ABCD-BIDS Preprocessing Pipeline: A Comprehensive Guide to Motion Denoising for Robust fMRI Analysis

Abstract

This article provides a comprehensive overview of the ABCD-BIDS preprocessing pipeline, with a specialized focus on motion denoising techniques for resting-state fMRI data. Tailored for researchers, scientists, and drug development professionals, it explores the critical challenge of head motion—the largest source of artifact in fMRI signals—and details how the ABCD-BIDS pipeline addresses this issue through integrated methods including global signal regression, respiratory filtering, motion parameter regression, and despiking. The content covers foundational principles, practical implementation, troubleshooting of common issues, and validation strategies based on recent large-scale studies including the Adolescent Brain Cognitive Development (ABCD) Study. By synthesizing methodological guidance with empirical evidence from processing thousands of datasets, this guide aims to empower researchers to optimize motion denoising protocols for more reliable brain-behavior association studies and clinical research applications.

Understanding the Motion Denoising Challenge in Large-Scale fMRI Studies

The Critical Impact of Head Motion on fMRI Functional Connectivity

Head motion is a major confounding factor that impairs the quality of functional magnetic resonance imaging (fMRI) data, particularly for resting-state functional connectivity (FC) studies. Even minor, sub-millimeter movements during image acquisition introduce spurious signal fluctuations that can systematically bias correlation measures between brain regions [1] [2]. These motion-induced artifacts represent a significant challenge for researchers, especially when studying populations prone to greater in-scanner movement, such as children, older adults, and individuals with neurological or psychiatric conditions [1] [3]. Within the context of the ABCD-BIDS preprocessing pipeline research, understanding and mitigating these artifacts is paramount for generating reliable, reproducible biomarkers for drug development and cross-sectional studies.

The impact of motion on FC is not random; it follows recognizable patterns, often decreasing long-distance connectivity while increasing short-range connectivity, most notably within the default mode network [3]. Despite the development of numerous retrospective denoising algorithms, residual motion artifact often persists, potentially leading to both false positive and false negative findings in brain-behavior association studies [3]. This application note details the quantitative impact of head motion, outlines validated protocols for its mitigation, and provides a framework for quality control within the ABCD-BIDS processing environment.

Quantitative Impact of Head Motion on fMRI Metrics

The effects of head motion on fMRI data quality and functional connectivity can be quantified through several key metrics. The tables below summarize these impacts and the corresponding quantitative measures used for assessment.

Table 1: Motion-Induced Effects on fMRI Data Quality

Affected Metric	Impact of Motion	Empirical Findings
Temporal Signal-to-Noise Ratio (tSNR)	Decrease	Prospective Motion Correction (PMC) reduced tSNR attenuation from 45% to 20% during large head movements [4].
Long-Distance Functional Connectivity	Decrease	Motion causes systematic reduction in long-distance FC [3].
Short-Range Functional Connectivity	Increase	Motion causes spurious increases in short-range connections [3].
Dissimilarity of Functional Connectivity (DiFC)	Increase	DiFC is positively associated with Framewise Displacement (FD), independent of population, scanner, or preprocessing method [1].
Resting-State Network (RSN) Spatial Definition	Degradation	Spatial definition of major RSNs (Default Mode, Visual) is impaired by motion and improved with PMC [4].

Table 2: Key Quantitative Metrics for Motion Artifact Assessment

Metric	Description	Interpretation
Framewise Displacement (FD)	Summarizes volume-to-volume head movement based on translational and rotational parameters [1].	Higher mean FD indicates greater overall motion. A common censoring threshold is FD > 0.2 mm [3].
DVARS	Measures the rate of change of BOLD signal across the entire brain at each frame.	Peaks in DVARS often coincide with motion spikes.
Dissimilarity of Functional Connectivity (DiFC)	A group-level QC metric that quantifies how an individual's motion affects the group's FC quality [1].	A strong association between motion and DiFC suggests significant residual motion artifact in the preprocessed data.
Motion Impact Score (SHAMAN)	A trait-specific score quantifying whether motion causes over- or under-estimation of a brain-behavior relationship [3].	A significant score indicates that a specific trait-FC finding is likely biased by motion.

Experimental Protocols for Motion Artifact Assessment

Protocol: Assessing Residual Motion with Dissimilarity of FC

This protocol evaluates the effectiveness of denoising at the group level by measuring the correlation between participants' motion and their functional connectivity profiles [1].

• Data Requirements: Preprocessed resting-state fMRI data and corresponding Framewise Displacement (FD) timeseries for a group of participants (N > 50 recommended).
• FC Calculation: For each participant, compute a whole-brain functional connectivity matrix (e.g., using a predefined atlas with 200-400 regions).
• Reference FC Profile: Calculate the group-average FC matrix by taking the mean of all participants' FC matrices.
• Individual Dissimilarity: For each participant, compute the correlation (e.g., Pearson's r) between their individual FC matrix and the group-average FC matrix. Transform this correlation into a dissimilarity metric (DiFC) for each subject i as: DiFC_i = 1 - r_i.
• Statistical Analysis: Perform a group-level regression analysis with mean FD as the independent variable and DiFC as the dependent variable.
• Interpretation: A significant positive association between FD and DiFC indicates that participants with higher motion are less representative of the group's connectivity, suggesting residual motion artifact. The goal of denoising is to flatten this regression slope as much as possible [1].

Protocol: The SHAMAN Framework for Trait-Specific Motion Impact

The Split-Half Analysis of Motion-Associated Networks (SHAMAN) method determines if a specific trait-FC relationship is confounded by motion [3].

• Data Preparation: For each participant, divide the preprocessed fMRI timeseries into two halves: a "low-motion" half (volumes with FD below the sample median) and a "high-motion" half (volumes with FD above the median).
• Split-Half FC Calculation: Compute separate FC matrices for the low-motion and high-motion halves for every participant.
• Trait-FC Effect Estimation: For a trait of interest (e.g., cognitive score), compute the correlation between the trait and each FC edge (connection) across participants, once using the FC values from the low-motion half and once using the high-motion half. This yields two trait-FC effect maps.
• Motion Impact Score: Calculate the difference between the two trait-FC effect maps (High-motion - Low-motion). A positive score indicates motion inflates the trait-FC effect (overestimation), while a negative score indicates motion suppresses it (underestimation) [3].
• Statistical Inference: Use permutation testing (e.g., shuffling trait labels) to generate a null distribution and compute a p-value for the observed motion impact score.

Figure 1: The SHAMAN workflow for calculating a trait-specific motion impact score.

The Researcher's Toolkit: Essential Reagents & Materials

Table 3: Key Software and Analytical Tools for Motion Mitigation

Tool/Solution	Function	Application Context
ABCD-BIDS Pipeline	A standardized preprocessing pipeline incorporating global signal regression, respiratory filtering, and motion censoring [5].	Default denoising for large-scale studies like the Adolescent Brain Cognitive Development (ABCD) Study.
Prospective Motion Correction (PMC)	Real-time tracking of head movement with markerless optical systems to update scan parameters during acquisition [6] [4].	Improving data quality at acquisition, especially in populations unable to remain still. Crucial for high-field (7T) fMRI [7].
ICA-AROMA	A robust ICA-based strategy for automatically identifying and removing motion-related components from fMRI data [1].	Aggressive denoising when prospective methods are unavailable or insufficient.
Framewise Displacement (FD)	A scalar summary of volume-to-volume head motion [1].	A universal metric for quantifying motion levels and setting censoring thresholds.
XCP (eXtensible Connectivity Pipeline)	Implements a validated, high-performance denoising strategy combining physiological signals, motion estimates, and mathematical expansions [2].	A flexible pipeline for mitigating motion artifact in FC studies, providing up to 100-fold improvement over minimal processing.

Integrated Denoising Protocol within the ABCD-BIDS Framework

The following workflow details the specific motion denoising steps as implemented in the ABCD-BIDS pipeline, which has been shown to achieve a 69% relative reduction in motion-related signal variance compared to minimal processing [3].

• Standard Preprocessing:

  • Nuisance Regression: A general linear model (GLM) is used to regress out several confounding signals from the fMRI data:
    • Signal Regressors: Mean time series from white matter (WM), cerebrospinal fluid (CSF), and the global signal (GSR). The inclusion of GSR is critical as it consistently reduces motion effects [5].
    • Motion Regressors: The 6 rigid-body realignment parameters (X, Y, Z, roll, pitch, yaw) and their Volterra expansion [5].
  • Band-Pass Filtering: The residuals from the denoising GLM are band-pass filtered (0.008 - 0.09 Hz) using a 2nd-order Butterworth filter to retain low-frequency fluctuations relevant to resting-state FC [5].

• Respiratory Motion Filtering:

  • The motion realignment parameters are filtered to remove artifactual motion caused by respiration, which is a common issue in multi-band data. This step produces more accurate Framewise Displacement estimates [5].

• Motion Censoring ("Scrubbing"):

  • Identification: Frames (volumes) where FD exceeds a threshold of 0.3 mm are labeled as "bad" [5].
  • Processing: These "bad" frames are removed when de-meaning and de-trending the data. The denoising GLM betas are calculated using only "good" frames. For band-pass filtering, interpolation is used to replace "bad" frames to avoid aliasing, and the final cleaned signal is derived from the GLM residuals [5].
  • Post-Censoring Check: It is recommended to exclude participants with an excessively high proportion of censored frames (e.g., >10-25%) from group-level analyses to prevent introducing noise [8].

Figure 2: Core denoising workflow of the ABCD-BIDS pipeline for mitigating motion artifact.

Head motion remains a critical challenge for robust functional connectivity analysis. While retrospective denoising pipelines like ABCD-BIDS provide powerful tools for mitigating these artifacts, a comprehensive approach is essential. This includes:

• Prospective Correction: Where feasible, using PMC during data acquisition to minimize the introduction of motion artifact [4].
• Aggressive Denoising: Employing standardized, validated pipelines that incorporate GSR and motion censoring [5].
• Rigorous Quality Control: Implementing both quantitative (e.g., DiFC, SHAMAN) and qualitative (visual inspection) checks to evaluate the success of denoising and identify potential confounds [1] [3] [8].

For researchers in drug development and clinical neuroscience, adhering to these protocols ensures that functional connectivity biomarkers derived from the ABCD-BIDS pipeline are more reliable and less susceptible to motion-induced false discoveries, thereby increasing the translational validity of the findings.

Historical Context and Motivation

The ABCD-BIDS pipeline did not emerge in isolation; it is a product of the broader neuroimaging community's decades-long effort to improve data sharing and reproducibility. The sharing of scientific neuroimaging data maximizes the knowledge derived from research, benefits funders and participants, enables reproduction of results, and levels the playing field for researchers from under-resourced environments [9]. Within neuroimaging, early data-sharing efforts began around 2000 with the fMRI Data Center, flourishing a decade later with the International Neuroimaging Data-sharing Initiative (INDI)/Functional Connectomes Project [9]. However, a significant challenge remained: making shared data truly FAIR (Findable, Accessible, Interoperable, and Reusable) [9].

The Brain Imaging Data Structure (BIDS) community standard was born from a 2015 meeting at Stanford University to address these challenges [9]. Its development was guided by two core principles: (1) that adoption is crucial, leading to a focus on engaging the research community and mirroring existing practices, and (2) that it should not "reinvent the wheel," favoring established, human-readable formats like NIfTI, JSON, and TSV files over more complex alternatives [9]. The ABCD-BIDS pipeline is a concrete implementation of these principles, designed as a BIDS App to process large-scale datasets according to this community-driven standard [5].

Core Philosophical Principles

The philosophical foundation of the ABCD-BIDS pipeline is built upon the core BIDS principles that prioritize practical adoption and technical pragmatism. This philosophy emerged from the practical limitations of earlier data-sharing efforts like OpenfMRI, where informal data organization schemes created significant curation burdens and limited scalability [9]. By developing a "detailed and general scheme" for data organization that supports automated validation, the ABCD-BIDS pipeline shifts the burden of curation from archives to data owners, enabling long-term sustainability [9].

A key philosophical decision was maintaining closeness to existing community practices rather than imposing entirely new workflows. This adoption-first mindset explains the pipeline's use of human-readable text formats instead of more technically complex but less accessible formats like RDF or XML [9]. Furthermore, the "don't reinvent the wheel" principle led to the incorporation of established standards and tools, including the NIfTI file format, JSON for metadata, and proven processing methods from the Human Connectome Project's minimal preprocessing pipelines [9] [5].

Technical Architecture and Workflow

The ABCD-BIDS pipeline implements a sophisticated, multi-stage architecture that processes MRI data from raw inputs to refined outputs. The technical implementation can be visualized as a sequential workflow with distinct processing stages:

The pipeline's architecture processes anatomical and functional data through parallel streams that eventually converge. The anatomical stream begins with PreFreeSurfer, which removes distortions from structural data and performs brain extraction in native space, utilizing Advanced Normalization Tools (ANTs) for improved performance on noisy data from various scanner types [5]. The FreeSurfer stage then segments brain structures and reconstructs cortical surfaces, largely following the HCP minimal preprocessing approach [5]. Finally, PostFreeSurfer generates CIFTI surface files and registers data to the Conte-69 surface template [5].

Simultaneously, the functional stream starts with FMRIVolume, which corrects for gradient nonlinearities and motion in the BOLD timeseries [5]. FSL's topup corrects distortions using spin-echo EPI scans with opposite phase encoding directions [5]. FMRISurface then maps the volume timeseries to standard CIFTI grayordinates space [5]. The crucial DCANBOLDProcessing stage performs nuisance regression, band-pass filtering (0.008-0.09 Hz), and motion censoring, incorporating global signal regression which has been "consistently shown to reduce the effects of motion on BOLD signals" [5].

Motion Denoising and Quality Control

Within the broader context of motion denoising research, the ABCD-BIDS pipeline implements a sophisticated approach to addressing head motion artifacts, which represent "the largest source of artifact in structural and functional MRI signals" [3]. The pipeline's DCANBOLDProcessing stage incorporates multiple strategies for motion mitigation, including a respiratory motion filter that removes frequencies between 18.582-25.726 breaths per minute from motion realignment data to produce better framewise displacement estimates [5].

A critical component is the pipeline's motion censoring procedure, which labels data as "bad" frames if they exceed a framewise displacement threshold of 0.3 mm [5]. These contaminated frames are removed during demeaning and detrending, with betas for denoising calculated using only "good" frames [5]. The pipeline generates temporal masks from 0-0.5 mm FD thresholds in 0.01 mm steps, providing flexibility for researchers to apply different censoring levels during analysis [5]. Research has demonstrated that after standard denoising with ABCD-BIDS, motion censoring at FD < 0.2 mm reduces significant motion overestimation in trait-functional connectivity relationships from 42% to just 2% of traits [3].

The ExecutiveSummary stage produces comprehensive visual quality control, displaying T1w and T2w segmentations, atlas registration overlays, and movement time series for each fMRI run both pre- and post-regression [5]. This transparency in quality assessment aligns with the pipeline's overarching philosophy of enabling reproducible research.

Implementation and Impact

The ABCD-BIDS pipeline has demonstrated significant real-world impact through its implementation in the ABCD-BIDS Community Collection (ABCC), "a rigorously curated MRI dataset derived from the ABCD Study" [10]. The pipeline processes data from over 11,000 participants in the ABCD study, with the latest release including BIDS inputs and processed derivatives for longitudinal timepoints [10].

Table 1: ABCD-BIDS Pipeline Data Availability in ABCC Release 3.0.0

Year	BIDS Inputs	ABCD-HCP Pipeline Derivatives	DWI Inputs	QSIPrep Derivatives
Baseline	11,753	11,751	9,564	8,852
2	8,086	8,085	7,669	7,273
4	6,355	6,351	6,207	6,066
6	3,820	3,820	3,748	3,676

The pipeline has supported numerous high-impact studies, including research on brain asymmetry and its association with inattention [11], investigations of motion impact on brain-behavior associations [3], and the creation of brain charts for the human lifespan [10]. Over 90 studies have been published referencing the ABCC, demonstrating the pipeline's substantial contribution to advancing neurodevelopmental science [10].

Table 2: Essential Research Reagents for ABCD-BIDS Pipeline Implementation

Resource	Function	Implementation Notes
BIDS Dataset	Input data organization	Must follow BIDS standard structure with appropriate file naming and metadata
FreeSurfer License	Required for anatomical processing	Must be acquired separately and provided to the pipeline via license.txt file
Gordon 333 ROI Atlas	Functional connectivity parcellation	Default template for generating parcellated timeseries
Power 264 ROI Atlas	Alternative connectivity parcellation	Included as additional template option
Conte-69 Surface Atlas	Surface registration target	Standard surface template for cortical alignment
DCAN Executive Summary	Quality control visualization	Generates HTML reports with BrainSprite viewer for visual QC
Container Technology	Pipeline execution environment	Docker images available from DCAN Docker Hub repository

The ABCD-BIDS pipeline continues to evolve, with ongoing developments including the integration of additional processing tools like fMRIPrep and XCP-D, expansion to specialized populations through variants like the infant-abcd-bids-pipeline [10] [12], and adherence to reproducibility standards through peer review under the NMIND infrastructure [10]. This commitment to continuous improvement while maintaining standardization embodies the pipeline's core philosophy of balancing innovation with practical utility for the research community.

Within the framework of research utilizing the ABCD-BIDS preprocessing pipeline for motion denoising, characterizing the specific nature and impact of motion artifacts is paramount. Head motion remains the largest source of artifact in functional MRI (fMRI) signals, introducing systematic bias rather than random noise into the data [3]. This systematic influence means that simply increasing sample sizes does not automatically mitigate the problem; it can, in fact, perpetuate and even amplify spurious findings if motion is correlated with the traits or behaviors under investigation [13] [14]. This application note details the characteristics of these biases, provides quantitative data on their effects, and outlines protocols for their detection and mitigation, with a specific focus on analyses within large-scale datasets like the Adolescent Brain Cognitive Development (ABCD) Study.

Quantitative Characterization of Motion-Induced Biases

Motion artifacts manifest in predictable patterns that can lead to both overestimation and underestimation of true brain-behavior relationships. The tables below summarize key quantitative findings on the prevalence and impact of these biases.

Table 1: Impact of Residual Motion on Trait-FC Associations After Standard Denoising (ABCD-BIDS)

Metric	Number of Traits Affected (out of 45)	Percentage of Traits
Significant Motion Overestimation Score	19	42%
Significant Motion Underestimation Score	17	38%
Total Traits with Significant Motion Impact	36	80%

Source: Adapted from Kay et al. [3] [15]

Table 2: Effect of Motion Censoring on Trait-FC Associations

Censoring Threshold (Framewise Displacement)	Traits with Significant Overestimation	Traits with Significant Underestimation
No Censoring	19/45 (42%)	17/45 (38%)
FD < 0.2 mm	1/45 (2%)	17/45 (38%)*

Source: Adapted from Kay et al. [3] [15]. *Note: Censoring did not reduce the number of traits with significant underestimation scores.

Table 3: Structural MRI (sMRI) Biases from Scan Quality

Scan Quality	Cortical Thickness	Cortical Surface Area	Significant Regions in Aggression Analysis
High-Quality Scans Only	Accurate Estimate	Accurate Estimate	3
Include Moderate-Quality Scans	Underestimation	Overestimation	21
Include All Scans	Further Underestimation	Further Overestimation	43

Source: Adapted from Roffman et al. [13]

Experimental Protocols for Motion Impact Assessment

Protocol: Split Half Analysis of Motion Associated Networks (SHAMAN)

The SHAMAN method assigns a motion impact score to specific trait-functional connectivity (FC) relationships, distinguishing between overestimation and underestimation [3].

Detailed Methodology:

• Data Prerequisites: Processed resting-state fMRI data from one or more scans per participant, along with trait data and framewise displacement (FD) timeseries.
• Data Splitting: For each participant's fMRI timeseries, split the data into high-motion and low-motion halves based on the Framewise Displacement (FD) of each volume.
• Connectivity Calculation: Compute separate functional connectivity matrices for the high-motion and low-motion halves for each participant.
• Trait-FC Effect Estimation: Calculate the correlation between the trait and each edge (connection) in the FC matrix. Do this separately for the high-motion and low-motion halves.
• Motion Impact Score:
  • For each edge, calculate the difference in trait-FC effect between the low-motion and high-motion halves.
  • A motion impact score that is aligned with the direction of the overall trait-FC effect indicates motion overestimation (i.e., motion inflates the observed effect).
  • A motion impact score opposite the direction of the overall trait-FC effect indicates motion underestimation (i.e., motion masks the true effect).
• Statistical Inference: Use permutation testing of the timeseries and non-parametric combining across edges to generate a p-value for the motion impact score, determining whether the trait-FC relationship is significantly influenced by motion.

Protocol: Evaluating Quality Control Trade-offs and Bias

This protocol assesses how quality control (QC) decisions, such as motion censoring thresholds, can bias sample representativeness and subsequently, brain-behavior associations [16] [17].

Detailed Methodology:

• Define QC Conditions: Establish a range of plausible QC thresholds (e.g., framewise displacement thresholds from 0.1 mm to 0.3 mm, or minimum frame retention after censoring).
• Apply Inclusion/Exclusion: For each QC threshold, determine which participants are included in the analysis based on their motion parameters.
• Characterize the Retained Sample: For each resulting sample, compare the distributions of key participant characteristics (e.g., demographic factors, socioeconomic status, clinical scores, cognitive performance) against the full baseline sample.
• Quantify Bias: Use statistical tests (e.g., chi-square, t-tests) to identify characteristics that are significantly over- or under-represented in the retained sample compared to the full sample. Calculate odds ratios for exclusion associated with each characteristic.
• Report and Mitigate: Document the biases introduced by each QC threshold. To mitigate bias, employ statistical methods such as multiple imputation or population weighting rather than simple listwise deletion of high-motion participants [16] [18].

The Scientist's Toolkit: Research Reagent Solutions

Table 4: Essential Tools for Motion Artifact Research in the ABCD Dataset

Tool / Resource	Function / Purpose	Application Notes
ABCD-BIDS Pipeline	Default preprocessing pipeline for ABCD data; includes global signal regression, motion parameter regression, despiking, and respiratory filtering.	Reduces motion-related signal variance by ~69% compared to minimal processing, but significant residual bias remains [3] [19].
Framewise Displacement (FD)	A scalar quantity summarizing volume-to-volume head motion.	Primary metric for quantifying in-scanner motion and setting censoring thresholds [3] [16].
SHAMAN (Split Half Analysis)	Computes a trait-specific motion impact score to flag spurious brain-behavior associations.	Critical for diagnosing whether motion causes over- or under-estimation of a specific trait-FC effect [3] [15].
Surface Hole Number (SHN)	An automated quality metric estimating imperfections in cortical surface reconstruction.	Serves as a proxy for manual quality ratings in sMRI; useful for stress-testing structural findings [13].
Motion-Ordering & Bagging	Mitigation methods that rank and resample fMRI timepoints from least to most motion.	Retains >99% of participants from high-motion groups, enhancing diversity and representation while controlling for motion [17].
Multiple Imputation / Population Weighting	Statistical techniques to account for data missing not-at-random (e.g., due to motion-related exclusion).	Corrects for biases introduced by listwise deletion of participants, improving generalizability [16] [18].

Integrated Analysis Workflow and Decision Pathway

The following diagram integrates the protocols and tools above into a coherent workflow for handling motion artifacts, from initial processing to final inference, ensuring both data quality and sample representativeness are considered.

The Adolescent Brain Cognitive Development (ABCD) Study represents an unprecedented large-scale longitudinal investigation of brain development and child health in the United States [20]. As the largest long-term study of brain development and behavior in adolescents, it tracks over 11,800 youth from ages 9-10 into young adulthood, employing state-of-the-art multimodal brain imaging, comprehensive behavioral assessments, and bioassays [21]. This application note details the scale, demographics, and data acquisition protocols of the ABCD Study, with particular emphasis on its relevance for researchers investigating motion denoising techniques within the ABCD-BIDS preprocessing pipeline framework.

The ABCD Study utilizes a longitudinal cohort design, following participants from pre-adolescence to young adulthood with annual lab-based assessments and biennial imaging acquisitions [20]. The study's initial primary objective was to examine risk and resilience factors associated with substance use disorders, particularly cannabis, though its aims have expanded to inform population-level inferences about the biopsychosocial correlates of mental and physical health throughout adolescence [20].

