Aurora fNIRS software [37] [31] [35]	Preprocessing, artifact removal, and signal analysis	Bandpass filtering (EEG: 0.5-40 Hz; fNIRS: 0.01-0.3 Hz); Independent Component Analysis

Adaptation for Special Populations

Motor imagery paradigm design requires specific adaptations for clinical populations such as intracerebral hemorrhage (ICH) patients and stroke survivors. These populations present unique challenges including faster fatigue rates, potential cognitive impairments, and altered neurovascular coupling.

For ICH patients, the HEFMI-ICH dataset demonstrated successful implementation with participants whose time since onset ranged from 2 days to 2 months [5]. Key adaptations include:

• Shorter Session Duration: Reduced total experiment time with more frequent rest intervals
• Enhanced Instruction Protocols: Additional practice sessions and simplified instructions
• Flexible Timing: Inter-session breaks adjusted based on participant readiness rather than fixed intervals
• Clinical Assessment Integration: Correlation with standardized scales (Fugl-Meyer Assessment, Modified Barthel Index, modified Rankin Scale) [5]

For acute stroke patients, research indicates that MI paradigms must account for potential disruptions in neurovascular coupling and altered neural activation patterns [35]. Transfer learning approaches using motor execution or motor observation to train MI classifiers have shown promise, particularly for low performers [4].

Validation and Quality Assessment

Data quality assessment should include both objective signal metrics and behavioral measures. For EEG data, Common Spatial Pattern (CSP) features combined with Linear Discriminant Analysis (LDA) typically achieve approximately 66.53% classification accuracy for two-class MI problems across healthy populations, with BCI poor performers representing approximately 36.27% of users [32]. Preprocessing choices significantly impact decoding performance, with studies showing that high-pass filtering and baseline correction generally improve performance, while aggressive artifact correction may remove task-relevant neural signals [37].

Hybrid systems typically achieve 5-10% improvement in classification accuracy compared to unimodal systems [5]. Effective paradigms should yield distinct hemodynamic responses characterized by increased oxygenated hemoglobin and decreased deoxygenated hemoglobin in the contralateral motor cortex during hand imagery tasks [5] [31].

The robustness of experimental paradigms can be enhanced through systematic variation of preprocessing steps and validation using multiple classifier approaches, including both neural networks (EEGNet) and time-resolved logistic regression models [37].

Motor imagery (MI) vividness, defined as the clarity and realism with which an individual can mentally simulate a movement without physical execution, is a critical determinant of success in MI-based brain-computer interface (BCI) and neurorehabilitation research [38] [39]. The ability to generate specific, task-related brain activity patterns in electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) signals directly depends on the participant's capacity to form accurate mental representations of movements [6]. Research indicates that approximately 30% of participants encounter difficulties in self-regulating brain activity during neurofeedback, often due to suboptimal MI techniques, highlighting the necessity for standardized calibration and instruction protocols [6]. Effective participant preparation ensures that the acquired neural signals are robust, interpretable, and suitable for decoding algorithms or therapeutic applications, ultimately enhancing the validity and reliability of research outcomes in multimodal EEG-fNIRS studies [5].

Theoretical Foundations and the Importance of Vividness

The neurological basis for MI training rests upon the principle of neuroplasticity and the overlapping neural networks activated during both motor execution and motor imagery [39]. Studies using functional magnetic resonance imaging (fMRI) have revealed that MI engages brain regions similar to those involved in actual movement, including the premotor area, supplementary motor area, primary motor cortex, and inferior parietal lobule [38]. First-person perspective MI, in particular, elicits brain activation patterns that closely mirror those of physical execution, making it the preferred approach for rehabilitation paradigms [38].

Enhancing MI vividness is not merely a procedural formality; it is fundamental to eliciting measurable and specific cortical responses. The hemodynamic response measured by fNIRS and the event-related desynchronization (ERD) measured by EEG are significantly modulated by the quality of the mental imagery performed [6] [5]. Successful modulation of these signals through vivid MI can lead to functional plasticity, which is the core objective in post-stroke motor rehabilitation research [6]. Furthermore, the efficacy of MI-based neurofeedback is contested in some circles, partly due to poor study design and a lack of standardized guidelines for participant training, underscoring the need for rigorous and evidence-based calibration protocols [6] [39].

Pre-Imagery Calibration and Preparation Techniques

Kinesthetic Cueing and Force Sensation

A primary challenge in MI research is ensuring participants understand the concept of "kinesthetic" MI, which involves feeling the sensations of movement rather than merely visualizing it [5]. To address this, a grip strength calibration procedure can be implemented before data acquisition. This protocol strengthens the tactile and force-related aspects of the movement, thereby enhancing MI vividness and inter-trial consistency [5].

Detailed Calibration Protocol:

• Equipment Setup: Prepare a hand dynamometer and a compliant stress ball.
• Maximal Force Exertion: The participant performs repeated (e.g., 3-5) maximal grip force exertions using the dynamometer with the target limb. This establishes a reference for maximal voluntary contraction.
• Tactile Reinforcement: The participant applies an equivalent, sub-maximal force (e.g., 5 kg or a perceived 50-70% of maximum) using the stress ball. This step reinforces the somatosensory and proprioceptive cues associated with the action.
• Temporal Rhythm Training: The participant practices gripping the stress ball at a standardized rate of one contraction per second, guided by a metronome. This rhythm is later used during the mental imagery task to standardize the imagined movement's timing [5].

Action Observation with Inverted Video Feedback

Action observation (AO) combined with MI (AO+MI) is a powerful technique for increasing MI vividness, particularly in clinical populations like stroke survivors who may have difficulty generating mental images of their paralyzed limb [38]. This method leverages the mirror neuron system to facilitate motor simulation.

Detailed Experimental Protocol:

• Video Recording: Film the patient's own non-paralyzed upper limb performing the target movement (e.g., grasping).
• Video Processing: Laterally invert (flip) the video footage so it appears as if the paralyzed limb is performing the movement.
• Instruction and Task: The patient is instructed to observe the inverted video while simultaneously imagining that the movement is being executed by their paralyzed limb. They should adopt a first-person perspective and focus on the kinesthetic sensations of the movement [38].
• Evidence Base: A study with 10 right-handed stroke patients demonstrated that MI vividness was significantly higher when an inverted video of the patient's own hand was presented, compared to another person's hand or no video. This condition also resulted in significantly enhanced cortical activity measured by fNIRS [38].

Table 1: Summary of Pre-Imagery Calibration Techniques

Technique	Primary Objective	Target Population	Key Outcome
Kinesthetic Cueing [5]	To reinforce tactile and force sensations of movement	All participants, especially those naive to MI	Improved understanding of kinesthetic MI; standardized movement rhythm
Inverted Video AO+MI [38]	To generate the illusion of a paralyzed limb moving	Stroke patients with upper limb hemiparesis	Significantly increased MI vividness and cortical activity

Instructional Elements for Effective Motor Imagery

The instructions provided to participants before and during MI tasks are critical. Best practices, derived from a systematic review of MI training across five disciplines, recommend a structured approach based on the PETTLEP model (Physical, Environment, Task, Timing, Learning, Emotion, Perspective) [39].

Detailed Instruction Protocol:

• Perspective and Modality: Instruct participants to use an internal perspective (first-person) and a kinaesthetic mode. They should feel the sensations of the movement "as if they are actually performing it," rather than watching themselves from an external, third-person viewpoint [39].
• Environment and Position: The MI environment and the participant's physical position should be as task-specific as possible. If the imagined task is performed while sitting, the participant should be seated similarly during MI practice [39].
• Timing and Dosage: MI should be performed in real-time, meaning the mental simulation should take the same amount of time as the physical execution. A typical successful MI training session lasts on average 17 minutes and includes about 34 MI trials [39].
• Task Focus and Delivery: Instructions should be acoustic, detailed, and standardized. They can be delivered live by an experimenter or via a recording to ensure consistency across participants and sessions [39].
• Eyes Closed vs. Open: During MI practice, participants are generally instructed to keep their eyes closed to minimize visual distractions and facilitate internal focus [39].

The following workflow diagram synthesizes the core signaling pathways and procedural steps involved in enhancing MI vividness, from external calibration to the resulting neural and functional outcomes.

Integration into Multimodal EEG-fNIRS Experimental Protocols

The calibration and instruction techniques described above must be seamlessly integrated into the broader experimental workflow of a multimodal EEG-fNIRS study. This ensures that the enhanced MI vividness directly translates into higher-quality, classifiable neural signals.

Detailed Experimental Session Protocol:

• Informed Consent and Familiarization: Conduct a pre-session familiarization to explain the concept of MI and the equipment, reducing anxiety [39].
• Calibration Phase: Execute the kinesthetic cueing and/or inverted video AO+MI protocols as detailed in Section 3.
• Baseline Recording (2 minutes): Record 1-minute of eyes-closed rest followed by 1-minute of eyes-open rest to establish baseline neural signals [5].
• MI Task Execution: Initiate the MI paradigm. A standard trial structure includes [5]:
  • Visual Cue (2 s): Presentation of a directional arrow (e.g., left/right) indicating the hand for MI.
  • Execution Phase (10 s): Participants perform kinesthetic MI of the cued hand movement (e.g., grasping at 1 Hz) while focusing on a fixation cross. An auditory cue (beep) marks the start.
  • Rest Interval (15 s): A blank screen allows neural activity to return to baseline.
• Data Acquisition: Synchronously record EEG and fNIRS data throughout the session. A typical setup may use a custom hybrid cap with 32 EEG electrodes and 62 optodes (32 sources, 30 detectors) configured over the sensorimotor cortices [6] [5].
• Feedback (Optional): In neurofeedback paradigms, provide real-time visual feedback (e.g., a ball moving on a gauge) based on the computed NF score from motor cortex activity [6].

Table 2: Key Reagents and Materials for EEG-fNIRS MI Research

Item	Specification / Example	Primary Function in Research
Hybrid EEG-fNIRS Cap	Custom design with integrated electrodes & optodes [5]	Simultaneous acquisition of electrical and hemodynamic brain activity.
fNIRS System	Continuous-wave system (e.g., NirScan) [5]	Measures concentration changes in oxygenated (HbO) and deoxygenated (HbR) hemoglobin.
EEG Amplifier	High-impedance amplifier (e.g., g.HIamp) [5]	Records electrical potentials from the scalp with high temporal resolution.
Grip Dynamometer	Standard hand-held device	Calibrates force sensation for kinesthetic MI cueing [5].
Stimulus Presentation Software	E-Prime, PsychoPy, Presentation	Presents visual cues and records behavioral responses; sends synchronization triggers.
Video Recording/Editing Tool	Common video software with flip/invert function	Creates inverted video stimuli for action observation in clinical populations [38].

The following diagram outlines the sequential workflow of a full experimental session, from participant preparation through to data acquisition, highlighting how calibration and instruction are foundational steps.

The simultaneous acquisition of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) data provides a powerful multimodal approach for investigating brain activity across complementary dimensions: millisecond-scale electrical potentials and hemodynamic responses reflecting metabolic demand. This integration is particularly valuable in motor imagery (MI) research, where understanding the complex interplay between rapid neuronal discharge and slower vascular changes can illuminate fundamental mechanisms of brain plasticity and functional reorganization [6] [40]. The technical challenge lies in achieving precise temporal synchronization between these modalities despite their different physiological origins, sampling rates, and potential sources of interference.

EEG measures electrical activity generated by synchronized neuronal firing, offering exceptional temporal resolution (milliseconds) but limited spatial precision due to signal dispersion through the skull and scalp [13] [41]. Conversely, fNIRS detects changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentrations associated with neural activity through near-infrared light, providing superior spatial localization (centimeter-level) but slower temporal resolution (seconds) due to the inherent latency of neurovascular coupling [13] [42]. When properly integrated, these modalities compensate for each other's limitations, enabling a more comprehensive characterization of brain dynamics during MI tasks [40] [34].

Technical Approaches to Synchronization

Synchronization Methodologies

Achieving temporal alignment between EEG and fNIRS data streams requires careful consideration of hardware capabilities and experimental constraints. Researchers primarily employ three synchronization approaches, each with distinct advantages and implementation requirements.

Table 1: Comparison of EEG-fNIRS Synchronization Methods

Synchronization Method	Technical Implementation	Temporal Precision	Implementation Complexity	Best Use Cases
Hardware Trigger Synchronization	Shared TTL pulses or parallel port signals sent to both systems	Millisecond precision	Moderate	Laboratory settings with compatible equipment
Software-based Synchronization	Shared software platform (e.g., E-Prime, PsychoPy) sending simultaneous triggers	Millisecond to sub-second precision	Low to Moderate	Studies with minimal timing constraints
External Recording Device	Dedicated synchronization unit or data acquisition card recording all signals	Sub-millisecond precision	High	Research requiring ultra-precise timing

Hardware trigger synchronization represents the gold standard for temporal precision, utilizing transistor-transistor logic (TTL) pulses or parallel port signals generated by stimulus presentation software and sent simultaneously to both acquisition systems [5]. This method ensures event markers are embedded in both data streams with millisecond accuracy, enabling precise alignment during subsequent analysis. This approach is particularly suitable for event-related paradigms common in MI research, where precise timing between task initiation and neural response is critical [6].

Software-based synchronization employs a unified software platform to control both stimulus presentation and data acquisition, sending synchronized triggers through a single computer interface [13]. While potentially offering slightly lower precision than hardware solutions, this method simplifies experimental setup and is sufficient for many study designs, particularly those examining sustained cognitive states rather than transient evoked responses [41].

External recording devices represent the most technically sophisticated approach, utilizing a dedicated synchronization unit or high-precision data acquisition card to record signals from both modalities alongside external triggers [13]. This method achieves the highest temporal precision (sub-millisecond) but requires specialized equipment and expertise, making it most appropriate for investigations requiring ultra-precise temporal resolution across modalities.

Integrated Hardware Configurations

The physical integration of EEG electrodes and fNIRS optodes presents significant engineering challenges, particularly regarding crosstalk minimization and signal quality preservation. Current systems employ several configurations to address these issues.

Integrated caps with pre-defined holder arrangements offer a practical solution for co-localizing sensors while maintaining consistent positioning across participants [43] [13]. These systems typically feature specialized holders that position EEG electrodes and fNIRS optodes at optimal distances (typically 30mm or less) to minimize crosstalk while capturing complementary signals from the same cortical regions [43]. Recent research demonstrates that with proper shielding and low-impedance EEG electrodes (<5 kΩ), high-quality signals can be obtained from both modalities without significant interference, even with closely positioned sensors [43].

Customized helmet designs using 3D printing or thermoplastic materials provide an alternative approach that offers improved individual fit and precise sensor placement [13]. These systems can be tailored to specific research requirements and head morphologies, potentially enhancing signal quality by ensuring consistent optode-scalp contact and reducing movement artifacts. However, these benefits must be balanced against higher fabrication costs and increased setup complexity [13].

The following diagram illustrates a standardized workflow for implementing synchronized EEG-fNIRS acquisition in motor imagery research:

Synchronization Implementation Workflow

Experimental Protocol for Motor Imagery Research

Participant Preparation and Sensor Placement

Proper participant preparation and sensor placement are critical for obtaining high-quality simultaneous EEG-fNIRS data during motor imagery tasks. The following protocol outlines standardized procedures based on current methodological recommendations [6] [5].

• Participant Positioning: Seat participants in a comfortable chair approximately 60-80 cm from the visual stimulus display screen. Ensure proper head and arm support to minimize movement artifacts during the recording session. For clinical populations with motor impairments, provide additional postural support as needed [5].
• EEG Cap Application: Apply integrated EEG-fNIRS cap according to the international 10-20 system, focusing coverage over sensorimotor regions (C3, Cz, C4 and surrounding areas). For EEG, achieve electrode impedances below 5 kΩ (preferably below 3 kΩ for combined holders) using appropriate electrolyte gels or pastes [43] [5].
• fNIRS Optode Placement: Position fNIRS sources and detectors over the primary motor cortex and supplementary motor areas with an inter-optode distance of 25-35 mm to ensure sufficient cortical penetration [6] [42]. Verify optimal optode-scalp coupling using signal quality indices where available (e.g., scalp-coupling index > 0.7) [42].
• Synchronization Setup: Implement hardware trigger synchronization by connecting the stimulus presentation computer to both EEG and fNIRS acquisition systems via TTL pulses or parallel port connections. Verify synchronization accuracy by sending test pulses and confirming simultaneous marker registration in both systems [13] [5].

Motor Imagery Paradigm and Data Acquisition

The following standardized motor imagery paradigm, adapted from established protocols in multimodal research [6] [5] [34], ensures consistent data quality while facilitating between-study comparisons.

• Baseline Recording: Begin with 2 minutes of resting-state data collection (1 minute eyes-open followed by 1 minute eyes-closed) to establish baseline brain activity for both modalities [5].
• Task Instruction: Provide clear instructions for kinesthetic motor imagery, emphasizing the mental simulation of movement sensations rather than visual representation. For hand motor imagery, instruct participants to imagine grasping and releasing a ball at a rate of approximately 1 repetition per second without executing actual movements [5] [34].
• Trial Structure: Implement a standardized trial structure consisting of: (1) 2-second visual cue presentation (directional arrow indicating left or right hand); (2) 10-second motor imagery execution period (fixation cross display); (3) 15-second inter-trial rest period (blank screen) [5]. This structure provides sufficient time for both EEG event-related desynchronization and the slower fNIRS hemodynamic response to develop.
• Data Acquisition Parameters: Configure EEG acquisition with a sampling rate ≥ 200 Hz and appropriate bandpass filtering (e.g., 0.1-100 Hz). Set fNIRS sampling rate ≥ 10 Hz using dual wavelengths (typically 760 nm and 850 nm) to compute HbO and HbR concentration changes [42] [5]. Record synchronization pulses simultaneously with both systems, embedding event markers for cue onset, imagery period, and trial completion.
• Data Quality Monitoring: Continuously monitor signal quality throughout acquisition, checking for excessive movement artifacts, EEG electrode drift, or fNIRS signal saturation. Implement real-time quality metrics where available, such as scalp-coupling indices for fNIRS [42] and impedance measurements for EEG [43].

Table 2: Standardized Motor Imagery Protocol Timeline

Phase	Duration	Visual Stimulus	Participant Task	Modality Recorded
Baseline (Eyes Open)	60 s	Fixation cross	Relaxed wakefulness	EEG + fNIRS
Baseline (Eyes Closed)	60 s	Blank screen	Relaxed wakefulness	EEG + fNIRS
Trial Cue	2 s	Directional arrow	Prepare for imagery	EEG + fNIRS
Motor Imagery	10 s	Fixation cross	Execute kinesthetic MI	EEG + fNIRS
Inter-trial Rest	15 s	Blank screen	Relax	EEG + fNIRS
Total Trials	30/session	-	-	-

The Scientist's Toolkit

Research Reagent Solutions

Successful implementation of synchronized EEG-fNIRS acquisition requires specific hardware and software components. The following table details essential resources for establishing a multimodal motor imagery research platform.

Table 3: Essential Materials for EEG-fNIRS Integration

Component	Specifications	Function	Example Products/Models
Integrated EEG-fNIRS Cap	Combined holders with EEG electrodes and fNIRS optodes	Co-localized signal acquisition from same cortical regions	Custom-designed hybrid caps [5], Artinis integrated systems [43]
EEG Amplifier	≥ 200 Hz sampling rate, low-noise design (< 1 µV)	Electrical potential measurement	g.HIamp [5], BrainAmp [40], APEX [43]
fNIRS System	Dual-wavelength (760/850 nm), ≥ 10 Hz sampling	Hemodynamic response measurement	NirScan [5], Hitachi ETG-4100 [34], NIRScout [13]
Synchronization Interface	TTL pulse generator, parallel port, or DAQ card	Temporal alignment of multimodal data	National Instruments DAQ, Arduino-based triggers
Stimulus Presentation Software	Millisecond precision, trigger output capability	Experimental paradigm delivery	E-Prime [5], PsychoPy, Presentation
Data Analysis Platform	Multimodal data integration capabilities	Signal processing and fusion analysis	MATLAB with toolboxes (EEGLAB, NIRS-KIT), MNE-Python [42], Brainstorm [42]

Technical Considerations and Troubleshooting

Optimizing Signal Quality

Several technical challenges require specific attention when implementing synchronized EEG-fNIRS acquisition. The following considerations address common issues encountered in multimodal motor imagery research.

• Crosstalk Mitigation: Electromagnetic interference from fNIRS optodes can introduce artifacts in EEG recordings, particularly when sensors are closely positioned [43]. Employ active shielding of fNIRS components, maintain low EEG electrode impedances (< 5 kΩ), and implement spectral filtering to remove interference at characteristic fNIRS firing frequencies (e.g., 17.4 Hz or 37 Hz, depending on system) [43].
• Motion Artifact Management: While fNIRS is relatively robust to movement artifacts, EEG is highly susceptible to motion-related signal corruption [41]. Use tight but comfortable cap fittings, instruct participants to minimize head movement, and implement motion correction algorithms during preprocessing. For studies involving substantial movement, consider incorporating accelerometers to quantify and correct for motion artifacts [42].
• Physiological Noise Separation: fNIRS signals are contaminated by systemic physiological noise (cardiac, respiratory, blood pressure oscillations) that can obscure task-evoked hemodynamic responses [44]. Incorporate short-separation (SS) fNIRS channels (< 15 mm source-detector distance) to measure superficial scalp hemodynamics, which can be used as regressors of no interest in general linear models to improve sensitivity to cortical signals [44].
• Temporal Precision Verification: Regularly validate synchronization accuracy by sending test pulses at known intervals and verifying precise temporal correspondence in recorded markers. For ultra-precise timing requirements, consider using an external data acquisition card to record all triggers and signals on a unified timeline [13].

The relationship between hardware configurations and data fusion approaches can be visualized as follows:

Hardware Impact on Data Fusion

Proper hardware integration and synchronization of EEG and fNIRS systems provide a powerful methodological foundation for advanced motor imagery research. By implementing the standardized protocols and technical solutions outlined in this application note, researchers can reliably capture complementary neural signatures of motor cognition—from millisecond-scale electrical oscillations to hemodynamic responses reflecting metabolic demand. The continued refinement of these multimodal approaches promises to accelerate developments in clinical neurorehabilitation, brain-computer interfaces, and fundamental neuroscience.

Within the field of motor imagery (MI) research using electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS), the precise placement of sensors on the scalp is a critical determinant of data quality and experimental validity. The sensorimotor cortex, which is the primary region activated during both imagined and executed movements, must be targeted with high specificity to accurately capture these neural events. This application note provides a detailed guide to optimizing cap layouts for comprehensive motor cortex coverage, framed within the broader context of standardizing motor imagery task protocols for neuroscientific and clinical research, including drug development studies where monitoring cortical activity is essential.

The complementary nature of EEG and fNIRS offers a powerful multimodal approach for investigating motor imagery. EEG provides millisecond-level temporal resolution of electrical brain activity, characterized by event-related desynchronization (ERD) in the mu (8-13 Hz) and beta (13-30 Hz) bands over contralateral sensorimotor areas during MI [45]. Meanwhile, fNIRS tracks hemodynamic responses by measuring changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentrations, offering superior spatial localization of cortical activation with approximately 5–10 mm resolution [5]. Optimizing sensor placement to leverage these complementary strengths requires careful consideration of anatomical landmarks, signal characteristics, and practical implementation factors.

Sensor Layout Specifications for Motor Cortex Coverage

Key Anatomical Landmarks and Positioning

Effective coverage of the motor cortex requires strategic placement of sensors based on the international 10-10 or 10-20 electrode positioning systems. These systems use consistent proportional measurements from anatomical landmarks (nasion, inion, and preauricular points) to ensure reproducible sensor placement across subjects and sessions [46]. For upper-limb motor imagery, which primarily engages the hand and arm representation areas, the following regions are particularly crucial:

• Primary Motor Cortex (M1): Located approximately at positions C3 (right hand) and C4 (left hand) in the 10-20 system
• Primary Somatosensory Cortex (S1): Adjacent to M1, with overlapping coverage
• Premotor Cortex and Supplementary Motor Area: Located anterior to the central sulcus

Research indicates that kinesthetic motor imagery reliably evokes mu and beta desynchronization over contralateral sensorimotor cortex that resembles the response seen during actual movement, albeit often with lower amplitude or broader spatial distribution [45]. This functional equivalence underscores the importance of comprehensive sensor coverage to capture the full extent of MI-related activation.

Quantitative Sensor Layout Specifications

Table 1: Exemplary EEG-fNIRS Integrated Cap Layout for Upper-Limb Motor Imagery

Modality	System Components	Channel Count	Key Positions	Targeted Regions
EEG	ActiCHamp (Brain Products GmbH)	32 electrodes (19 over sensorimotor)	FC5, FC3, FC1, FC2, FC4, FC6, C5, C3, C1, C2, C4, C6, CP5, CP3, CP1, CP2, CP4, CP6 [10]	Bilateral sensorimotor cortices
fNIRS	NIRScout XP (NIRx)	16 sources, 16 detectors, 8 short channels	Source-detector pairs with 3 cm separation [5]	Primary motor cortex, Premotor areas
Integrated	Custom EasyCap	Combined layout	Optimized topography covering prefrontal, motor, association cortices [5]	Comprehensive motor network

Table 2: Alternative Configurations from Hybrid BCI Datasets

Dataset/Study	EEG Configuration	fNIRS Configuration	Coverage Focus	Subject Population
HEFMI-ICH [5]	32 channels (g.tec)	32 sources, 30 detectors (90 channels)	Prefrontal, motor, association cortices	17 healthy, 20 ICH patients
Muller et al. [10]	19 sensorimotor channels	16 sources, 16 detectors	Contralateral sensorimotor cortex	30 healthy (planned)

The specifications in Table 1 and Table 2 demonstrate two optimized approaches for integrated EEG-fNIRS cap layouts. The exemplary layout from Muller et al. emphasizes focused coverage of the sensorimotor cortices with 19 EEG electrodes positioned over key areas, while the HEFMI-ICH dataset employs a more extensive configuration covering multiple cortical regions. The 3 cm source-detector distance for fNIRS represents a standard compromise between sensitivity to cortical signals and adequate signal strength [5].

Experimental Protocol for Sensor Placement and Motor Imagery

Cap Application and Signal Quality Verification

Materials Required:

• Integrated EEG-fNIRS cap with appropriate size for head circumference
• Electrolyte gel (for EEG) and fiber optic check equipment (for fNIRS)
• Impedance checker and fNIRS signal quality software
• Measuring tape for anatomical landmark identification

Procedure:

• Measure head circumference and identify nasion, inion, and preauricular points according to the 10-10 international system [46].
• Position the integrated cap with Cz at the vertex, ensuring proper alignment with anatomical landmarks.
• For EEG: Apply electrolyte gel to achieve impedance below 10 kΩ for each electrode, with particular attention to sensorimotor positions (C3, C4, Cz, etc.) [10].
• For fNIRS: Verify optical coupling by ensuring signal quality indices are within manufacturer specifications for all channels.
• Perform simultaneous signal verification to check for cross-talk between modalities and ensure both systems are acquiring data properly.

Motor Imagery Task Protocol with Verification Steps

Task Structure:

• Baseline Recording (2 minutes): Collect resting-state data with eyes open and closed [5].
• Task Paradigm (30 trials):
  • Visual cue presentation (2 s): Arrow indicating left or right hand MI
  • Execution phase (10 s): Kinesthetic motor imagery with central fixation cross
  • Inter-trial interval (15 s): Rest period with blank screen [5]
• Quality Checks During Acquisition:
  • Monitor EEG for artifacts (blinks, muscle movement)
  • Verify fNIRS signal quality through HbO/HbR correlation
  • Ensure real-time feedback accuracy in neurofeedback paradigms

Enhancing Motor Imagery Vividness: Studies show that incorporating a grip strength calibration procedure before data acquisition—using a dynamometer and stress ball—enhances participants' perception of the imagined motion and improves MI vividness and consistency across trials [5]. This preparatory phase strengthens the tactile and force-related aspects of the grasping movement and standardizes its temporal rhythm.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Equipment for EEG-fNIRS Motor Imagery Research

Item	Specification	Research Function
Integrated EEG-fNIRS Cap	32-channel EEG + 16-source/16-detector fNIRS [10]	Simultaneous electrophysiological and hemodynamic recording
Amplifier Systems	EEG: ActiCHamp or g.HIamp; fNIRS: NIRScout or NirScan [10] [5]	Signal acquisition and digitization
Electrode Gel	High-conductivity, chloride-based	Ensures low impedance for EEG signal quality
fNIRS Optodes	Wavelengths: 760 nm & 850 nm [10]	Measures oxygenated and deoxygenated hemoglobin
Calibration Tools	Dynamometer, stress ball [5]	Enhances motor imagery vividness through physical calibration
Software Platforms	Real-time signal processing, NF score calculation [10]	Data analysis and neurofeedback presentation

Implementation Workflow and Data Integration

Discussion and Best Practices

Optimizing sensor placement for motor cortex coverage requires balancing spatial coverage with practical considerations of signal quality and participant comfort. The integrated layouts presented here have demonstrated efficacy in capturing MI-related brain activity while minimizing cross-talk between modalities. Researchers should note that approximately 15-30% of naïve users may struggle to generate distinguishable EEG patterns above the 70% classification accuracy threshold typically required for usable BCI control [45], emphasizing the need for optimal sensor placement to maximize signal quality.

For clinical populations, particularly stroke patients with intracerebral hemorrhage, sensor placement may need adjustment to accommodate lesion locations while still targeting viable motor pathways [5]. The functional equivalence between motor imagery and motor execution—where MI recapitulates the spatiotemporal dynamics of ME in EEG, albeit with lower amplitude and broader distribution [45]—supports the use of these optimized layouts across both research and clinical applications.

Future developments in sensor technology and cap design will likely continue to improve the spatial precision and comfort of these integrated systems. The ongoing standardization of placement protocols will enhance reproducibility across studies and facilitate more direct comparison of findings in the field of motor imagery research.

Protocol Adaptations for Upper-Limb vs. Lower-Limb Motor Imagery

Motor imagery (MI), the mental rehearsal of a movement without any physical execution, is a foundational paradigm for brain-computer interface (BCI) applications, especially in neurorehabilitation. The design of an effective MI-BCI protocol is highly dependent on the target limb, as upper-limb and lower-limb motor tasks engage distinct neural substrates and present unique challenges for signal detection and classification. Electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) are two non-invasive neuroimaging techniques often used singly or in combination to decode MI tasks. This application note details the critical protocol adaptations required when designing MI experiments for the upper versus lower limbs, providing a structured guide for researchers and clinicians. The content is framed within a broader thesis on optimizing MI task protocols to enhance the efficacy and reliability of BCIs for motor recovery.

Neurophysiological and Hemodynamic Basis for Protocol Adaptation

The primary rationale for adapting MI protocols based on the target limb lies in the fundamental organization of the motor cortex. The primary motor cortex (M1) is somatotopically organized according to the cortical homunculus, where different body parts are mapped to specific, spatially distinct regions [47]. Figure 1 illustrates how this organization directly influences the design of MI-BCI protocols.

Upper-limb movements, particularly of the hands, are represented over large, laterally positioned areas in M1. In contrast, lower-limb movements are mapped to more medial, parasagittal cortical areas, which have a smaller surface representation and are more challenging to access with scalp-based sensors [47]. Furthermore, a key finding from fNIRS studies is that left and right upper-limb activation patterns are highly distinguishable, whereas left and right lower-limb activation patterns are highly similar, making their discrimination a significant challenge [47]. These anatomical and functional differences necessitate tailored approaches to sensor placement, task design, and signal processing.

Comparative Analysis: Upper-Limb vs. Lower-Limb MI

Table 1 summarizes the core experimental and performance differences between upper-limb and lower-limb MI paradigms, synthesizing findings from current literature.

Table 1: Comparative Protocol Specifications for Upper-Limb vs. Lower-Limb Motor Imagery

Protocol Component	Upper-Limb MI	Lower-Limb MI	Key References
Typical MI Tasks	Hand grasping, finger tapping, ball squeezing	Foot tapping, toe movement, walking imagery	[47] [5]
Primary Cortical Areas	Lateral Primary Motor Cortex (M1), Premotor Cortex	Medial Primary Motor Cortex (M1), Supplementary Motor Area	[47]
Optimal Sensor Placement (EEG)	C3 (right hand), C4 (left hand) positions	Cz (vertex) position	[47]
Optimal Sensor Placement (fNIRS)	Over lateral sensorimotor cortex	Over medial/parasagittal sensorimotor cortex	[47] [48]
Spatial Discriminability (Left vs. Right)	High. Clear differentiation in cortical activation patterns.	Low. Highly similar bilateral activation patterns.	[47]
Typical Performance (Classification Accuracy)	Up to 89% for left vs. right hand discrimination with fNIRS; ~76% with hybrid EEG-fNIRS	Lower than upper-limb; requires advanced methods for left-right discrimination	[47] [17]
Key Challenges	Subject's MI ability, signal contamination from muscle artifacts	Low spatial discrimination, smaller cortical representation, proximity to sagittal sinus	[47]

Detailed Experimental Protocols

Standardized Upper-Limb Motor Imagery Protocol

A robust protocol for upper-limb MI, suitable for both healthy subjects and patient populations like those with intracerebral hemorrhage (ICH), is detailed below [5].

• Participant Preparation and Calibration: Seat the participant comfortably 25 cm from a monitor. To enhance kinesthetic MI vividness, introduce a grip strength calibration phase using a dynamometer or stress ball. This involves:
  • Repeated maximal force exertions with the dynamometer.
  • Equivalent force applications using a stress ball.
  • Grip training at a rate of one contraction per second to standardize the temporal rhythm of the imagined movement.
• Experimental Paradigm: The paradigm should be structured into sessions, each containing multiple trials (e.g., 15-30 trials per session for left- and right-hand MI).
  • Visual Cue (2 s): A yellow directional arrow (pointing left or right) is displayed on a blue background.
  • Execution Phase (10 s): The arrow is replaced by a central yellow fixation cross, accompanied by an auditory cue (200 ms beep). The participant performs kinesthetic MI of hand grasping with the cued hand at ~1 Hz.
  • Inter-Trial Interval (15 s): A blank screen indicates a rest period.
• Data Acquisition: Synchronized EEG and fNIRS data should be acquired. A typical setup uses a 32-channel EEG configuration and an fNIRS system with sources and detectors arranged to cover the sensorimotor cortex, creating numerous measurement channels [5]. Temporal synchronization between modalities is critical and can be achieved using event markers from stimulus presentation software like E-Prime.

Adapted Lower-Limb Motor Imagery Protocol

Protocols for lower-limb MI must account for its medial cortical representation and lower left-right discriminability.

• Task Selection: Simple, repetitive tasks such as unilateral foot tapping or toe movement are recommended. The kinesthetic sensation of pushing a pedal or tapping the foot should be emphasized over visual imagery.
• Sensor Placement: The fNIRS optode arrays and EEG electrodes must be positioned to cover the medial aspects of the motor cortex, centered on the Cz electrode according to the International 10-20 system [47].
• Paradigm Structure: The trial structure from the upper-limb protocol (cue-execution-rest) can be retained, with instructions modified for foot/ankle movement imagery.
• Signal Processing Considerations: As simple sensor placement may be insufficient for left-right discrimination, researchers should plan for advanced signal processing techniques, such as hyper-scanning or the use of high-density arrays, to improve classification performance [47].

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 2 lists key materials and their functions for setting up a hybrid EEG-fNIRS MI-BCI experiment.

Table 2: Essential Research Reagents and Solutions for Hybrid EEG-fNIRS MI Research

Item	Function/Application in MI Research	Specific Examples / Notes
EEG Amplifier	Records electrical brain activity with high temporal resolution.	g.HIamp (g.tec), BrainAmp. Sampling rate ≥ 200 Hz is typical [5] [17].
fNIRS System	Measures hemodynamic changes (HbO/HbR) with better spatial resolution than EEG.	NirScan (Danyang Huichuang), Hitachi ETG-4000/4100. Continuous-wave systems are common [47] [5] [34].
Hybrid Cap	Integrates EEG electrodes and fNIRS optodes in a single assembly for co-localized measurement.	Custom designs with 32+ EEG channels and 30+ fNIRS detectors [5]. Optode separation ~3 cm [5].
Stimulus Presentation Software	Prescribes the experimental paradigm and delivers synchronized event markers.	E-Prime (Psychology Software Tools) [5].
Electrode Gel/Grounding	Ensures high-quality, low-impedance electrical contact for EEG.	Standard EEG conductive gel and skin preparation supplies.
Calibration Tools	Enhances kinesthetic vividness and standardizes the imagined movement.	Dynamometer, stress ball [5].
Data Analysis Platform	For pre-processing, feature extraction, and classification of multimodal data.	MATLAB (e.g., MartMi-BCI, EEGNet), Python (MNE, PyRiemann) [49] [17] [50].