Table 1: ABCD Study Core Characteristics

Characteristic	Specification
Sample Size	~11,800 youth [20]
Age at Baseline	9-10 years old [21]
Planned Duration	10 years [21]
Number of Sites	21 sites across the US [20]
Imaging Frequency	Biennial [20]
Behavioral Assessment	Annual [20]
Special Populations	Includes over 800 twin pairs [21]

The recruitment strategy aimed to create a population-level, socio-demographically diverse sample through probability sampling within a combined catchment area encompassing approximately 20% of all 9- and 10-year-olds in the United States [16]. This sampling approach facilitates broader generalizability of findings while acknowledging that the sample is not perfectly representative of the entire U.S. population [16].

Participant Demographics and Characterization

The ABCD Study cohort includes substantial diversity across socioeconomic, racial, and ethnic backgrounds, though some demographic factors consistently relate to data quality and participant retention [16]. Prior to the COVID-19 pandemic, parental education level and employment status were the most consistent predictors of missed visits and study withdrawal [20]. These challenges were likely exacerbated during the pandemic, with non-White and/or Spanish-speaking families experiencing more financial worry and increased food insecurity [20].

Table 2: Key Demographic Considerations and Relationships with Data Quality

Demographic Factor	Relationship to Data Quality & Retention
Parental Education	Consistent predictor of missed visits and study withdrawal [20]
Employment Status	Associated with risk for missed visits [20]
Race/Ethnicity	Differential pandemic effects on participation [20]
Socioeconomic Status	Associated with both data quality and retention [16]
Family Income	Related to exclusion probability in fMRI analyses [16]

Researchers should note that systematic relationships exist between participant characteristics and data exclusion due to quality control procedures, particularly for resting-state fMRI data [16]. These biases become more pronounced with stricter motion scrubbing thresholds, potentially affecting the generalizability of findings if not properly accounted for statistically [16].

Data Acquisition Protocols

Multimodal Imaging Acquisition

The ABCD imaging protocol represents a significant advancement in large-scale, multi-site neuroimaging, building upon state-of-the-art protocols from the Pediatric Imaging, Neurocognition, and Genetics (PING) study and the Human Connectome Project (HCP) [21]. Through close collaboration with three major MRI system manufacturers (Siemens, General Electric, and Philips), the ABCD protocol achieves HCP-style temporal and spatial resolution on all three manufacturers' 3 Tesla systems without requiring non-commercially available upgrades [21].

The comprehensive imaging protocol includes:

• High-resolution structural MRI: T1-weighted (T1w) and T2-weighted (T2w) structural images [21] [22]
• Diffusion MRI (dMRI): Multiple b-values and directions for assessing brain microstructure [21] [22]
• Resting-state fMRI (rs-fMRI): For investigating functional connectivity networks [22]
• Task-based fMRI: Including Monetary Incentive Delay (MID), Stop Signal Task (SST), and Emotional N-Back (EN-back) paradigms to probe reward processing, executive control, and working memory, respectively [21] [22]

Motion Mitigation During Acquisition

A critical consideration for pediatric neuroimaging, particularly in the context of motion denoising research, is the implementation of real-time motion monitoring and correction technologies [21]. The ABCD Study employs several sophisticated approaches to address the challenge of head motion:

• Prospective motion correction for sMRI acquisitions on Siemens (using navigator-enabled sequences) and GE (using PROMO sequences) scanners, with plans for implementation on Philips platforms [21]
• Real-time motion monitoring of fMRI acquisitions at Siemens sites using the Frame-wise Integrated Real-time Motion Monitoring (FIRMM) software, which provides operators with real-time head motion estimates and allows for scan adjustment based on whether data quality criteria have been met [21]
• Structured discarding of initial volumes: For Siemens and Philips scanners, the first 8 volumes are removed, while different schemes are applied for GE scanners depending on software version [23]

Figure 1: ABCD Real-time Motion Monitoring and Acquisition Workflow

ABCD-BIDS Preprocessing Pipeline

The ABCD-BIDS pipeline is a BIDS App that processes BIDS-formatted MRI datasets using methods from both the Human Connectome Project's minimal preprocessing pipeline and DCAN Labs resting state fMRI analysis tools [24] [5]. The pipeline outputs preprocessed MRI data in both volume and surface spaces and follows NMIND guidelines for reproducibility and standardization, having undergone formal peer review and meeting at least Bronze badge standards in the NMIND rating system [24].

Pipeline Stages and Methodologies

Table 3: ABCD-BIDS Pipeline Processing Stages

Stage	Primary Function	Key Methodologies
PreFreeSurfer	Remove distortions from anatomical data; align and extract brain in native volume space	ANTs for denoising and N4 bias field correction; Enables processing without T2w image [5]
FreeSurfer	Brain segmentation; cortical surface reconstruction	Standard HCP pipeline largely unchanged from Glasser et al. 2013 [5]
PostFreeSurfer	Generate CIFTI surface files; apply surface registration	Uses refined brain mask from FreeSurfer; ANTs registration outperforms FNIRT-based approach [5]
FMRIVolume	Start functional processing; correct distortions and motion	FSL topup for distortion correction; FSL FLIRT for motion correction; non-linear registration to MNI space [5]
FMRISurface	Map volume time series to CIFTI grayordinates space	Unchanged from original HCP pipeline [5]
DCANBOLDProcessing	Nuisance regression; motion censoring	Respiratory motion filtering; frame censoring (FD threshold 0.3mm); global signal regression [5]
ExecutiveSummary	Generate HTML visual quality control page	BrainSprite viewer; movement and time series visualization [5]

Motion Denoising in the DCAN BOLD Processing Stage

The DCANBOLDProcessing (DBP) stage implements sophisticated motion denoising procedures that are particularly relevant for research on motion artifact correction [5]. This stage involves four broad steps:

• Standard pre-processing: Includes de-meaning and de-trending of fMRI data, followed by denoising using a general linear model with regressors comprising signal variables (white matter, CSF, and global signal) and movement variables (translational and rotational measures with Volterra expansion) [5]

• Respiratory motion filtering: Implements a specialized filter to address respiratory artifacts in multi-band data that can affect frame alignment estimates, using frequency ranges of 18.582 to 25.726 breaths per minute [24] [5]

• Motion censoring: Identifies "bad" frames exceeding a framewise displacement (FD) threshold of 0.3 mm, which are removed during demeaning and detrending, with interpolation used initially to replace bad frames during band-pass filtering [5]

• Generation of parcellated timeseries: Constructs time series for pre-defined atlases including Gordon's 333 ROI, Power's 264 ROI, Yeo's 118 ROI, and HCP's 360 ROI atlas templates [5]

Figure 2: DCAN BOLD Processing (DBP) Motion Denoising Workflow

Quality Control Procedures

The ABCD Study employs comprehensive quality control procedures that combine automated metrics with crowdsourced visual inspection [25]. These QC measures are particularly important for motion denoising research, as they help identify residual artifacts and ensure data quality.

Automated Quality Assurance

Automated QC metrics evaluate both structural and functional data quality [25]:

• Structural metrics: Assess brain coverage requirements (must have less than 10% cutoff in both ventral/inferior and dorsal/superior planes) and registration quality
• Functional metrics: Focus on motion and signal issues, particularly the presence of parcellated connectivity matrices derived from 5 and 10 minutes of functional data after motion correction using a frame displacement threshold of 0.2 mm
• Population-based outlier detection: Identifies outliers falling more than 3 standard deviations from the population mean for structural measures (subcortical segmentation volumes, cortical morphometry) and functional measures (connectivity matrices)

Visual Quality Control via BrainSwipes

The study implements BrainSwipes, a gamified, crowdsourced QC platform built on the open-source Swipes for Science framework, which engages users in evaluating brain image quality through an intuitive interface [25]. This platform includes:

• Surface delineation assessment: Evaluation of T1w and T2w surface delineations in differentiating gray and white matter across coronal, axial, and sagittal planes
• Atlas registration quality: Assessment of alignment between subject images and atlas templates
• Functional registration: Evaluation of functional-to-structural alignment and artifacts such as signal dropout
• Diffusion direction encoding: Quality assessment of processed DWI images

Data Derivatives and Access

The ABCD Data Release 6.0 contains minimally processed neuroimaging data formatted according to BIDS specifications [22] [26]. These derivatives include:

• High-resolution structural data (3D T1w and T2w scans) corrected for gradient nonlinearity distortions and intensity inhomogeneity
• Advanced diffusion MRI with multiple b-values and directions
• Resting-state and task-fMRI data that have undergone distortion correction, movement correction, and alignment to standard space
• Complete FreeSurfer output directories using FreeSurfer version 7.1.1 for cortical surface reconstruction [22]

Data is available through the NIH Brain Development Cohorts (NBDC) Data Hub, which provides scalable data integration, customized query tools, and streamlined data use certification workflows [26]. Researchers are encouraged to account for the systematic relationships between participant characteristics and data quality when analyzing these data, employing appropriate statistical methods such as multiple imputation to address potential biases introduced by quality-based exclusions [16].

The Scientist's Toolkit: Research Reagent Solutions

Table 4: Essential Research Tools for ABCD Data Analysis

Tool/Resource	Function	Relevance to Motion Denoising
ABCD-BIDS Pipeline	Docker container for standardized processing of ABCD MRI data	Primary pipeline implementing motion correction and denoising [24]
QSIPrep	Containerized pipeline for diffusion MRI data preprocessing	Provides automated QC metrics for diffusion data [24] [25]
BrainSwipes	Gamified platform for visual quality control	Enables crowdsourced QC of processed data [25]
FIRMM	Real-time head motion monitoring during fMRI acquisition	Provides motion estimates during data collection [21]
fMRIPrep	Robust preprocessing pipeline for fMRI data	Alternative pipeline option; requires raw DICOM conversion [23]
HeuDiConv	DICOM to BIDS converter	Essential for converting ABCD "fast track" data to BIDS format [23]
ABCD Community Collection	BIDS-formatted derivatives	Preprocessed data for analysis [26]
FreeSurfer	Automated cortical surface reconstruction	Used for segmentation and surface reconstruction [22]

Why Motion is Particularly Problematic in Developmental and Clinical Populations

In-scanner head motion presents a fundamental challenge for functional magnetic resonance imaging (fMRI), but its impact is profoundly magnified in developmental and clinical populations. This document details why motion artifact is not merely a technical nuisance but a significant source of systematic bias that can compromise the validity of neuroscientific findings in these groups. Research consistently demonstrates that motion systematically alters functional connectivity (FC) measures, reducing long-distance connections and increasing short-range connectivity [27] [28]. Because both developmental status and clinical conditions are often correlated with increased head motion, studies comparing children to adults or clinical groups to healthy controls are uniquely vulnerable to spurious findings where motion, rather than the variable of interest, drives observed differences in brain function [27] [28] [17]. This Application Note synthesizes current evidence and provides practical protocols for mitigating these biases, with a specific focus on the ABCD-BIDS preprocessing pipeline, a community standard for large-scale developmental neuroimaging studies like the Adolescent Brain Cognitive Development (ABCD) Study [10].

The Nature and Magnitude of the Problem

Systematic Bias in Group Comparisons

Motion artifacts introduce non-random noise that can fundamentally alter scientific inference. In developmental and clinical neuroscience, the variables of primary interest—such as age, diagnosis of a neurodevelopmental disorder, or cognitive ability—are frequently correlated with the amount of in-scanner head motion [27] [17]. For instance, younger children, individuals with attention-deficit/hyperactivity disorder (ADHD), or those with autism spectrum disorder typically exhibit greater motion than comparison groups [28] [17]. This correlation creates a systematic bias wherein motion-induced changes in functional connectivity can be misattributed to the group difference or trait under investigation.

A landmark 2025 study analyzing data from 7,270 participants in the ABCD Study found that even after standard denoising with the ABCD-BIDS pipeline, residual motion had a pronounced effect on trait-FC associations [28]. Without motion censoring, 42% (19/45) of behavioral traits showed significant motion overestimation scores, while 38% (17/45) showed significant underestimation scores [28]. This indicates that motion can artificially inflate or mask genuine brain-behavior relationships, leading to both false positive and false negative findings.

Quantitative Impact of Motion on fMRI Signals

The following table summarizes the quantitative effects of motion on fMRI data, as revealed by empirical studies:

Table 1: Quantitative Impact of Motion on fMRI Metrics

Metric	Impact of Motion	Population Studied	Source
Signal Variance Explained by Motion	73% of variance after minimal processing; 23% after ABCD-BIDS denoising (69% relative reduction)	ABCD Study participants (n=9,652)	[28]
Correlation between Motion-FC effect and Average FC matrix	Spearman ρ = -0.58 (strong negative correlation)	ABCD Study participants	[28]
Disproportionate Exclusion of Minoritized Youth	Black and Hispanic youth exhibited excess head motion and were disproportionately discarded by conventional QC	ABCD Study (N=5,733)	[17]
Impact of Censoring on Motion Overestimation	Censoring at FD < 0.2 mm reduced significant overestimation from 42% to 2% of traits	ABCD Study (n=7,270)	[28]

The Exclusion Paradox and Representativeness

Conventional quality control practices that exclude high-motion participants create a significant ethical and methodological dilemma: they disproportionately remove data from underrepresented groups. Analysis of ABCD Study data confirms that Black and Hispanic youth exhibit greater head motion relative to White youth, leading to their disproportionate exclusion when standard framewise displacement (FD) thresholds are applied [17]. This practice potentially biases samples and limits the generalizability of findings. When standard FD thresholds were applied, more than 16% of the ABCD sample was excluded [17]. Methods such as motion-ordering and bagging can retain over 99% of Black and Hispanic youth while producing reproducible brain-behavior associations [17], offering a more inclusive approach without sacrificing scientific rigor.

Experimental Protocols for Motion Mitigation and Assessment

Protocol 1: Implementing the ABCD-BIDS Preprocessing Pipeline

The ABCD-BIDS Community Collection (ABCC) provides a rigorously curated MRI dataset derived from the ABCD Study, leveraging BIDS standards and NMIND-reviewed pipelines for reproducible neuroimaging [10]. The following workflow details the key stages of the ABCD-HCP pipeline, which incorporates robust motion denoising procedures [5].

Key Denoising Components:

• FMRIVolume Stage: Performs rigid-body motion correction via FSL FLIRT, realigning each volume to an initial reference frame. It also employs FSL's topup using spin-echo EPI pairs with opposite phase encoding directions to correct for distortions. Motion parameters (3 translations, 3 rotations) are output for subsequent denoising [5].

• DCANBOLDProcessing (DBP) Stage: This critical stage handles nuisance regression and motion censoring [5]:

  • Respiratory Motion Filter: An optional but recommended step that filters frequencies associated with respiratory artifacts (18.582-25.726 breaths per minute) from the motion realignment data, producing more accurate framewise displacement estimates [5].
  • Standard Pre-processing: Involves de-meaning, de-trending, and denoising via a general linear model (GLM). Nuisance regressors include:
    • Signal from white matter, CSF, and global signal (GSR)
    • 6 motion parameters and their Volterra expansion
  • Motion Censoring: Frames exceeding an FD threshold of 0.3 mm are labeled "bad" and removed during de-meaning and de-trending. The GLM for denoising is calculated using only "good" frames. Band-pass filtering (0.008-0.09 Hz) is applied with interpolation to replace censored frames, preventing aliasing [5].

Protocol 2: Assessing Trait-Specific Motion Impact with SHAMAN

For studies investigating specific brain-behavior relationships, the Split Half Analysis of Motion Associated Networks (SHAMAN) method provides a way to quantify whether a particular trait-FC association is contaminated by residual motion artifact [28]. This is particularly crucial when studying traits known to correlate with motion, such as inattention symptoms.

Table 2: Research Reagent Solutions for Motion Management

Reagent/Solution	Function in Motion Management	Implementation Example
Framewise Displacement (FD)	Quantifies volume-to-volume head motion; used for censoring thresholds	Calculated from motion realignment parameters [27] [5]
Global Signal Regression (GSR)	Nuisance regressor that reduces motion-related artifacts	Included in ABCD-BIDS denoising GLM [5]
Respiratory Motion Filter	Removes respiratory artifacts from motion estimates	Applied to realignment parameters in DBP stage [5]
Motion Censoring ( Scrubbing)	Removes high-motion frames from analysis	Implemented at FD threshold of 0.3 mm in DBP [5]
SHAMAN Algorithm	Quantifies trait-specific motion impact	Computes motion overestimation/underestimation scores [28]

Procedure:

• Data Requirements: One or more resting-state fMRI scans per participant, along with trait data of interest (e.g., cognitive scores, clinical symptoms).

• Split-Half Analysis:

  • For each participant, divide the preprocessed fMRI timeseries into high-motion and low-motion halves based on framewise displacement.
  • Compute separate functional connectivity matrices for each half.

• Motion Impact Score Calculation:

  • Calculate the correlation between the trait and each FC edge in both the high-motion and low-motion halves.
  • The motion impact score is derived from the difference in trait-FC correlations between the two halves.
  • A significant positive score aligned with the trait-FC effect indicates motion overestimation.
  • A significant negative score opposite the trait-FC effect indicates motion underestimation.

• Statistical Inference:

  • Use permutation testing of the timeseries and non-parametric combining across connections to generate a p-value for the motion impact score [28].

Protocol 3: Inclusive Analysis with Motion-Ordering and Bagging

To address the disproportionate exclusion of high-motion participants from minoritized groups, the following protocol provides an alternative to traditional censoring:

Motion-Ordering Procedure:

• For each participant's motion-limited fMRI data, rank all timepoints from least motion to most motion based on FD.
• Compute the functional connectivity matrix using only the top N timepoints (e.g., the 5-10 minutes with least motion) from this ordered list [17].

Motion-Ordering with Bagging (Resampling):

• After motion-ordering, generate multiple bootstrapped samples (e.g., 500) with replacement from the top timepoints.
• Compute a functional connectivity matrix for each bootstrapped sample.
• Generate a final, bagged functional connectivity matrix by averaging across all bootstrap iterations [17].

Validation: Assess the reproducibility of brain-behavior associations by examining confidence intervals and area under the curve (AUC) metrics across racial/ethnic groups. Comparable results between standard and motion-ordering methods indicate that motion-limited data can yield valid inferences while maximizing sample inclusivity [17].

Discussion and Recommendations

The evidence unequivocally demonstrates that motion artifact presents a particularly severe challenge in developmental and clinical neuroimaging. The correlation between motion and variables of interest creates systematic bias that can undermine the validity of research findings. Based on current evidence, we recommend:

• Mandatory Motion Mitigation: Always employ comprehensive denoising pipelines like ABCD-BIDS that include global signal regression, respiratory filtering, and motion censoring [10] [5].

• Trait-Specific Motion Assessment: For studies focusing on specific brain-behavior relationships, implement methods like SHAMAN to quantify and account for trait-specific motion impacts [28].

• Inclusive Quality Control: Adopt motion-ordering or bagging methods when studying diverse populations to prevent disproportionate exclusion of minoritized participants and maintain sample representativeness [17].

• Transparent Reporting: Clearly document motion thresholds, exclusion rates, and denoising procedures, and report potential correlations between motion and key demographic or clinical variables.

As neuroimaging continues to advance our understanding of brain development and pathology, rigorous attention to motion artifact remains essential for producing valid, reproducible, and inclusive science.

Implementing the ABCD-BIDS Pipeline: Core Denoising Components and Commands

The ABCD-BIDS pipeline represents a sophisticated neuroimaging preprocessing framework that integrates the robust anatomical processing of the Human Connectome Project (HCP) minimal preprocessing pipeline with specialized functional magnetic resonance imaging (fMRI) denoising tools from DCAN Labs. This integrated architecture is specifically designed to handle large-scale datasets like the Adolescent Brain Cognitive Development (ABCD) Study, which includes multimodal MRI data from over 11,000 participants [28] [10]. The pipeline is distributed as a BIDS Application, requiring only data organized according to the Brain Imaging Data Structure (BIDS) specification with minimal user configuration [5] [19]. This architectural approach prioritizes reproducibility, standardization, and motion artifact mitigation—critical factors in developmental neuroimaging where head motion correlates with many traits of interest and can introduce spurious brain-behavior associations [28].

Within the context of motion denoising research, the ABCD-BIDS pipeline implements a multi-stage strategy to address the confounding effects of in-scanner head motion. Even with state-of-the-art denoising, residual motion artifacts persist and can significantly impact functional connectivity estimates. Recent research utilizing the Split Half Analysis of Motion Associated Networks (SHAMAN) method on ABCD data revealed that after standard denoising with ABCD-BIDS without motion censoring, 42% (19/45) of traits showed significant motion overestimation scores, while 38% (17/45) showed significant underestimation scores [28]. This underscores the critical importance of the pipeline's architectural decisions in minimizing motion-related false positives in brain-wide association studies.

Pipeline Architecture: Stage-by-Stage Breakdown

The ABCD-BIDS pipeline processes neuroimaging data through a sequential, modular architecture that transforms raw structural and functional inputs into fully processed outputs in both volume and surface spaces. The table below summarizes the core processing stages:

Table 1: ABCD-BIDS Pipeline Processing Stages

Stage Number	Stage Name	Primary Function	Key Tools/Algorithms	Outputs
1	PreFreeSurfer	Remove distortions from anatomical data, align and extract brain in native volume space	ANTs (denoising, N4 bias field correction)	Distortion-corrected structural images, initial brain extraction
2	FreeSurfer	Brain segmentation, cortical surface reconstruction	FreeSurfer	Surface models, anatomical parcellations
3	PostFreeSurfer	Generate CIFTI files, apply surface registration to Conte-69 template	ANTs (registration)	CIFTI surface files, atlas registration
4	FMRIVolume	Functional data distortion correction, motion realignment	FSL (topup, FLIRT)	Motion-corrected BOLD volumes, motion parameter files
5	FMRISurface	Map volume time series to CIFTI grayordinates space	HCP Workbench	CIFTI timeseries data
6	DCANBOLDProcessing	Nuisance regression, motion censoring, bandpass filtering	DCAN Labs algorithms	Fully denoised BOLD timeseries, Framewise Displacement (FD) calculations
7	ExecutiveSummary	Generate visual quality control reports	HTML, BrainSprite viewer	Quality assessment reports

Structural Processing Stages (1-3)

The initial pipeline stages establish the structural foundation for functional analysis. PreFreeSurfer addresses gradient nonlinearity distortions and intensity inhomogeneity using Advanced Normalization Tools (ANTs), providing superior performance for data from General Electric and Philips scanners that often exhibit more noise [5]. A significant adaptation from the original HCP pipeline is the deferral of Montreal Neurological Institute (MNI) standard space registration to the PostFreeSurfer stage, allowing utilization of the refined brain mask created during FreeSurfer processing [5].

The FreeSurfer stage remains largely unchanged from the HCP implementation, performing automated segmentation of brain structures and reconstruction of white and pial cortical surfaces [5]. PostFreeSurfer then generates CIFTI surface files and registers data to the Conte-69 surface template using ANTs' diffeomorphic symmetric image normalization method, which demonstrates superior performance compared to FSL's FNIRT-based registration [5].

Functional Processing Stages (4-6)

The FMRIVolume stage initiates functional processing with distortion correction using FSL's topup, followed by rigid-body motion correction via FSL FLIRT with the initial frame as reference [5] [24]. This stage produces comprehensive motion parameter files including translations, rotations, and their derivatives. The FMRISurface stage projects volume-based timeseries data to the standard CIFTI grayordinates space, enabling combined surface-volume analysis [5].

The DCANBOLDProcessing stage implements critical denoising procedures including:

• General Linear Model (GLM) denoising with regressors for white matter, CSF, and global signal
• Motion parameter regression including Volterra expansion
• Band-pass filtering (0.008-0.09 Hz) using a 2nd order Butterworth filter
• Motion censoring based on framewise displacement (FD) thresholds [5] [29]

A key innovation in this stage is the respiratory motion filter that removes respiratory artifacts (18.582-25.726 breaths per minute) from motion realignment data, addressing a specific artifact in multi-band sequences that can corrupt motion estimates [5] [24].

Quality Assurance and Output Stages (7)

The ExecutiveSummary stage generates comprehensive HTML visual quality control reports containing BrainSprite viewers for anatomical segmentation, atlas registration overlays, and visualizations of movement and grayordinate timeseries before and after regression [5]. This implements the "glass box" philosophy essential for verifying processing quality and identifying outliers.