Workflow for a Hybrid EEG-fNIRS MI Experiment

Implementing a hybrid MI-BCI study involves a sequence of critical steps, from participant preparation to data fusion. Figure 2 outlines a generalized workflow that can be adapted for both upper- and lower-limb studies.

The development of effective motor imagery BCIs requires a limb-specific approach. Upper-limb protocols benefit from high left-right discriminability and well-established sensor placements over lateral cortical areas. In contrast, lower-limb protocols must overcome the challenges of a smaller medial cortical representation and highly bilateral activation, often necessitating advanced imaging and signal processing for successful differentiation. The integration of EEG and fNIRS in a hybrid BCI offers a promising path forward, leveraging their complementary strengths to achieve more accurate and robust classification. The protocols and adaptations outlined in this application note provide a framework for researchers to design more effective MI-BCI systems, ultimately advancing their application in clinical neurorehabilitation and drug development research. Future work should focus on improving real-time classification of lower-limb MI and developing more engaging feedback paradigms to enhance user learning and adherence.

Incorporating Action Observation with MI (AO+MI) for Enhanced Cortical Activation

Action Observation combined with Motor Imagery (AO+MI) represents an advanced neurorehabilitation and research technique where an individual observes a movement while simultaneously imagining performing that same movement. This dual-approach capitalizes on the brain's mirror neuron system (MNS), which activates during both action observation and execution, creating a powerful synergistic effect on cortical activation [51] [52]. While traditional Motor Imagery (MI) and Action Observation (AO) have been used independently in rehabilitation, emerging evidence demonstrates that their combined use elicits superior cortical activation compared to either technique alone [51] [53].

The theoretical foundation of AO+MI lies in the principle of functional equivalence, suggesting that the neural substrates involved in motor execution are also recruited during motor simulation states like observation and imagery [45]. Neurophysiological studies have consistently shown that combined AO+MI produces significantly increased and more widespread activity throughout the motor execution network compared to independent AO or MI, sometimes even exceeding what would be expected from simply summing their individual effects [51].

Quantitative Evidence of Enhanced Cortical Activation

Table 1: Neurophysiological Evidence for AO+MI Enhanced Cortical Activation

Measurement Technique	Key Findings	Research Context
fNIRS [54]	VR-based AO+MI significantly enhanced functional connectivity in bilateral primary sensory cortex (S1), premotor cortex, and supplementary motor area (PM&SMA).	50 healthy participants using VR tasks
fMRI [51]	AO+MI increased neural activity in caudal SMA, basal ganglia, cerebellum, inferior frontal gyrus, and inferior parietal cortex beyond AO or MI alone.	Multiple neuroimaging studies
TMS [53]	Meta-analysis confirmed AO+MI significantly increased corticospinal excitability (MEP amplitudes) compared to control and AO conditions.	Synthesis of TMS studies
EEG [51] [45]	Stronger event-related desynchronization (ERD) in theta, alpha, and beta bands over primary sensorimotor areas during AO+MI compared to independent AO or MI.	Healthy adults and clinical populations
EEG Classification [4]	Transfer learning from motor execution to MI tasks achieved 65.93% accuracy, demonstrating shared neural mechanisms.	28 subjects performing ME, MO, and MI tasks

Table 2: Behavioral Outcomes from AO+MI Interventions in Clinical Populations

Population	Intervention Type	Outcomes	Source
Stroke Patients (13 studies, n=399) [55]	AO+MI vs. routine rehabilitation	Significant improvement in upper extremity function (SMD=1.02) and lower extremity function (SMD=6.31)	Systematic Review & Meta-analysis
Stroke Patients [55]	AO+MI vs. MI alone	Superior improvement in upper extremity function (SMD=0.97)	Systematic Review & Meta-analysis
Young Healthy Adults [56]	MI + tDCS vs. control	Significant improvements in obstacle course time-to-completion immediately post-intervention and at 1-week follow-up	Randomized Controlled Trial
Parkinson's Disease [52]	AOT + MI + Conventional treatment	Protocol for evaluating functional gains in FIM, UPDRS scales and gait analysis (Study ongoing)	Study Protocol

Detailed Experimental Protocols for EEG/fNIRS Research

Core Synchronous AO+MI Protocol for Upper Limb Rehabilitation

This protocol is adapted for EEG/fNIRS studies investigating cortical activation during AO+MI tasks [55] [52]:

Phase 1: Preparation (15 minutes)

• Participant preparation with EEG/fNIRS equipment according to manufacturer specifications
• Electrode placement focusing on motor areas (C3, Cz, C4 according to 10-20 system)
• Signal quality verification and impedance checking
• Baseline recording: 5 minutes resting state with eyes open

Phase 2: Task Block Structure (Total duration: 45-60 minutes)

• Rest Period: 30 seconds of fixation cross display
• Cue Period: 2-second visual or auditory cue indicating upcoming task
• Observation + Imagery Period: 30-second combined AO+MI task
• Response Period: Optional motor execution or rating of imagery vividness
• Inter-trial Interval: 30 seconds rest between trials
• Total Trials: 6-8 blocks of 5-10 trials each with brief breaks between blocks

Phase 3: Task Instructions for Participants

• "Observe the action carefully while simultaneously imagining yourself performing the exact same movement"
• "Focus on the kinesthetic sensations of movement rather than just visual imagery"
• "Maintain the imagery throughout the entire movement demonstration"
• "Avoid actual muscle contraction during the task"

Virtual Reality Enhanced AO+MI Protocol

Leveraging virtual reality (VR) technology enhances immersion and cortical engagement [54] [57]:

Apparatus Setup

• Head-mounted display (VR goggles)
• Motion tracking sensors for hand position monitoring
• Virtual hand rendering synchronized with user's actual hand position
• fNIRS/EEG compatible VR setup

Illusion Induction Phase (10 minutes)

• Virtual hand illusion procedure: participants observe virtual hand moving in synchrony with their own hidden hand
• Reaching tasks to virtual targets to enhance embodiment
• Ownership questionnaire administration

Imagery Session (30 minutes)

• Observation of virtual hand performing target movements (e.g., hand opening/closing)
• Synchronous MI of the observed actions
• Continuous cortical monitoring via fNIRS/EEG
• This approach has demonstrated amplified cortical signals comparable to actual execution and superior to pure MI [57]

Technical Implementation and Research Toolkit

Research Reagent Solutions and Essential Materials

Table 3: Essential Research Materials for AO+MI Studies

Item Category	Specific Examples	Research Function
Neuroimaging Hardware	fNIRS systems (Octamon, Artinis); EEG systems (g.Nautilus PRO); High-density EEG	Monitoring hemodynamic responses and electrical activity during AO+MI tasks [54] [58] [56]
Stimulation Presentation	Head-mounted VR displays (e.g., Huawei VR); Monitors with 60Hz+ refresh rate; Motion tracking sensors	Presenting AO stimuli and creating immersive MI environments [54] [58] [57]
Stimulation Software	Custom VR environments; Video presentation software (e.g., E-Prime, Presentation)	Controlling stimulus timing, duration, and randomization [58] [57]
Peripheral Equipment	Motorized rubber hand setups; Motion actuators (e.g., MX106R, ROBOTIS); Brush stroking kits for RHI	Providing multisensory integration for enhanced embodiment [57]
Validation Tools	Motor Evoked Potential (MEP) measurement via TMS; Electromyography (EMG) systems	Verifying absence of overt movement and measuring corticospinal excitability [53]
Subjective Measures	Vividness of Movement Imagery Questionnaire (VMIQ); Kinesthetic and Visual Imagery Questionnaire (KVIQ); Ownership questionnaires	Assessing imagery ability and embodiment illusion strength [52] [57]

Paradigm Variants for Specific Research Applications

Object-Oriented AO+MI Protocol

• Incorporates goal-directed actions with objects (e.g., grasping a cup)
• Enhances ERD patterns and classification accuracy [45]
• Particularly effective for stroke rehabilitation applications

Gait and Locomotion AO+MI Protocol

• Features lower limb movements and walking observations
• Demonstrates clear alpha desynchronization and movement-related cortical potentials
• Classification accuracies exceeding 80% reported [45]

Adaptive Protocol for BCI Illiteracy/MI Difficulty

• Implements transfer learning from motor execution or observation to MI tasks
• 90% of low performers (accuracy <70%) showed improvement with ME-to-MI transfer [4]
• Effective solution for participants struggling with pure MI

Neural Mechanisms and Experimental Workflow

Neural Mechanisms of AO+MI Cortical Activation

Experimental Workflow for AO+MI EEG/fNIRS Research

Methodological Considerations for Research Applications

Optimization Strategies for Enhanced Cortical Activation

Synchronization Timing

• Ensure precise temporal alignment between observed actions and motor imagery
• Use metronome or rhythmic cues for complex movement sequences
• Implement real-time motion tracking for virtual hand paradigms [57]

Modality Selection

• Kinesthetic MI (feeling the movement) produces stronger sensorimotor activation than visual MI
• First-person perspective videos yield greater motor resonance than third-person
• Object-oriented actions enhance ERD patterns and classification accuracy [45]

Participant Training and Screening

• Assess baseline imagery ability using standardized questionnaires (KVIQ, VMIQ)
• Provide practice sessions with feedback before experimental trials
• Screen for BCI illiteracy and implement transfer learning protocols when needed [4]

Technical Recommendations for EEG/fNIRS Studies

EEG Specific Parameters

• Focus on mu (8-13 Hz) and beta (13-30 Hz) frequency bands for ERD analysis
• Ensure sufficient electrode coverage over sensorimotor areas (C3, Cz, C4)
• Implement EMG monitoring to exclude trials with overt muscle activity
• Target classification accuracy benchmarks: ~70% for usable BCI control [45]

fNIRS Specific Parameters

• Monitor oxygenated hemoglobin (HbO) and total hemoglobin (HbT) as primary indicators
• Position optodes over premotor, primary motor, and prefrontal regions
• Account for hemodynamic response delay (typically 4-6 seconds) in analysis
• VR-compatible systems minimize movement artifacts [54] [56]

The incorporation of AO+MI protocols in EEG and fNIRS research provides a robust method for enhancing cortical activation through synergistic engagement of the mirror neuron system and motor execution networks. The standardized protocols and technical guidelines presented here facilitate implementation across various research contexts from basic motor neuroscience to clinical rehabilitation applications.

Optimizing Signal Quality and Overcoming Common Protocol Challenges

Addressing Participant Non-Responsiveness and the 'BCI-Illiteracy' Problem

BCI illiteracy represents a significant challenge in brain-computer interface research, affecting approximately 15-30% of users who fail to achieve effective control of BCI systems despite adequate training [59] [60]. This phenomenon is particularly prevalent in motor imagery (MI)-based BCIs, where users must generate specific neural patterns through mental simulation of movements without physical execution. The inability of a substantial minority of participants to produce classifiable EEG signals severely limits the practical implementation and reliability of BCI technologies for both assistive devices and research applications [61].

The term "BCI illiteracy" refers to the lack of knowledge and proficiency in using a BCI system within a standard training period [59]. Studies indicate that 55.6% of BCI-naïve participants were unable to achieve 70% performance accuracy across one training session, with 42.9% still failing to reach this threshold after three sessions [60]. This performance variability stems from complex interactions between neurophysiological, psychological, and technical factors that must be addressed through multimodal approaches [61].

This application note synthesizes current methodologies for identifying, addressing, and potentially mitigating BCI illiteracy within the context of motor imagery task protocols for EEG and fNIRS research, providing researchers with practical tools to enhance participant responsiveness and data quality.

Quantifying the BCI-Illiteracy Problem

Table 1: Prevalence and Performance Metrics of BCI Illiteracy Across Studies

Study Reference	Subject Pool	Illiteracy Rate	Performance Threshold	Key Assessment Metric
BCI Competition IV-2a [59]	9 subjects	15-30%	<70% accuracy	MI classification accuracy
Lee et al., 2019 [60]	BCI-naïve participants	55.6%	<70% accuracy across 1 session	Failure rate after initial training
Meng & He, 2019 [60]	BCI-naïve participants	42.9%	<70% accuracy across 3 sessions	Failure rate after multiple sessions
Somatosensory-Motor Imagery Study [60]	14 participants	~64% (9 poor performers)	<70% accuracy	3-class MI classification

Table 2: Performance Comparison Between Good and Poor Performers

Performance Category	Average Classification Accuracy	Improvement Potential	Characteristic Features
Good Performers	80.93% (MI) to 81.79% (SMI) [60]	Minimal with intervention	Consistent ERD/ERS patterns, better focus
Poor Performers	51.45% (MI) to 62.18% (SMI) [60]	~10% with targeted training	Weak ERD/ERS, high variability, attention lapses

Neurophysiological and Psychological Underpinnings

Neural Correlates of MI Performance

Successful motor imagery generates distinct event-related desynchronization (ERD) and event-related synchronization (ERS) patterns in sensorimotor areas, particularly in mu (8-13 Hz) and beta (13-30 Hz) frequency bands [59]. Poor performers demonstrate attenuated or inconsistent ERD/ERS responses, making their neural patterns difficult to distinguish across different MI tasks [60]. The somatosensory cortex activation, when combined with motor cortex signals, has shown potential for improving classification accuracy in poor performers by approximately 10% [60].

Psychological Predictors of Performance

Table 3: Psychological Factors Correlating with BCI Performance

Factor Category	Specific Predictors	Impact Direction	Neural Correlates
User-Technology Relationship	Computer anxiety, fear of incompetence, tension [61]	Negative correlation	Increased right frontal alpha power
Attentional Capacity	Attention span, avoidance of distraction, absorption traits [61]	Positive correlation	Theta oscillations in frontal regions
Spatial Abilities	Mental rotation performance, spatial visualization [61]	Positive correlation	Parietal and occipital activation

Research indicates that attentional traits and spatial abilities constitute the most significant psychological predictors of MI-BCI performance. Users with higher capacities in these domains typically achieve better classification accuracy, likely due to their enhanced ability to generate and maintain vivid motor imagery [61].

Technical Approaches to Mitigate BCI Illiteracy

Signal Processing and Classification Advances

Subject-to-subject semantic style transfer networks (SSSTN) represent a promising approach that transfers class discrimination styles from high-performing subjects (BCI experts) to low-performing subjects through specialized style and content loss functions [59]. This method converts high-dimensional EEG data into images using continuous wavelet transform, then applies transfer learning at the feature level to enhance classification performance for BCI-illiterate users.

Deep learning approaches using convolutional neural networks have shown particular promise for addressing BCI illiteracy, as they can extract relevant features from noisy EEG signals without relying on strong ERD/ERS patterns that poor performers often lack [60].

Multimodal Fusion Approaches

Combining EEG with fNIRS creates complementary information streams that can improve classification robustness. EEG provides millisecond-level temporal resolution of electrical activity, while fNIRS measures hemodynamic responses with better spatial resolution, together offering a more complete picture of neural processing during motor imagery tasks [62] [19].

Diagram 1: Multimodal fusion workflow for addressing BCI illiteracy

Experimental Protocols for Enhanced MI Training

Somatosensory-Motor Imagery (SMI) Protocol

The SMI approach combines traditional motor imagery with somatosensory attentional orientation using tangible objects to enhance neural responses in poor performers [60].

Materials Required:

• EEG acquisition system (64 channels recommended)
• Tangible objects with distinct textures (hard/rough balls)
• Visual cueing system (arrow directions or hand pictures/videos)

Procedure:

• Participant Preparation: Apply EEG cap according to 10-20 system, focusing on sensorimotor areas (C3, C4, Cz)
• Motor Execution Task (MET): Participants physically perform movements (left hand, right hand, right foot) following visual cues (3s duration)
• Motor Imagery Task (MIT): Participants imagine the same movements kinesthetically without physical execution
• Somatosensory-Motor Imagery (SMI): Participants handle textured objects while performing motor execution, then imagine both the movement and tactile sensation simultaneously
• Feedback Phase: Provide neurofeedback of cortical activations 0.5s after cue presentation

Timing Parameters:

• Fixation cross: 2s
• Cue presentation: 3s
• Task execution: 3-5s (depending on paradigm)
• Inter-trial interval: 2-4s with relaxation screen

This protocol demonstrated a 10.73% performance improvement in poor performers compared to traditional MI approaches [60].

Enhanced Paradigm Visualization Protocol

Traditional arrow cues can be supplemented or replaced with more intuitive visualization paradigms to improve imagery quality in naïve users [58].

Table 4: Paradigm Comparison for MI-BCI Acquisition

Paradigm Type	Description	Advantages	Limitations
Traditional Arrow	Arrow pointing left/right indicates imagined hand [58]	Standardized, widely used	Abstract, less intuitive for some users
Hand Picture	Image of hand serves as cue for imagination [58]	More concrete representation	Static, lacks movement context
Hand Video	Video demonstration of hand movement [58]	Provides dynamic movement reference	Potentially more distracting

Procedure:

• Initial Setup: Present fixation cross on black screen (2s)
• Cue Phase: Display paradigm-specific cue (arrow, picture, or video) for 2s
• Imagery Phase: Participants perform MI task for 5s
• Relaxation Phase: Display "Relax" message (2s+ before next trial)
• Randomization: Present cues randomly across trials (40 trials per class recommended)

Studies implementing this multi-paradigm approach achieved up to 97.5% accuracy in naïve subjects, significantly outperforming traditional single-paradigm approaches [58].

The Scientist's Toolkit: Research Reagent Solutions

Table 5: Essential Materials and Solutions for BCI Illiteracy Research

Item Category	Specific Examples	Research Function	Implementation Notes
EEG Systems	g.Nautilus PRO, BioSemi ActiveTwo, Neuracle wireless EEG [60] [58] [24]	Neural signal acquisition	64-channel systems recommended for full sensorimotor coverage
fNIRS Systems	Wearable continuous-wave fNIRS [19]	Hemodynamic response measurement	Complementary to EEG for multimodal approaches
Stimulation Materials	Textured balls, tactile objects [60]	Somatosensory stimulation	Vary texture, hardness for distinct sensory input
Visual Cueing Systems	Arrow cues, hand pictures, instructional videos [58]	Paradigm presentation	Video demonstrations show highest efficacy for naïve users
Classification Algorithms	EEGNet, DeepConvNet, SSSTN, CSP+LDA [59] [24]	Signal classification	Deep learning approaches particularly effective for poor performers
Data Fusion Tools	Joint decomposition, classifier ensembles [19]	Multimodal integration	Addresses limitations of single-modality approaches

Integrated Protocol for Addressing BCI Illiteracy

Diagram 2: Comprehensive assessment and intervention protocol for BCI illiteracy

This integrated protocol recommends a systematic approach to identifying and addressing BCI illiteracy through targeted interventions. Implementation requires approximately 3-5 sessions across different days to account for inter-session variability, with each session lasting 35-48 minutes including breaks to maintain participant focus [24].