Motion Denoising: Quantitative Assessment and Protocols

The ABCD-BIDS pipeline's effectiveness in motion denoising has been rigorously quantified through large-scale validation studies. The following table summarizes key performance metrics:

Table 2: Motion Denoising Effectiveness in ABCD-BIDS Pipeline

Processing Stage	Variance Explained by Motion	Relative Reduction vs. Minimal Processing	Key Denoising Components
Minimal Processing (motion correction only)	73% of signal variance	Baseline	Frame realignment
ABCD-BIDS Standard Denoising	23% of signal variance	69% reduction	Global signal regression, respiratory filtering, motion timeseries regression, despiking
ABCD-BIDS + Motion Censoring (FD < 0.2 mm)	Further reduction in overestimation artifacts	42% to 2% of traits with significant overestimation	Framewise displacement thresholding, outlier detection

Motion Impact Assessment Protocol

The SHAMAN (Split Half Analysis of Motion Associated Networks) methodology provides a robust framework for quantifying motion impact on trait-FC relationships [28]:

• Data Requirements: One or more resting-state fMRI scans per participant; extensive demographic, biophysical, and behavioral data (45 traits assessed in ABCD study)

• Split-Half Processing:

  • Divide each participant's fMRI timeseries into high-motion and low-motion halves
  • Calculate correlation structure differences between halves
  • Permutation testing with non-parametric combining across connections

• Impact Score Calculation:

  • Motion Overestimation Score: Motion impact direction aligned with trait-FC effect direction
  • Motion Underestimation Score: Motion impact direction opposite to trait-FC effect direction

• Validation: Application to n=7,270 participants from ABCD Study assessing 45 behavioral and cognitive traits

• Censoring Optimization: Evaluation of different FD thresholds (0.1-0.3mm) for balancing false positive reduction and sample retention

This protocol revealed that censoring at FD < 0.2mm reduced significant overestimation from 42% to 2% of traits but did not decrease the number of traits with significant motion underestimation scores, highlighting the complex relationship between motion correction and effect estimation [28].

Visualization: End-to-End Pipeline Architecture

The pipeline processes structural and functional data in parallel pathways before integrating them for surface-based functional analysis. The structural pathway (yellow) transforms raw anatomical images into segmented surfaces and atlases, while the functional pathway (green) processes BOLD data through increasingly sophisticated denoising stages. Convergence occurs when functional data is mapped to the structural surface models, enabling comprehensive denoising in DCANBOLDProcessing before final quality assessment and output generation.

The Researcher's Toolkit: Essential Components

Table 3: Research Reagent Solutions for ABCD-BIDS Pipeline Implementation

Component Category	Specific Tool/Resource	Function in Pipeline	Implementation Details
Containerization	Docker/Singularity	Reproducible execution environment	Pre-built images available at DCAN Docker Hub [5]
Structural Processing	FreeSurfer 5.3.0-HCP	Cortical surface reconstruction	Requires license acquisition [19]
Functional Processing	FSL (topup, FLIRT)	Distortion correction, motion realignment	Integrated with HCP minimal preprocessing [5]
Registration & Normalization	ANTs	Improved nonlinear registration	Replaces FSL FNIRT for superior performance [5]
Motion Denoising	DCAN BOLD Processing	Nuisance regression, motion censoring	Respiratory motion filtering, FD calculation [29]
Quality Assessment	ExecutiveSummary	Visual quality control	HTML reports with BrainSprite viewer [5]
Data Management	CustomClean	Output size optimization	Removes non-critical files to reduce storage footprint [5]
Diffusion Processing	QSIPrep	Diffusion MRI preprocessing	Alternative pipeline for DWI data [24]

Implementation Protocols

Minimum System Requirements:

• Computational Resources: Minimum 4 cores, 12GB RAM (3GB per core recommended)
• Storage: Substantial scratch space for intermediate processing files
• Time: 24+ hours per subject on single core [19]

Example Execution Command (Docker):

Critical Parameters for Motion Denoising:

• Bandstop Filter: 18.582-25.726 breaths per minute (ABCD participant interquartile range)
• FD Threshold: 0.3mm for default censoring, with masks generated from 0-0.5mm in 0.01mm steps
• Respiratory Filtering: Notch or lowpass filter applied to motion regressors [29] [19]

Discussion and Research Implications

The integrated architecture of the ABCD-BIDS pipeline represents a significant advancement in addressing the critical challenge of head motion in large-scale neurodevelopmental studies. By combining the rigorous anatomical processing of the HCP minimal preprocessing pipeline with specialized denoising tools from DCAN Labs, this framework provides a comprehensive solution for mitigating motion artifacts that disproportionately affect clinical and developmental populations [28] [10].

The persistent finding that residual motion impacts trait-FC relationships even after sophisticated denoising underscores the importance of the pipeline's modular architecture. Researchers can implement additional motion censoring strategies based on framewise displacement thresholds, with the optimal threshold (FD < 0.2mm) empirically shown to reduce motion overestimation artifacts from 42% to 2% of traits [28]. However, the limited effect on motion underestimation scores highlights the need for continued methodological refinement.

The ABCD-BIDS pipeline's adherence to BIDS standards and implementation as a containerized application ensures reproducibility and facilitates widespread adoption across research laboratories. This standardization is particularly valuable for multi-site studies like ABCD, where scanner and acquisition protocol differences can introduce additional variance. The pipeline's continuous development within the NMIND framework further ensures rigorous peer review and adherence to evolving best practices in neuroimaging preprocessing [24] [10].

For researchers investigating brain-behavior relationships in motion-prone populations, the ABCD-BIDS pipeline provides an essential foundation for distinguishing genuine neurobiological associations from motion-induced artifacts. The integration of assessment tools like SHAMAN further enables quantification of motion impact on specific trait-FC relationships, creating a more comprehensive approach to one of the most persistent challenges in functional connectivity research.

Within the framework of the ABCD-BIDS preprocessing pipeline, denoising is a critical step to mitigate the influence of non-neural signals on resting-state functional Magnetic Resonance Imaging (fMRI) data. Head motion represents the largest source of artifact, systematically altering functional connectivity (FC) measures by, for example, decreasing long-distance connectivity and increasing short-range connectivity [28]. Failure to adequately address these artifacts can lead to spurious brain-behavior associations, a particular concern when studying traits often linked to greater motion, such as certain psychiatric disorders [28]. This Application Note details the protocols and quantitative outcomes for three core denoising components within the ABCD-BIDS context: Global Signal Regression (GSR), Motion Parameter Regression, and Despiking. The integration of these methods is designed to reduce spurious findings while preserving the integrity of neural signals.

Core Denoising Components: Protocols and Applications

Global Signal Regression (GSR)

Global Signal Regression (GSR) remains a contentious but widely used denoising step. It involves regressing out the global signal—the average time course across all voxels within the brain—from each voxel's time series.

• Rationale: The global signal is a "catch-all" signal composed of contributions from various sources, including low-frequency drifts, motion, physiological noise (cardiac and respiratory cycles), and potentially neural activity [30]. GSR aims to remove global-scale fluctuations that are presumed to be non-neural in origin.
• Controversy: The primary debate surrounds the fact that GSR can introduce negative correlations (anti-correlations) in functional connectivity matrices and may remove neural signal of interest alongside noise [30] [31]. However, evidence suggests that systemic physiological fluctuations account for a significantly larger fraction of global signal variance compared to electrophysiological fluctuations, and that GSR effectively reduces artifactual connectivity related to heart rate and breathing while preserving connectivity patterns linked to electrophysiological activity [32].
• ABCD-BIDS Context: The ABCD-BIDS pipeline incorporates GSR as part of its standard denoising strategy [28].

Table 1: Variance Explained in the Global Signal by Different Nuisance Regressor Sets [30]

Nuisance Regressor Set	Average Variance Explained in Global Signal
Low-frequency & Motion Regressors	48%
+ Physiological Noise Regressors	31% (Cumulative 79%)
+ White Matter & CSF Regressors	14% (Cumulative 93%)
Residual Variance	7%

Experimental Protocol: Global Signal Regression

• Input Data: Use minimally preprocessed BOLD time series (after motion correction, slice-timing correction, and potentially spatial smoothing).
• Computation: Calculate the global signal for each time point (volume) as the mean signal across all voxels within a pre-defined brain mask.
• Regression: Include the global signal time course as an additional column in the design matrix of nuisance regressors.
• Model Fitting: For each voxel, fit a general linear model (GLM) with the global signal and other nuisance regressors (e.g., motion parameters, white matter signal). The residuals from this regression constitute the GSR-processed BOLD data.
• Output: The denoised voxel-wise BOLD time series, now with the variance associated with the global signal projected out.

Motion Parameter Regression

Motion Parameter Regression addresses the direct confound of head motion by regressing out time series derived from head realignment.

• Rationale: During preprocessing, head motion is estimated for each volume as six parameters (three translations and three rotations). These motion parameters are included as nuisance regressors to remove signal variance that is linearly correlated with head position [28].
• Considerations: This method primarily captures linear relationships between motion and the BOLD signal. Non-linear effects and spin-history artifacts may not be fully removed by this approach alone.

Experimental Protocol: Motion Parameter Regression

• Parameter Extraction: Obtain the six motion parameter time series (x, y, z translations and pitch, roll, yaw rotations) from the volume realignment step during initial preprocessing.
• Expanded Regressors: To model more complex motion-related artifacts, it is common practice to expand the six basic parameters. This includes:
  • The squared parameters for each of the six (total 12).
  • The temporal derivatives of the original six and the squared parameters (total 24).
  • The mean signal from white matter and cerebrospinal fluid (CSF) compartments [33].
• Regression: Include the full set of expanded motion parameters (e.g., 24- or 36-parameter model) in the nuisance GLM.
• Output: The residuals of the GLM, representing the BOLD data with motion-related variance removed.

Despiking

Despiking is a procedure to identify and mitigate the impact of extreme motion outliers, or "spikes," in the BOLD signal.

• Rationale: Sudden, large head movements can cause signal disruptions that are not fully corrected by motion parameter regression. These high-motion frames can disproportionately influence connectivity estimates [28].
• Implementation: The ABCD-BIDS pipeline, and post-processing tools like XCP-D, use framewise displacement (FD) as a scalar measure of volume-to-volume head movement. Volumes with FD exceeding a specified threshold (e.g., 0.2 mm or 0.3 mm) are flagged as outliers [28] [33].

Table 2: Impact of Motion Censoring (Despiking) on Trait-FC Associations in ABCD Data [28]

Condition	Traits with Significant Motion Overestimation	Traits with Significant Motion Underestimation
After ABCD-BIDS Denoising (No Censoring)	42% (19/45 traits)	38% (17/45 traits)
After Censoring at FD < 0.2 mm	2% (1/45 traits)	38% (17/45 traits)

Experimental Protocol: Despiking and Censoring

• Calculate Framewise Displacement (FD): Compute FD for each volume from the six motion parameters [28].
• Identify Outliers: Flag volumes where FD exceeds a chosen threshold (e.g., 0.2 mm). Additional criteria, such as differential standardized signal (DVARS) thresholds or identifying volumes 2 standard deviations from the mean, can be applied [10].
• Censoring (Volume Removal): Remove the flagged high-motion volumes from subsequent connectivity analysis. This is the most direct despiking method.
• Interpolation (Optional): In some pipelines, like XCP-D, the censored volumes can be replaced via interpolation (e.g., using linear or spectral methods) to maintain a continuous time series for certain analyses [33].

Integrated Workflow and Research Toolkit

The denoising components are not applied in isolation but as part of a coordinated sequence within a larger processing workflow. The following diagram illustrates the logical relationship and position of these key denoising steps within a standardized pipeline like ABCD-BIDS or its common post-processing extensions.

Diagram 1: Denoising steps in the ABCD-BIDS pipeline.

Table 3: Key Software and Data Resources for fMRI Denoising Research

Resource Name	Type	Primary Function / Description	Relevance to Denoising
ABCD-BIDS Pipeline	Pre-processing Pipeline	The standardized pipeline for processing ABCD Study data, incorporating GSR, motion regression, and despiking [28].	Provides the foundational implementation of the denoising components discussed.
XCP-D	Post-processing Pipeline	A robust BIDS-App that consumes preprocessed data (e.g., from fMRIPrep, ABCD-BIDS) for further denoising and derivative generation [33] [34].	Applies top-performing denoising strategies, including censoring and confound regression, in a flexible and standardized manner.
fMRIPrep / fMRIPrep Lifespan	Pre-processing Pipeline	A robust, standardized pipeline for fMRI preprocessing that produces data ready for post-processing tools like XCP-D [35].	Generates high-quality, preprocessed data and confound files that are essential inputs for specialized denoising.
ABCD-BIDS Community Collection (ABCC)	Data Resource	A rigorously curated, BIDS-standardized dataset derived from the ABCD Study, including preprocessed derivatives [10].	Provides ready-to-use data for developing, testing, and validating denoising methods on a large scale.
Framewise Displacement (FD)	Quality Metric	A scalar summary of volume-to-volume head motion [28].	The key metric for identifying motion-contaminated volumes for despiking/censoring.
SHAMAN	Analytical Method	A method (Split Half Analysis of Motion Associated Networks) to compute a trait-specific motion impact score for brain-behavior associations [28].	A tool for post-denoising validation to assess if residual motion is biasing specific trait-FC relationships.

Respiratory motion poses a significant challenge in magnetic resonance imaging (MRI), particularly in fast acquisition sequences such as multiband functional MRI (fMRI). During data acquisition, chest movement from breathing generates magnetic field (B0) fluctuations that corrupt head motion estimates, creating factitious head motion not associated with true movement or reductions in fMRI data quality [36]. These respiratory artifacts are especially prominent in multiband sequences with high temporal resolution (short repetition time, or TR), which have become standard in large-scale studies like the Adolescent Brain and Cognitive Development (ABCD) Study and the Human Connectome Project (HCP) [36].

The physiological basis of this artifact stems from the physics of echo-planar imaging (EPI). On a 3T scanner, a magnetic field perturbation of approximately 0.07 parts per million can generate a 1 mm shift of the reconstructed image in the phase encoding direction [36]. Respiratory movements create precisely such perturbations, which are then misinterpreted as head motion during rigid body motion correction algorithms. These factitious motion estimates can subsequently trigger inappropriate motion censoring during processing, potentially leading to the loss of valuable data and introducing biases in functional connectivity analyses [5].

The band-stop filter represents a targeted solution to this specific problem. Rather than attempting to eliminate all respiratory influences from the MRI signal—which would be impractical—this approach focuses on removing respiration-frequency contamination specifically from the head motion estimates themselves [36]. This precise intervention preserves true head motion information while eliminating the respiratory artifact, resulting in more accurate framewise displacement (FD) calculations and improved downstream processing outcomes.

Filter Parameters and Specifications

The respiratory band-stop filter implemented in the ABCD-BIDS pipeline is precisely defined by frequency boundaries derived from population-level respiratory characteristics. These parameters were empirically determined to cover the interquartile range (25th to 75th percentile) of respiratory rates for the ABCD study participant demographic [24] [19].

Table 1: Band-Stop Filter Specifications for Respiratory Motion Filtering

Parameter	Value	Unit	Interpretation
Lower Bound	18.582	breaths per minute (bpm)	25th percentile of participant respiratory rate
Upper Bound	25.726	breaths per minute (bpm)	75th percentile of participant respiratory rate
Frequency Range	0.310–0.429	Hz	Equivalent frequency range
Filter Type	Band-stop (Notch)		Attenuates frequencies within specified range
Primary Application	Multiband fMRI with TR < 1s		Especially effective for fast acquisition protocols

The filter functions by scanning the motion parameter timeseries for frequency components falling within the specified 18.582–25.726 bpm range (0.310–0.429 Hz) and attenuating these components [36] [5]. This process effectively removes the respiratory contamination while preserving both lower-frequency (e.g., true head motion) and higher-frequency components of the motion signal. For context, the upper bound of 25.726 bpm cannot exceed the Nyquist folding frequency, which is determined by the acquisition TR (0.5 × (60 / TR)) [19].

Implementation Protocols

Integration within ABCD-BIDS Pipeline

The respiratory motion filter is implemented during the DCANBOLDProcessing (DBP) stage of the ABCD-BIDS pipeline [5]. This strategic placement ensures that the motion regressors used for censoring have been corrected for respiratory artifacts prior to determining which frames exceed the FD threshold.

Table 2: Implementation Methods for Respiratory Band-Stop Filter

Method	Usage Context	Command Specification	Key Considerations
Default Pipeline	Standard ABCD data processing	Automatically applied with default parameters (18.582–25.726 bpm)	No user intervention required; uses population-based parameters
Manual Specification	Custom studies or atypical populations	--bandstop 18.582 25.726	Allows adjustment for different demographic groups
Real-time Application	FIRMM (Real-time monitoring)	Integrated into real-time motion tracking	Enables immediate quality assessment during scanning

The following workflow diagram illustrates how the respiratory filter integrates within the broader DCANBOLDProcessing stage:

Command Line Implementation

For researchers implementing the filter in their own processing workflows, the ABCD-BIDS pipeline provides explicit options for respiratory motion filtering:

This command executes the full ABCD-HCP pipeline with the respiratory band-stop filter applied to the motion regressors prior to FD calculation and motion censoring [19]. The --ncpus 4 flag accelerates processing by utilizing multiple cores, which is particularly beneficial when processing multiple fMRI runs in parallel.

The Scientist's Toolkit

Table 3: Essential Research Reagents and Solutions for Implementation

Item	Function/Description	Implementation Role
ABCD-BIDS Pipeline	BIDS App built on HCP minimal preprocessing pipelines with DCAN Labs fMRI tools	Primary software environment for implementing the respiratory filter [5]
Siemens Prisma Scanner	3.0 T MRI system with 32-channel head coil	Acquisition platform used in ABCD study; multiband sequences prone to respiratory artifacts [36]
Respiratory Monitoring Belt	Physiologic monitor to track respiratory patterns	Used to validate respiratory rates and filter efficacy [36]
FIRMM Software	Framewise Integrated Real-Time MRI Monitoring	Enables real-time application of respiratory filtering during scanning [36]
DCANBOLDProcessing	Signal processing software for nuisance regression and motion censoring	Pipeline stage where respiratory filter is implemented [5]

Integration with Broader Denoising Strategy

The respiratory band-stop filter functions as one component within a comprehensive motion denoising strategy in the ABCD-BIDS pipeline. It is crucial to recognize that this filter specifically addresses only the respiratory contamination of motion estimates, not other motion-related artifacts in the BOLD signal itself.

The complete denoising approach incorporates multiple complementary techniques:

• Global Signal Regression (GSR): Effectively reduces motion effects on BOLD signals and eliminates known batch effects impacting group comparisons [5]
• Motion Parameter Regression: Removes variance associated with the estimated head motion parameters [28]
• Motion Censoring (Flagging): Excludes high-motion frames exceeding specific FD thresholds (typically 0.3 mm) from analysis [5]
• Spectral Filtering: Applies temporal band-pass filtering (0.008-0.09 Hz) to retain frequencies of interest while reducing high-frequency noise [5]

The following diagram illustrates how respiratory filtering integrates with these other denoising components:

This integrated approach addresses the multifaceted nature of motion artifacts: the band-stop filter ensures accurate motion estimation, while subsequent processing steps address motion-induced artifacts in the BOLD signal itself.

Validation and Efficacy Assessment

The implementation of the respiratory band-stop filter within the ABCD-BIDS pipeline represents a significant advancement in motion denoising methodology, specifically addressing a previously underappreciated source of artifact in multiband fMRI data. Extensive validation has demonstrated that this approach improves the accuracy of framewise displacement estimates and subsequently enhances the quality of post-processed fMRI data [36].

For researchers working with the ABCD dataset or similar multiband fMRI data, implementing this specific band-stop filter (18.582-25.726 bpm) provides a crucial preprocessing step that reduces inappropriate motion censoring and improves the fidelity of motion estimates. The filter's integration within the established ABCD-BIDS pipeline ensures accessibility and reproducibility, while the option for manual parameter adjustment maintains flexibility for studies involving populations with different respiratory characteristics.

As neuroimaging continues to advance toward higher temporal resolutions and larger-scale datasets, specialized techniques such as respiratory motion filtering will play an increasingly important role in ensuring data quality and analytical validity. The successful implementation of this approach in the ABCD study serves as a model for addressing sequence-specific artifacts through targeted, physiologically-informed solutions.

The ABCD-BIDS pipeline is a standardized framework for processing neuroimaging data from the Adolescent Brain Cognitive Development (ABCD) Study, leveraging the Brain Imaging Data Structure (BIDS) to ensure reproducibility and facilitate large-scale analysis [10]. This open-source pipeline is built upon the Human Connectome Project's (HCP) minimal processing pipelines and is distributed as a Docker or Singularity container, ensuring a consistent computational environment across different computing platforms [19].

Deploying this pipeline requires attention to computational resources. It is recommended to use at least 4 CPU cores and allow for at least 12 GB of memory (approximately 3 GB per core) to ensure efficient operation. The pipeline can process multiple fMRI runs in parallel, so using a number of cores that evenly divides the number of runs is optimal for performance. All intermediate processing occurs in the designated output folder, which must have sufficient disk space and read/write performance [19].

Research Reagent Solutions

Table 1: Essential Software Tools and Their Functions in the ABCD-BIDS Pipeline

Tool Name	Function	Key Relevance to Motion Denoising
ABCD-BIDS Pipeline [10] [19]	End-to-end BIDS application for fMRI processing.	Incorporates denoising algorithms like global signal regression, motion timeseries regression, and despiking.
FreeSurfer [19]	Automated cortical surface reconstruction and anatomical segmentation.	Provides anatomical context for functional data; a license is required.
FSL [19]	FMRIB Software Library for MRI data analysis.	Used for tools like TOPUP for susceptibility distortion correction.
ANTs [19]	Advanced Normalization Tools for image registration.	Enables accurate registration to template spaces, improving group analysis.
SHAMAN [28]	Split Half Analysis of Motion Associated Networks.	A novel method for assigning a motion impact score to specific trait-FC relationships.

Containerized Implementation Commands

Docker Execution

A minimal run command to execute the ABCD-BIDS pipeline using Docker is as follows:

Singularity Execution

For High-Performance Computing (HPC) environments where Singularity is preferred, the equivalent command is:

The --rm flag in Docker and the env -i command in Singularity help to ensure a clean runtime environment. If you encounter file permission issues with the outputs, a Docker workaround is to add the flag --user "$(id -u):$(id -g)" to the docker run command [19].

Critical Parameter Configuration for Motion Denoising

Configuring the pipeline's parameters is crucial for effective mitigation of motion artifacts, which is a primary source of spurious brain-behavior associations [28].

Framewise Displacement Censoring

Framewise displacement (FD) quantifies head motion between consecutive volumes. Motion censoring, the practice of excluding high-motion frames from analysis, is a key post-processing step. Evidence from the ABCD study shows that without censoring, a significant proportion of traits (42%) exhibited motion-related overestimation of functional connectivity effects. Censoring at a threshold of FD < 0.2 mm was highly effective, reducing significant overestimation to just 2% of traits [28]. This threshold should be explicitly set in the downstream connectivity analysis scripts that use the pipeline's outputs.

Band-Stop Filtering for Respiratory Artifacts

For data acquired with a fast repetition time (TR < 1 second), it is highly recommended to employ a band-stop filter on the motion regressors to mitigate artifactually high motion estimates caused by respiration [19]. The filter boundaries should match the inter-quartile range (25th to 75th percentile) for the participant group's respiratory rate in breaths per minute (bpm).

In this example, --bandstop 18.582 25.726 applies a filter to remove frequencies between 18.582 and 25.726 bpm. The upper limit must not exceed the Nyquist frequency, calculated as 0.5 * (60 / TR) [19].

Distortion Correction Methods

The pipeline supports different methods for susceptibility distortion correction, which can be specified with the --dcmethod flag. The available options are TOPUP (for spin-echo field maps with reverse phase encoding), FIELDMAP, and NONE. If no method is specified, the pipeline will attempt to auto-detect the appropriate method based on the contents of the BIDS fmap/ directory [19]. Proper configuration of the IntendedFor metadata in the fieldmap JSON files is critical for this auto-detection to work correctly [37].