Addressing BCI illiteracy requires a multifaceted approach that combines technical innovations in signal processing, enhanced training protocols that incorporate multimodal cues, and individualized approaches based on participant performance tiers. The strategies outlined in this application note demonstrate that performance improvements of 10% or more are achievable in poor performers through targeted interventions.

Future research directions should focus on real-time adaptive systems that can dynamically adjust training paradigms based on ongoing performance, as well as standardized evaluation metrics for BCI literacy across different populations. As these approaches mature, the proportion of users excluded from BCI applications due to illiteracy should substantially decrease, expanding the potential user base for both clinical and research applications.

In the context of motor imagery (MI) task protocols for electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) research, achieving clean neural data is paramount. Both modalities are susceptible to various physiological and motion artifacts that can obscure the genuine brain activity signals of interest. EEG measures electrical potentials on the scalp with millisecond temporal resolution but is highly susceptible to ocular and muscle activity [19]. fNIRS measures hemodynamic changes associated with neural activity with better spatial localization than EEG but is contaminated by systemic physiological noises such as cardiac pulsation, respiration, Mayer waves, and motion artifacts [19] [63] [64]. Effective artifact rejection and signal processing are therefore critical for the validity of subsequent neural decoding analyses, especially in hybrid EEG-fNIRS systems developed for MI-based brain-computer interfaces (BCIs) and neurorehabilitation [6] [5]. This document outlines standardized protocols for tackling these contaminants.

A thorough understanding of potential noise sources is the first step in designing an effective filtering strategy. The following table categorizes common artifacts in EEG and fNIRS recordings during MI tasks.

Table 1: Common Physiological and Motion Artifacts in EEG and fNIRS

Modality	Artifact Category	Specific Source	Typical Frequency Characteristics	Impact on Signal
EEG	Physiological	Ocular (EOG)	Low-frequency (＜ 4 Hz)	High-amplitude, slow drifts
		Cardiac (ECG)	~1 Hz	Periodic spike-like interference
		Muscle (EMG)	Broadband (30+ Hz)	High-frequency, non-stationary
	Motion	Head movement	Variable	Abrupt signal shifts or breaks
		Jaw clenching	High-frequency (＞ 20 Hz)	EMG contamination
fNIRS	Physiological	Cardiac	~1 Hz [63] [64]	Periodic oscillation (0.8-1.2 Hz)
		Respiration	~0.25 Hz [63] [64]	Periodic oscillation (0.2-0.3 Hz)
		Mayer Waves	~0.1 Hz [63] [64]	Low-frequency oscillation (0.07-0.13 Hz)
		Very Low-Frequency Drift	＜ 0.01 Hz [63]	Baseline wander
	Motion	Head movement	Variable, abrupt	Signal spikes and baseline shifts

Signal Processing Techniques for Artifact Rejection

A multi-stage processing pipeline is recommended to address the diverse noise sources outlined above.

fNIRS-Specific Filtering Methods

fNIRS signals require specialized approaches to separate the task-evoked hemodynamic response from pervasive physiological noise.

Table 2: Summary of Advanced Filtering Techniques for fNIRS Physiological Noise

Method	Underlying Principle	Key Parameters	Reported Efficacy	Best Use Cases
MODWT-LSTM [63]	Decomposes resting-state signal via wavelet; trains LSTM to predict & subtract noise in task data.	MODWT Levels: 5-9; Prediction Window: 15s	Effective for low-frequency noise (0.01 Hz) overlapping with HRF [63]	Unknown trial periods; passive BCI
RLSE with EFF [64]	Adaptive filter modeling HR, its derivatives, short-separation, and physiological noises.	Forgetting Factor (λ): 0.99-1; Physiological Frequencies: 1, 0.25, 0.1 Hz	77-99% channel improvement in CNR for HbO/HbR vs. Kalman/ICA [64]	Online/offline processing; known stimulus timing
ML-GESG Filter [65]	Maximum likelihood estimation to filter multivariate disturbances (heartbeat, breathing, shivering).	Multivariate parameter estimation	Better HbO correlation with psychoacoustic indices vs. conventional filters [65]	Complex environments with multiple, concurrent noise sources

EEG-Specific Filtering Methods

While this application note focuses on the MI context shared by both modalities, standard EEG preprocessing remains essential. Robust artifact removal techniques for EEG include Independent Component Analysis (ICA) for isolating and removing ocular and cardiac artifacts, and advanced algorithms like Artifact Subspace Reconstruction (ASR) for handling high-amplitude, transient motion artifacts.

Multimodal Fusion for Enhanced Artifact Handling

The complementary nature of EEG and fNIRS can be leveraged for improved artifact rejection. Data-driven unsupervised symmetric fusion techniques, such as joint source decomposition, can help disentangle shared physiological confounders from underlying neuronal activity, which is particularly beneficial for data from naturalistic environments [19].

Detailed Experimental Protocols

Protocol 1: MODWT-LSTM for fNIRS Physiological Noise Prediction

This protocol is suitable for tasks where the exact timing of brain activation is unknown, making traditional model-driven approaches unsuitable [63].

Application Note: This method is ideal for passive BCIs or studies exploring spontaneous cognitive processes during MI.

Workflow Diagram:

Step-by-Step Procedure:

• Data Acquisition: Collect a 10-minute resting-state fNIRS recording prior to the main task [63].
• Signal Decomposition: Apply the Maximal Overlap Discrete Wavelet Transform (MODWT) to the resting-state data to decompose it into nine levels of wavelet coefficients.
• Noise Component Selection: Identify the wavelet levels corresponding to low-frequency physiological noise (typically levels 5 to 9, which capture frequencies around 0.01 Hz to 0.1 Hz) [63].
• LSTM Training: Train a Long Short-Term Memory (LSTM) network using the selected low-frequency wavelet coefficients from the resting-state data. The LSTM learns to predict the temporal structure of the physiological noise.
• Noise Prediction & Subtraction: During the task period, use the trained LSTM model to predict the physiological noise component for the subsequent task data window (e.g., 15 seconds). Subtract the predicted noise signal from the actual task fNIRS recording.
• Validation: Compare the cleaned signal with the original and validate using performance metrics like Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).

Protocol 2: Recursive Least-Squares (RLS) Adaptive Filtering

This protocol uses an adaptive filter to model and remove known physiological noises and is effective for both online and offline processing [64].

Application Note: Best applied when the experimental paradigm has known trial timing and access to short-separation channels.

Workflow Diagram:

Step-by-Step Procedure:

• Model Construction: Formulate a linear regression model for the fNIRS signal y(t) that includes [64]:
  • The expected hemodynamic response (HR) u(t) and its first and second derivatives (Δu(t), Δ²u(t)).
  • Short-separation channel data y_SS(t) to account for superficial scalp hemodynamics.
  • A sum of sinusoidal functions Σ b_m sin(2πf_m t) modeling cardiac (~1 Hz), respiratory (~0.25 Hz), and Mayer wave (~0.1 Hz) noises.
  • A baseline drift term b_0.
• Parameter Estimation: Employ the Recursive Least-Squares Estimation (RLSE) algorithm with an exponential forgetting factor (e.g., λ = 0.99) to adaptively estimate the unknown parameters (weights) of the linear model.
• Noise Removal: The cleaned hemodynamic response is derived from the estimated parameters associated with the expected HR component. The other modeled components (short-separation, physiological noises, baseline) are effectively regressed out.

The Scientist's Toolkit

Table 3: Essential Research Reagent Solutions for Hybrid EEG-fNIRS MI Research

Item	Specification / Function	Example Use Case
Hybrid EEG-fNIRS Cap	Integrated cap with EEG electrodes and fNIRS optodes positioned over sensorimotor cortices [6] [5].	Standardized scalp positioning for MI studies (e.g., C3, Cz, C4 locations).
Short-Separation fNIRS Channels	Optode pairs with ＜ 1 cm separation to measure extracerebral hemodynamics [64].	Used as a regressor in RLS filtering to separate scalp from brain activity [64].
Synchronization Trigger Box	Hardware to send simultaneous event markers to both EEG and fNIRS recording systems.	Temporal alignment of EEG epochs and fNIRS blocks with task events (e.g., MI cue onset) [5].
Validated Motor Imagery Paradigm	Software for presenting visual cues (e.g., left/right hand arrows) with fixed trial timing [5].	Elicits reproducible event-related desynchronization in EEG and hemodynamic responses in fNIRS.
High-Density DOT Arrays	High-density arrays of multiple source–detector separations for 3D image reconstruction [19].	Enables improved spatial localization of MI-activated brain regions when using HD-DOT.

In electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) research, the integrity of motor imagery (MI) data is contingent on the participant's adherence to the protocol, specifically the absence of overt or covert motor execution. Covert movements, defined as subtle, sub-threshold muscle activations without visible motion, can significantly contaminate neural signals by introducing movement-related artifacts and confounding the interpretation of sensorimotor rhythms. Electromyography (EMG) monitoring provides an objective, quantitative method to ensure task compliance and validate the purity of MI paradigms. This document outlines the application and protocols for integrating EMG monitoring into MI research to safeguard data quality and bolster the validity of findings in brain-computer interface (BCI) and neurorehabilitation studies.

The Critical Need for EMG Monitoring in MI Research

Motor imagery and motor execution share overlapping neural substrates, eliciting similar patterns of sensorimotor rhythm desynchronization, such as mu (8-13 Hz) and beta (13-30 Hz) event-related desynchronization (ERD) [45]. However, MI-induced ERD is typically of smaller amplitude and broader spatial distribution than that of motor execution (ME). The inadvertent occurrence of covert movements during MI tasks can blur this distinction, leading to erroneous conclusions.

Evidence indicates that the reliability of MI-ME convergence improves with kinesthetic instructions and higher imagery vividness [45]. Nevertheless, without rigorous monitoring, it is impossible to distinguish between genuine cortical activity from pure imagery and that contaminated by subtle motor commands. Quantitative EMG monitoring is therefore recommended to:

• Verify Task Compliance: Objectively confirm the absence of muscle activity during MI blocks.
• Prevent Data Contamination: Mitigate the risk of movement artifacts in EEG/fNIRS signals.
• Explain Performance Variability: Identify if poor BCI classifier performance (e.g., below the ~70% usability threshold) or anomalous neural patterns are linked to non-compliance or covert movement [45].
• Enhance Protocol Standardization: Address a key methodological inconsistency noted across MI studies, many of which lack proper EMG monitoring [45].

Quantitative EMG Metrics for Compliance Verification

EMG data should be quantitatively assessed to establish a baseline and define thresholds for acceptable muscle activity during MI. The following table summarizes key metrics and parameters for setting up EMG monitoring in an MI paradigm.

Table 1: EMG Monitoring Parameters and Metrics for MI Task Compliance

Parameter/Metric	Specification	Rationale & Application
Muscle Groups	Hand: First Dorsal Interosseous, Abductor Pollicis Brevis. Forearm: Flexor Digitorum Superficialis, Extensor Digitorum.	Covers primary muscles involved in hand and finger movements commonly used in upper-limb MI tasks [66].
Baseline Recording	≥ 2 minutes of rest (muscle fully relaxed).	Establishes individual baseline EMG amplitude and noise floor for each session.
Maximum Voluntary Contraction (MVC)	3 trials of maximum force exertion, e.g., gripping a dynamometer, each held for 3-5 seconds with rest intervals.	Provides a reference for normalizing EMG activity and setting participant-specific thresholds [5].
Compliance Threshold	Root Mean Square (RMS) of EMG signal during MI task must be < 5% of MVC RMS and within 3 standard deviations of the baseline rest period.	A conservative, quantitative threshold to flag trials with significant covert activation.
Data Quality Metric	Signal-to-Noise Ratio (SNR) should be maximized through proper electrode placement and skin preparation.	Ensures EMG signals are detectable above electrical and environmental noise [66].

Experimental Protocol for Integrated EEG/fNIRS-EMG

This protocol details the simultaneous acquisition of EEG, fNIRS, and EMG during a standardized left-right hand motor imagery task.

Participant Preparation and Equipment

• Participants: Recruit participants based on approved ethics. Screen for imagery ability using the Kinesthetic and Visual Imagery Questionnaire (KVIQ-20), with a score of ≥60 indicating good imagery ability [67].
• EEG/fNIRS Setup: Utilize a hybrid EEG-fNIRS cap (e.g., 32 EEG electrodes, 90 fNIRS channels) positioned according to the international 10-20 system, ensuring coverage over the sensorimotor cortex (C3, Cz, C4) [5].
• EMG Setup:
  • Electrodes: Apply disposable, pre-gelled surface EMG electrodes in a bipolar configuration.
  • Placement: After light abrasion and cleaning of the skin, place electrodes on the muscle belly of the target muscles (e.g., right and left First Dorsal Interosseous) with a ground electrode on a bony prominence (e.g., wrist or elbow).
  • Impedance Check: Ensure electrode-skin impedance is below 10 kΩ for EEG and 5 kΩ for EMG.
  • Amplifier: Use an EMG amplifier with a sampling rate of at least 1000 Hz.

Calibration and Baseline Recording

• Grip Strength Calibration: To enhance kinesthetic vividness, have participants perform repeated grip force exertions (e.g., 5 kg maximum force) using a dynamometer or stress ball. This reinforces the tactile and force sensations of the movement to be imagined [5].
• EMG Baseline & MVC: Record a 2-minute baseline with the participant at rest. Subsequently, record three trials of Maximum Voluntary Contraction (MVC) for the target muscles.
• Signal Synchronization: Use a common trigger system (e.g., from E-Prime or similar presentation software) to send event markers simultaneously to the EEG, fNIRS, and EMG recording systems [5].

Motor Imagery Task Execution

The following workflow diagram summarizes the experimental procedure from preparation to data validation.

• Paradigm Structure: Each trial should be structured as follows [5]:
  • Visual Cue (2s): A directional arrow (left or right) is displayed.
  • Execution Phase (10s): A fixation cross appears. Participants perform kinesthetic MI of a grasping movement with the cued hand at a rate of one imagined grasp per second. EMG monitoring is critical during this phase.
  • Inter-Trial Interval (15s): A blank screen is shown for rest.
• Session Details: Conduct at least two sessions, each containing 15 trials per hand (60 trials total), with breaks between sessions to prevent fatigue.

Data Analysis and Compliance Check

• EMG Processing: Filter the raw EMG signal (e.g., band-pass 20-450 Hz). Calculate the Root Mean Square (RMS) in epochs time-locked to the MI execution phase.
• Compliance Validation: For each trial, compare the RMS during the MI phase to the baseline and MVC values. A trial is considered valid only if the EMG RMS remains below the 5% MVC threshold.
• Neural Signal Analysis: Proceed with the analysis of EEG ERD/ERS and fNIRS hemodynamic responses only for validated trials.

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 2: Essential Materials for EEG/fNIRS-EMG Motor Imagery Research

Item	Function & Application
Hybrid EEG-fNIRS Cap	Integrated headgear providing simultaneous electrophysiological and hemodynamic monitoring of the sensorimotor cortex [5].
Surface EMG Electrodes & Amplifier	Disposable Ag/AgCl electrodes and a dedicated amplifier for capturing muscle electrical activity. Critical for quantifying covert movement [68].
Electrode Abrasive Gel & Skin Prep	Used to reduce skin impedance at electrode sites, improving signal quality for both EEG and EMG.
Trigger Interface Box	Hardware to synchronize event markers from the task presentation software (e.g., E-Prime) with all data acquisition systems (EEG, fNIRS, EMG) [5].
Grip Dynamometer / Stress Ball	Calibration tool for standardizing the motor command and enhancing the kinesthetic vividness of motor imagery before the task [5].
Kinesthetic and Visual Imagery Questionnaire (KVIQ-20)	A validated tool to screen and select participants with adequate motor imagery ability, reducing the incidence of BCI "illiteracy" [67].

Motor imagery (MI)-based brain-computer interface (BCI) technology has emerged as a transformative approach for motor rehabilitation in clinical populations with neurological damage, particularly stroke and intracerebral hemorrhage (ICH) patients. These systems establish direct communication pathways between the brain and external devices, bypassing damaged neural pathways by decoding cortical activity to facilitate functional recovery [27]. The integration of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) has created powerful hybrid BCI systems that leverage complementary neurophysiological signals, offering both millisecond temporal resolution from EEG and superior spatial localization from fNIRS [5] [27]. This article provides detailed application notes and experimental protocols for adapting MI tasks for stroke and ICH populations within EEG and fNIRS research paradigms, addressing the unique pathophysiological challenges these patients present.

Key Challenges in Clinical Populations

Stroke and ICH patients exhibit fundamental divergences from neurotypical populations in both neuroelectrophysiological and hemodynamic signatures, creating significant challenges for BCI-based rehabilitation [5]. Table 1 summarizes the primary clinical challenges and adaptation strategies for these populations.

Table 1: Clinical Challenges and Adaptation Strategies for Stroke and ICH Patients

Challenge	Clinical Impact	Adaptation Strategy
Neurovascular Uncoupling	Altered relationship between neural activity and blood flow in ICH patients [5]	Multimodal monitoring combining EEG (neural) and fNIRS (hemodynamic) [5]
Cortical Underactivity	Reduced ERD/ERS patterns during MI tasks [69] [27]	Robotic assistance and tactile feedback to enhance cortical engagement [69]
BCI Illiteracy	Inability to effectively generate/control MI signals [70]	Pre-training with kinesthetic calibration and stress ball exercises [5]
Individual Variability	Significant differences in neural distribution patterns [27]	Transfer learning algorithms using normal subject templates [27]
Cognitive Impairment	Difficulty understanding MI concepts [5]	Simplified instructions with physical movement simulation [5]

Protocol Adaptations for Clinical Populations

Participant Screening and Preparation

The successful implementation of MI protocols for clinical populations requires careful participant selection and preparation. For ICH patients, specific inclusion criteria should encompass first-time ICH with unilateral upper limb motor dysfunction, time since onset ranging from 2 days to 2 months, and sufficient cognitive capacity to follow instructions (MMSE ≥ 18) [69] [5]. Critical exclusion criteria include other neurological disorders, sensory or mixed aphasia, history of epilepsy, and significant skull defects affecting signal acquisition [69].

A crucial preparatory phase involves kinesthetic calibration to enhance MI vividness. This protocol includes:

• Repeated 5 kg maximal force exertions using a dynamometer
• Equivalent force applications using a stress ball
• Grip training at a rate of one contraction per second [5]

This calibration reinforces tactile and force-related aspects of grasping movements, standardizing temporal rhythm and improving consistency across MI trials.