Table 2: Quantitative Impact of Denoising Strategies on Motion Artifact (ABCD Study Data)

Processing Strategy	% of Signal Variance Explained by Motion	% of Traits with Significant Motion Overestimation	% of Traits with Significant Motion Underestimation
Minimal Processing (Motion-Correction Only)	73% [28]	Not Reported	Not Reported
ABCD-BIDS Denoising (Standard)	23% [28]	42% (19/45 traits) [28]	38% (17/45 traits) [28]
Standard Denoising + Censoring (FD < 0.2 mm)	Not Reported	2% (1/45 traits) [28]	38% (17/45 traits) [28]

Experimental Protocol for Motion Denoising Research

This protocol outlines the steps for a study investigating the efficacy of the ABCD-BIDS pipeline and different denoising parameters in mitigating motion artifacts.

Data Acquisition and Inputs

• Dataset: Utilize the ABCD-BIDS Community Collection (ABCC), a rigorously curated MRI dataset from the ABCD Study [10]. For Release 3.0.0, this includes data from over 11,750 participants at baseline.
• Input Data: The pipeline requires BIDS-formatted input data. The INPUTS directory must contain T1-weighted anatomical images (T1w.nii.gz) and distorted BOLD images (BOLD_d.nii.gz). The participants.tsv file should be checked for subject inclusion and key demographics [10] [38].

Experimental Groups and Processing

• Group 1: Process data through the standard ABCD-BIDS pipeline without additional motion censoring.
• Group 2: Process data with the standard pipeline, then apply motion censoring at FD < 0.2 mm during functional connectivity matrix generation.
• Group 3: Process data using the standard pipeline with the --bandstop filter applied, tailored to the cohort's respiratory rate.

Outcome Measures and Analysis

• Primary Outcome: Motion Impact Score. Calculate for each trait-FC relationship using the SHAMAN method [28]. SHAMAN distinguishes between motion causing overestimation (a positive motion impact score aligned with the trait-FC effect) or underestimation (a score opposite the trait-FC effect) of the true effect.
• Secondary Outcomes:
  • Framewise Displacement (FD): Average and maximum FD for each participant.
  • Temporal Signal-to-Noise Ratio (tSNR): Measure of data quality [39].
  • Functional Connectivity (FC) Matrices: Compare matrices generated under different denoising conditions.

Workflow Visualization

Container Execution and Denoising Workflow

Motion Impact Assessment with SHAMAN

Integration with Fieldmap-Based Distortion Correction Using FSL's topup

Magnetic field inhomogeneities represent a significant source of artifact in functional magnetic resonance imaging (fMRI) data acquired using echo-planar imaging (EPI) sequences. These inhomogeneities cause geometric distortions and signal losses that degrade data quality and complicate the alignment of functional data with anatomical images [40]. Within the ABCD-BIDS preprocessing pipeline, which is designed for the high-quality analysis of large-scale neuroimaging datasets such as the Adolescent Brain Cognitive Development (ABCD) Study, the correction of these distortions is a critical preprocessing step [24] [10]. FSL's topup tool provides a robust method for estimating and correcting these susceptibility-induced distortions using pairs of images acquired with opposing phase-encoding (PE) directions, a approach that has been integrated into the standardized ABCD-BIDS processing workflow [40] [24]. This application note details the theoretical basis, practical implementation, and experimental validation of topup within the context of the ABCD-BIDS pipeline, with particular emphasis on its role in motion denoising research.

The core physical principle underlying topup is the presence of an off-resonance field, which is the difference between the actual magnetic field experienced by the subject and the scanner's nominal field strength. This field, typically measured in Hz, causes spins to precess at frequencies that deviate from expectations. In EPI sequences, which acquire all k-space data following a single excitation, the long time between the acquisition of the first and last echoes makes the signal particularly sensitive to these off-resonance effects [40]. The consequence is a spatial displacement of signal along the phase-encoding direction. The magnitude of this displacement (in voxels) is determined by the ratio of the off-resonance field (in Hz) to the bandwidth per voxel in the PE-direction (in Hz/voxel). Critically, this displacement will be in opposite directions for images acquired with reversed PE gradients, a property that topup exploits to estimate the underlying off-resonance field [40].

topup Integration in the ABCD-BIDS Pipeline

Implementation and Workflow

The ABCD-BIDS pipeline incorporates topup as its primary method for fieldmap-based distortion correction [24]. The pipeline is designed to automatically detect and utilize a pair of spin-echo EPI images with opposed phase-encoding directions (typically posterior-anterior (PA) and anterior-posterior (AP)) from the BIDS-structured fmap directory [19]. The operational logic is that a single optimal pair of these images is used to correct for distortions in the phase encoding direction of all anatomical and functional acquisitions within a session [24]. This centralized correction ensures consistency across modalities.

A key feature of topup that makes it suitable for use in large-scale studies like ABCD is its integrated movement model. Unlike other reverse PE-direction methods, topup can simultaneously estimate the off-resonance field and any rigid-body movement that occurred between the two acquisitions. Without this model, any subject movement would be misinterpreted as part of the off-resonance field, leading to inaccurate estimation and correction [41]. The pipeline executes topup in a single command that handles both the estimation of the distortion field and its application to the functional images.

The following diagram illustrates the complete integration path of topup within the broader ABCD-BIDS preprocessing workflow, from data input to the generation of distortion-corrected functional outputs:

Acquisition Parameters and Configuration

Successful execution of topup requires precise specification of acquisition parameters in a text file (typically named acqp.txt). This file defines the phase-encoding direction and total readout time for each input image. The ABCD-BIDS pipeline automates the generation of this file based on metadata extracted from the BIDS JSON sidecar files, which are populated during the DICOM to NIfTI conversion process [24] [41].

Table 1: Phase-Encoding Vectors for --acqp File

Phase Encoding Direction (in FSLeyes)	Phase Encoding (Siemens Protocol)	--acqp File Vector	Example Calculation (Echo Spacing=0.75ms, EPI Factor=128)
P >> A	Phase enc. dir P >> A	0 1 0 0.095	Total Readout Time = 0.75 ms * 128 = 0.096 s ≈ 0.095
A >> P	Phase enc. dir A >> P	0 -1 0 0.095	Total Readout Time = 0.75 ms * 128 = 0.096 s ≈ 0.095
R >> L	Phase enc. dir R >> L	1 0 0 0.122	Total Readout Time = 0.96 ms * 128 = 0.123 s ≈ 0.122
L >> R	Phase enc. dir L >> R	-1 0 0 0.122	Total Readout Time = 0.96 ms * 128 = 0.123 s ≈ 0.122

The total readout time (ROT) is a critical parameter that can be directly obtained from the TotalReadoutTime field in the BIDS JSON sidecar file generated by dcm2niix [41]. If this field is unavailable, ROT can be calculated from the echo spacing ((SE)) and the EPI factor ((F{EPI})) using the formula: ( ROT = SE \times F{EPI} ). It is important to note that while an accurate ROT is necessary to express the distortion field in Hz, topup primarily estimates a displacement field. If the output of topup is to be used exclusively with applytopup or eddy (as in the ABCD-BIDS pipeline), any error in the ROT specification will effectively cancel out during the application of the correction, and the final distortion correction will remain accurate [41].

Experimental Protocols and Validation

Quantitative Performance Metrics

The efficacy of topup integration within the ABCD-BIDS pipeline has been evaluated through large-scale validation efforts. A recent study analyzing data from 7,270 participants in the ABCD Study quantified the pervasive impact of head motion, even after comprehensive denoising [28]. The findings underscore the critical importance of robust distortion correction as a foundational step for mitigating motion-related artifacts.

Table 2: Validation of Motion Denoising in ABCD-BIDS Pipeline (n=7,270)

Processing Stage	Signal Variance Explained by Motion (FD)	Reduction vs. Minimal Processing	Traits with Significant Motion Overestimation	Traits with Significant Motion Underestimation
Minimal Processing	73%	-	Not Assessed	Not Assessed
After ABCD-BIDS Denoising	23%	69% Relative Reduction	42% (19/45 traits)	38% (17/45 traits)
+ Motion Censoring (FD < 0.2 mm)	Further Reduced	-	2% (1/45 traits)	38% (17/45 traits)

The data reveals that while the ABCD-BIDS denoising, which includes topup distortion correction, achieves a substantial 69% relative reduction in motion-related variance, significant residual motion artifacts persist, affecting a large proportion of trait-FC relationships [28]. This highlights that while topup is a powerful tool for correcting geometric distortions, it is part of a larger arsenal needed to combat motion artifacts.

Protocol for topup Execution

The following protocol details the steps for running topup both as a standalone tool and as an integrated component within the full ABCD-BIDS pipeline.

Standalone topup Protocol:

• Data Preparation: Ensure your data includes at least two spin-echo EPI images (e.g., b0 images from a diffusion sequence or dedicated fieldmaps) acquired with opposite phase-encoding directions.
• Parameter File Creation: Create a datain file (e.g., acqp.txt) specifying the phase-encoding vectors and total readout time for each volume, as detailed in Table 1.
• Command Execution: Run topup using a standard command structure:

• Application of Correction: Apply the calculated field to your target EPI data using applytopup:

Full ABCD-BIDS Pipeline Protocol:

• BIDS Validation: Ensure your dataset is BIDS-validated and includes the necessary fmap directory with correctly labeled AP and PA SE-EPI images.
• Containerized Execution: Run the pipeline using Docker or Singularity. A minimal Docker command is:

• Advanced Execution (Recommended): For optimal results, especially with multiband data, include a bandstop filter for respiratory motion and specify multiple CPU cores.
  The bandstop filter (18.582 25.726 bpm, the interquartile range for the ABCD cohort) mitigates artifactually high motion estimates by filtering respiratory frequencies from the motion regressors prior to framewise displacement calculation [24] [19].

The Scientist's Toolkit

Table 3: Essential Research Reagents and Software Solutions

Tool/Reagent	Function in Distortion Correction & Motion Denoising
FSL (FMRIB Software Library)	Provides the core software suite containing topup, applytopup, and eddy for field estimation, correction, and diffusion data processing [40] [42].
ABCD-BIDS Pipeline (abcd-hcp-pipeline)	A BIDS App that integrates HCP-style minimal preprocessing with DCAN Labs tools, automatically handling topup execution for structural and functional data [24] [19].
BIDS Dataset	A standardized neuroimaging data structure that organizes AP/PA spin-echo fieldmaps, functional, and anatomical data, enabling automatic pipeline configuration [24] [10].
dcm2niix	A DICOM to NIfTI converter that generates BIDS-compliant JSON sidecar files containing critical metadata like TotalReadoutTime and PhaseEncodingDirection [41].
Synth	An alternative software package for distortion correction that generates a synthetic reference image from T1w and T2w anatomicals, useful when fieldmap data is missing or corrupted [43].
SHAMAN (Split Half Analysis of Motion Associated Networks)	A novel method for computing a trait-specific motion impact score, used to validate the success of denoising pipelines in removing residual motion artifacts from functional connectivity analyses [28].

The integration of FSL's topup into the ABCD-BIDS pipeline provides a robust, automated, and theoretically sound method for correcting susceptibility-induced distortions in EPI data. Its internal movement model makes it particularly suitable for large-scale studies and populations prone to motion, such as pediatric cohorts [41]. However, validation data confirms that distortion correction is a necessary but insufficient step for the complete removal of motion artifact from final brain-behavior analyses [28]. Consequently, topup is most effective when deployed as part of a comprehensive strategy that includes optimized acquisition (e.g., multi-echo sequences), post-processing denoising (e.g., bandstop filters, global signal regression), and rigorous statistical checks for residual motion bias (e.g., SHAMAN) [28] [24].

Future developments in the NiPreps ecosystem, such as the fMRIPrep Lifespan extension, promise to generalize these robust correction frameworks to a wider age range, from infancy to old age [35]. Furthermore, tools like Synth offer a promising path for accurate distortion correction in legacy datasets or studies where acquiring high-quality fieldmap data is not feasible, ensuring that the high standards of the ABCD-BIDS pipeline can be applied even more broadly [43].

The Brain Imaging Data Structure (BIDS) is a community-driven standard that provides a simple and intuitive way to organize neuroimaging and associated data. By using a consistent framework for file naming, directory structure, and metadata documentation, BIDS aims to make datasets more findable, accessible, interoperable, and reusable (FAIR), thereby lowering scientific waste and improving research efficiency [44]. The standard has been successfully extended to numerous data modalities, including a recent extension for motion data, Motion-BIDS, which is particularly relevant for motion denoising research [45].

In functional MRI (fMRI), subject motion is considered one of the major confounding factors affecting data quality. Even in cooperative subjects, involuntary motion of 1-2 mm and rotations of approximately 1° are not uncommon, while patients, elderly, and children often exhibit more substantial motion [46]. These movements introduce a variety of physical phenomena that cause undesired temporal variations in the voxel signal evolution, potentially leading to reduced statistical significance of activation maps and increased prevalence of false activations [46]. The ABCD-BIDS preprocessing pipeline represents a sophisticated framework for addressing these challenges through rigorous data standardization and processing methodologies.

Motion-BIDS Specification for Data Organization

Core Principles and File Structure

Motion-BIDS is designed to organize motion data recorded alongside human brain imaging data, making it interoperable across laboratories and with other data modalities [45]. The standard accommodates data from diverse motion tracking systems, including optical systems (with or without markers), inertial measurement units (IMUs), electromagnetic systems, and even virtual motion in simulated environments [45].

In accordance with BIDS standards, motion data for each subject is stored in a modality-specific subdirectory. Motion data must be shared in a motion subdirectory within the subject or session folder [45]. The core file structure follows this pattern:

• sub-<label>[_ses-<label>]_task-<label>[_tracksys-<label>][_acq-<label>][_run-<index>]_motion.tsv - Tab-separated values file containing the actual motion data timeseries without a header row.
• sub-<label>[_ses-<label>]_task-<label>[_tracksys-<label>][_acq-<label>][_run-<index>]_motion.json - Sidecar JSON file containing metadata about the task, hardware, and recording parameters.
• sub-<label>[_ses-<label>]_task-<label>[_tracksys-<label>][_acq-<label>][_run-<index>]_channels.tsv - File describing each data channel in the corresponding motion.tsv file.

A tracking system is defined as a group of channels that synchronously sample motion data from one or multiple tracked points, sharing core sampling parameters (sampling rate, duration) and hardware/software properties [45]. Each tracking system must have its own set of files distinguished by the tracksys-<label> entity.

Motion Data Types and Metadata Requirements

Motion-BIDS encompasses various types of motion data, primarily focused on time series of position or orientation samples and their derivatives. The standard specifies several channel types with distinct metadata requirements:

Table 1: Motion Data Channel Types in BIDS Specification

Channel Type	Description	Units	Required Metadata
POS	Position coordinates in up to three spatial axes	meters (m)	TrackedPointID, component (x, y, z)
ORNT	Orientation data as Euler angles or quaternions	radians, degrees, or unitless for quaternions	TrackedPointID, component, rotation_order (for Euler)
JNTANG	Joint angles	radians or degrees	TrackedPointID, component
ACCEL	Acceleration (first derivative of position)	meters per second squared (m/s²)	TrackedPointID, component
GYRO	Angular velocity (first derivative of orientation)	radians per second or degrees per second	TrackedPointID, component
ANGACCEL	Angular acceleration	radians per second squared or degrees per second squared	TrackedPointID, component
LATENCY	Time information relative to recording onset	seconds (s)	-
MISC	Miscellaneous channels not covered by other types	varies	description

The accompanying JSON sidecar file must contain specific metadata fields, categorized as follows:

Table 2: Required and Recommended Metadata Fields in Motion-BIDS

Category	Field	Requirement Level	Description
Task Information	TaskName	REQUIRED	Name of the task (e.g., "walking", "head nodding")
	TaskDescription	RECOMMENDED	Longer description of the task
	Instructions	RECOMMENDED	Text of instructions given to participants
Hardware Information	Manufacturer	RECOMMENDED	Manufacturer of the motion tracking equipment
	ManufacturersModelName	RECOMMENDED	Model name of the equipment
	DeviceSerialNumber	RECOMMENDED	Serial number or pseudonym for the device
	SoftwareVersions	RECOMMENDED	Software version of the equipment
Motion-Specific Fields	SamplingFrequency	REQUIRED	Nominal sampling frequency (in Hz)
	SamplingFrequencyEffective	RECOMMENDED	Effective sampling frequency if different from nominal
	TrackedPointsCount	RECOMMENDED	Number of distinct points being tracked
	MissingValues	RECOMMENDED	How missing values are represented (e.g., "NaN", "0")

For orientation data, the standard allows representation as either Euler angles or quaternions. When using quaternions (particularly relevant for virtual reality applications), they should be distinguished from Euler angles through the component and units columns in the channels.tsv file, with quaternion components typically labeled as x, y, z, and w [47].

Motion Data Processing in the ABCD-HCP Pipeline

The ABCD-HCP BIDS fMRI pipeline represents a comprehensive implementation for processing BIDS-formatted MRI datasets, with specific stages dedicated to motion management [5]. This pipeline utilizes methods from both the Human Connectome Project's minimal preprocessing pipeline and DCAN Labs resting state fMRI analysis tools.

Pipeline Architecture and Motion Processing Stages

The ABCD-HCP pipeline processes data through multiple serial stages, with several stages specifically addressing motion correction and denoising:

Stage 4: FMRIVolume initiates the functional processing with motion parameter estimation. Each volume in the time series is aligned using FSL FLIRT rigid-body registration (six degrees of freedom) to the initial frame, which typically has greater anatomical contrast [5]. This registration produces a twelve-column text file containing translations and rotations along each axis and their derivatives. Additionally, the stage applies distortion correction using a pair of spin echo EPI scans with opposite phase encoding directions through FSL's topup tool [5].

Stage 6: DCANBOLDProcessing (DBP) represents the core motion denoising component, implementing sophisticated signal processing for nuisance regression and motion censoring [5]. This stage operates through four broad steps:

• Standard pre-processing: Data is de-meaned and de-trended, followed by denoising using a general linear model with regressors comprising signal variables (mean time series for white matter, CSF, and global signal) and movement variables (translational and rotational measures with Volterra expansion).
• Respiratory motion filter: An optional but recommended step that filters respiratory frequencies (18.582-25.726 breaths per minute) from motion realignment data to produce better estimates of framewise displacement.
• Motion censoring: Identifies "bad" frames exceeding a framewise displacement (FD) threshold of 0.3 mm, removing them during demeaning and detrending while using interpolation to avoid aliasing during band-pass filtering.
• Generation of parcellated time series: Constructs time series for pre-defined atlases (Gordon's 333 ROI, Power's 264 ROI, Yeo's 118 ROI, HCP's 360 ROI) to facilitate correlation matrix construction and time series analysis.

Motion Denoising Methodologies

The DCANBOLDProcessing stage implements several critical methodologies for effective motion denoising:

Global Signal Regression (GSR) has been consistently shown to reduce the effects of motion on BOLD signals and eliminate known batch effects that impact group comparisons [5]. The combination of motion censoring with GSR represents the most effective existing method for eliminating artifacts produced by motion.

Motion censoring follows the procedure established by Power et al. (2014), labeling data as "bad" frames if they exceed an FD threshold of 0.3 mm [5]. These frames are removed during demeaning and detrending, while betas for denoising are calculated using only "good" frames. For band-pass filtering (0.008-0.09 Hz using a 2nd-order Butterworth filter), interpolation replaces "bad" frames, and residuals are extracted from the denoising GLM. This approach uses only quality data while avoiding potential aliasing from missing timepoints.

The respiratory motion filter addresses artifacts in multi-band data, where respiratory artifacts outside the brain can affect estimates of frame alignment, leading to inappropriate motion censoring [5]. By filtering respiratory frequencies from motion realignment data, the pipeline produces more accurate FD estimates.

Experimental Protocols for Motion Data Collection and Processing

Data Acquisition Protocol

For researchers collecting new motion data compliant with Motion-BIDS, the following protocol ensures data quality and standardization:

• System Selection and Validation: Choose motion tracking technology appropriate for your research context (optical, IMU, electromagnetic, or virtual). Validate system precision and accuracy for expected motion ranges.
• Participant Preparation: Apply sensors or markers according to manufacturer specifications. For markerless systems, ensure adequate camera coverage and lighting conditions.
• Synchronization Setup: Establish precise timing synchronization between motion tracking systems and neuroimaging equipment using hardware triggers or software timestamps.
• Recording Parameters: Configure sampling frequency to capture relevant motion dynamics (typically ≥100 Hz for human movement). Set appropriate filtering parameters to avoid aliasing.
• Calibration Procedure: Perform system-specific calibration following manufacturer guidelines. Record calibration data in the session metadata.
• Task Administration: Implement standardized motion tasks with clear instructions. Include resting periods, controlled movements, and task-specific activities relevant to your research questions.
• Data Export: Export data in manufacturer's format (source data) while maintaining all original metadata.

BIDS Conversion Protocol

Converting acquired motion data to Motion-BIDS format involves these critical steps:

• File Organization: Create the BIDS directory structure with dataset_description.json, participants.tsv, and session-specific subdirectories containing motion folders.
• Data Transformation: Convert proprietary data formats to TSV files without headers. Ensure column order matches the corresponding channels.tsv file.
• Metadata Extraction: Populate required and recommended JSON sidecar fields with task, hardware, and motion-specific information.
• Channel Description: Create channels.tsv file specifying name, type, component, units, and tracking point ID for each data channel.
• Synchronization Documentation: Record acquisition times in scans.tsv file for temporal alignment across modalities.
• Data Validation: Run BIDS Validator to ensure compliance and identify any formatting issues.

Table 3: Essential Tools and Resources for BIDS Motion Data Research

Tool/Resource	Type	Purpose	Implementation
BIDS Validator	Quality Assurance Tool	Checks dataset compliance with BIDS standards	Web-based, command-line, or JavaScript library [48] [49]
ABCD-HCP BIDS fMRI Pipeline	Processing Pipeline	Processes BIDS-formatted MRI datasets with motion denoising	Docker container available at DCAN Docker Hub [5]
Motion-BIDS Extension	Specification	Standard for organizing motion data in BIDS	Reference specification for data organization [45] [47]
DCANBOLDProcessing	Software Module	Performs nuisance regression and motion censoring	Part of ABCD-HCP pipeline [5]
BIDS-Apps	Software Ecosystem	Portable pipelines that understand BIDS datasets	Collection of containerized applications [44]
OpenNeuro	Data Repository	Public database for BIDS formatted datasets	Data sharing and discovery platform [44]
FNIRT & FLIRT	Registration Tools	Image registration and motion correction	FSL tools used in ABCD-HCP pipeline [5]
ANTs	Normalization Tools	Advanced Normalization Tools for image registration	Used in PreFreeSurfer stage for denoising and bias correction [5]

For researchers working within the ABCD-BIDS framework, the ABCD-BIDS Community Collection (ABCC) provides a rigorously curated MRI dataset derived from the ABCD Study, leveraging BIDS standards and NMIND-reviewed pipelines [10]. This resource offers ready-to-use MRI raw data and derivatives that have passed quality control, facilitating reproducible research in motion denoising.

When implementing motion denoising protocols, particular attention should be paid to the respiratory motion filtering option in DCANBOLDProcessing, especially when working with multi-band data where respiratory artifacts can affect motion estimates [5]. The motion censoring threshold of 0.3 mm FD represents a validated starting point, though researchers may need to adjust this based on their specific data characteristics and research questions.

The BIDS Validator is available through multiple interfaces to suit different workflows, including a web application that runs entirely in the browser (ensuring data privacy), a command-line version for scripting and automation, and a JavaScript library for integration into custom applications [48] [49]. Regular validation throughout the data preparation process ensures compliance and prevents formatting issues that could disrupt analysis pipelines.

Solving Common ABCD-BIDS Implementation Challenges and Advanced Optimization

Resolving Fieldmap Dimension Mismatches and Assignment Errors

In magnetic resonance imaging (MRI) preprocessing, particularly within the ABCD-BIDS pipeline for motion denoising research, fieldmaps are essential for correcting spatial distortions caused by magnetic field inhomogeneities. However, fieldmap dimension mismatches and incorrect fieldmap assignment introduce significant artifacts that compromise data integrity and complicate the separation of true neural signal from motion-induced noise. The ABCD-BIDS pipeline, which incorporates tools from both the Human Connectome Project's minimal preprocessing pipeline and DCAN Labs resting-state fMRI analysis tools, relies heavily on accurate distortion correction to generate reliable functional connectivity metrics [50]. When fieldmap errors occur, they can systematically bias motion-FC (functional connectivity) effects, potentially leading to both overestimation and underestimation of trait-brain relationships in large-scale studies like the Adolescent Brain Cognitive Development (ABCD) Study [28]. Even after comprehensive denoising with the ABCD-BIDS pipeline, which includes global signal regression, respiratory filtering, and motion timeseries regression, residual motion can still explain approximately 23% of signal variance [28], highlighting the critical need for precise fieldmap handling to prevent additional artifacts in motion denoising research.

Understanding Fieldmap Dimension Mismatches

Common Dimension Mismatch Scenarios

Fieldmap dimension mismatches occur when the acquired fieldmap and the target EPI (Echo Planar Imaging) volumes have differing spatial properties. Based on common preprocessing challenges reported in neuroimaging communities, these mismatches typically manifest in several scenarios:

• Voxel size differences: Fieldmap and EPI volumes have identical matrix sizes (e.g., 64×64×36) but different voxel dimensions (e.g., 3.516×3.516×3mm vs. 3×3×3mm) [51].
• Image dimension discrepancies: The fieldmap and EPI volumes share the same voxel size but have different matrix sizes (e.g., 64×64×36 for EPI vs. 64×64×38 for fieldmap) [51].
• Acquisition parameter variations: Functional runs acquired within the same session but with slightly different parameters (e.g., different TR values: 2.1s vs. 2.0s) [51].

Impact on Distortion Correction Quality

Dimension mismatches can lead to imperfect distortion correction, which manifests as misalignment between functional and anatomical data, residual distortions in regions with magnetic susceptibility gradients (particularly near sinuses and ear canals), and introduced noise that correlates with motion parameters. In severe cases, these errors can cause complete preprocessing failures in the TopupPreprocessingAll.sh stage for GE scanners or during the FMRIVolume step [50]. Visual inspection of the fMRIPrep HTML reports is essential for identifying these issues, as the pipeline may not raise explicit errors despite suboptimal correction [51].

Table: Common Fieldmap Dimension Mismatch Scenarios and Solutions

Mismatch Type	Example	Potential Impact	Recommended Solution
Voxel Dimensions	EPI: 3×3×3mm, FMap: 3.516×3.516×3mm	Misaligned distortion field application	Resize fieldmap using EPI as reference [50]
Image Dimensions	EPI: 64×64×36, FMap: 64×64×38	Incomplete brain coverage in correction	Interpolation during registration [51]
Acquisition Parameters	Different TR between tasks	Inaccurate distortion modeling	Use session-specific fieldmaps when possible [51]

Protocols for Resolving Fieldmap Dimension Mismatches

Fieldmap Resizing Protocol for ABCD-BIDS Pipeline

For dimension mismatches encountered in the ABCD-BIDS pipeline, particularly with GE scanners where spin echo fieldmaps may have different dimensions than the scout image, the following protocol has been successfully implemented:

• Identify mismatched pairs: Check log files for TopupPreprocessingAll.sh errors indicating dimension mismatches [50].
• Resize fieldmap to match BOLD reference: Use FSL's FLIRT to resample the fieldmap to the BOLD reference space with isotropic 2.4mm resolution [50]:

• Update JSON metadata: Ensure the resized fieldmap's JSON sidecar file contains appropriate metadata, including IntendedFor fields pointing to all relevant functional runs [50] [52].
• Rerun preprocessing: Continue with the ABCD-BIDS pipeline from the distortion correction stage.

This approach specifically addresses the "Fieldmap Dimension Mismatch in TopupPreprocessingAll.sh (GE Only)" issue documented in the ABCC processing notes [50].

Interpolation-Based Protocol for General Dimension Mismatches

For more general dimension mismatch scenarios beyond specific ABCD-BIDS errors, an interpolation-based approach can be applied:

• Visual quality assessment: First, inspect the fMRIPrep HTML reports before and after susceptibility distortion correction to determine if the automatic interpolation produces acceptable results [51].
• Validate fieldmap applicability: Check if the transformation matrices in the NIfTI headers result in adequate brain overlap between the fieldmap and EPI data, despite dimension differences [51].
• Proceed with caution: If visual inspection confirms reasonable distortion correction, the interpolated fieldmap can be used, as fieldmaps are fundamentally capable of undergoing interpolation to handle dimension differences [51].
• Consider alternative approaches: For severe mismatches or when visual inspection reveals poor results, consider implementing fieldmap-less distortion correction methods like Synth, as used in the non-human primate ABCD-BIDS pipeline variant [53].

Addressing Fieldmap Assignment Errors

Root Causes of Assignment Errors

Fieldmap assignment errors typically occur when the preprocessing pipeline incorrectly associates fieldmaps with specific functional runs. In the ABCD-BIDS context, this manifests as "Incorrect Fieldmap Assignment in FMRIVolume Step" [50]. Primary causes include:

• Missing IntendedFor metadata: Fieldmap JSON files lack explicit specification of which functional runs they should correct [52].
• Incomplete BIDS validation: The BIDS validator may not catch all missing metadata, leading to preprocessing errors downstream [52].
• Session organization issues: Complex session structures with multiple functional runs can confuse the fieldmap assignment logic [50].

Protocol for Correcting Fieldmap Assignment

The following protocol ensures proper fieldmap assignment in ABCD-BIDS processing:

• Preprocessing JSON validation: Before running the full pipeline, verify that all fieldmap JSON files contain the IntendedFor field with correct paths to target functional runs [50] [52].
• Update missing IntendedFor specifications: Use tools like mne_bids.update_sidecar_json or manually edit JSON files to include appropriate IntendedFor entries [54]:

• Session-specific fieldmap assignment: For datasets with multiple sessions, ensure fieldmaps are properly linked to same-session functional runs [52].
• Run limited preprocessing: Test the pipeline with a subset of data to verify correct fieldmap assignment before processing the full dataset.

Integration with Motion Denoising in ABCD-BIDS Pipeline

Interplay Between Fieldmap Correction and Motion Denoising

Proper fieldmap correction is an essential prerequisite for effective motion denoising in the ABCD-BIDS pipeline. Spatial distortions caused by field inhomogeneities interact non-linearly with motion-induced artifacts, creating complex spatiotemporal noise patterns that challenge standard denoising approaches. The pipeline employs a respiratory motion filter to address a specific respiratory artifact found in multi-band ABCD data, which filters frequencies between 18.582 to 25.726 breaths per minute from the motion realignment data during the DCANBoldProc stage [50]. This produces better estimates of framewise displacement (FD), but its effectiveness depends on accurate distortion correction first being applied.

Quantitative Impact on Motion-Related Variance

Recent research utilizing the ABCD dataset demonstrates that even after comprehensive denoising with ABCD-BIDS (including respiratory filtering, motion timeseries regression, and despiking), head motion still explains approximately 23% of signal variance [28]. This represents a 69% reduction compared to minimal processing alone (where motion explained 73% of variance), but highlights the need for every optimization, including precise fieldmap correction [28]. The strong negative correlation (Spearman ρ = -0.58) between the motion-FC effect matrix and average FC matrix underscores how residual motion artifacts systematically decrease long-distance connectivity [28], an effect that could be exacerbated by imperfect fieldmap correction.

Table: Motion-Related Variance Explained at Different Processing Stages

Processing Stage	Variance Explained by Motion	Relative Reduction	Notes
Minimal Processing	73%	Baseline	Motion-correction by frame realignment only [28]
ABCD-BIDS Full Denoising	23%	69%	Includes respiratory filtering, motion regression, despiking [28]
With Censoring (FD < 0.2mm)	Further reduction	Additional ~20%	Censoring reduces significant motion overestimation to 2% of traits [28]

Visualizing the Integrated Solution Workflow

The following workflow diagram illustrates the comprehensive approach to resolving fieldmap issues within the ABCD-BIDS motion denoising pipeline:

Workflow for Fieldmap Issue Resolution in ABCD-BIDS Motion Denoising

The Researcher's Toolkit: Essential Solutions for Fieldmap Processing

Table: Key Research Reagent Solutions for Fieldmap Processing

Tool/Resource	Function	Application Context
FLIRT (FSL)	Image registration and resizing	Resolving fieldmap dimension mismatches [50]
mnebids.updatesidecar_json	BIDS metadata management	Adding IntendedFor fields to fieldmap JSON files [54]
BIDS Validator	Dataset quality control	Identifying missing required metadata before processing [55]
fMRIPrep HTML Reports	Visual quality assessment	Evaluating distortion correction effectiveness [51]
SHAMAN Method	Motion impact quantification	Assessing residual motion effects after full processing [28]
Synth (DCAN Labs)	Fieldmap-less distortion correction	Alternative when fieldmaps are unavailable or problematic [53]
dcm2bids	DICOM to BIDS conversion	Proper initial organization with fieldmap assignment [52]

Successful fieldmap processing within the ABCD-BIDS motion denoising framework requires a systematic approach that begins with proper BIDS organization and validation. Researchers should prioritize comprehensive BIDS validation before processing, paying special attention to IntendedFor fields in fieldmap JSON files. For dimension mismatches, proactive resizing using established tools like FLIRT provides a robust solution, while visual quality assessment remains essential for verifying distortion correction effectiveness. Most importantly, researchers should recognize fieldmap processing as an integral component of the motion denoising pipeline rather than an independent preprocessing step, as accurate distortion correction significantly impacts downstream motion artifact removal and the validity of functional connectivity analyses. By implementing these protocols, researchers can minimize fieldmap-related artifacts and enhance the reliability of motion denoising in large-scale studies like the ABCD project.

Framewise displacement (FD) is a scalar quantity that summarizes the estimated frame-to-frame head movement of a participant during functional magnetic resonance imaging (fMRI). It is derived from the six rigid-body realignment parameters (translations in X, Y, Z and rotations around X, Y, Z) obtained during image registration, with rotations converted to distance using simplifying assumptions about head shape [56]. In contemporary fMRI preprocessing pipelines, motion censoring (also known as "scrubbing") is a widely adopted strategy to mitigate the detrimental effects of head motion on Blood Oxygen Level Dependent (BOLD) signal analysis. This technique identifies and excludes motion-contaminated volumes from statistical analysis, typically by including scan-nulling regressors in the general linear model [57].

The selection of an appropriate FD threshold represents a critical methodological decision that balances data quality against data retention. Overly stringent thresholds may discard excessive data, reducing statistical power and potentially introducing bias by excluding participants with naturally higher movement (e.g., specific clinical populations or children). Overly lenient thresholds risk retaining motion-contaminated data, which can introduce spurious correlations in functional connectivity analyses and obscure true neural signals [57] [58]. Within the specific context of the ABCD-BIDS preprocessing pipeline—a standardized framework for processing large-scale neuroimaging datasets—the FD < 0.2 mm threshold has been implemented as a default parameter for generating specific derivatives, establishing it as a community standard for balancing these competing demands in population neuroscience research [10].

Quantitative Comparison of FD Thresholds

Table 1: Comparative Analysis of Framewise Displacement Thresholds in fMRI Processing

FD Threshold	Typical Data Loss	Primary Use Cases	Key Advantages	Reported Performance
0.2 mm	Moderate	• ABCD-BIDS pipeline derivatives• Large-scale consortia studies• Developmental populations	• Balanced approach• Community standard in ABCD• Preserves more data than stricter thresholds	• Used for Gordon connectivity matrices in ABCD [10]• Identifies subtle motion in high-quality data
0.3 mm	Low-Moderate	• General task-based fMRI• Initial processing steps• Adult populations	• Retains more data points• Common default in many pipelines• Reduced exclusion rates	• Modest censoring (1-2% data loss) shows consistent improvements [57]• Threshold for "bad" frames in DCAN BOLD Processing standard preprocessing [5]
0.5 mm	Low	• Minimal censoring approaches• Initial quality assessment• High-motion populations	• Maximizes data retention• Minimizes subject exclusion• Useful for observing motion patterns	• Often insufficient for removing motion artifacts [58]• May retain significantly contaminated volumes

Table 2: Alternative Motion Censoring Strategies Beyond FD Thresholding

Method	Basis	Key Features	Considerations
Data-Driven Scrubbing (Projection Scrubbing)	Statistical outlier detection in ICA component space	• Flags only volumes with abnormal patterns• Not dependent on motion parameters alone• Adapts to specific data characteristics	• Dramatically increases retained sample size [58]• Improves functional connectivity identifiability [58]
DVARS	Root mean square of voxel-wise intensity differences	• Based on observed signal changes• Sensitive to abrupt intensity fluctuations	• Can flag non-motion-related artifacts• Requires appropriate threshold setting
Multi-Metric Approaches	Combination of FD and DVARS	• Leverages complementary information• More comprehensive artifact detection	• Increased complexity in implementation• Potential for over-censoring

Implementation in the ABCD-BIDS Pipeline

The ABCD-BIDS pipeline implements a sophisticated, multi-stage processing workflow that incorporates motion censoring at critical junctures. The pipeline utilizes methods from both the Human Connectome Project's minimal preprocessing pipeline and the DCAN Labs resting state fMRI analysis tools to output preprocessed MRI data in both volume and surface spaces [5]. Within this framework, the DCANBOLDProcessing (DBP) stage is particularly relevant for motion censoring implementation.

The DBP stage involves four broad steps: (1) standard pre-processing, (2) optional application of a respiratory motion filter, (3) motion censoring followed by standard re-processing, and (4) construction of parcellated timeseries [5]. During the motion censoring procedure, data are labeled as "bad" frames if they exceed an FD threshold of 0.3 mm for standard pre-processing. However, for the final construction of parcellated timeseries, the pipeline generates temporal masks from 0 ("No censoring") to 0.5 mm FD thresholds in steps of 0.01 mm, allowing researchers to apply the specific FD < 0.2 mm threshold that has been established for ABCD study connectivity matrices [5] [10].

A key innovation in the ABCD-BIDS pipeline is the implementation of a respiratory motion filter, which addresses the specific finding that respiratory artifacts in multi-band data can affect estimates of frame alignment, leading to inappropriate motion censoring [5]. By filtering frequencies corresponding to respiratory rates (18.582 to 25.726 breaths per minute) from the motion realignment data, this filter produces more accurate estimates of FD, thereby refining the censoring process.

Diagram 1: ABCD-BIDS Pipeline Workflow with Integrated Motion Censoring. The DCAN BOLD Processing (DBP) stage contains the critical motion censoring procedures, including the application of the respiratory motion filter and FD-based censoring.

Experimental Protocols for FD Threshold Evaluation

Protocol 1: Multi-Dataset Evaluation of Frame Censoring

Objective: To systematically evaluate the performance of frame censoring across multiple FD thresholds in diverse task-based fMRI datasets [57].

Datasets: Eight publicly available studies from OpenNeuro representing 11 distinct tasks in child, adolescent, and adult participants (total n > 300). Datasets include flanker tasks, n-back paradigms, motor tasks, and word recognition tasks with varying motion characteristics (median FD range: 0.08-0.30 mm) [57].

Processing Pipeline:

• Image Preprocessing: Implement standard preprocessing pipeline using Automatic Analysis version 5.4.0 scripting SPM12, including realignment, normalization, and spatial smoothing.
• Motion Parameter Calculation: Compute framewise displacement (FD) from rigid-body realignment parameters.
• Censoring Implementation: Apply frame censoring at multiple FD thresholds (0.1-0.5 mm in 0.1 mm increments) by including scan-nulling regressors in first-level general linear models.
• Comparison Conditions: Evaluate against alternative motion correction approaches:
  • 6 canonical head motion regressors (RP6)
  • 24-term expansions of head motion estimates (RP24)
  • Wavelet despiking (WDS)
  • Robust weighted least squares (rWLS)
  • Untrained independent component analysis (uICA)

Outcome Metrics:

• Maximum group t-statistic across whole brain and in task-relevant ROI
• Mean parameter estimates within ROI
• Test-retest consistency of subject-level parametric maps
• Spatial overlap of thresholded group-level statistical maps (Dice coefficient)

Key Findings: Modest amounts of frame censoring (1-2% data loss, typically corresponding to FD thresholds around 0.3 mm) showed consistent improvements over standard motion regressors. However, no single approach consistently outperformed others across all datasets and tasks, highlighting the importance of context-specific threshold selection [57].

Protocol 2: Data-Driven Scrubbing Comparison

Objective: To compare motion-based scrubbing (using FD thresholds) with data-driven alternatives in terms of functional connectivity validity, reliability, and identifiability [58].

Data: Resting-state fMRI data from the Human Connectome Project (HCP).

Methods:

• Apply three scrubbing approaches:
  • Motion Scrubbing: Using conventional FD thresholds (0.2 mm, 0.3 mm, 0.5 mm)
  • DVARS: Based on RMS of voxel-wise intensity differences
  • Projection Scrubbing: Novel data-driven method using statistical outlier detection in ICA component space
• Calculate functional connectivity matrices for each approach.
• Evaluate performance using three metrics:
  • Validity: Correspondence between observed functional connectivity and established neurobiological knowledge
  • Reliability: Consistency of functional connectivity across measurements
  • Identifiability: Ability to identify individuals from their functional connectivity patterns

Results: Data-driven scrubbing methods (particularly projection scrubbing) yielded greater improvements to fingerprinting while not generally worsening validity or reliability. Importantly, data-driven scrubbing excluded a fraction of the number of volumes or entire sessions compared to motion scrubbing, thereby dramatically increasing effective sample size [58].

Diagram 2: Framewise Displacement Threshold Selection Decision Framework. This flowchart guides researchers in selecting appropriate FD thresholds based on study population, data quality, and analysis goals.

The Scientist's Toolkit: Essential Research Reagents and Solutions

Table 3: Essential Tools for Motion Censoring Implementation in fMRI Research

Tool/Solution	Function	Implementation Considerations
FIRMM Software	Provides real-time motion feedback during scanning sessions; calculates FD in real-time to alert participants and technicians about excessive movement.	• Particularly effective for reducing high-motion events• Shown to reduce average FD from 0.347 to 0.282 in task-based fMRI [56]
ABCD-HCP BIDS Pipeline	Standardized processing pipeline for large-scale neuroimaging datasets; implements motion censoring at multiple stages with configurable FD thresholds.	• Community standard for reproducible processing• Includes respiratory motion filtering to improve FD estimates [5] [10]
DCAN BOLD Processing (DBP)	Specialized module for nuisance regression and motion censoring; generates temporal masks across a range of FD thresholds (0-0.5 mm in 0.01 mm steps).	• Enables precise threshold optimization• Includes global signal regression which combined with motion censoring provides best motion artifact reduction [5]
Projection Scrubbing Algorithm	Data-driven scrubbing method using statistical outlier detection in ICA component space.	• Retains more data than motion-based scrubbing• Improves functional connectivity fingerprinting [58]
Real-time Motion Feedback	Experimental setup providing visual feedback to participants about their head motion during scanning.	• Uses color-coded cross (white < 0.2 mm, yellow 0.2-0.3 mm, red ≥ 0.3 mm)• Effective for participant training and motion reduction [56]

The selection of an FD threshold of 0.2 mm represents a balanced approach that has been adopted as a community standard within the ABCD-BIDS preprocessing pipeline for generating specific derivatives, particularly connectivity matrices [10]. This threshold effectively mitigates motion artifacts while preserving sufficient data for robust statistical analysis in large-scale population studies. However, the optimal censoring strategy remains context-dependent, influenced by factors including participant population, acquisition parameters, and specific research questions.

Future methodological developments are likely to focus on personalized censoring approaches that adapt to individual motion patterns and data quality characteristics, rather than applying fixed thresholds across entire datasets. The integration of real-time motion monitoring with prospective correction techniques [56], combined with data-driven scrubbing methods [58], offers promising avenues for enhancing data quality while maximizing data retention. Furthermore, as multi-band acquisition protocols become increasingly standard, continued refinement of respiratory artifact correction methods will be essential for accurate motion estimation and appropriate censoring implementation [5].

Addressing Respiratory Artifacts in Multi-Band Data

Respiratory artifacts represent a persistent and pervasive source of noise in functional Magnetic Resonance Imaging (fMRI) data, introducing systematic distortions that can confound the interpretation of results [59]. In multi-band fMRI acquisitions, which utilize simultaneous multi-slice (SMS) imaging to accelerate data collection, these artifacts manifest as complex physiological noise that corrupts the blood-oxygen-level-dependent (BOLD) signal [59] [28]. The ABCD-BIDS preprocessing pipeline specifically addresses this challenge through specialized denoising techniques, including respiratory motion filtering, to mitigate artifacts that otherwise introduce spurious correlations and reduce the validity of neuroimaging studies [5] [28].

The physiological basis of respiratory artifacts stems from both direct and indirect mechanisms. Direct effects include frequency and amplitude changes in the BOLD signal due to respiratory cycles, while indirect effects involve magnetic field fluctuations caused by chest motion and subsequent magnetic susceptibility variations [59]. These artifacts are particularly problematic in multi-band acquisitions because acceleration techniques can amplify subtle physiological fluctuations, potentially degrading data quality despite reduced scan times [59].

Quantitative Characterization of Respiratory Artifacts

Impact of Acquisition Parameters on Artifact Severity

Table 1: Effects of Acquisition Parameters on Respiratory Artifacts in Multi-Band fMRI

Parameter	Level	Impact on Respiratory Artifacts	Recommended Settings
Multiband Factor	4-6 (Moderate)	Improved data quality with RETROICOR correction [59]	MB ≤ 6 for optimal correction
	8 (High)	Significant degradation of data quality [59]	Avoid when possible
Flip Angle	45°	Better performance with RETROICOR models [59]	Recommended for moderate MB factors
	20°	Reduced effectiveness of artifact correction [59]	Use with caution
Echo Time (TE)	Multiple (Multi-echo)	Enables advanced denoising approaches [59] [60]	Multi-echo acquisition preferred

Efficacy of Denoising Approaches

Table 2: Performance Comparison of Respiratory Artifact Correction Methods

Method	Principle	Advantages	Limitations
RETROICOR	Retrospective correction using physiological monitoring [59]	Significantly improves tSNR and SFS; compatible with fast acquisitions [59]	Requires additional physiological data collection
Respiratory Motion Filter (ABCD-BIDS)	Filters respiratory frequency band from motion estimates [5]	Reduces spurious motion estimates; integrated into standardized pipeline [5]	May require parameter tuning for specific populations
Motion Censoring	Exclusion of high-motion frames [5] [28]	Effectively reduces motion-related artifacts [28]	Can introduce bias by excluding specific participant groups [28]
Multi-Echo ICA	Separates BOLD and non-BOLD components using echo time differences [59]	Data-driven approach; no external monitoring required [59]	Computational intensive; requires specialized expertise

Experimental Protocols for Respiratory Artifact Mitigation

RETROICOR Implementation for Multi-Band Data

Purpose: To implement RETROICOR (Retrospective Image Correction) for physiological artifact mitigation in multi-echo fMRI data [59].

Materials and Equipment:

• Siemens Prisma 3T scanner or equivalent
• 64-channel head-neck coil
• Physiological monitoring system (pulse oximeter, respiratory belt)
• Multi-band EPI BOLD sequence from CMRR, University of Minnesota

Procedure:

• Physiological Data Acquisition: Simultaneously record cardiac and respiratory signals throughout fMRI acquisition using pulse oximeter and respiratory belt [59].
• fMRI Data Collection: Acquire multi-echo fMRI data with the following typical parameters:
  • Resolution: 3×3×3.5 mm
  • Multiband factors: 1, 4, 6, 8
  • TR: 400-3050 ms (varies by run)
  • Flip angles: 20°, 45°, 80°
  • TE: 17.00, 34.64, and 52.28 ms [59]
• RETROICOR Application:
  • Option A (RTCind): Apply RETROICOR correction to individual echoes before multi-echo combination [59].
  • Option B (RTCcomp): Apply RETROICOR correction after multi-echo data combination [59].
• Quality Assessment: Calculate temporal signal-to-noise ratio (tSNR), signal fluctuation sensitivity (SFS), and variance of residuals to evaluate correction efficacy [59].

Validation: Compare quantitative metrics between RTCind and RTCcomp implementations. Both approaches show similar performance, with moderately accelerated runs (MB factors 4 and 6) demonstrating the most significant improvements [59].

ABCD-BIDS Respiratory Motion Filtering Protocol

Purpose: To implement respiratory artifact filtering within the ABCD-BIDS preprocessing pipeline [5].