Experimental Paradigm Design

The MI paradigm should be structured to accommodate the fatigue patterns and attention limitations of clinical populations. A standardized protocol consists of:

• Baseline Recording: 1-minute eyes-closed followed by 1-minute eyes-open states [5]
• Trial Structure:
  • Visual cue presentation (2s): Directional arrow indicating left/right hand MI
  • Execution phase (10s): Kinesthetic MI of grasping movement at 1Hz rate
  • Inter-trial interval (15s): Blank screen for rest [5]
• Session Configuration: Minimum of 2 sessions with 15 trials each, with intersession breaks adjusted based on participant readiness [5]

For stroke patients with more severe impairments, incorporating robotic hand assistance provides critical closed-loop feedback. When EEG features match MI characteristics, the system triggers robotic hand movement that executes the corresponding action, providing tactile feedback alongside ongoing auditory and visual cues [69].

Detailed Experimental Protocols

Hybrid EEG-fNIRS Acquisition Protocol

The integration of EEG and fNIRS requires careful technical configuration to ensure optimal signal quality and temporal synchronization.

Table 2: Hybrid EEG-fNIRS System Configuration

Component	Specification	Function
EEG System	g.HIamp amplifier (g.tec) [5]	Records cortical electrophysiological activity
EEG Sampling Rate	256 Hz [5]	Captures millisecond temporal dynamics
fNIRS System	NirScan (Danyang Huichuang) [5]	Monitors hemodynamic changes
fNIRS Sampling Rate	11 Hz [5]	Tracks slower hemodynamic responses
EEG Electrodes	32-channel configuration [5]	Comprehensive cortical coverage
fNIRS Channels	32 sources, 30 detectors (90 channels) [5]	Hemodynamic monitoring across motor cortex
Synchronization	E-Prime 3.0 event markers [5]	Temporal alignment of multimodal data

The hybrid cap should integrate EEG electrodes and fNIRS optodes in an optimized topography, with fNIRS source-detector pairing at controlled separation distances (3 cm) to enable simultaneous electrophysiological and hemodynamic monitoring across prefrontal, motor, and association cortices [5].

Data Processing and Analysis Framework

The analysis of hybrid EEG-fNIRS data from clinical populations requires specialized processing approaches:

EEG Feature Extraction:

• Focus on event-related desynchronization (ERD) in α (8-12 Hz) and β (12-27 Hz) bands over sensorimotor regions [69] [27]
• Calculate laterality indices to assess interhemispheric balance [71]
• Compute power spectral density (PSD) and brain symmetry index (BSI) for prognostic value [72]

fNIRS Feature Extraction:

• Monitor changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) [5] [6]
• Extract statistical descriptors (mean, kurtosis) for classification [27]

Multimodal Fusion:

• Implement transfer learning algorithms to address individual variability [27]
• Apply Wasserstein metric-driven source domain selection to quantify inter-subject neural distribution divergence [27]
• Utilize feature-level fusion with adaptive weighting mechanisms to enhance multimodal data synergy [5]

The following workflow diagram illustrates the comprehensive experimental procedure for hybrid EEG-fNIRS MI protocols with clinical populations:

Neurophysiological Mechanisms and Biomarkers

Understanding the neural mechanisms underlying MI-based rehabilitation is essential for protocol optimization. In clinical populations, key biomarkers provide critical insights into recovery trajectories and treatment efficacy.

Event-Related Desynchronization/Synchronization (ERD/ERS): During MI tasks, stroke patients typically exhibit ERD in the α and β frequency bands over sensorimotor regions, reflecting suppression of motor cortex activity as the brain prepares for movement without actual execution [69]. This ERD serves as a crucial indicator of motor cortex activation and has been studied as a biomarker of cortical reorganization in stroke rehabilitation [69]. In contrast, ERS commonly occurs after movement cessation or during recovery phases, reflecting re-engagement and reorganization of the motor cortex [69].

Hemodynamic Response Patterns: fNIRS captures complementary hemodynamic responses during MI tasks, with increased oxygen consumption in activated regions leading to elevated HbR levels and decreased HbO levels [27]. In ICH patients, neurovascular uncoupling may alter these typical patterns, necessitating multimodal assessment approaches [5].

Predictive Biomarkers for Recovery: Resting-state EEG parameters show significant prognostic value for upper limb motor recovery:

• Beta band power spectral density (PSD) correlates positively with FMA-UL and ARAT scores [72]
• Brain Symmetry Index (BSI) demonstrates negative correlation with Brunnstrom-UL, FMA-UL, and MBI scores [72]
• Baseline PSD predicts hand function recovery (FMA-Hand), while PSD at week 1 predicts ARAT improvement [72]

The following diagram illustrates the neurofeedback loop that facilitates motor recovery through operant conditioning of brain signals:

The Scientist's Toolkit: Research Reagent Solutions

Implementing effective MI protocols for clinical populations requires specific tools and assessment instruments. Table 3 details essential research reagents and their applications in stroke and ICH rehabilitation studies.

Table 3: Essential Research Reagents and Tools for Clinical MI Studies

Tool/Assessment	Specification	Research Application
RxHEAL BCI System	BCI Hand Rehabilitation Training System [69]	Closed-loop MI training with robotic assistance
g.HIamp Amplifier	256 Hz sampling rate, 32 channels [5]	High-quality EEG signal acquisition
NirScan fNIRS	11 Hz sampling rate, 90 measurement channels [5]	Hemodynamic response monitoring
Fugl-Meyer Assessment (FMA-UE)	Performance-based quantitative measure [5]	Assesses motor function, coordination, and reflexes
Modified Barthel Index (MBI)	Activities of daily living assessment [5]	Measures functional independence
Modified Rankin Scale (mRS)	Global disability measurement [5]	Evaluates overall functional recovery
Brain Symmetry Index (BSI)	EEG-derived asymmetry parameter [72]	Quantifies interhemispheric power distribution
Wasserstein Metric Algorithm	Transfer learning method [27]	Quantifies inter-subject neural distribution divergence

Adapting motor imagery protocols for stroke and ICH patients requires specialized approaches that address the unique pathophysiological signatures of these clinical populations. The integration of hybrid EEG-fNIRS systems provides complementary neural and hemodynamic data that enhances classification accuracy and enables more robust decoding of motor intentions in compromised brains. Critical success factors include kinesthetic calibration to enhance MI vividness, closed-loop feedback systems incorporating robotic assistance, transfer learning algorithms to address individual variability, and standardized assessment protocols to quantify functional outcomes. Future research directions should focus on optimizing multimodal integration, developing more adaptive classification algorithms that accommodate dynamic neural reorganization, and establishing standardized protocols that can be deployed across diverse clinical settings. By implementing these tailored strategies, researchers can advance the development of personalized rehabilitation systems that significantly improve upper limb motor recovery for stroke and ICH patients.

Motor imagery (MI), the mental simulation of a movement without physical execution, is a cornerstone of modern neurorehabilitation and brain-computer interface (BCI) research. This application note details the methodology for leveraging real-time neurofeedback (NF) to enhance MI performance, with a specific focus on integrated electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) protocols. We provide a comprehensive framework for researchers and clinicians, including standardized experimental setups, quantitative data comparisons, and visualization of signaling pathways. The protocols are designed to be implemented within a broad thesis on motor imagery task protocols, offering a validated approach for improving volitional control and inducing neuroplastic changes in both healthy and clinical populations, such as post-stroke patients.

Neurofeedback operates on the principle of operant conditioning, where individuals learn to self-regulate their brain activity through real-time feedback derived from neuroimaging data [73]. This process is thought to induce functional plasticity, leading to lasting changes in brain function and potential improvements in motor recovery [6]. While unimodal NF has shown promise, it faces challenges, including a roughly 30% non-response rate often attributed to the limitations of any single neuroimaging modality [6].

Integrating EEG and fNIRS creates a synergistic multimodal NF system. EEG provides excellent temporal resolution (milliseconds), capturing rapid neuronal oscillations during MI tasks. Conversely, fNIRS measures hemodynamic responses—changes in oxygenated hemoglobin (HbO2) concentration—offering superior spatial specificity to the sensorimotor cortex [5] [6]. This combination mitigates the weaknesses of each standalone method and has been shown to improve classification accuracy in MI-BCI systems by 5-10% compared to unimodal systems [5]. This protocol outlines the application of this multimodal approach to enhance MI performance.

Experimental Protocols

Participant Preparation and Calibration

Best Practices for Data Fidelity:

• Scheduling: Conduct recordings in the morning to minimize the impact of daily fatigue and stressors [74].
• Substance Restrictions: To prevent signal alteration, participants should avoid caffeine for at least 12 hours, cannabis for 24-48 hours, and psychostimulants (e.g., Adderall, Ritalin) for 48 hours prior to recording [74].
• Grip Strength Calibration: To enhance the kinesthetic vividness of MI, introduce a pre-acquisition calibration procedure. This involves:
  • Repeated maximal force exertions using a dynamometer.
  • Equivalent force applications using a stress ball.
  • Grip training at a rate of one contraction per second [5]. This procedure reinforces tactile and force-related aspects of movement, standardizing the temporal rhythm and improving MI consistency across trials.

Data Acquisition Hardware Setup

Integrated EEG-fNIRS Cap Assembly:

• Use a custom-designed hybrid cap (e.g., Model M for head circumferences 54-58 cm) integrating both modalities.
• EEG Configuration: Arrange 32 electrodes according to the international 10-20 system, ensuring coverage of the sensorimotor cortices (C3, Cz, C4). Use a sampling rate of 256 Hz [5].
• fNIRS Configuration: Arrange 32 laser sources and 30 photodetectors to create approximately 90 measurement channels via source-detector pairing at a fixed distance of 3 cm. This topology should cover prefrontal, motor, and association cortices. Use a sampling rate of 11 Hz for hemodynamic signals [5].
• Synchronization: Link the EEG amplifier (e.g., g.HIamp) and fNIRS system (e.g., NirScan) to a stimulus presentation computer (e.g., E-Prime 3.0) that sends simultaneous event markers to both recording systems to ensure temporal alignment of data streams [5].

Motor Imagery Neurofeedback Paradigm

The following protocol, adapted from established studies, involves a blocked design executed over multiple sessions [5] [6].

• Baseline Recording (2 minutes):

  • Record neural signals during a 1-minute eyes-closed state, followed by a 1-minute eyes-open state. Demarcate with an auditory cue (200 ms beep).

• Trial Structure (per trial):

  • Visual Cue (2 seconds): Present a yellow directional arrow (left/right) on a blue background indicating the hand for MI.
  • Execution/Feedback Phase (10 seconds): Display a central yellow fixation cross accompanied by an auditory cue. During this period:
    • The participant performs kinesthetic MI of grasping with the cued hand at ~1 grasp/second.
    • Real-time NF is presented as a visual gauge (e.g., a ball moving vertically). The position is controlled by a computed NF score derived from target region activity [6].
  • Inter-Trial Interval (15 seconds): Show a blank screen for rest.

• Session Structure: Each session should contain a minimum of 30 trials (15 left/right hand each). Participants should undergo at least two sessions, with rest intervals between to mitigate fatigue. The order of NF conditions (EEG-only, fNIRS-only, EEG-fNIRS) should be randomized across participants to control for order effects [6].

Real-Time Signal Processing and Feedback Calculation

Data processing must occur in near real-time to provide instantaneous feedback. The following workflow and diagram outline this process.

Processing Steps:

• EEG Processing: Apply spatial and temporal filters. Compute the Event-Related Desynchronization (ERD)—a decrease in oscillatory power within the mu (8-12 Hz) and beta (12-30 Hz) frequency bands over the contralateral sensorimotor cortex during MI [75] [6].
• fNIRS Processing: Convert raw light intensity signals into concentration changes for oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (HbR). Focus on HbO2 increases in the contralateral primary motor cortex as the primary hemodynamic feature [5] [6].
• NF Score Calculation: Develop an algorithm to combine the processed EEG and fNIRS features into a single, normalized NF score. This can involve feature-level fusion with adaptive weighting to leverage the spatiotemporal synergy of the two modalities [5].

The tables below summarize key metrics and parameters from relevant studies and this protocol to facilitate comparison and replication.

Table 1: Key Metrics from Multimodal MI-NF Studies

Study / Dataset	Participants	Modalities	Key Performance Metrics	Clinical Outcome Measures
HEFMI-ICH Dataset [5]	20 ICH Patients, 17 Healthy	EEG & fNIRS	Improved classification accuracy vs. unimodal systems (5-10%)	FMA-UE, MBI, mRS
Muller et al. (2025) Protocol [6]	30 Healthy	EEG & fNIRS	NF score modulation, task-related brain activity specificity	N/A (Protocol)
fMRI-NF Protocol [76]	Healthy & Tinnitus	fMRI	Volitional control of target ROI (e.g., auditory cortex)	Tinnitus symptom reduction

Table 2: Acquisition Parameters for Multimodal Setup

Parameter	EEG Specifications	fNIRS Specifications
Sampling Rate	256 Hz [5]	11 Hz [5]
Key Channels/Regions	C3, Cz, C4 (Sensorimotor)	Primary Motor Cortex (Contralateral)
Primary Features	ERD in Mu/Beta Bands [6]	HbO2 Concentration [5]
Temporal Resolution	High (Milliseconds) [6]	Moderate (Seconds) [6]
Spatial Resolution	Low (Centimeters) [5]	Moderate (5-10 mm) [5]

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Essential Materials for EEG-fNIRS MI Neurofeedback

Item	Function & Specification	Example Product/Note
Hybrid EEG-fNIRS Cap	Integrates sensors for simultaneous data acquisition; follows 10-20 system for EEG.	Custom design with 32 EEG electrodes, 32 fNIRS sources, 30 detectors [5].
Amplifier & fNIRS Unit	Amplifies, digitizes, and records raw neural signals.	g.HIamp amplifier (EEG); NirScan fNIRS system [5].
Stimulus Presentation Software	Presents visual cues and NF display; sends synchronization triggers.	E-Prime 3.0 [5].
Real-Time Processing Software	Performs signal preprocessing, feature extraction, and NF score calculation.	Custom software in MATLAB/Python; OpenVIBE.
Normative Database & Analysis Tool	Provides reference for QEEG feature comparison and analysis.	Neuroguide database [74].
Grip Strength Calibration Tools	Enhances kinesthetic vividness and standardizes MI tempo.	Dynamometer, stress ball [5].

Signaling Pathways and Logical Workflows

The efficacy of MI-NF is rooted in the neurovascular coupling mechanism and the closed-loop feedback system. The following diagram illustrates the pathway from mental strategy to measurable neural response and feedback.

Pathway Explanation: The process begins with the participant engaging in Motor Imagery. This mental act drives Neural Firing in the sensorimotor cortex, which is directly captured by EEG as a desynchronization in specific rhythms (ERD). Concurrently, this neural activity triggers a Hemodynamic Response via Neurovascular Coupling, leading to an increase in cerebral blood flow and oxygenated hemoglobin (HbO2) in the active region, which is measured by fNIRS. Both the EEG and fNIRS features are then fused into a single NF Score that is presented to the user as Feedback. This feedback loop enables Operant Learning, allowing the user to subconsciously or consciously refine their mental strategy to better modulate the target brain signals.

Benchmarking Performance and Validating Protocols for Clinical Translation

Motor Imagery (MI) decoding is a fundamental component of modern Brain-Computer Interface (BCI) systems, enabling direct communication between the brain and external devices without peripheral nervous system involvement. Establishing clear performance benchmarks for MI classification accuracy is crucial for evaluating algorithmic advances and facilitating clinical translation. This document outlines established accuracy benchmarks, detailed experimental protocols, and essential methodological considerations for EEG and fNIRS-based MI decoding within hybrid BCI systems, providing researchers with standardized reference points for evaluating system performance.

The complementary nature of EEG and fNIRS signals offers significant advantages for MI-BCIs. EEG provides millisecond-level temporal resolution ideal for capturing rapid neuronal activation patterns during MI tasks, while fNIRS delivers superior spatial localization (5–10 mm resolution) of hemodynamic responses associated with cortical activation [5]. This spatiotemporal synergy enables hybrid systems to achieve classification accuracies surpassing unimodal approaches by 5%–10% [5], making them particularly valuable for clinical applications such as stroke rehabilitation and assistive technologies for motor-impaired individuals.

Current Classification Accuracy Benchmarks

Comparative Performance Across Methodologies

Table 1: Classification accuracy benchmarks for MI decoding across modalities and methods

Modality	Methodology	Dataset	Subjects	Accuracy	Reference
EEG-fNIRS Fusion	Deep Learning + Evidence Theory	TU-Berlin-A	29	83.26%	[29]
EEG-fNIRS Fusion	Early-Stage Fusion Y-Network	Shin Dataset A	29	76.21%	[17]
fNIRS Only	TopoTempNet	MA, WG, UFFT	85 total	Up to 90.04% ± 3.53%	[77]
Hybrid NIRS-EEG	Linear Discriminant Analysis	4-direction decoding	12	>80%	[78]
EEG Only	Conventional Methods	Shin Dataset A	29	~65%	[17]
fNIRS Only	Conventional Methods	Shin Dataset A	29	~57%	[17]

Performance in Clinical Populations

The HEFMI-ICH dataset, incorporating data from 20 intracerebral hemorrhage (ICH) patients, provides crucial benchmarks for clinical applications. While specific accuracy figures for patient populations are not provided in the available literature, the dataset enables development of models accounting for neurovascular uncoupling and other pathophysiological factors distinct from neurotypical cohorts [5]. This resource is particularly valuable for advancing personalized rehabilitation BCI development tailored to patients with motor impairments resulting from stroke.