Materials and Equipment:

• ABCD-BIDS pipeline software
• BIDS-formatted MRI datasets
• High-performance computing environment

Procedure:

• Data Preparation: Ensure fMRI data is formatted according to BIDS specification with appropriate metadata [5] [10].
• Pipeline Execution: Run the ABCD-BIDS pipeline through the FMRIVolume stage to obtain motion realignment parameters [5].
• Respiratory Motion Filter Application:
  • Apply respiratory motion filter to the motion realignment data [5].
  • Filter frequencies between 18.582 and 25.726 breaths per minute (0.31-0.43 Hz) to remove respiratory artifacts from motion estimates [5].
  • Optionally specify custom upper and lower frequencies for specialized breathing patterns.
• Motion Censoring: Implement framewise displacement (FD) censoring with threshold of 0.3 mm for standard preprocessing [5].
• Quality Control: Utilize the ExecutiveSummary stage to generate visual quality control pages displaying movement and grayordinate time series pre- and post-regression [5].

Validation: The respiratory motion filter produces better estimates of framewise displacement, reducing inappropriate motion censoring caused by respiratory artifacts [5].

Figure 1: Integrated Workflow for Respiratory Artifact Correction in Multi-Band fMRI

Multi-Echo Acquisition and Denoising Protocol

Purpose: To leverage multi-echo acquisition for improved respiratory artifact removal [59] [60].

Materials and Equipment:

• Multi-echo fMRI sequence capability
• Processing tools supporting ME-ICA (Multi-Echo Independent Component Analysis)
• NORDIC (Noise Reduction with Distribution Corrected PCA) for thermal denoising [60]

Procedure:

• Multi-Echo Data Acquisition: Acquire fMRI data with multiple echo times (e.g., TE: 17.00, 34.64, and 52.28 ms) [59].
• Echo Combination: Optimally combine echoes based on T2* relaxation times of underlying tissue [60].
• ME-ICA Processing: Apply Multi-Echo Independent Component Analysis to differentiate BOLD and non-BOLD components in fMRI time series [59].
• Thermal Denoising: Implement NORDIC PCA to remove thermal noise, improving temporal signal-to-noise ratio (tSNR) [60].
• Validation: Assess split-half reliability of functional connectivity matrices and quantitative tSNR improvements [60].

Application Notes: Multi-echo acquisitions show particular promise for developmental populations where T2* relaxation times vary significantly from adults [60].

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Tools for Respiratory Artifact Research

Tool/Reagent	Function	Application Notes
ABCD-BIDS Pipeline	Standardized preprocessing for BIDS-formatted data [5] [10]	Includes integrated respiratory motion filtering; recommended for multi-site studies
RETROICOR	Retrospective physiological noise correction [59]	Requires cardiac and respiratory monitoring; compatible with multi-echo data
ME-ICA	Multi-Echo Independent Component Analysis [59]	Data-driven approach; effective for separating respiratory artifacts from BOLD signal
NORDIC PCA	Thermal noise removal [60]	Improves tSNR without spatial smoothing; compatible with multi-echo data
SHAMAN	Motion impact scoring for trait-FC relationships [28]	Quantifies residual motion effects after denoising; identifies overestimation/underestimation
Physiological Monitoring System	Records cardiac and respiratory waveforms during scanning [59]	Essential for RETROICOR; should include pulse oximeter and respiratory belt

Integrated Analysis Framework

Figure 2: ABCD-BIDS Pipeline with Integrated Respiratory Artifact Correction

Validation and Quality Control

Performance Metrics and Benchmarking

Robust validation of respiratory artifact correction requires multiple complementary approaches. The ABCD-BIDS pipeline incorporates quantitative quality control measures through its ExecutiveSummary stage, which provides visual assessment of movement and grayordinate time series before and after regression [5]. For quantitative benchmarking, temporal signal-to-noise ratio (tSNR) and signal fluctuation sensitivity (SFS) provide standardized metrics for comparing denoising efficacy across different methods and acquisition parameters [59].

Framewise displacement (FD) thresholds require careful consideration in study design. Research demonstrates that censoring at FD < 0.2 mm reduces significant motion overestimation from 42% to 2% of traits, but does not decrease the number of traits with significant motion underestimation scores [28]. This highlights the importance of trait-specific motion impact assessment using methods like SHAMAN (Split Half Analysis of Motion Associated Networks), which distinguishes between motion causing overestimation or underestimation of trait-FC relationships [28].

Advanced Integration Strategies

For optimal respiratory artifact correction, integrated approaches combining multiple methods show superior performance. The combination of RETROICOR with multi-echo acquisitions provides complementary advantages - RETROICOR effectively models periodic physiological fluctuations, while multi-echo approaches enable separation of BOLD and non-BOLD components based on their distinct T2* decay characteristics [59]. Furthermore, incorporating thermal denoising with NORDIC PCA following multi-echo combination provides additional tSNR improvements and enhanced functional connectivity reliability [60].

When implementing these advanced integrated approaches, researchers should consider computational requirements and sequence optimization. Multi-echo acquisitions require careful parameter selection, including appropriate echo time spacing and readout optimization to maintain spatial and temporal resolution [59] [60]. The ABCD-BIDS pipeline provides a standardized framework for implementing these integrated solutions while maintaining reproducibility across sites and scanners [5] [10].

Hardware and Software Resource Specifications

Efficient execution of the ABCD-BIDS preprocessing pipeline requires careful consideration of computational resources. The following table summarizes the key resource requirements and performance characteristics based on operational documentation.

Table 1: Computational Resource Requirements for ABCD-BIDS Pipeline

Resource Type	Minimum Specification	Recommended Specification	Performance Impact
CPU Cores	1 core (highly discouraged)	≥4 cores [19]	Enables parallel processing of multiple fMRI runs; reduces processing time from >24 hours to more manageable durations [19]
Memory (RAM)	Not specified	≥12 GB total (≥3 GB per core) [19]	Prevents memory bottlenecks during computationally intensive stages (e.g., segmentation, registration)
Storage	Sufficient for final outputs	High capacity and I/O performance for intermediate files [19]	Intermediate processing occurs in output directory; sufficient scratch space critical for pipeline stability [19]
Container Technology	Docker or Singularity [19]	Docker or Singularity with GPU support (if applicable)	Standardizes software environment, ensuring reproducibility across different computing systems

Parallel Processing and Performance Optimization Strategies

Performance optimization for the ABCD-BIDS pipeline primarily involves strategic allocation of computational resources and configuration of parallel processing capabilities.

CPU Core Utilization and Parallelization

The pipeline is designed to leverage multiple CPU cores through two primary mechanisms:

• Concurrent Processing: Different stages of the pipeline can utilize multiple cores for algorithmic speedups, particularly in tools like ANTs (Advanced Normalization Tools) and FreeSurfer [19].
• Parallel Run Processing: For sessions containing multiple fMRI runs, the pipeline can process these runs in parallel. The optimal number of CPU cores should evenly divide the number of runs for maximum efficiency [19].

The command-line option --ncpus specifies the number of cores allocated, significantly reducing processing time compared to single-core execution [19].

Computational Intensity Across Pipeline Stages

The ABCD-BIDS pipeline comprises multiple sequential stages, each with varying computational demands:

• PreFreeSurfer: Involves initial structural image processing and alignment.
• FreeSurfer: Performs cortical surface reconstruction - typically one of the most computationally intensive stages.
• PostFreeSurfer: Maps surface data to standard spaces.
• FMRIVolume: Processes functional MRI data in volume space.
• FMRISurface: Projects functional data to surface space.
• DCANBOLDProcessing: Applies final BOLD processing steps, including motion denoising.

For incomplete runs or specific research needs, the --stage flag allows running a subset of stages, saving substantial computational resources when full reprocessing isn't necessary [19].

Experimental Protocols for Performance Benchmarking

Resource Utilization Profiling Protocol

Objective: To quantitatively assess computational resource requirements across different pipeline configurations.

Methodology:

• System Configuration: Prepare identical ABCD-BIDS datasets on systems with varying core counts (1, 4, 8, 16) and memory allocations.
• Pipeline Execution: Run the full pipeline using the standard command structure with varying --ncpus parameters [19]:

• Metrics Collection: Monitor and record execution time, peak memory usage, CPU utilization, and disk I/O for each stage.
• Analysis: Calculate speedup efficiency as cores increase and identify computational bottlenecks.

Motion Denoising-Specific Optimization Protocol

Objective: To optimize computational parameters specifically for motion denoising research within the ABCD-BIDS framework.

Methodology:

• Bandstop Filter Implementation: Apply respiratory artifact correction using the --bandstop filter with parameters (e.g., --bandstop 18.582 25.726) matching the interquartile range of participant respiratory rates [24] [19].
• Selective Stage Execution: Use --stage to execute only motion-relevant processing stages (e.g., --stage DCANBOLDProcessing) for iterative algorithm development [19].
• Resource Allocation Testing: Compare processing time and memory usage for motion correction modules under different core allocations.
• Output Verification: Validate motion-corrected outputs including:
  • *_desc-filteredincludingFD_motion.tsv - Motion parameters with frame-wise displacement [10]
  • *_space-MNI_bold.nii.gz - Processed BOLD data in standard space [10]

Workflow Visualization

The Scientist's Toolkit

Table 2: Essential Research Reagent Solutions for ABCD-BIDS Pipeline Optimization

Tool/Resource	Function	Implementation Example
ABCD-HCP Pipeline Software	Core processing pipeline implementing HCP-style minimal preprocessing for ABCD data	Docker: dcanumn/abcd-hcp-pipeline [19]Singularity: Built from Docker image
FreeSurfer License	Enables automated cortical surface reconstruction	File: license.txt mounted to /opt/freesurfer/license.txt in container [12] [19]
Computational Container	Standardizes software environment across computing platforms	Docker or Singularity runtime environment [19]
Bandstop Filter	Reduces respiratory motion artifacts in motion regressors	Command parameter: --bandstop 18.582 25.726 (bpm range) [24] [19]
Study-Specific Templates	Improves registration for non-typical populations (e.g., pediatric, elderly)	Command parameter: --study-template /path/head.nii.gz /path/brain.nii.gz [19]
Stage Selection	Enables partial pipeline execution for iterative development	Command parameter: --stage PreFreeSurfer:PostFreeSurfer [19]

Within the framework of motion denoising research utilizing the ABCD-BIDS preprocessing pipeline, managing incomplete or corrupted data is a critical operational challenge. A specific and frequently encountered obstacle is the FNIRT read error, a failure in FSL's Nonlinear Image Registration Tool (FNIRT) that can halt pipeline progression. Such errors are particularly problematic in large-scale, automated processing environments like the ABCD-BIDS pipeline, where manual intervention is impractical. This application note details the primary causes of FNIRT read errors and other common data issues, providing structured protocols for their identification, resolution, and prevention to ensure research continuity and data integrity.

The table below summarizes the most frequent data-related errors encountered during the ABCD-BIDS preprocessing lifecycle, their root causes, and immediate solutions.

Table 1: Common Data Errors in ABCD-BIDS Preprocessing

Error Type	Primary Cause	Immediate Solution	Preventive Measure
FNIRT Read Error [24]	Insufficient volumes in functional data (FNIRT requires ≥3 volumes).	Skip functional pipeline for affected session; run anatomical-only processing.	Implement pre-processing data validation to check for truncated scans.
Fieldmap Dimension Mismatch [24]	Spin-echo fieldmap dimensions do not match the scout/reference image.	Resize fieldmap to match the BOLD reference using FLIRT: flirt -in "$fmap" -ref "$bold" -out "$output" -applyisoxfm 2.4.	Ensure consistent acquisition parameters between fieldmaps and functional scans.
Incorrect Fieldmap Assignment [24]	Pipeline misassignment of fieldmaps to corresponding functional runs.	Manually update the JSON metadata files for the fieldmaps to specify the correct "IntendedFor" field.	Curate BIDS dataset meticulously to ensure clear metadata linkages.
Index Error in anat_only Cases [24]	Interference from fieldmap JSON files when only anatomical data is processed.	Temporarily sync only the anat directory for processing.	Implement conditional logic in the pipeline to ignore irrelevant functional files for anat-only runs.

Detailed Experimental Protocols for Error Resolution

Protocol: Diagnosis and Resolution of FNIRT Read Errors

This protocol addresses the specific error where FNIRT fails due to insufficient data, a known issue within the ABCD-BIDS pipeline [24].

• Error Identification: The pipeline will halt and report a FNIRT read error, typically in the FMRIVolume stage. The log file will often contain a message indicating a problem reading the functional data.
• Root Cause Analysis:
  • Navigate to the subject's functional data directory.
  • Use fslinfo or check the BIDS sidecar JSON file to determine the number of time-points (volumes) in the corrupted functional run.
  • Confirm Cause: If the number of volumes is less than three, this is the definitive cause of the FNIRT error [24].
• Immediate Resolution:
  • For the affected subject and session, bypass the full pipeline execution.
  • Run the ABCD-BIDS pipeline in anat_only mode to at least generate the structural derivatives and FreeSurfer reconstructions, which are prerequisites for many subsequent analyses.
• Data Exclusion and Reporting:
  • Log the subject ID, session ID, and specific run that failed.
  • Clearly document this data loss in your study's methods and quality control records. The data from this run should be excluded from all group-level analyses.

Protocol: Correction of Fieldmap Dimension Mismatch (GE Scanners)

This protocol resolves a fieldmap scaling issue prevalent in data from certain scanner manufacturers [24].

• Identify Mismatch: The pipeline, specifically the TopupPreprocessingAll.sh script, will fail with a dimension mismatch error between the fieldmap and the scout image.
• Resize Fieldmap:
  • Locate the problematic fieldmap file ($fmap_file) and the BOLD run it is intended for ($func_run).
  • Execute the following FSL command to resample the fieldmap to match the BOLD reference dimensions and resolution [24]:

• Update Pipeline Input: Replace the original fieldmap in the BIDS directory with the resampled version, or redirect the pipeline to use the corrected output.
• Rerun Pipeline: Restart the pipeline from the stage that initially failed.

Workflow for Systemic Data Handling

The following diagram illustrates a robust workflow for handling incomplete or corrupted data, from error detection to final analysis.

Systematic workflow for handling preprocessing errors

The Scientist's Toolkit: Essential Research Reagents and Software

The following table catalogues the key software tools and resources essential for implementing the ABCD-BIDS pipeline and troubleshooting the errors described herein.

Table 2: Research Reagent Solutions for ABCD-BIDS Preprocessing

Tool/Resource	Function in Pipeline	Relevance to Error Handling
ABCD-BIDS Pipeline [5]	Primary processing pipeline integrating HCP minimal preprocessing and DCAN BOLD tools.	Understanding its multi-stage structure (PreFreeSurfer, FMRIVolume, etc.) is essential for pinpointing and isolating errors.
FSL (FMRIB Software Library)	Provides core registration (FLIRT, FNIRT) and distortion correction (topup) tools.	FNIRT read errors originate here. FLIRT is used for fieldmap resizing and motion correction.
FreeSurfer	Performs automated cortical reconstruction and volumetric segmentation.	Generates high-quality brain masks and anatomical models used in later registration stages.
BIDS Validator	Validates dataset compliance with the Brain Imaging Data Structure standard.	Critical for preventing metadata-related errors, such as incorrect fieldmap assignment, before processing begins.
tedana	A Python package for TE-dependent analysis of multi-echo fMRI data.	While not part of the standard ABCD pipeline, it represents an advanced tool for denoising, which is the broader research context [61].
fMRIPrep	A robust fMRI preprocessing pipeline.	Serves as a comparison pipeline; studies show preprocessing choices (like one-step interpolation in OGRE) affect inter-individual variability and noise [62] [63].

Proactive management of data integrity is paramount for the success of motion denoising research using the ABCD-BIDS pipeline. By understanding common failure modes like the FNIRT read error, and by implementing the detailed diagnostic and resolution protocols outlined in this document, researchers can transform data crises into manageable workflow events. Adherence to systematic workflows and robust quality control ensures the reliability of derived data for downstream analysis, safeguarding the scientific validity of large-scale neuroimaging studies.

Within the broader context of ABCD-BIDS preprocessing pipeline research for motion denoising, the use of study-specific templates represents a crucial methodological advancement for enhancing data quality in special populations. The standard adult brain templates used in conventional neuroimaging pipelines often fail to accommodate the substantial anatomical variability found in populations such as infants, elderly individuals with large ventricles, or clinical groups with significant structural deviations [12]. This mismatch can introduce systematic errors in spatial normalization, tissue segmentation, and functional data alignment—errors that compound with existing motion-related artifacts and ultimately compromise the fidelity of functional connectivity measures.

The ABCD-BIDS pipeline, which integrates methods from both the Human Connectome Project's minimal preprocessing pipeline and DCAN Labs resting state fMRI analysis tools, provides researchers with configurable options for implementing study-specific templates [5] [24]. This capability is particularly valuable for motion denising research because improved anatomical alignment through appropriate templates enhances the accuracy of motion correction algorithms and nuisance regression techniques applied later in the processing stream. When functional data is mapped more precisely to anatomical space, motion-related artifacts can be more effectively identified and removed, leading to cleaner estimates of true neural signals.

Technical Specifications: Template Configuration Options

The ABCD-BIDS pipeline offers researchers specific command-line options for implementing custom templates when processing data from special populations. These options enable replacement of default registration targets with study-appropriate alternatives at critical stages of the anatomical processing workflow.

Table 1: Study-Template Configuration Options in ABCD-BIDS Pipeline

Configuration Option	Input Requirements	Processing Stage	Target Population Examples	Impact on Motion Denoising
--t1-study-template	Head and brain images for T1w template [12]	PreFreeSurfer: Intermediate nonlinear registration [12]	Elderly with large ventricles [12]	Improves tissue segmentation accuracy for better CSF/WM signal regression
--t2-study-template	Head and brain images for T2w template [12]	PreFreeSurfer: Intermediate nonlinear registration [12]	Neonates, infants [12]	Enhances distortion correction for field mapping, reducing motion-related misalignment
--hyper-normalization-method	ADULTGMIP, ROI_IPS, or NONE [12]	FreeSurfer: Intensity normalization [12]	Populations with different tissue intensity profiles	Creates more reliable tissue masks for global signal regression

For special populations including human infants, elderly individuals, or those with clinical conditions affecting brain structure, these template options enable researchers to address the fundamental anatomical differences that standard pipelines might misinterpret as artifacts [12]. The technical implementation involves providing separate head and brain images for both T1w and T2w templates, which the pipeline then uses for intermediate nonlinear registration steps instead of the default adult templates. This approach is particularly effective "where population differs greatly from average adult, e.g., in elderly populations with large ventricles" [12].

Diagram 1: Study-template integration workflow in ABCD-BIDS pipeline. Custom templates are incorporated during PreFreeSurfer stage to improve subsequent motion denoising.

Implementation Protocol: Custom Template Integration

Template Creation and Validation

The successful implementation of study-specific templates begins with the creation of population-appropriate reference images. Researchers should select a representative subset of high-quality anatomical scans from the target population, ensuring exclusion of images with excessive motion artifact or other quality issues. The recommended approach involves:

• Subject Selection: Choose 20-50 participants with minimal motion artifact, balanced for any relevant demographic variables (age, sex, clinical status) within the special population.
• Template Construction: Use advanced normalization tools (ANTs) or similar software to create symmetric, population-specific T1w and T2w templates through iterative nonlinear registration.
• Quality Assurance: Visually inspect the resulting templates for anatomical fidelity and absence of artifacts using tools like BrainSwipes, which is employed in the ABCD-BIDS Community Collection for quality control [10].

Pipeline Configuration

Once custom templates are created, researchers can implement them within the ABCD-BIDS pipeline using specific command-line arguments. The following protocol details the complete integration process:

This configuration directs the pipeline to utilize the provided templates during the critical PreFreeSurfer stage, where nonlinear registration to an intermediate space occurs [12]. For infant populations, additional considerations may include adjusting the --atropos-mask-method to CREATE rather than REFINE and modifying the --max-cortical-thickness parameter to reflect developmental differences [12].

Impact on Motion Denoising: Empirical Evidence

The integration of study-specific templates creates foundational improvements that enhance subsequent motion denoising procedures in the ABCD-BIDS pipeline. When anatomical alignment is optimized for special populations, several key motion denoising benefits emerge:

Table 2: Template Effects on Motion Denoising Performance

Denoising Component	Standard Template Performance	Study-Template Enhanced Performance	Impact on Functional Connectivity
Framewise Displacement (FD) Estimation	Potential respiratory artifacts in motion estimates [24]	Improved FD estimates through better registration [24]	More accurate motion censoring decisions
Global Signal Regression	Suboptimal tissue segmentation affects nuisance regressors	Enhanced WM/CSF masks improve signal regression [5]	Reduced motion-related false positives in FC [3]
Respiratory Motion Filter	23% signal variance explained by motion after denoising [3]	Better motion estimates enhance filtering efficiency	Decreased overestimation of trait-FC effects (42% to 2%) [3]

Empirical evidence from the ABCD Study demonstrates that even with advanced denoising including global signal regression, respiratory filtering, and motion censoring, approximately 23% of signal variance remains explained by head motion [3]. The implementation of study-specific templates addresses foundational anatomical alignment issues that contribute to this residual motion artifact. Research shows that when custom templates improve registration accuracy, the subsequent motion denoising procedures—particularly the respiratory motion filter that targets frequencies between 18.582-25.726 breaths per minute—become more effective at reducing both overestimation and underestimation of trait-functional connectivity relationships [24] [3].

The Scientist's Toolkit: Essential Research Reagents

Table 3: Essential Resources for Template-Based Motion Denoising Research

Resource Category	Specific Tool/Solution	Function in Research	Implementation Notes
Processing Software	ABCD-HCP BIDS Pipeline [5]	Integrated anatomical/functional processing with template support	Docker images available at DCAN Docker Hub; requires FreeSurfer license
Template Construction	Advanced Normalization Tools (ANTs)	Creation of population-specific templates	Used in PreFreeSurfer for denoising and N4 bias field correction [5]
Quality Assessment	BrainSwipes [10]	Visual inspection of derivatives	Community-driven QC for template validation
Data Standardization	BIDS Validator	Verification of input dataset structure	Ensures compatibility with ABCD-BIDS template options
Motion Quantification	Framewise Displacement (FD) Calculation	Measures head motion between volumes	Respiratory filtering applied to reduce artifacts in FD estimates [24]

The implementation of study-specific templates within the ABCD-BIDS preprocessing pipeline represents a sophisticated methodological approach for enhancing motion denoising in special populations. By addressing fundamental anatomical mismatches at the initial registration stages, researchers can create a more robust foundation for all subsequent motion correction procedures—from improved framewise displacement estimation through respiratory motion filtering to more precise nuisance regression and motion censoring. The configurable template options provided in the pipeline enable researchers to adapt these advanced neuroimaging methods to diverse populations including infants, elderly individuals, and clinical groups, thereby reducing motion-related artifacts that might otherwise confound functional connectivity findings. As the field moves toward increasingly precise brain-behavior association studies, such anatomical customization will play a crucial role in ensuring that observed effects reflect genuine neural phenomena rather than methodological artifacts.

Evaluating Denoising Efficacy: Performance Benchmarks and Alternative Methods

In-scanner head motion represents the most substantial source of artifact in functional MRI (fMRI) signals, introducing systematic bias to resting-state functional connectivity (FC) that is not completely removed by standard denoising algorithms [28]. This technical challenge is particularly critical for researchers studying traits associated with motion, such as psychiatric disorders, who need to determine whether their observed trait-FC relationships reflect genuine neural associations or spurious motion-related artifacts [28]. The Split Half Analysis of Motion Associated Networks (SHAMAN) framework was developed to address this need by assigning a motion impact score to specific trait-FC relationships, distinguishing between motion causing overestimation or underestimation of effects [28] [64]. Within the context of the ABCD-BIDS preprocessing pipeline, SHAMAN provides a crucial method for quantifying residual motion artifact after initial denoising, enabling researchers to avoid reporting false positive results in brain-wide association studies (BWAS) [28] [24].