Detailed Experimental Protocols

Standardized Motor Imagery Paradigm

The following protocol, validated across multiple studies including the HEFMI-ICH dataset, provides a robust framework for MI data acquisition:

Participant Preparation and Calibration:

• Conduct grip strength calibration using a dynamometer and stress ball to enhance MI vividness
• Perform repeated 5 kg maximal force exertions (or voluntary maximum efforts)
• Conduct equivalent force applications using a stress ball
• Implement grip training at one contraction per second to standardize temporal rhythm [5]

Experimental Sequence:

• Position participants 25 cm from display monitor in ergonomic seating
• Record baseline signals: 1-minute eyes-closed followed by 1-minute eyes-open states
• Demarcate phases with auditory cue (200 ms beep)
• Structure each trial as follows:
  • Visual cue presentation (2 s): Display yellow directional arrow (left/right) on blue background
  • Execution phase (10 s): Present central yellow fixation cross following auditory cue; participants perform kinesthetic MI of grasping movement at ~1 Hz
  • Inter-trial interval (15 s): Blank screen for rest period [5]
• Conduct minimum of 2 sessions per subject with 15 trials each (30 trials total per hand)
• Adjust intersession breaks based on participant readiness to mitigate fatigue

Data Acquisition Specifications

Table 2: Research reagent solutions for hybrid EEG-fNIRS MI studies

Equipment Category	Specific Product/Model	Key Parameters	Function in Research
EEG Amplifier	g.HIamp amplifier (g.tec)	32 channels, 256 Hz sampling rate	Records electrical brain activity with millisecond temporal resolution
fNIRS System	NirScan (Danyang Huichuang)	11 Hz sampling rate, 90 measurement channels	Measures hemodynamic responses via near-infrared light
Hybrid Cap	Custom Model M (54-58 cm)	32 EEG electrodes, 32 optical sources, 30 photodetectors	Enables synchronized multimodal data acquisition
Stimulus Presentation	E-Prime 3.0	N/A	Prescribes experimental paradigm and triggers synchronization
Grip Strength Calibration	Dynamometer & Stress Ball	5 kg maximal exertions	Enhances kinesthetic MI vividness through tactile reinforcement

Synchronization and Setup:

• Transmit event markers from E-Prime 3.0 to simultaneously trigger both recording systems
• Arrange fNIRS optodes in anatomically-guided configuration with source-detector separation of 3 cm
• Position EEG electrodes using international 10-20 system expansion for comprehensive cortical coverage [5]
• For specific investigations, position fNIRS over prefrontal region and EEG over motor cortex to leverage regional specialization [78]

Signal Processing Pipeline

EEG Preprocessing:

• Downsample from 200 Hz to 128 Hz
• Remove EOG channels and re-reference to common average
• Apply band-pass filter (8–25 Hz) to preserve μ-band and low-β band [17]
• Extract spatiotemporal features using dual-scale temporal convolution and depthwise separable convolution
• Implement hybrid attention module to enhance sensitivity to salient neural patterns [29]

fNIRS Preprocessing:

• Convert raw optical density changes to HbO and HbR concentration changes using modified Beer-Lambert law
• Apply task-specific filtering (bandpass 0.01–0.1 Hz for MA; lowpass 0.2 Hz cutoff for WG)
• Segment signals into non-overlapping 1 s epochs with pre- and post-stimulus periods
• Perform baseline correction using pre-stimulus reference interval [77]
• Employ spatial convolution across channels to explore regional activation differences
• Implement parallel temporal convolution with GRU to capture hemodynamic response dynamics [29]

Methodological Considerations for Accuracy Optimization

Fusion Strategies for Hybrid BCI

The stage of multimodal fusion significantly impacts classification performance. Early-stage fusion of EEG and fNIRS has demonstrated statistically superior accuracy (76.21%) compared to middle-stage and late-stage fusion approaches (N=57, P<0.05) [17]. The Y-shaped network architecture, employing separate encoders for each modality with fusion before final classification, has proven particularly effective.

For decision-level fusion, advanced methods include:

• Quantifying decision outputs using Dirichlet distribution parameter estimation to model uncertainty
• Implementing two-layer reasoning processes using Dempster-Shafer Theory (DST)
• Fusing evidence from basic belief assignment (BBA) methods across modalities [29]

Interpretation and Clinical Translation

Beyond raw accuracy metrics, interpretable decoding methods provide insights into neurophysiological mechanisms underlying MI. The analysis of functional connectivity metrics (connection strength, density, global efficiency) reveals task-specific brain network patterns [77]. For clinical applications, incorporate standardized assessments:

• Fugl-Meyer Assessment for Upper Extremities (FMA-UE) to quantify motor impairment
• Modified Barthel Index (MBI) to measure functional independence
• Modified Rankin Scale (mRS) to evaluate disability degrees [5]

Experimental Workflow and Signaling Pathways

Experimental Workflow for Hybrid EEG-fNIRS MI Studies

Neurophysiological Signaling Pathways in MI Decoding

Brain-computer interfaces (BCIs) have emerged as a transformative technology for facilitating direct communication between the brain and external devices. In the specific domain of motor imagery (MI) task protocols, electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) have stood as the two predominant non-invasive neuroimaging modalities. EEG measures the electrical activity of neurons with excellent temporal resolution, while fNIRS measures hemodynamic responses correlated with neural activity with better spatial resolution and less susceptibility to noise. Historically, MI-based BCI systems were developed using either EEG or fNIRS in a unimodal configuration. However, a growing body of research indicates that hybrid systems, which integrate these complementary modalities, can overcome the inherent limitations of each unimodal approach and yield superior performance. This application note provides a detailed comparative analysis, framed within a broader thesis on motor imagery protocols, to guide researchers and scientists in the field. We present quantitative performance data, detailed experimental methodologies, and essential technical protocols for implementing both unimodal and hybrid BI systems.

Extensive research has demonstrated that hybrid EEG-fNIRS systems consistently outperform their unimodal counterparts in classification accuracy for motor imagery tasks. The table below summarizes key performance metrics from recent studies.

Table 1: Performance Comparison of Unimodal vs. Hybrid BCIs for Motor Imagery Tasks

Modality	Classification Accuracy (%)	Information Transfer Rate (bit/min)	Key Advantages	Main Limitations
EEG-only	~65.0 [17]	Not Specified	High temporal resolution (milliseconds), well-established analysis methods [6]	Low spatial resolution, sensitive to noise (e.g., EMG, EOG) [17]
fNIRS-only	~57.0 [17]	Not Specified	Better spatial resolution, less susceptible to movement artifacts [6]	Low temporal resolution (seconds), slow hemodynamic response [6]
Hybrid EEG-fNIRS	76.2 - 77.6 [79] [17]	4.70 ± 1.92 [79]	Complementary information enhances accuracy and robustness, higher ITR [79]	Increased system complexity, more data channels, complex data fusion [79]

The performance gain in hybrid systems stems from the complementary nature of electrical (EEG) and hemodynamic (fNIRS) signals. EEG captures fast neural oscillations like event-related desynchronization/synchronization (ERD/ERS), while fNIRS provides a delayed but more localized measure of cortical activation through changes in oxygenated (HbO) and deoxygenated hemoglobin (HbR) concentrations [6] [34]. This synergy makes the hybrid system more resilient to the limitations of either modality alone.

Detailed Experimental Protocols

To ensure reproducible results in BCI research, standardized experimental protocols are crucial. The following section details a common framework for conducting a motor imagery study comparing unimodal and hybrid systems.

Participant Preparation and Setup

• Participants: Recruit healthy, right-handed adults with normal or corrected-to-normal vision. Obtain informed consent as approved by an institutional review board (IRB).
• Sensor Placement: Use a cap that integrates both EEG electrodes and fNIRS optodes. For upper-limb MI, focus on the sensorimotor cortex.
  • EEG: Position electrodes according to the international 10-20 system (e.g., over C3, Cz, C4) [47] [17].
  • fNIRS: Place sources and detectors over the same region to form channels covering the primary motor cortex and adjacent areas. Inter-optode distance should typically be 3 cm to ensure sufficient cortical penetration [47].
• Signal Acquisition:
  • EEG: Record with a sampling rate ≥ 200 Hz. Include a bipolar channel for electrooculogram (EOG) to monitor and later remove ocular artifacts [79].
  • fNIRS: Record optical densities at multiple wavelengths (e.g., 695 nm, 830 nm, 780 nm, 805 nm) [79] [47]. Convert these to concentration changes of HbO and HbR using the modified Beer-Lambert law [47].

Motor Imagery Task Paradigm

A typical trial-based paradigm for a left-hand vs. right-hand MI classification task is described below and visualized in Figure 1.

Table 2: Standardized Motor Imagery Trial Structure

Period	Duration	Participant Task	Visual/Auditory Cue
Instruction / Cue	2 seconds	Prepare for the task	A black arrow pointing left or right is displayed [17].
Task (MI Execution)	10 seconds	Perform kinesthetic motor imagery of the cued hand (e.g., imagining grasping a ball) [79] [17].	A fixation cross is displayed.
Rest	10-20 seconds (randomized)	Remain relaxed and avoid specific mental tasks.	A blank screen or steady fixation cross.

• The experiment should consist of multiple runs, with each run containing 20-30 trials per task (e.g., left hand, right hand) presented in a randomized order [79] [17].
• It is critical to instruct participants to minimize physical movement during the task period to avoid confounding artifacts.

Data Processing and Analysis Workflow

The following diagram illustrates the core data processing workflow, highlighting the divergence and fusion points for hybrid systems.

Figure 1: Data processing workflow for a hybrid EEG-fNIRS BCI, showing unimodal streams and fusion points.

Preprocessing Protocols

• EEG Preprocessing:
  • Downsampling: Reduce the sampling rate to 128-200 Hz to minimize computational load [17].
  • Filtering: Apply a bandpass filter (e.g., 1-50 Hz) to remove slow drifts and high-frequency noise [79]. A notch filter (e.g., 50/60 Hz) can be applied to remove line noise.
  • Artifact Removal: Use techniques like independent component analysis (ICA) or regression to remove artifacts from eye blinks (EOG) and muscle activity (EMG) [79].
• fNIRS Preprocessing:
  • Optical Density Conversion: Convert raw light intensity signals to optical density (OD) [79].
  • Hemoglobin Conversion: Apply the modified Beer-Lambert law to convert OD changes to concentration changes of HbO and HbR [79] [47].
  • Filtering: Apply a bandpass filter (e.g., 0.01 - 0.1 Hz) to isolate the task-related hemodynamic response from physiological noise like heart rate and respiration [79].

Feature Extraction and Fusion Protocols

• EEG Feature Extraction: For MI, the most relevant features are power in specific frequency bands.
  • Protocol: Calculate the log-variance of signals in the μ (8-13 Hz) and low β (13-30 Hz) bands. Alternatively, use the Common Spatial Patterns (CSP) algorithm to find spatial filters that maximize the variance difference between two classes of MI tasks [79] [17].
• fNIRS Feature Extraction: Features are typically based on the temporal dynamics of hemoglobin.
  • Protocol: For each HbO and HbR signal in the task window, calculate the mean value and the slope (linear regression coefficient) [17]. HbO is often the more informative feature.
• Data Fusion Strategies: Fusion can occur at different stages, with early-stage fusion (as shown in Figure 1) proving particularly effective [17].
  • Protocol (Early Fusion): Concatenate the extracted EEG and fNIRS feature vectors into a single, high-dimensional feature vector before feeding it into a classifier [17].

Classification and Validation

• Classifier Training: Use standard machine learning classifiers such as Linear Discriminant Analysis (LDA) or Support Vector Machines (SVM). Deep learning models like convolutional neural networks (CNNs) are also increasingly used, especially for early fusion [17].
• Validation: Always use cross-validation (e.g., 5-fold or leave-one-out) to obtain unbiased estimates of classification accuracy. Report both average accuracy and standard deviation across participants or validation folds [79] [17].

The Scientist's Toolkit: Essential Research Reagents & Materials

Successful implementation of a hybrid BCI requires specific hardware and software components. The following table details the essential "research reagent solutions."

Table 3: Essential Materials and Tools for Hybrid EEG-fNIRS Research

Item Name	Function / Description	Example Specifications / Notes
Integrated EEG-fNIRS Cap	Provides a stable platform with co-located electrodes and optodes, ensuring simultaneous data acquisition from the same cortical regions.	Custom-made or commercially available solutions with EEG electrodes positioned according to the 10-20 system and fNIRS optodes over the motor cortex.
EEG Amplifier	Amplifies microvolt-level electrical signals from the scalp for digitization.	High-input impedance, > 24-bit resolution, sampling rate ≥ 200 Hz. (e.g., BrainAmp) [17]
fNIRS System	Emits near-infrared light and detects attenuated light returning from the cortex to measure hemoglobin concentrations.	A continuous-wave system with multiple wavelengths (e.g., 695 & 830 nm [47] or 780 & 805 nm [79]). (e.g., Hitachi ETG-4000) [47]
Stimulus Presentation Software	Prescribes the experiment timeline and delivers visual/auditory cues to the participant.	Software like Psychtoolbox (MATLAB), E-Prime, or Presentation.
Data Analysis Suite	Provides tools for preprocessing, feature extraction, fusion, and classification of multimodal data.	MATLAB with toolboxes (EEGLAB, BBCI Toolbox [79]), Python (MNE, scikit-learn, PyTorch).
Structured Sparse Multiset CCA (ssmCCA)	An advanced data fusion algorithm used to identify brain regions consistently activated in both EEG and fNIRS modalities [34].	Used for analyzing shared neural correlates during motor execution, observation, and imagery [34].

This application note provides a rigorous comparative analysis and detailed protocols for unimodal and hybrid EEG-fNIRS systems within the context of motor imagery research. The quantitative evidence unequivocally demonstrates that hybrid systems offer a significant performance advantage, achieving higher classification accuracy and information transfer rates by leveraging the complementary strengths of EEG and fNIRS. While the increased complexity of hybrid systems presents challenges in data fusion and hardware integration, the protocols and toolkit outlined herein provide a clear roadmap for researchers. The adoption of hybrid BCIs is poised to enhance the efficacy of applications in neurorehabilitation, assistive technology, and fundamental neuroscience research, offering more robust and accurate decoding of user intent. Future work should focus on standardizing fusion algorithms and developing even more compact, user-friendly hardware to facilitate translation from the laboratory to clinical and real-world settings.

The rigorous validation of predictive models in patient cohorts is a cornerstone of robust clinical research. This process ensures that developed models are not only statistically sound but also generalizable and clinically applicable. Research on spontaneous intracerebral hemorrhage (sICH), a condition with high mortality and disability rates, provides a compelling paradigm for examining these validation methodologies. sICH accounts for 10-20% of all stroke subtypes and is the second leading cause of death globally, with a one-month mortality rate of 30-40% [80] [81] [82]. The exploration of validation frameworks within sICH research offers valuable insights that can be strategically applied to other domains, including the development of motor imagery task protocols for electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) research. This application note details the validation approaches derived from sICH datasets and proposes their adaptation to multimodal neuroimaging studies.

Key Validation Metrics from ICH Prognostic Models

Recent studies developing machine learning (ML) models for sICH prognosis provide a clear framework for evaluating predictive performance in clinical cohorts. The consistent metrics used across these studies offer a template for validation in neuroimaging research. Key performance indicators from sICH models are summarized in Table 1.

Table 1: Performance Metrics of Machine Learning Models for sICH Prognosis Prediction

Study Reference	Model Type	Sample Size (Training/Validation)	Primary Metric (AUROC)	Key Predictors Identified
Multi-center NCCT Study [80]	LightGBM	1091/102	0.813 ± 0.012 (External Validation)	GCS, IVH, NIHSS, hematoma volume, mean CT value, black hole sign
Clinical & Lab Feature Model [81]	Random Forest	413/74	0.817 (External Validation)	NIHSS, AST, Age, white blood cell count, hematoma volume
Clinical-Based Signature [82]	Logistic Regression (Nomogram)	835/287	0.868 (External Validation)	Age, ICH volume, GCS, IVH, SAH, hypodensities, PHE volume, temperature, SBP, WBC, neutrophil, NLR

These models demonstrate several consistent validation strengths: use of multicenter data for enhanced generalizability, implementation of external validation cohorts to test real-world performance, and reporting of confidence intervals for performance metrics [80] [81] [82]. The sICH research emphasizes that models achieving high performance on their training data must still be validated on independent, external cohorts to prove their clinical utility.

Experimental Protocols for Model Validation

Patient Cohort Selection and Data Acquisition

The validation protocols begin with rigorous patient selection. The standard inclusion criteria for sICH prognosis studies typically comprise: (1) age ≥18 years; (2) meeting diagnostic criteria for sICH consistent with cranial NCCT scan; (3) initial NCCT performed within 24-48 hours of symptom onset; and (4) first-ever acute-onset sICH [80] [81]. Exclusion criteria generally remove: (1) hemorrhage due to trauma, vascular malformation, or tumor; (2) isolated intraventricular hemorrhage; (3) loss to follow-up; and (4) patients with pre-existing disability (modified Rankin scale score ≥3) [80].

Data collection encompasses several domains:

• Demographic and clinical characteristics: Age, gender, medical history (hypertension, diabetes), medication history (anticoagulation, antiplatelet), smoking, drinking [80] [81].
• Clinical scales: Glasgow Coma Scale (GCS) score at admission, National Institutes of Health Stroke Scale (NIHSS) score at admission, Intracerebral Hemorrhage (ICH) score [80] [82].
• Laboratory parameters: White blood cells, neutrophils, lymphocytes, neutrophil-lymphocyte ratio (NLR), aspartate aminotransferase (AST) [81] [82].
• Imaging features: Hematoma volume, location, intraventricular hemorrhage (IVH), subarachnoid hemorrhage (SAH), midline shift, hypodensities, perihematomal edema (PHE) volume, mean CT value, black hole sign [80] [82].

Machine Learning Model Development and Validation Workflow

The validation workflow for sICH models follows a structured pipeline that can be adapted for neuroimaging research, as visualized below:

Diagram 1: Validation workflow for clinical prediction models

The workflow employs several feature selection techniques to optimize model performance and interpretability. Recursive Feature Elimination (RFE) is commonly used, which "removes features that are not important for the ending variables, and ultimately obtains the optimal combination of variables for the best performance of the model" [81]. Sequential Forward Selection (SFS) is another approach used to screen core features [80].

Multiple machine learning algorithms are typically compared to identify the best-performing model for the specific dataset. Commonly employed algorithms include:

• Random Forest (RF): "An integrated learning method based on decision trees that operates on the logic of improving the accuracy and robustness of the model by constructing multiple decision trees based on random samples and random features" [81].
• Light Gradient Boosting Machine (LightGBM): "A high-performance gradient boosting decision tree based running framework whose unique histogram gradient boosting method and leaf-wise learning strategy make it perform well in large datasets" [80] [81].
• Other algorithms such as Support Vector Machines (SVM), Logistic Regression (LR), and XGBoost are also frequently evaluated [80] [81].

Validation Methods

The sICH models employ rigorous validation approaches:

• Internal Validation: Implemented through k-fold cross-validation (typically 5-fold) where "the dataset within the training set is divided in a 7:3 ratio" for training and testing, with internal five-fold cross-validation employed "to discern the most suitable hyperparameters for each distinct model" [81].
• External Validation: Performed using "an external validation cohort" from completely different centers to "independently verify the model" and assess its generalizability [80] [82].
• Performance Evaluation: Comprehensive assessment using "ROC curves, accuracy and other related indicators" with calculation of area under the curve (AUC) values and 95% confidence intervals [80] [81].
• Model Interpretability: Application of SHapley Additive exPlanations (SHAP) diagrams "to illustrate the importance of variables in the model" [80] [81].
• Clinical Utility Assessment: Use of "decision curve analysis (DCA) to evaluate the clinical utility of the model" and calibration curves to assess agreement between predicted and observed outcomes [82].