The ABCD-BIDS pipeline represents the default denoising algorithm for pre-processed data from the Adolescent Brain Cognitive Development (ABCD) Study, incorporating global signal regression, respiratory filtering, spectral filtering, despiking, and motion parameter timeseries regression [28] [5]. Despite this comprehensive processing, residual motion artifact persists, explaining approximately 23% of signal variance after ABCD-BIDS processing compared to 73% with minimal processing alone [28]. This residual motion systematically decreases long-distance connectivity and increases short-range connectivity, most notably in the default mode network, creating spatially specific bias patterns that can be mistaken for genuine neurobiological effects [28].

Table 1: Key Characteristics of Motion Artifact in fMRI Data

Aspect	Impact	Quantitative Evidence
Residual Motion After ABCD-BIDS	23% of signal variance explained by motion	Reduced from 73% with minimal processing [28]
Spatial Pattern	Decreased long-distance connectivity, increased short-range connectivity	Strong negative correlation (Spearman ρ = -0.58) between motion-FC effect matrix and average FC matrix [28]
Trait Impact	42% of traits show significant motion overestimation; 38% show underestimation	Assessment of 45 traits from n=7,270 ABCD participants [28]

SHAMAN Framework: Theoretical Foundations and Quantitative Assessment

Core Theoretical Principles

SHAMAN capitalizes on a fundamental physiological observation: traits such as cognitive abilities or clinical characteristics remain stable over the timescale of an MRI scan, while motion is a state-dependent variable that fluctuates from second to second [28]. This theoretical foundation enables the framework to distinguish motion-related artifacts from genuine trait-FC relationships by measuring differences in correlation structure between split high-motion and low-motion halves of each participant's fMRI timeseries [28]. When trait-FC effects are independent of motion, the difference between halves will be non-significant due to trait stability. A significant difference indicates that state-dependent motion variations impact the trait's connectivity measures [28].

The directionality of the motion impact score provides critical information about the nature of potential bias. A motion impact score aligned with the direction of the trait-FC effect indicates motion overestimation, where residual motion artifact inflates the apparent strength of the relationship. Conversely, a motion impact score opposite to the trait-FC effect direction indicates motion underestimation, where artifact masks or reduces the genuine effect size [28]. This directional discrimination helps researchers interpret whether motion might be creating false positive findings or obscuring true effects.

Quantitative Assessment of Motion Impact

Application of SHAMAN to 45 traits from n=7,270 participants in the ABCD Study revealed substantial motion impact across diverse measures [28]. After standard denoising with ABCD-BIDS without motion censoring, 42% (19/45) of traits demonstrated significant (p < 0.05) motion overestimation scores, while 38% (17/45) showed significant underestimation scores [28]. This comprehensive assessment demonstrates that nearly 80% of traits examined showed statistically significant impact from residual motion, highlighting the critical importance of motion impact quantification in neuroimaging studies.

Table 2: Motion Impact on Traits Before and After Censoring in ABCD Study

Condition	Motion Overestimation	Motion Underestimation
After ABCD-BIDS denoising (no censoring)	42% (19/45 traits)	38% (17/45 traits)
With censoring (FD < 0.2 mm)	2% (1/45 traits)	No decrease observed
Key Finding	Censoring effectively reduces overestimation	Underestimation effects persist despite censoring

The effectiveness of motion censoring as a supplementary approach demonstrates asymmetric effects on different types of motion impact. Implementing censoring at framewise displacement (FD) < 0.2 mm dramatically reduced significant overestimation to only 2% (1/45) of traits, but did not decrease the number of traits with significant motion underestimation scores [28]. This finding indicates that while censoring effectively controls for inflation of effect sizes, it does not address situations where motion artifact masks genuine relationships, potentially leading to false negative conclusions.

Experimental Protocols and Implementation

SHAMAN Implementation Protocol

The SHAMAN analytical workflow can be implemented through the following detailed protocol:

Step 1: Data Preparation and Preprocessing

• Process raw fMRI data through the ABCD-BIDS pipeline, which includes the following stages [5]:
  • PreFreeSurfer: Anatomical distortion correction and brain extraction
  • FreeSurfer: Cortical surface reconstruction and segmentation
  • PostFreeSurfer: CIFTI surface file generation and surface registration
  • FMRIVolume: Functional distortion correction and motion realignment
  • FMRISurface: Volume to surface mapping and CIFTI conversion
  • DCANBOLDProcessing: Nuisance regression and motion censoring
• Apply global signal regression, respiratory filtering (18.582-25.726 breaths per minute), and motion parameter timeseries regression [5] [24]
• Generate framewise displacement (FD) timeseries from motion parameters, applying respiratory motion filtering to reduce artifactual motion estimates [5]

Step 2: Timeseries Segmentation

• For each participant, divide the preprocessed fMRI timeseries into two halves based on motion state
• Identify high-motion and low-motion frames using framewise displacement threshold (typically FD = 0.2 mm)
• Ensure sufficient frames in each half for reliable connectivity estimation (minimum 4-5 minutes of data)

Step 3: Connectivity Calculation

• Calculate separate functional connectivity matrices for high-motion and low-motion halves
• Use Pearson correlation between regional timeseries from predefined atlas parcellations (e.g., Gordon's 333 ROI atlas)
• Apply Fisher's z-transform to correlation coefficients for normalization

Step 4: Motion Impact Scoring

• For each trait of interest, compute trait-FC correlations separately for high-motion and low-motion halves
• Calculate motion impact score as the difference between trait-FC correlations in high-motion versus low-motion conditions
• Determine statistical significance through permutation testing (typically 1,000 permutations) with non-parametric combining across connections

Step 5: Interpretation and Directionality Assessment

• Align motion impact score direction with observed trait-FC effect
• Classify as motion overestimation when directions align
• Classify as motion underestimation when directions oppose

ABCD-BIDS Preprocessing Protocol

The ABCD-BIDS pipeline provides the essential foundation for SHAMAN analysis through its comprehensive processing approach:

Anatomical Processing Stages:

• PreFreeSurfer: Utilizes Advanced Normalization Tools (ANTs) for denoising and N4 bias field correction, significantly improving results for GE and Philips scanners [5]
• FreeSurfer: Segments brain structures, reconstructs white and pial cortical surfaces, and performs folding-based surface registration [5]
• PostFreeSurfer: Generates CIFTI surface files and applies surface registration to Conte-69 template using ANTs' diffeomorphic symmetric normalization [5]

Functional Processing Stages:

• FMRIVolume: Corrects gradient-nonlinearity distortions, performs motion realignment via FSL FLIRT rigid-body registration, and applies FSL's topup for distortion correction [5]
• FMRISurface: Maps volume time series to standard CIFTI grayordinates space [5]
• DCANBOLDProcessing (DBP): Executes four key steps [5]:
  • Standard preprocessing (de-meaning, de-trending)
  • Respiratory motion filtering (optional)
  • Motion censoring with FD threshold of 0.3 mm for standard preprocessing
  • Generation of parcellated timeseries for standard atlases

Critical Processing Parameters:

• Framewise displacement threshold: 0.3 mm for censoring during standard preprocessing
• Band-pass filter: 0.008-0.09 Hz using 2nd order Butterworth filter
• Respiratory frequency filtering: 18.582-25.726 breaths per minute
• Global signal regression: Included as essential component for motion reduction

Research Reagents and Computational Tools

Table 3: Essential Research Reagents and Computational Solutions

Resource	Type	Function	Implementation Details
ABCD-BIDS Pipeline	Software Pipeline	Comprehensive fMRI preprocessing	BIDS App utilizing HCP minimal preprocessing + DCAN Labs tools [5]
Gordon 333 ROI Atlas	Brain Parcellation	Regional timeseries extraction	Cortical parcellation with 333 regions across functional networks [5]
Framewise Displacement	Motion Metric	Quantifying head motion between volumes	Derived from motion parameters, respiratory-filtered [28] [5]
SHAMAN Algorithm	Analytical Framework	Motion impact quantification	Split-half analysis of motion-associated networks [28]
Permutation Testing	Statistical Method	Significance assessment	Non-parametric combining across connections [28]

Workflow Integration and Quality Assessment

Quality Control Integration

The integration of SHAMAN within the ABCD-BIDS preprocessing workflow requires systematic quality assessment at multiple stages:

Preprocessing Quality Metrics:

• Executive Summary stage generates HTML visual quality control pages with BrainSprite viewer for T1w/T2w segmentation [5]
• Atlas registration overlays on single band reference created by FSL's slicer
• Visualization of movement and grayordinate timeseries pre- and post-regression [5]

Data Sufficiency Checks:

• Minimum of 8 minutes of resting-state fMRI data recommended for inclusion in SHAMAN analysis [28]
• Verification of adequate frames after censoring (minimum 4-5 minutes per split half)
• Assessment of motion-FC effect matrix correlation with average FC matrix (expected Spearman ρ ≈ -0.58 before censoring) [28]

SHAMAN-Specific Quality Indicators:

• Permutation testing consistency across multiple iterations
• Comparison of motion impact scores across different FD thresholds
• Assessment of spatial consistency in motion-associated networks

Interpretation Guidelines

Substantial Motion Impact Indicators:

• Motion impact scores with p < 0.05 following permutation testing
• Consistent motion effects across multiple functional connections
• Alignment of motion impact spatial pattern with known motion artifact patterns (e.g., default mode network susceptibility)

Reporting Recommendations:

• Include motion impact scores for all primary trait-FC findings
• Differentiate between motion overestimation and underestimation scenarios
• Report both pre-censoring and post-censoring motion impact assessments
• Acknowledge potential residual motion impact even after SHAMAN application

The SHAMAN framework, when implemented within the comprehensive ABCD-BIDS preprocessing pipeline, provides researchers with a robust method for quantifying and addressing the pervasive challenge of motion-related artifact in functional connectivity studies, particularly in large-scale brain-wide association studies where subtle effects may be disproportionately influenced by residual motion.

Large-scale validation using data from the Adolescent Brain Cognitive Development (ABCD) Study demonstrates the efficacy of the ABCD-BIDS preprocessing pipeline in mitigating motion-related variance in resting-state functional MRI (rs-fMRI). Quantitative assessments reveal that the pipeline achieves a 69% relative reduction in motion-related signal variance compared to minimally processed data. However, residual motion continues to significantly impact functional connectivity (FC) measures, influencing the detection of brain-behavior associations. This application note details the experimental protocols and quantitative metrics essential for evaluating pipeline performance, providing researchers with a framework for assessing denoising efficacy in large-scale neuroimaging studies.

Head motion represents the most substantial source of artifact in functional MRI, systematically altering functional connectivity measurements and potentially leading to spurious brain-behavior associations [28]. The ABCD-BIDS pipeline was developed to address these challenges within the context of the large-scale, multi-site ABCD Study, implementing a comprehensive suite of denoising strategies grounded in the Human Connectome Project's minimal preprocessing pipeline and DCAN Labs processing tools [5] [10]. This protocol outlines standardized methodologies for quantifying the pipeline's efficacy in reducing motion-related variance, with particular emphasis on validation approaches applicable to large-scale pediatric neuroimaging datasets.

Experimental Protocols

Data Acquisition and Participant Selection

Dataset: Adolescent Brain Cognitive Development (ABCD) Study data

• Sample Size: n = 9,652 children (ages 9-10) with at least 8 minutes of rs-fMRI data [28]
• Exclusion Criteria: Insufficient rs-fMRI data (<8 minutes)
• Dataset Characteristics: Multi-site acquisition across 21 sites, multiple scanner manufacturers

Imaging Parameters:

• Modality: Resting-state fMRI
• Minimal Data Requirements: >8 minutes of resting-state fMRI data per participant
• Motion Quantification: Framewise displacement (FD) calculated from rigid-body realignment parameters

ABCD-BIDS Processing Pipeline Protocol

The ABCD-BIDS pipeline implements a multi-stage processing workflow optimized for large-scale datasets [5] [24]:

Table 1: ABCD-BIDS Pipeline Stages

Pipeline Stage	Primary Function	Key Tools/Algorithms
PreFreeSurfer	Anatomical distortion removal and brain extraction	ANTs denoising, N4 bias field correction
FreeSurfer	Cortical surface reconstruction and segmentation	FreeSurfer 5.3.0-HCP
PostFreeSurfer	CIFTI file generation and surface registration	ANTs registration to Conte-69 template
FMRIVolume	Functional distortion correction and motion realignment	FSL topup, FLIRT rigid-body registration
FMRISurface	Projection to CIFTI grayordinates space	HCP-style processing
DCANBOLDProcessing	Nuisance regression and motion censoring	Global signal regression, band-pass filtering

Critical Denoising Parameters [5] [24]:

• Global Signal Regression: Included to reduce motion effects and eliminate batch effects
• Respiratory Motion Filter: Filters frequencies between 18.582-25.726 breaths per minute from motion realignment data
• Band-Pass Filtering: 2nd order Butterworth filter (0.008-0.09 Hz)
• Motion Censoring Threshold: FD < 0.3 mm for standard preprocessing
• Nuisance Regressors: Mean white matter, CSF, and global signal timeseries; motion parameters with Volterra expansion

Motion Impact Assessment Protocol (SHAMAN Method)

The Split Half Analysis of Motion Associated Networks (SHAMAN) method quantifies trait-specific motion impacts [28]:

Procedure:

• Data Splitting: Divide each participant's fMRI timeseries into high-motion and low-motion halves based on framewise displacement
• Trait-FC Effect Calculation: Compute correlation structure differences between split halves
• Motion Impact Scoring:
  • Motion Overestimation Score: Motion impact direction aligns with trait-FC effect direction
  • Motion Underestimation Score: Motion impact direction opposes trait-FC effect direction
• Statistical Validation: Permutation testing with non-parametric combining across connections

Application:

• Traits Assessed: 45 demographic, cognitive, and mental health measures from ABCD Study
• Statistical Threshold: p < 0.05 for significant motion impact scores
• Censoring Comparison: Evaluate motion impacts with and without strict censoring (FD < 0.2 mm)

Quantitative Efficacy Metrics

Variance Reduction Performance

Large-scale validation demonstrates significant motion artifact reduction through ABCD-BIDS processing:

Table 2: Motion-Related Variance Reduction Metrics

Processing Stage	Variance Explained by Motion	Relative Reduction	Sample Size
Minimal Processing (Motion correction only)	73%	Baseline	n = 9,652
ABCD-BIDS Full Denoising	23%	69%	n = 9,652

The pipeline achieves substantial motion variance reduction through its integrated denoising approach, combining global signal regression, respiratory filtering, motion parameter regression, and spectral filtering [28].

Impact on Functional Connectivity

Residual motion effects persist despite comprehensive denoising:

Table 3: Motion Effects on Functional Connectivity

Metric	Value	Interpretation
Correlation between motion-FC effect matrix and average FC matrix	Spearman ρ = -0.58	Strong negative relationship
Motion-FC correlation after FD < 0.2 mm censoring	Spearman ρ = -0.51	Persistent systematic bias
Traits with significant motion overestimation (no censoring)	42% (19/45 traits)	Substantial false positive risk
Traits with significant motion underestimation (no censoring)	38% (17/45 traits)	Effect masking concerns
Traits with significant motion overestimation (FD < 0.2 mm)	2% (1/45 traits)	Censoring reduces overestimation

The data demonstrate that motion systematically decreases long-distance connectivity, with stronger associations observed in participants with higher motion, even after denoising [28].

Trait-Specific Motion Impacts

Motion disproportionately affects traits correlated with head movement:

High-Risk Traits: Attention measures, psychiatric symptoms (ADHD, autism features) Impact Mechanism: Participants with these traits exhibit higher in-scanner motion, creating systematic bias in trait-FC relationships [28]

Visualization of Processing Workflows

ABCD-BIDS Processing Pipeline

ABCD-BIDS Processing Stages

Motion Impact Assessment (SHAMAN)

SHAMAN Motion Impact Assessment

The Scientist's Toolkit: Research Reagent Solutions

Table 4: Essential Research Tools for ABCD-BIDS Validation

Tool/Resource	Function	Application in Validation
ABCD-BIDS Community Collection (ABCC)	Curated, BIDS-standardized dataset	Provides quality-controlled input data for pipeline validation
SHAMAN (Split Half Analysis of Motion Associated Networks)	Trait-specific motion impact quantification	Measures residual motion effects on brain-behavior associations
Framewise Displacement (FD)	Head motion quantification	Standardized metric for motion censoring and quality control
Global Signal Regression (GSR)	Nuisance regression component	Reduces motion effects and eliminates scanner-related batch effects
Respiratory Motion Filter	Artifact correction in motion estimates	Removes respiratory-induced motion artifacts from realignment data
BrainSwipes Platform	Crowdsourced quality control	Enables visual quality assessment of pipeline derivatives
Gordon Atlas (333 ROI)	Functional parcellation template	Standardized network definition for connectivity analysis
Power Atlas (264 ROI)	Functional parcellation template	Alternative network definition for connectivity analysis

Discussion and Implementation Guidelines

Interpretation of Efficacy Metrics

The demonstrated 69% relative reduction in motion-related variance confirms the substantial efficacy of ABCD-BIDS denoising, yet the persistent motion-FC correlations highlight the necessity for complementary motion mitigation strategies. Researchers should interpret brain-behavior associations with caution, particularly for motion-correlated traits, and implement rigorous motion censoring thresholds (FD < 0.2 mm) to minimize spurious findings.

Recommended Implementation Protocol

For optimal motion denoising performance:

• Apply Respiratory Motion Filtering: Essential for accurate FD estimation in multi-band data
• Implement Strict Censoring: FD < 0.2 mm threshold significantly reduces motion overestimation effects
• Validate Trait-Specific Motion Impacts: Utilize SHAMAN methodology for high-motion-correlated traits
• Leverage Automated QC Metrics: Incorporate population-based outlier detection for systematic quality assessment

The ABCD-BIDS pipeline represents a robust, standardized approach for large-scale functional connectivity analysis, with quantified efficacy metrics supporting its utility for developmental neuroimaging research.

In-scanner head motion is a significant source of artifact in functional magnetic resonance imaging (fMRI) data, particularly for resting-state functional connectivity (FC) studies. Motion artifacts can systematically bias results, especially when studying populations prone to higher motion, such as children, older adults, or individuals with certain neurological or psychiatric conditions [3]. The Adolescent Brain Cognitive Development (ABCD) Study—one of the largest longitudinal studies of child brain development—addresses this challenge through the ABCD-BIDS preprocessing pipeline, a standardized approach for mitigating motion artifacts in neuroimaging data [5] [10]. This pipeline incorporates methods from both the Human Connectome Project's (HCP) minimal preprocessing pipeline and the DCAN Labs resting-state fMRI analysis tools [5]. Understanding the efficacy and specific components of ABCD-BIDS relative to other denoising approaches is crucial for researchers interpreting ABCD data and selecting appropriate preprocessing strategies for their specific research questions.

The ABCD-BIDS Pipeline: Architecture and Motion Denoising Components

The ABCD-BIDS pipeline is a BIDS App that processes MRI data through a series of sequential stages, outputting preprocessed data in both volume and surface spaces [5]. Its motion denoising strategy is not confined to a single step but is rather distributed across multiple specialized stages.

Key Denoising Stages in ABCD-BIDS

• Stage 4: FMRIVolume: This stage initiates the functional processing pipeline. It corrects for gradient-nonlinearity-induced distortions in the EPI (Echo Planar Imaging) data and performs rigid-body motion correction by aligning each volume in the time series to the initial frame using FSL FLIRT. The motion parameters generated (translations and rotations along each axis and their derivatives) form the foundation for subsequent denoising steps. Furthermore, it employs FSL's topup with a pair of spin-echo EPI scans having opposite phase-encoding directions to correct for distortions in the phase encoding direction of each fMRI volume [5].

• Stage 6: DCANBOLDProcessing (DBP): This is the core signal processing stage for motion denoising, comprising four broad steps [5]:

  • Standard Pre-processing: The fMRI data is de-meaned and de-trended. A general linear model (GLM) is then applied for denoising, which includes regressing out signal variables (mean time series for white matter, cerebrospinal fluid, and the global signal) and movement variables (the estimated motion parameters and their Volterra expansion). The pipeline explicitly includes Global Signal Regression (GSR), which has been shown to reduce motion effects and eliminate batch impacts on group comparisons [5].
  • Respiratory Motion Filter: To address a respiratory artifact found in multi-band ABCD data, an optional filter removes specific frequency bands (18.582 to 25.726 breaths per minute) from the motion realignment data. This produces more accurate estimates of framewise displacement (FD), preventing inappropriate motion censoring [5] [24].
  • Motion Censoring: Frames exceeding an FD threshold of 0.3 mm are labeled as "bad" and are removed during the demeaning and detrending phases of the standard pre-processing. The denoising GLM betas are calculated using only the "good" frames. The pipeline also generates comprehensive temporal masks for FD thresholds from 0 to 0.5 mm in 0.01 mm steps, facilitating post-hoc analysis [5].
  • Generation of Parcellated Time Series: The cleaned data is used to extract time series for predefined brain atlases (e.g., Gordon, Power, Yeo), enabling the construction of correlation matrices for network analysis [5].

Integrated Anatomical Processing

The anatomical processing stages (PreFreeSurfer, FreeSurfer, PostFreeSurfer) also contribute indirectly to robust functional denoising. By producing high-quality anatomical segmentations, brain masks, and surface reconstructions, these stages ensure that the functional data is accurately aligned and mapped, thereby improving the specificity of nuisance signal regression from tissues like white matter and CSF [5].

Table 1: Core Motion Denoising Components of the ABCD-BIDS Pipeline

Pipeline Stage	Component/Tool	Primary Denoising Function	Key Parameters
FMRIVolume	FSL FLIRT	Rigid-body motion correction	6 degrees-of-freedom
FMRIVolume	FSL topup	Distortion correction in phase encoding direction	Paired spin-echo field maps
DCANBOLDProcessing	General Linear Model (GLM)	Nuisance variable regression	WM/CSF/Global signal, motion parameters
DCANBOLDProcessing	Global Signal Regression (GSR)	Mitigation of motion-related and global artifacts	Mandatory inclusion
DCANBOLDProcessing	Respiratory Motion Filter	Removes respiratory artifact from motion estimates	18.582 - 25.726 breaths per minute
DCANBOLDProcessing	Motion Censoring	Removal of high-motion frames ("scrubbing")	FD threshold: 0.3 mm (default)

Quantitative Efficacy and Comparative Performance

Evaluations of the ABCD-BIDS pipeline and other denoising strategies reveal their relative effectiveness in mitigating motion artifacts and preserving biological signals.

Motion Reduction Efficacy

The ABCD-BIDS pipeline achieves a substantial reduction in motion-related variance. One study reported that after processing with ABCD-BIDS (which includes respiratory filtering, motion timeseries regression, and despiking), only 23% of the signal variance was explained by head motion. This represents a 69% relative reduction compared to the 73% of variance explained by motion after only minimal processing (motion-correction by frame realignment only) [3].

Impact on Trait-FC Relationships and the Role of Censoring

Even after comprehensive denoising, residual motion can bias trait-functional connectivity relationships. The SHAMAN (Split Half Analysis of Motion Associated Networks) method has been used to quantify this "motion impact score" on specific trait-FC relationships within ABCD data [3].

• Without motion censoring: After standard ABCD-BIDS denoising, 42% (19/45) of examined traits showed significant motion overestimation scores, while 38% (17/45) showed significant underestimation scores [3].
• With additional censoring (FD < 0.2 mm): The number of traits with significant motion overestimation scores was drastically reduced to 2% (1/45). However, this stringent censoring did not reduce the number of traits with significant underestimation scores [3].

This highlights a critical trade-off: while censoring effectively controls for false positives due to motion overestimation, it may not address, and could even exacerbate, biases leading to effect underestimation. It also risks biasing sample distributions by systematically excluding high-motion individuals [3].

Comparison with Other Denoising Pipelines

No single pipeline universally excels at both motion mitigation and the enhancement of brain-behavior associations. A 2025 comparative study found that pipelines combining ICA-AROMA (Independent Component Analysis - Automatic Removal Of Motion Artifacts) and GSR offered a reasonable trade-off between these two objectives [65]. The optimal denoising strategy can also be population-dependent; for instance, ICA-AROMA-based strategies may be most effective for non-lesional encephalopathic conditions, while anatomical Component Correction (aCompCor) strategies might be superior for lesional conditions like brain tumors [66].