The Scientist's Toolkit: Research Reagent Solutions

The validation approaches from sICH research provide essential methodological "reagents" that can be applied across clinical prediction domains. Table 2 outlines these key methodological components.

Table 2: Essential Methodological Components for Clinical Prediction Model Validation

Component Category	Specific Tool/Method	Function in Validation	Example from ICH Research
Study Design	Multicenter Retrospective Design	Enhances generalizability and sample diversity	Data from 3-4 independent medical centers [80] [82]
Feature Selection	Recursive Feature Elimination (RFE)	Identifies optimal feature combinations	Used to select top predictors like NIHSS, hematoma volume [81]
Model Interpretability	SHAP (SHapley Additive exPlanations)	Illustrates variable importance in complex models	Identified GCS, IVH, NIHSS as core predictors [80]
Validation Framework	External Validation Cohort	Tests model performance on independent data	Single-center cohort used for external validation [80]
Clinical Implementation	Nomogram / Web Platform	Translates model to clinical practice	Publicly accessible online prognosis platform [80]

Application to Motor Imagery Protocol Validation

The validation principles derived from sICH research can be directly adapted to strengthen the development of motor imagery (MI) task protocols for EEG and fNIRS research. The integration of EEG and fNIRS is particularly promising as it "capitalizes on EEG high temporal resolution and fMRI superior spatial resolution, providing a complementary approach" [10]. The application of sICH-derived validation frameworks can enhance the reliability of such multimodal approaches.

Adaptation of Validation Workflows

The structured validation workflow from sICH models can be translated to MI protocol development as follows:

Diagram 2: Adapted validation workflow for motor imagery protocols

Implementing Multicenter Validation for Neuroimaging

Similar to sICH studies that utilized "1091 sICH patients from three centers as the training cohort, and 102 patients from a single center as the external validation cohort" [80], MI protocol research should incorporate data from multiple research sites. This approach tests the robustness of signal processing pipelines and classification algorithms across different equipment and populations.

The fusion of EEG and fNIRS presents particular validation challenges that benefit from the sICH framework. Research indicates that "early-stage fusion of EEG and fNIRS have significantly higher performance compared to middle-stage and late-stage fusion network configuration" [17]. This finding parallels the feature fusion approaches in sICH models that combined "clinical feature set, NCCT imaging feature set and fusion feature set" [80].

Performance Metrics and Clinical Translation

Mirroring the clinical implementation of sICH models through "a convenient and practical prognosis prediction platform" [80], validated MI protocols should ultimately deploy user-friendly interfaces for clinical applications. The translation of MI protocols to clinical populations (e.g., stroke rehabilitation) requires the same rigorous validation standards demonstrated in sICH research, particularly external validation in independent patient cohorts.

Application of model interpretability methods like SHAP analysis to MI classification models can identify the most important neural features for successful classification, similar to how sICH research identified "Glasgow Coma Scale (GCS) score at admission, intraventricular hemorrhage (IVH), National Institutes of Health Stroke Scale (NIHSS) score at admission, hematoma volume, mean CT value, and black hole sign" as core predictors [80]. This approach can reveal whether EEG, fNIRS, or their combination provides the most significant features for specific clinical applications.

The validation methodologies refined through sICH research provide a robust framework for developing reliable motor imagery protocols in EEG and fNIRS research. The core principles of multicenter data collection, rigorous feature selection, comprehensive internal and external validation, and clinical implementation translate effectively to the neuroimaging domain. By adopting these structured validation approaches, researchers can enhance the reliability, generalizability, and clinical utility of motor imagery protocols, ultimately advancing the field of multimodal brain-computer interfaces and their applications in neurorehabilitation.

Cross-Subject vs. Within-Subject Validation Frameworks

In motor imagery (MI) research utilizing electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS), the choice of validation framework fundamentally shapes experimental design, algorithmic development, and clinical translation. These frameworks determine how well findings generalize across populations and how reliably they can be applied in real-world settings. Within-subject validation assesses model performance when trained and tested on data from the same individual, maximizing personalization but potentially limiting broad applicability. In contrast, cross-subject validation evaluates how well models generalize to new, unseen individuals, addressing scalability but facing challenges from significant physiological and anatomical variability between subjects.

The complementary nature of EEG and fNIRS has positioned multimodal approaches as particularly valuable for addressing validation challenges. EEG provides millisecond-level temporal resolution crucial for capturing rapid neural dynamics during MI tasks, while fNIRS offers superior spatial localization (5-10 mm resolution) of hemodynamic responses in activated cortical regions [13]. This spatiotemporal synergy enables more robust feature extraction that can better accommodate inter-subject differences while maintaining task-specific sensitivity.

Conceptual Foundations and Performance Comparisons

Fundamental Characteristics of Validation Approaches

Within-subject frameworks operate on the principle of individualized modeling, where algorithms are tailored to the unique neurophysiological signatures of each participant. This approach typically involves collecting multiple sessions from the same individual and using k-fold cross-validation or held-out session testing within that individual's data. The primary strength of this framework lies in its ability to account for subject-specific patterns in both electrical and hemodynamic responses, potentially achieving higher classification accuracy by accommodating unique neurovascular coupling characteristics.

Cross-subject frameworks aim to develop generalized models that perform reliably across diverse populations. These approaches train models on data from multiple subjects and evaluate performance on completely unseen individuals. This validation paradigm directly tests real-world applicability but must contend with significant challenges including anatomical differences (cortical folding, scalp thickness), physiological variations (neurovascular coupling efficiency, baseline hemodynamics), and technical factors (optode placement variability, impedance characteristics) [13] [83].

Quantitative Performance Comparisons

Table 1: Comparative Performance Metrics of Validation Frameworks in MI Research

Study	Modality	Within-Subject Accuracy	Cross-Subject Accuracy	Performance Gap	Key Methodology
HA-FuseNet (2025) [84]	EEG	77.89%	68.53%	9.36%	Feature fusion + attention mechanism
CHTLM (2025) [83]	fNIRS	-	83.1% (pre-rehab) 91.3% (post-rehab)	-	Heterogeneous transfer learning from EEG
Early Fusion (2023) [17]	EEG+fNIRS	76.21%	-	-	Y-shaped neural network with early fusion
fNIRS Reproducibility (2025) [85]	fNIRS	High session consistency	Notable variability	Significant	Source localization with digitized optodes

The performance disparities between frameworks highlight both the challenge and necessity of cross-subject approaches. The approximately 10% accuracy reduction in cross-subject versus within-subject validation observed in HA-FuseNet exemplifies the "individual differences penalty" that researchers must overcome [84]. Heterogeneous transfer learning approaches like CHTLM demonstrate promising pathways for bridging this gap by leveraging knowledge from related modalities and populations [83].

Experimental Protocols for Validation Frameworks

Within-Subject Validation Protocol

Participant Preparation and Calibration: Effective within-subject protocols begin with comprehensive participant preparation. For motor imagery tasks, researchers should implement a grip strength calibration procedure using a dynamometer and stress ball to enhance kinesthetic imagery vividness [5]. This involves: (1) repeated 5 kg maximal force exertions (or voluntary maximum efforts), (2) equivalent force applications using a stress ball, and (3) grip training at a rate of one contraction per second. This calibration phase strengthens tactile and force-related aspects of the movement, standardizing temporal rhythm and improving MI consistency across trials.

Experimental Paradigm and Data Acquisition: Participants should be seated approximately 25 cm from a display monitor in an ergonomic position. The recommended trial structure comprises [5]:

• Baseline recording: 1-minute eyes-closed followed by 1-minute eyes-open states
• Visual cue presentation (2 s): A directional arrow indicating the required MI task
• Execution phase (10 s): Participants perform kinesthetic MI at approximately 1 repetition per second
• Inter-trial interval (15 s): Rest period with blank screen This structure should be repeated across multiple sessions (minimum 2 recommended), with each session containing 15 trials per MI task (e.g., left hand, right hand). Sufficient inter-session rest intervals are critical to mitigate fatigue effects.

Data Acquisition Specifications: For multimodal studies, synchronized acquisition is essential. The HEFMI-ICH protocol uses a g.HIamp amplifier for EEG (256 Hz sampling) and a continuous-wave fNIRS system (11 Hz sampling) with a custom hybrid cap integrating 32 EEG electrodes and 90 fNIRS measurement channels through source-detector pairing [5]. Temporal synchronization should be established using event markers transmitted from experimental software (e.g., E-Prime 3.0) to simultaneously trigger both recording systems.

Cross-Subject Validation Protocol

Population Selection and Domain Adaptation: Cross-subject validation requires careful population selection to ensure representative sampling. The CHTLM protocol recommends including both healthy participants (12 males, 5 females, mean age 23.6 ± 1.8 years) and clinical populations (e.g., 20 ICH patients: 17 males, 3 females, mean age 50.8 ± 10.3 years) to test generalization across physiological and pathological variations [5]. Clinical assessments should include standardized measures like Fugl-Meyer Assessment for Upper Extremities (FMA-UE), Modified Barthel Index (MBI), and modified Rankin Scale (mRS) to characterize participant capabilities.

Transfer Learning Implementation: For effective cross-subject generalization, implement heterogeneous transfer learning where labeled EEG data from healthy individuals serves as the source domain, and target fNIRS data comes from the clinical population [83]. The CHTLM framework utilizes:

• Wavelet transformation to convert raw fNIRS signals into image data, enhancing clarity of frequency components and temporal changes
• Adaptive feature matching network to explore correlations between source (EEG) and target (fNIRS) domains
• Sparse Bayesian extreme learning machine to classify fused deep learning features while preventing overfitting

Data Acquisition and Preprocessing: For fNIRS acquisition, use a medical-grade portable system (e.g., NirSmart 6000B) with optodes placed symmetrically on the motor cortex around C3 and C4 positions [83]. Standard configuration should include:

• 7 light sources and 7 detectors forming 16 channels
• Source-detector distance: 30 mm
• Wavelengths: 730 nm and 850 nm
• Sampling rate: 11 Hz For EEG, follow the BCI Competition IV Dataset 2a protocol with 22 channels at 250 Hz sampling rate [83].

Table 2: Detailed Experimental Parameters for Cross-Subject Validation

Parameter	Healthy Participants	Clinical Population	Technical Specifications
Sample Size	17 subjects [5]	20 ICH patients [5]	Minimum 15-20 per group
Age Range	23.6 ± 1.8 years [5]	50.8 ± 10.3 years [5]	Document variance
MI Tasks	Left/right hand imagery [17]	Hemiplegic hand imagery [83]	15-30 trials per task
Session Structure	2+ sessions, 30 trials each [5]	Pre- and post-rehabilitation [83]	Counterbalance order
fNIRS Setup	90 channels, 3cm spacing [5]	16 channels, C3/C4 coverage [83]	11 Hz sampling
EEG Setup	32 channels, 10-20 system [5]	22 channels, 250 Hz [83]	256 Hz sampling

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 3: Key Research Reagent Solutions for MI-EEG/fNIRS Research

Category	Specific Product/Model	Key Function	Research Application
EEG Systems	g.HIamp amplifier (g.tec) [5]	Signal amplification and digitization	High-quality EEG acquisition at 256 Hz
fNIRS Systems	NirScan (Danyang Huichuang) [5]	Hemodynamic response measurement	Continuous-wave fNIRS at 11 Hz sampling
Hybrid Caps	Custom Model M cap [5]	Integrated EEG/fNIRS positioning	32 EEG electrodes + 90 fNIRS channels
Calibration Tools	Dynamometer + Stress ball [5]	MI vividness enhancement	Grip strength calibration for kinesthetic imagery
Stimulus Presentation	E-Prime 3.0 [5]	Experimental paradigm control	Precise timing and synchronization
Portable fNIRS	NirSmart 6000B [83]	Clinical-grade fNIRS acquisition	Bedside monitoring in patient populations
Computational Framework	HA-FuseNet [84]	MI-EEG classification	Feature fusion with attention mechanisms
Transfer Learning	CHTLM framework [83]	Cross-subject knowledge transfer	Heterogeneous EEG to fNIRS adaptation

Implementation Considerations and Best Practices

Technical Considerations for Multimodal Integration

Successful implementation of validation frameworks requires careful attention to technical integration challenges. For hybrid EEG-fNIRS systems, researchers must address helmet design optimization to ensure proper optode and electrode placement. Current approaches include using flexible EEG electrode caps as a foundation with punctures at specific locations for fNIRS probe fixtures [13]. Alternatively, 3D-printed customized joint-acquisition helmets offer improved positioning accuracy but at higher cost, while composite polymer cryogenic thermoplastic sheets provide a cost-effective middle ground [13].

Signal quality management must address the distinctive artifacts of each modality. For fNIRS, ensure consistent optode-scalp contact pressure to minimize motion artifacts, while for EEG, implement appropriate filtering (8-25 Hz for MI tasks) and reference schemes (common average referencing) [17]. Synchronization between modalities should utilize unified processor systems rather than post-hoc alignment whenever possible, as this achieves more precise temporal correspondence [13].

Methodological Recommendations

Based on current research, the following evidence-based recommendations optimize validation framework implementation:

• Prioritize early-stage fusion for multimodal integration, as Y-shaped neural networks combining EEG and fNIRS at initial processing stages demonstrate significantly higher performance compared to middle-stage and late-stage fusion (76.21% accuracy in left/right MI classification) [17].

• Implement source localization for fNIRS reproducibility, as digitized optode positions improve spatial accuracy and reliability across sessions [85].

• Address inter-subject variability through transfer learning architectures like CHTLM, which improves cross-subject accuracy by 8.6-10.5% in pre-rehabilitation and 11.3-15.7% in post-rehabilitation scenarios [83].

• Leverage hybrid feature characteristics by recognizing that oxyhemoglobin (HbO) demonstrates higher reproducibility than deoxyhemoglobin (HbR) in within-subject contexts, informing feature selection priorities [85].

• Account for clinical considerations in patient populations by incorporating appropriate rest intervals and task modifications to accommodate fatigue and attention limitations, particularly in post-stroke individuals [5] [83].

Stroke remains a principal contributor to chronic disability worldwide, with 55%-75% of stroke survivors developing motor deficits that critically impair functional independence and socioeconomic function. Intracerebral hemorrhage (ICH), a specific stroke subtype, accounts for 6.5%-19.6% of all stroke cases while contributing disproportionately to more than 40% of stroke-related mortality, posing exceptional neurorehabilitation challenges. Upper limb motor impairment, a major sequela among ICH survivors, necessitates innovative neurorehabilitation strategies to address the limitations of conventional therapies, which often result in suboptimal recovery due to disrupted corticospinal pathways [5].

In this context, motor imagery (MI)-based brain-computer interfaces (MI-BCIs) have emerged as a transformative approach, leveraging neuroplasticity to facilitate motor network reorganization through closed-loop feedback mechanisms. The integration of electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) creates hybrid BCIs that capitalize on their complementary strengths: EEG provides millisecond temporal resolution for capturing rapid neuronal activation patterns during MI tasks, while fNIRS offers superior spatial localization (5–10 mm resolution) and tracks slower hemodynamic changes associated with cortical reorganization. This multimodal approach has demonstrated 5%-10% improvement in classification accuracy compared to unimodal systems in normal subjects, though it remains underexplored in ICH populations where neurovascular uncoupling may alter signal dynamics [5].

Laboratory Foundations: Evidence for Multimodal Integration

Technical Validation and Performance Metrics

Recent advances in hybrid EEG-fNIRS systems have demonstrated significant improvements in motor imagery classification accuracy, a critical foundation for clinical translation. Research on deep learning approaches for signal integration shows that novel fusion methods combining evidence theory with hierarchical attention mechanisms can achieve average accuracy of 83.26% on standardized datasets, representing a 3.78% improvement over state-of-the-art methods [29]. Even more impressively, hierarchical attention-enhanced deep learning architectures have reached unprecedented accuracy levels of 97.25% on custom four-class motor imagery datasets comprising 4,320 trials from 15 participants [86].

Table 1: Performance Comparison of Motor Imagery Decoding Approaches

Method Category	Representative Architecture	Average Accuracy	Key Advantages	Clinical Translation Challenges
Traditional Machine Learning	SVM, LDA, CSP	65%-80%	Computational efficiency, interpretability	Performance plateaus with high-dimensional data
Deep Learning (Unimodal)	CNN-LSTM hybrids	75%-85%	Automatic feature extraction, temporal modeling	High computational demand, inter-subject variability
Multimodal Fusion (EEG-fNIRS)	Evidence theory with DST	83.26%	Uncertainty quantification, handles heterogeneous data	System complexity, calibration time
Attention-Enhanced Architectures	Hierarchical CNN-LSTM with attention	97.25%	Interpretable features, selective processing	Model complexity, data hunger

Real-world evaluations of deep learning decoders reveal critical considerations for clinical implementation. When tested under soft real-time protocols with 2-second windows—reflecting clinical reality—compact spectro-temporal CNN backbones with lightweight temporal context sustain performance more consistently than deeper attention and Transformer stacks, which show greater sensitivity to subject and session differences. This highlights the importance of balancing model complexity with robustness for therapeutic applications [87].

Neurophysiological Mechanisms of Motor Imagery

Kinesthetic motor imagery (kMI), the mental simulation of movement without execution, activates similar neural pathways to actual movement, making it particularly valuable for neurorehabilitation. Enhancing kMI quality through optimized protocols significantly impacts neural activation patterns. Recent research demonstrates that providing visual guidance with color-coded muscle activation cues in demonstration models enhances kMI vividness by 19.9% compared to controls without such cues [88].

This enhancement manifests in distinct neurophysiological signatures: selective inhibition in frontal-central-temporal brain regions coupled with activation in occipital-parietal regions within brain rhythmic activity. Additionally, source-functional connectivity analyses reveal increased information flow between occipital-parietal and frontal-parietal brain regions, suggesting more efficient sensorimotor integration during high-quality kMI [88]. These findings provide crucial insights for designing effective therapeutic protocols that maximize neuroplasticity.

Experimental Protocols and Methodologies

Participant Recruitment and Clinical Characterization

The HEFMI-ICH dataset provides a exemplary recruitment model, comprising 17 normal subjects (right-handed; 12 males and 5 females; mean age 23.6 ± 1.8 years) and 20 patients with ICH (right-handed; 17 males and 3 females; mean age 50.8 ± 10.3 years; time since onset ranging from 2 days to 2 months). Comprehensive clinical assessments should include:

• Fugl-Meyer Assessment for Upper Extremities (FMA-UE): A standardized, performance-based quantitative measure widely employed to assess motor function, coordination, and reflex activity in patients with hemiplegia after stroke
• Modified Barthel Index (MBI): Assesses activities of daily living (mean score 85.8 ± 28.7 in HEFMI-ICH cohort)
• Modified Rankin Scale (mRS): Measures disability and functional independence (mean score 2.0 ± 1.7 in HEFMI-ICH cohort)

All participants and their families must provide written informed consent after being fully informed of procedures and objectives. Ethical approval should be obtained from institutional review boards, with registration in national medical research registration systems where applicable [5].

Motor Imagery Paradigm Design

A standardized MI experimental paradigm comprises at least two consecutive sessions for each subject, with each session including 15 trials of either left-hand or right-hand MI. Sufficient inter-session rest intervals are imperative to mitigate fatigue effects and ensure data quality. The protocol sequence should include:

• Baseline Recording: 1-minute eyes-closed followed by 1-minute eyes-open states, demarcated by an auditory cue (200 ms beep)
• Trial Structure:
  • Visual cue presentation (2 s): A yellow directional arrow (left/right) on a blue background indicating the required MI
  • Execution phase (10 s): Central yellow fixation cross display following auditory cue (200 ms beep), during which participants conduct kinesthetic MI of grasping movements at a rate of one imagined grasp per second
  • Inter-trial interval (15 s): Blank screen indicating rest period

For patients struggling with MI concepts, a grip strength calibration procedure using a dynamometer and stress ball enhances MI vividness by reinforcing tactile and force-related aspects of grasping movements and standardizing temporal rhythm [5].

Figure 1: Experimental workflow for motor imagery protocols showing trial structure and timing

Multimodal Data Acquisition Setup

Integrated data acquisition requires synchronized neurophysiological recording using:

• EEG System: g.HIamp amplifier (g.tec medical engineering GmbH) or equivalent with 32-electrode configuration following international 10-20 system, sampling rate 256 Hz
• fNIRS System: Continuous-wave multifunctional fNIRS system (NirScan, Danyang Huichuang Medical Equipment Co. Ltd.) or equivalent with 32 optical sources and 30 photodetectors, sampling rate 11 Hz
• Cap Design: Custom-designed hybrid EEG-fNIRS cap with optimized topography for 90 fNIRS measurement channels through source-detector pairing at 3 cm separation distances
• Synchronization: Event markers transmitted from E-Prime 3.0 or equivalent experimental control software simultaneously triggering both recording systems

Optode and electrode placement should ensure comprehensive coverage of major functional areas, particularly sensorimotor cortices, with detailed documentation of 3D coordinates under MNI template for reproducibility [5] [10].

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials and Equipment for Hybrid EEG-fNIRS Research

Category	Specific Item	Function/Application	Implementation Example
Neuroimaging Hardware	32-channel EEG amplifier with electrode cap	Measures electrical brain activity with millisecond temporal resolution	g.HIamp amplifier (g.tec) with 32-electrode configuration [5]
Neuroimaging Hardware	Continuous-wave fNIRS system with optodes	Measures hemodynamic responses with superior spatial localization	NirScan system with 32 sources, 30 detectors [5]
Experimental Control	Stimulus presentation software	Precisely controls experimental paradigm and synchronization	E-Prime 3.0 for visual cue presentation and event marking [5]
Calibration Tools	Dynamometer and stress ball	Enhances MI vividness through tactile reinforcement and force calibration	Pre-acquisition grip strength training [5]
Signal Processing	Deep learning frameworks with multimodal fusion	Classifies motor imagery intent from heterogeneous neural signals	Evidence theory with Dirichlet distribution and DST [29]
Quality Assessment	Motor imagery questionnaires	Quantifies subjective MI experience and vividness	Vividness of Motor Imagery Questionnaire-2 (VMIQ-2) [88]

Data Processing and Analytical Framework

Signal Processing Pipeline

Multimodal data integration requires sophisticated processing approaches that respect the unique characteristics of each signal type. For EEG signals, effective processing involves spatiotemporal feature extraction using dual-scale temporal convolution and depthwise separable convolution, with hybrid attention modules enhancing network sensitivity to salient neural patterns. For fNIRS signals, spatial convolution across all channels explores activation differences among brain regions, while parallel temporal convolution combined with gated recurrent units (GRUs) captures richer temporal dynamics of the hemodynamic response [29].

At the decision fusion stage, decision outputs from both modalities should be quantified using Dirichlet distribution parameter estimation to model uncertainty, followed by a two-layer reasoning process using Dempster-Shafer Theory (DST) to fuse evidence from basic belief assignment methods and both modalities. This approach has demonstrated significant improvements in classification accuracy while providing principled uncertainty quantification [29].

Figure 2: Multimodal signal processing pipeline showing integration of EEG and fNIRS data streams

Quantitative Performance Metrics

Comprehensive evaluation of hybrid BCI systems requires multiple performance dimensions:

• Classification Metrics: Accuracy, sensitivity, precision, F1-score
• Real-time Performance: Information transfer rate (ITR), false-alarm rate (FAR), miss-as-neutral rate (MANR)
• Workload Assessment: NASA-TLX or similar subjective measures
• Physiological Validation: Lateralization indices, C3 vs C4 mu-band power, MI-versus-neutral topographic contrasts

These metrics provide complementary insights into system performance from both technical and user-centered perspectives, enabling balanced evaluation of clinical viability [87].

Clinical Implementation Pathway

Translation to Rehabilitation Practice

The pathway from laboratory validation to clinical implementation requires addressing several critical challenges. First, systems must be validated with actual patient populations, as BCI signals from patients with ICH exhibit fundamental divergences from normative baselines due to neurostructural compromise, neurocognitive reorganization, and age-related cerebrovascular alterations. Second, protocols must be adaptable to individual capabilities and recovery trajectories, potentially using adaptive learning algorithms that adjust task difficulty based on performance [5].

Neurofeedback paradigms represent a particularly promising clinical application, where real-time feedback of brain activity enables patients to voluntarily modulate their sensorimotor rhythms. Studies investigating EEG-fNIRS-based neurofeedback during upper-limb motor imagery tasks have developed customized experimental platforms that compute integrated NF scores from both modalities, presenting participants with visual representations (e.g., a ball moving along a one-dimensional gauge) that reflect their brain activity level during MI tasks [10].

Protocol Adaptation for Clinical Populations

Successful clinical translation requires specific adaptations for patient populations:

• Session Duration: Shorter sessions with more frequent breaks to accommodate fatigue
• Instruction Modifications: Simplified instructions with concrete metaphors for abstract concepts like kinesthetic motor imagery
• Calibration Procedures: Extended practice sessions with tactile feedback to enhance MI vividness
• Feedback Design: Simplified, intuitive visual feedback that maintains patient engagement without causing frustration
• Progress Monitoring: Integrated tracking of both neural and functional metrics to demonstrate clinical progress

These adaptations ensure that the technological sophistication of hybrid BCI systems remains accessible and therapeutic for patients with varying levels of cognitive and motor impairments [5] [88].

The integration of EEG and fNIRS for motor imagery-based BCIs represents a paradigm shift in neurorehabilitation, offering a multimodal window into brain function that captures both electrical and hemodynamic aspects of neuroplasticity. While significant technical challenges remain in signal processing, system integration, and clinical validation, the current evidence supports optimistic projections for hybrid systems becoming clinically viable tools for post-stroke rehabilitation within the coming decade.

Future developments should focus on personalizing decoding algorithms to individual patient characteristics, simplifying system calibration for clinical use, and demonstrating functional gains in rigorously controlled trials. As these technologies mature, they hold the potential to transform rehabilitation for the millions affected by stroke and other neurological disorders each year, finally bridging the gap between laboratory proof-of-concept and meaningful functional recovery.

Conclusion

Motor imagery protocols for hybrid EEG-fNIRS systems represent a powerful and evolving toolkit for clinical research and therapeutic development. The synergy between EEG's millisecond temporal resolution and fNIRS's superior spatial localization offers a more complete picture of brain activity during MI, leading to more robust brain-computer interfaces. Success hinges on standardized methodologies, meticulous attention to participant instruction and signal quality, and protocols adaptable to both healthy and clinical populations, such as those recovering from stroke. Future directions should focus on developing personalized and adaptive training systems that leverage deep learning for improved classification, integrating MI-BCIs with assistive robotics and functional electrical stimulation in multisite trials, and establishing rigorous, large-scale datasets to validate durable functional outcomes in neurorehabilitation and drug development pathways.

References

• 1. Motor imagery and action observation: cognitive tools for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC2797860/]
• 2. Characterization of kinesthetic motor imagery compared ... [https://www.nature.com/articles/s41598-021-82241-0]
• 3. Best practice for motor imagery: a systematic literature review ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC3141540/]
• 4. Motor task-to-task transfer learning for motor imagery brain- ... [https://www.sciencedirect.com/science/article/pii/S1053811924004038]
• 5. HEFMI-ICH: a hybrid EEG-fNIRS motor imagery dataset for ... [https://www.nature.com/articles/s41597-025-06100-7]
• 6. Evaluating the effects of multimodal EEG-fNIRS ... [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0331177]
• 7. Neural correlates of motor imagery and execution in real- ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2024.1412307/full]
• 8. Neural correlates of action: Comparing meta-analyses ... [https://www.sciencedirect.com/science/article/abs/pii/S0149763417309284]
• 9. The role of fNIRS in assessing motor function: A bibliometric ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12338178/]
• 10. Evaluating the effects of multimodal EEG-fNIRS ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12425317/]
• 11. Distinct patterns of whole-body coding in human motor ... [https://www.medrxiv.org/content/10.1101/2025.08.01.25332521v2.full-text]
• 12. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOorviBOvhTdm6RWVOEFf6YNumdW4-c27GnRmp0OdhjdtgRiXVTRi]
• 13. A Cross-Disciplinary Review of the fNIRS-EEG Dual-Modality ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11506513/]
• 14. fNIRS-EEG BCIs for Motor Rehabilitation: A Review - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC10740927/]
• 15. Enhanced spatiotemporal resolution imaging of neuronal ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7778454/]
• 16. Functional Near-Infrared Spectroscopy and Its Clinical ... [https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2020.00724/full]
• 17. Early-stage fusion of EEG and fNIRS improves ... [https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2022.1062889/full]
• 18. Optimizing spatial specificity and signal quality in fNIRS [https://www.frontiersin.org/journals/neuroergonomics/articles/10.3389/fnrgo.2024.1286586/full]
• 19. Multimodal fNIRS–EEG sensor fusion: Review of data-driven ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12592382/]
• 20. EFRM: A Multimodal EEG–fNIRS Representation-learning ... [https://www.sciencedirect.com/science/article/abs/pii/S0010482525016464]
• 21. Neurovascular coupling in humans: Physiology ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4821024/]
• 22. Concurrent fNIRS and EEG for Brain Function Investigation [https://www.mdpi.com/1424-8220/22/15/5865]
• 23. Review of public motor imagery and execution datasets in ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2023.1134869/full]
• 24. A multi-day and high-quality EEG dataset for motor imagery ... [https://www.nature.com/articles/s41597-025-04826-y]
• 25. Multimodal fNIRS-EEG measurements — Analysis ... [https://www.artinis.com/blogpost-all/multimodal-fnirs-eeg-measurements-analysis-approaches]
• 26. fNIRS Analysis | NIRS Analysis Techniques, Homer2, NIRS ... [https://nirx.net/fnirs-analysis]
• 27. A Novel Transfer Learning‐Based Hybrid EEG‐fNIRS Brain ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12631896/]
• 28. An fNIRS and EEG Review [https://pubmed.ncbi.nlm.nih.gov/41008372/]
• 29. EEG–fNIRS signal integration for motor imagery ... [https://www.sciencedirect.com/science/article/pii/S2772528625000299]
• 30. EEG-based brain-computer interface enables real-time ... [https://www.nature.com/articles/s41467-025-61064-x]
• 31. Open-Access fNIRS Dataset for Motor Imagery of Lower- ... [https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2025.1695169/full]
• 32. Review of public motor imagery and execution datasets in ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10101208/]
• 33. How to design real-world functional near-infrared ... [https://www.spiedigitallibrary.org/journals/neurophotonics/volume-13/issue-S1/S10701/How-to-design-real-world-functional-near-infrared-spectroscopy-studies/10.1117/1.NPh.13.S1.S10701.full]
• 34. Simultaneous multimodal fNIRS-EEG recordings reveal ... [https://www.nature.com/articles/s41598-023-31609-5]
• 35. An EEG motor imagery dataset for brain computer interface ... [https://www.nature.com/articles/s41597-023-02787-8]
• 36. Evaluating the effects of multimodal EEG-fNIRS ... [https://pubmed.ncbi.nlm.nih.gov/40934195/]
• 37. How EEG preprocessing shapes decoding performance [https://www.nature.com/articles/s42003-025-08464-3]
• 38. A Method for Using Video Presentation to Increase Cortical ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9855988/]
• 39. Best practice for motor imagery: a systematic literature review ... [https://bmcmedicine.biomedcentral.com/articles/10.1186/1741-7015-9-75]
• 40. Early-stage fusion of EEG and fNIRS improves ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9869134/]
• 41. fNIRS vs EEG: Comparing Non-Invasive Brain Imaging ... [https://imotions.com/blog/learning/research-fundamentals/fnirs-vs-eeg/?srsltid=AfmBOoo369RBbBE8MSrgZDFNiDo7FAIYvIoWvySLQF6gwoOmAHjGvSgP]
• 42. Comparing structure–function relationships in brain ... [https://www.nature.com/articles/s41598-024-79817-x]
• 43. Ensuring high-quality EEG recordings in integrated fNIRS ... [https://www.artinis.com/blogpost-all/ensuring-high-quality-eeg-recordings-in-integrated-fnirs-eeg]
• 44. Quantitative comparison of correction techniques for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7511246/]
• 45. Motor Imagery and Motor Execution: A Narrative Review of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12550537/]
• 46. Exploring Imagined Movement for Brain–Computer ... [https://www.mdpi.com/2076-3425/15/9/1013]
• 47. Comparison of Brain Activation during Motor Imagery and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5435907/]
• 48. fNIRS-EEG BCIs for Motor Rehabilitation: A Review [https://www.mdpi.com/2306-5354/10/12/1393]
• 49. MartMi-BCI: A matlab-based real-time motor imagery brain- ... [https://www.sciencedirect.com/science/article/pii/S2352711023000675]
• 50. An Online Data Visualization Feedback Protocol for Motor ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8215169/]
• 51. Motor Imagery during Action Observation: A Brief Review of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5116576/]
• 52. Motor imagery and action-observation in neurorehabilitation [https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2022.990618/full]
• 53. The effects of combined action observation and motor ... [https://www.sciencedirect.com/science/article/pii/S0149763422004006]
• 54. The Brain Mechanisms of Music Stimulation, Motor ... [https://www.eneuro.org/content/12/7/ENEURO.0557-24.2025]
• 55. Action Observation Combined With Motor Imagery Training to ... [https://rehab.jmir.org/2025/1/e75705]
• 56. Improving locomotor performance with motor imagery and ... [https://www.nature.com/articles/s41598-025-86039-2]
• 57. Development of a Guidance System for Motor Imagery ... [https://www.mdpi.com/1424-8220/21/6/2197]
• 58. Motor Imagery Acquisition Paradigms: In the Search to ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12526967/]
• 59. Bridging the BCI illiteracy gap: a subject-to- ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2023.1194751/full]
• 60. Improving Motor Imagery-Based Brain-Computer Interface ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8602688/]
• 61. Advances in user-training for mental-imagery-based BCI ... [https://www.sciencedirect.com/science/article/abs/pii/S0079612316300061]
• 62. Simultaneous EEG and fNIRS recordings for semantic ... [https://www.nature.com/articles/s41597-025-04967-0]
• 63. Physiological Noise Filtering in Functional Near-Infrared ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10295283/]
• 64. Adaptive filtering of physiological noises in fNIRS data [https://biomedical-engineering-online.biomedcentral.com/articles/10.1186/s12938-018-0613-2]
• 65. Multivariate disturbance filtering in auditory fNIRS signals ... [https://www.sciencedirect.com/science/article/pii/S0010482524009259]
• 66. Surface EMG Sensing and Granular Gesture Recognition ... [https://www.mdpi.com/2313-7673/10/4/229]
• 67. Motor Imagery Training on Muscle Strength and Gait ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4843370/]
• 68. Now is the time to adopt EMG monitoring: Game-changing ... [https://clinicalview.gehealthcare.com/article/now-time-adopt-emg-monitoring-game-changing-benefits]
• 69. Exploring neural activity changes during motor imagery-based ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12592122/]
• 70. Progress in the combined application of Brain-Computer ... [https://www.sciencedirect.com/science/article/abs/pii/S1388245725012350]
• 71. EEG-fMRI neurofeedback versus motor imagery after stroke, a ... [https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01598-9]
• 72. Resting-state EEG associated with clinical measures to ... [https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2025.1577393/full]
• 73. A Guide to Literature Informed Decisions in the Design of ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2020.00060/full]
• 74. Reading Your Own QEEG Brain Map [https://peakbraininstitute.com/reading-your-own-qeeg-brain-map/]
• 75. What Data is Used in Neurofeedback Training? | NHA Health [https://nhahealth.com/what-data-is-used-in-neurofeedback-training/]
• 76. A Protocol for the Administration of Real-Time fMRI ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5614365/]
• 77. A High-Accuracy and Interpretable Decoding Method for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12431000/]
• 78. Decoding of four movement directions using hybrid NIRS ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC4009438/]
• 79. Toward a compact hybrid brain-computer interface ( BCI )... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7080269/]
• 80. Development and validation of a machine learning-based ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12572785/]
• 81. Development and validation of a machine learning-based ... [https://translational-medicine.biomedcentral.com/articles/10.1186/s12967-024-04896-3]
• 82. Development and Validation of a Clinical-Based Signature ... [https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2022.904085/full]
• 83. Heterogeneous transfer learning model for improving the ... [https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2025.1555690/full]
• 84. A study of motor imagery EEG classification based on ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12307489/]
• 85. Reproducibility of fNIRS within Subject for Visual and Motor ... [https://www.sciencedirect.com/science/article/pii/S1053811925004951]
• 86. Hierarchical attention enhanced deep learning achieves ... [https://www.nature.com/articles/s41598-025-17922-1]
• 87. Real-World Evaluation of Deep Learning Decoders for ... [https://www.frontiersin.org/journals/systems-neuroscience/articles/10.3389/fnsys.2025.1718390/full]
• 88. Enhancing the quality of kinesthetic motor imagery for ... [https://www.sciencedirect.com/science/article/pii/S1053811925000539]