Table 2: Comparative Performance of fMRI Denoising Pipelines

Denoising Pipeline / Strategy	Reported Motion Reduction	Impact on Brain-Behavior Prediction	Notable Strengths & Limitations
ABCD-BIDS (with GSR & resp. filter)	69% relative variance reduction [3]	Data not provided in search results	Integrated, standardized; performance varies by trait.
ICA-AROMA + GSR	Data not provided	Reasonable trade-off between motion reduction and prediction [65]	Balanced performance across objectives.
Minimal Processing (Motion correction only)	Baseline (73% variance explained by motion) [3]	Presumably poor due to high residual noise	Inadequate for most research applications.
Motion Censoring (FD < 0.2 mm) post-ABCD-BIDS	Eliminates most overestimation artifacts [3]	May introduce bias by excluding high-motion subjects [3]	Highly effective for specific biases; risk of sample bias.

Experimental Protocols for Motion Denoising Evaluation

For researchers seeking to implement or evaluate these methods, the following protocols detail standard operational procedures.

Protocol 1: Executing the ABCD-BIDS Pipeline

This protocol outlines the core command for running the full ABCD-BIDS pipeline on a BIDS-formatted dataset [24].

• Software Requirement: The pipeline is run as a Docker container, requiring Docker to be installed on the host system.
• Data Preparation: Ensure the input dataset is properly formatted according to the BIDS standard. Acquire and locally save a FreeSurfer license.
• Execution Command:

• Key Options:
  • --participant-label: To process only specific participants.
  • --stage: To restart the pipeline from a specific stage in case of interruption.
  • --anat-only: To run only the anatomical pipeline [12].

Protocol 2: Assessing Motion Impact with SHAMAN

This protocol describes the methodology for applying the SHAMAN approach to quantify trait-specific motion impact, as implemented in a recent study [3].

• Input Data Requirement: Preprocessed resting-state fMRI data and phenotypic measures for the traits of interest.
• Data Splitting: For each participant, split the preprocessed fMRI timeseries into high-motion and low-motion halves based on the Framewise Displacement (FD) timecourse.
• Connectivity Calculation: Compute separate functional connectivity matrices for the high-motion and low-motion halves for each participant.
• Trait-FC Effect Estimation: Calculate the correlation between each trait and FC strength for every connection (edge) in the network, separately for the high-motion and low-motion halves.
• Statistical Testing and Score Calculation:
  • For each trait-edge pair, compute the difference in trait-FC effects between the high-motion and low-motion halves.
  • Use permutation testing (e.g., shuffling the timeseries) and non-parametric combining across edges to generate a single motion impact score (and p-value) for the effect of motion on each trait-FC relationship.
  • A score aligned with the trait-FC effect direction indicates motion overestimation; a score in the opposite direction indicates motion underestimation.

Visualization and Workflow Diagrams

ABCD-BIDS Motion Denoising Workflow

The following diagram illustrates the sequential flow of data through the key stages of the ABCD-BIDS pipeline that are dedicated to motion denoising, from raw data input to the final denoised output.

Diagram Title: ABCD-BIDS Motion Denoising Workflow

For researchers working with the ABCD-BIDS pipeline or developing comparative motion denoising analyses, the following tools and data resources are indispensable.

Table 3: Essential Research Resources for Motion Denoising Analysis

Resource Name	Type	Primary Function in Research	Access/Reference
ABCD-BIDS Community Collection (ABCC)	Data Resource	Provides standardized, BIDS-formatted ABCD Study data and derivatives for analysis and method benchmarking.	NBDC Data Hub [10]
FSL (FMRIB Software Library)	Software Library	Provides critical tools used within ABCD-BIDS, including topup for distortion correction and FLIRT for motion realignment.	https://fsl.fmrib.ox.ac.uk/fsl/fslwiki [5]
FreeSurfer	Software Suite	Provides anatomical processing pipeline for brain segmentation and surface reconstruction, integrated into ABCD-BIDS.	https://surfer.nmr.mgh.harvard.edu/ [5]
SHAMAN Analysis Code	Analytical Method	Enables quantification of motion impact scores on specific trait-functional connectivity relationships.	Nielsen et al., 2025 [3]
DCAN Labs Executive Summary	Quality Control Tool	Generates an HTML QC report visualizing T1w/T2w segmentation, atlas registration, and movement time series.	Included in ABCD-BIDS pipeline [5]
ICA-AROMA	Denoising Tool	Provides an alternative denoising strategy based on independent component analysis for motion artifact removal.	https://github.com/maartenmennes/ICA-AROMA [65] [66]

Residual motion effects refer to the systematic biases that persist in functional Magnetic Resonance Imaging (fMRI) data even after the application of standard denoising pipelines. Within the context of the ABCD-BIDS preprocessing pipeline, these effects represent a significant challenge for researchers studying brain-behavior relationships, particularly in populations prone to higher in-scanner movement, such as children, older adults, and individuals with psychiatric or neurological disorders [28]. Despite rigorous denoising procedures like those implemented in the ABCD-BIDS community collection (ABCC), residual motion artifact can lead to both overestimation and underestimation of trait-functional connectivity (trait-FC) associations, potentially resulting in false positive or false negative findings [28]. This application note details the quantitative impact, detection methodologies, and mitigation strategies for residual motion effects, providing a structured framework for researchers and drug development professionals to enhance the robustness of their neuroimaging findings.

Quantitative Evidence of Residual Motion Effects

Empirical evidence from large-scale studies demonstrates that residual motion remains a potent source of artifact after standard denoising. Analyses of the Adolescent Brain Cognitive Development (ABCD) Study data, which utilizes the ABCD-BIDS denoising algorithm, reveal the pervasive nature of this problem.

Table 1: Impact of Residual Motion on Functional Connectivity (FC) After ABCD-BIDS Denoising

Metric	Value Before Denoising	Value After ABCD-BIDS Denoising	Notes
Signal Variance Explained by Motion	73% [28]	23% [28]	69% relative reduction, but substantial variance remains
Correlation (Motion-FC effect vs Avg FC)	-0.58 [28]	-0.51 (after FD < 0.2mm censoring) [28]	Strong negative correlation indicates decreased long-distance connectivity with more motion
Traits with Significant Motion Overestimation	Not Applicable	42% (19/45 traits) [28]	Before additional motion censoring
Traits with Significant Motion Underestimation	Not Applicable	38% (17/45 traits) [28]	Before additional motion censoring

The data in Table 1 underscores a critical finding: residual motion introduces systematic bias that is often larger than the trait-FC effects of interest [28]. The motion-FC effect matrix shows a strong, negative correlation with the average FC matrix, meaning participants who moved more tended to show weaker connection strengths across the brain. This effect was so pronounced that the largest motion-FC effect size for a single connection was larger than the largest trait-FC effect size [28].

Table 2: Impact of Motion Censoring on Trait-FC Associations

Censoring Threshold (Framewise Displacement)	Traits with Significant Overestimation	Traits with Significant Underestimation
No Censoring (Post-ABCD-BIDS only)	42% (19/45) [28]	38% (17/45) [28]
FD < 0.2 mm	2% (1/45) [28]	38% (17/45) [28]

As shown in Table 2, while stringent motion censoring (FD < 0.2 mm) effectively reduces false positives due to motion overestimation, it does not address the problem of motion-induced underestimation of true effects. This highlights a fundamental limitation of relying solely on censoring as a mitigation strategy.

Protocols for Detecting and Quantifying Residual Motion

The SHAMAN Framework for Trait-Specific Motion Impact

The Split Half Analysis of Motion Associated Networks (SHAMAN) is a novel method designed to assign a motion impact score to specific trait-FC relationships [28].

Experimental Protocol:

• Data Input: Processed resting-state fMRI timeseries for each participant, along with their trait measures and framewise displacement (FD) timeseries.
• Timeseries Splitting: For each participant, split the fMRI timeseries into two halves: a "high-motion" half (volumes with higher FD) and a "low-motion" half (volumes with lower FD).
• Connectivity Calculation: Calculate separate functional connectivity matrices for the high-motion and low-motion halves for each participant.
• Trait-FC Effect Estimation: Compute the correlation between a participant's trait score and the connectivity strength for each network edge (connection) separately in the high-motion and low-motion halves.
• Motion Impact Score Calculation:
  • A significant difference in trait-FC effects between the two halves indicates a motion impact.
  • A motion impact score aligned with the direction of the trait-FC effect is classified as a "motion overestimation score."
  • A motion impact score opposite the direction of the trait-FC effect is classified as a "motion underestimation score."
• Statistical Significance: Permutation testing and non-parametric combining across connections are used to generate a p-value for the motion impact score.

The following diagram illustrates the logical workflow and output interpretation of the SHAMAN framework:

Data-Driven Scrubbing as an Alternative to Motion Censoring

Projection scrubbing is a data-driven method that flags individual fMRI volumes (timepoints) displaying abnormal patterns, offering an alternative to motion-based censoring [58].

Experimental Protocol:

• Dimension Reduction: Apply a strategic dimension reduction technique, such as Independent Component Analysis (ICA), to the fMRI data to isolate major sources of variance.
• Outlier Detection: Within this reduced subspace, use a statistical outlier detection framework (e.g., based on robust Mahalanobis distance) to identify volumes that are significant multivariate outliers.
• Flagging Volumes: Flag these outlier volumes as potential artifacts. The method only censors volumes that display statistically abnormal patterns, avoiding the exclusion of data based solely on an arbitrary head motion threshold.
• Validation Metric: The success of scrubbing should be measured by its ability to maximize data retention while maintaining or improving the validity, reliability, and identifiability (fingerprinting) of functional connectivity [58].

Advanced Mitigation Strategies and Inclusive Analysis

Motion-Ordering and Bagging for Inclusive Sampling

Strict motion exclusion criteria disproportionately discard data from minoritized populations, such as Black and Hispanic youth, who often exhibit greater in-scanner head motion [67]. Motion-ordering and bagging are resampling techniques designed to retain these participants.

Experimental Protocol:

• Scrubbing: Identify and remove motion-corrupted timepoints (FD > 0.20 mm) and their adjacent volumes (one preceding, two succeeding) [67].
• Minimum Timepoint (minTP) Matching: Rank the remaining scrubbed timepoints by their FD values. For each participant, select the top N timepoints (e.g., minTP = 100) with the lowest FD to create a standardized, motion-limited timeseries [67].
• Motion-Ordering: Compute the functional connectivity matrix directly from the minTP-matched timepoints.
• Bagging (Motion-Ordering + Resampling):
  • From the minTP-matched timepoints, generate a large number (e.g., 500) of bootstrapped samples (sampling with replacement).
  • For each bootstrap sample, compute a functional connectivity matrix.
  • Generate a final, bagged functional connectivity matrix for the participant by averaging all bootstrap matrices [67].
• Analysis: Use the motion-ordered or bagged connectivity matrices to compute brain-behavior associations, thereby including data from high-motion participants who would otherwise be excluded.

Emerging Acquisition and Processing Technologies

Table 3: Advanced Methods for Motion Mitigation

Method	Function	Key Advantage
Multi-Echo (ME) Acquisition [68]	Acquires fMRI data at multiple echo times; allows optimal combination and separation of BOLD from non-BOLD signals.	Increases temporal signal-to-noise ratio (tSNR) and split-half reliability of functional connectivity.
Thermal Denoising (NORDIC) [68]	Removes zero-mean, unstructured thermal noise from fMRI data.	Improves SNR without sacrificing spatial precision, beneficial for precision functional mapping.
Deep Learning Residual Correction [69] [70]	A 3D CNN learns the residual error between motion-affected quantitative maps and their motion-free references.	Provides retrospective motion correction for quantitative MRI parameters (e.g., T1, T2), improving image quality.

The Scientist's Toolkit: Research Reagent Solutions

Table 4: Essential Tools for Motion Denoising Research

Research Reagent	Function in Motion Research	Example Use Case
Framewise Displacement (FD) [28]	Quantifies head motion from volume-to-volume realignment parameters.	Primary metric for motion censoring (e.g., flag volumes with FD > 0.2 mm).
ABCD-BIDS Pipeline [10] [5]	Standardized preprocessing pipeline for ABCD Study data, includes motion parameter regression, global signal regression, and respiratory filtering.	Baseline denoising to remove gross motion artifacts before specialized analyses.
SHAMAN Algorithm [28]	Computes a trait-specific motion impact score to quantify over-/under-estimation of brain-behavior associations.	Diagnosing whether a significant trait-FC finding is likely driven by residual motion.
Projection Scrubbing [58]	Data-driven outlier detection method to flag artifact-contaminated fMRI volumes.	An alternative to motion-based censoring that avoids excessive data loss.
Gordon/Fan Atlas [67]	A predefined brain parcellation scheme (e.g., 333 or 352 regions) used to extract regional timeseries.	Constructing whole-brain functional connectivity matrices for network-based analysis.
Motion-Ordered Resampling (Bagging) [67]	Bootstrap aggregation method to generate reliable connectivity estimates from motion-limited data.	Maximizing participant inclusivity and representation in population neuroscience studies.

Residual motion is a formidable confound that persists after standard denoising with pipelines like ABCD-BIDS. It can inflate (overestimate) or mask (underestimate) true brain-behavior associations, threatening the validity of neuroimaging findings. Based on the current evidence, the following protocols are recommended:

• Go Beyond Standard Denoising: Do not assume standard preprocessing has fully removed motion artifacts. Implement additional, targeted strategies like data-driven scrubbing [58] or trait-specific motion impact analysis [28].
• Adopt Inclusive Analysis Techniques: To mitigate selection bias and increase the generalizability of results, employ methods like motion-ordering and bagging to incorporate data from high-motion participants, who are often from underrepresented groups [67].
• Leverage Advanced Acquisition: Where feasible, utilize multi-echo acquisitions combined with thermal denoising (e.g., NORDIC) to achieve higher signal quality and reliability from the outset, reducing the impact of motion and other artifacts [68].
• Quantify, Don't Assume: Use frameworks like SHAMAN to quantitatively evaluate the impact of residual motion on the specific trait-FC relationships under investigation, rather than making assumptions about its influence [28].

By integrating these strategies, researchers can produce more robust, reproducible, and inclusive brain-wide association studies, advancing the translation of neuroimaging findings into clinical and drug development applications.

In-scanner head motion represents the most substantial source of artifact in functional magnetic resonance imaging (fMRI) signals, introducing systematic bias into resting-state functional connectivity (FC) measurements that persists despite denoising algorithms [3]. This technical challenge is particularly acute in studies examining traits inherently associated with motion, such as psychiatric disorders, where researchers require robust methods to determine whether observed trait-FC relationships reflect genuine neural correlates or motion-induced artifacts [3]. The development of Split Half Analysis of Motion Associated Networks (SHAMAN) provides a methodological framework for assigning motion impact scores to specific trait-FC relationships, distinguishing between motion causing overestimation or underestimation of trait-FC effects [3]. Within the context of ABCD-BIDS preprocessing pipeline research, understanding and quantifying these motion impacts is essential for avoiding false positive results in brain-wide association studies (BWAS) involving thousands of participants [3] [10].

The problem of motion artifact is especially relevant for functional connectivity analysis because resting-state fMRI is particularly vulnerable to motion effects due to the unknown timing of underlying neural processes [3]. Motion systematically alters fMRI data through spatially specific patterns, notably decreasing long-distance connectivity while increasing short-range connectivity, with pronounced effects in the default mode network [3]. Early studies of clinical populations including children, older adults, and patients with neurological or psychiatric disorders have demonstrated spurious findings linked to motion, such as erroneous conclusions that autism decreases long-distance FC when the results actually reflected increased head motion in autistic participants [3]. These findings have motivated extensive methodological developments in motion mitigation, yet residual motion artifacts continue to pose challenges for neuroimaging research, particularly in large-scale studies like the Adolescent Brain Cognitive Development (ABCD) Study [3] [10].

The SHAMAN Methodology

Theoretical Foundation

The SHAMAN method capitalizes on a crucial observation regarding the temporal stability of traits versus the dynamic nature of motion: traits such as cognitive abilities or demographic characteristics remain stable over the timescale of an MRI scan, while motion represents a state that varies from second to second [3]. This fundamental difference enables the detection of motion-related artifacts by examining differences in correlation structure between split high- and low-motion halves of each participant's fMRI timeseries [3]. When trait-FC effects operate independently of motion, the difference in connectivity between high- and low-motion halves will be non-significant due to trait stability over time. A statistically significant difference emerges only when state-dependent motion variations impact the trait's connectivity pattern [3].

The directionality of the motion impact score provides critical information about the nature of the bias. A motion impact score aligned with the direction of the trait-FC effect indicates motion causing overestimation of the trait-FC effect, resulting in a "motion overestimation score." Conversely, a motion impact score opposite to the direction of the trait-FC effect suggests motion causing underestimation of the trait-FC effect, termed a "motion underestimation score" [3]. Through permutation of the timeseries and non-parametric combining across pairwise connections, SHAMAN generates a motion impact score with an associated p-value that distinguishes significant from non-significant motion impacts on trait-FC effects [3].

Workflow and Implementation

Table 1: Key Stages in SHAMAN Motion Impact Analysis

Stage	Description	Output
Data Preparation	Application of ABCD-BIDS denoising with optional motion censoring	Preprocessed fMRI timeseries
Timeseries Splitting	Division of each participant's data into high-motion and low-motion halves based on framewise displacement	Paired high-motion and low-motion datasets
Trait-FC Analysis	Calculation of trait-functional connectivity relationships within each motion half	Trait-FC effect matrices for high and low motion conditions
Motion Impact Scoring	Comparison of trait-FC effects between high and low motion halves	Motion impact scores with directionality (overestimation/underestimation)
Statistical Significance Testing	Permutation testing and non-parametric combining across connections	p-values for motion impact scores
Interpretation	Classification of significant motion effects based on directionality	Identification of traits vulnerable to motion bias

The SHAMAN workflow operates on one or more resting-state fMRI scans per participant and can be adapted to incorporate covariates in the analytical model [3]. Implementation begins with standard denoising procedures, such as those provided by the ABCD-BIDS pipeline, which includes global signal regression, respiratory filtering, spectral filtering, despiking, and motion parameter timeseries regression [3] [5]. The method then proceeds with motion censoring at user-defined thresholds, typically using framewise displacement (FD) metrics, though SHAMAN provides flexibility in censoring stringency [3].

Workflow Visualization

SHAMAN Motion Impact Analysis Workflow

Quantitative Assessment in ABCD Study

Prevalence of Motion Bias

Application of SHAMAN to 45 traits from n = 7,270 participants in the Adolescent Brain Cognitive Development (ABCD) Study revealed substantial prevalence of motion-related biases even after standard denoising procedures [3]. The findings demonstrate that motion artifacts systematically affect trait-functional connectivity relationships across a broad spectrum of measures, with concerning implications for brain-wide association studies [3].

Table 2: Motion Impact on Traits in ABCD Study After Standard Denoising

Motion Impact Type	Percentage of Traits Affected	Number of Traits (out of 45)	Primary Characterization
Significant Overestimation	42%	19/45	Motion inflates observed trait-FC effects
Significant Underestimation	38%	17/45	Motion obscures genuine trait-FC effects
No Significant Impact	20%	9/45	Trait-FC effects robust to motion artifacts

The data reveal that a majority of traits (80%) exhibited significant motion impacts after standard denoising with ABCD-BIDS without motion censoring, highlighting the pervasive nature of residual motion artifact in large-scale neuroimaging studies [3]. These effects were not random but displayed systematic patterns of either exaggerating or diminishing observed brain-behavior relationships.

Impact of Motion Censoring

Motion censoring, the process of excluding high-motion fMRI frames (timepoints) from analysis, represents a post-hoc approach to reduce residual motion artifact [3]. The effectiveness of censoring depends critically on the selected framewise displacement threshold, with different thresholds exhibiting distinct effects on overestimation versus underestimation biases [3].

Table 3: Effect of Motion Censoring on Motion Impact Scores

Censoring Threshold (Framewise Displacement)	Traits with Significant Overestimation	Traits with Significant Underestimation	Key Findings
No Censoring	42% (19/45)	38% (17/45)	Baseline condition after ABCD-BIDS denoising
FD < 0.2 mm	2% (1/45)	38% (17/45)	Dramatically reduces overestimation but does not address underestimation
FD < 0.3 mm	Not reported in results	Not reported in results	Standard threshold in DCANBoldProcessing for defining "bad" frames [5]

The differential impact of censoring on overestimation versus underestimation biases reveals a complex relationship between motion artifact and trait-FC effect estimation. While stringent censoring at FD < 0.2 mm effectively addresses overestimation biases, reducing significant overestimation from 42% to just 2% of traits, it fails to ameliorate underestimation biases, which persist at the same rate (38%) observed without censoring [3]. This finding highlights the nuanced nature of motion artifacts and suggests that different mechanisms may underlie overestimation versus underestimation effects.

Experimental Protocols

SHAMAN Implementation Protocol

Objective: To implement the Split Half Analysis of Motion Associated Networks (SHAMAN) for quantifying trait-specific motion impact scores in resting-state fMRI data.

Materials and Software Requirements:

• ABCD-BIDS pipeline (or comparable preprocessing tools)
• Framewise displacement (FD) calculation utilities
• Permutation testing framework
• Statistical analysis software (R, Python, or MATLAB)

Procedure:

• Data Preprocessing: Process raw fMRI data through the ABCD-BIDS pipeline, which includes:
  • Gradient nonlinearity distortion correction
  • Motion correction via rigid-body registration
  • Susceptibility distortion correction using FSL's topup with spin-echo field maps
  • Brain extraction and tissue segmentation
  • MNI space registration [5] [24]

• Denoising Application: Apply comprehensive denoising through DCANBOLDProcessing (DBP) stage, including:

  • Global signal regression (GSR) to reduce motion effects
  • Nuisance regression of white matter, CSF, and motion parameters
  • Band-pass filtering (0.008-0.09 Hz) using 2nd order Butterworth filter
  • Optional respiratory motion filter (18.582-25.726 breaths per minute) to improve FD estimates [5] [24]

• Motion Censoring: Generate motion censoring masks using framewise displacement metrics:

  • Calculate FD from translational (X, Y, Z) and rotational (roll, pitch, yaw) parameters
  • Apply censoring thresholds from 0-0.5 mm in 0.01 mm steps
  • Implement outlier detection to exclude frames >2 standard deviations from mean [5]

• Timeseries Splitting: For each participant:

  • Calculate median framewise displacement across the entire scan
  • Assign individual frames to high-motion or low-motion categories based on median split
  • Ensure equal number of frames in each half to maintain statistical power

• Trait-FC Analysis:

  • Extract parcellated timeseries using standardized atlases (Gordon's 333 ROI, Power's 264 ROI, Yeo's 118 ROI, or HCP's 360 ROI)
  • Compute correlation matrices for each motion half separately
  • Calculate trait-FC effects by correlating trait measures with FC strengths

• Motion Impact Scoring:

  • Compare trait-FC effects between high-motion and low-motion halves
  • Determine directionality of effect differences
  • Calculate motion impact scores using non-parametric combining across connections

• Statistical Testing:

  • Perform permutation testing (recommended: 10,000 permutations)
  • Adjust for multiple comparisons using false discovery rate (FDR) correction
  • Classify significant effects as overestimation or underestimation biases

Quality Control:

• Visual inspection of Executive Summary outputs from ABCD-BIDS pipeline
• Verify successful surface reconstruction and registration
• Check motion parameter distributions for outliers
• Confirm adequate frame retention after censoring (>8 minutes of data recommended) [3] [5]

Motion Censoring Optimization Protocol

Objective: To determine optimal framewise displacement thresholds for minimizing both overestimation and underestimation biases in trait-FC analyses.

Procedure:

• Threshold Sweep: Process data across a range of FD thresholds (0.1-0.5 mm in 0.01 mm increments)
• SHAMAN Application: Apply SHAMAN methodology at each threshold level
• Impact Assessment: Record the proportion of traits showing significant overestimation and underestimation at each threshold
• Optimal Threshold Identification: Select threshold that minimizes both types of bias while retaining sufficient data for analysis
• Validation: Verify optimal threshold in independent sample or through cross-validation

Interpretation: Based on ABCD Study findings, researchers should anticipate that lower thresholds (FD < 0.2 mm) will effectively control overestimation bias but may not address underestimation bias [3]. The optimal threshold may vary depending on the specific traits under investigation and the motion characteristics of the study population.

The Scientist's Toolkit

Table 4: Essential Research Reagents and Computational Tools

Tool/Reagent	Function	Implementation Notes
ABCD-BIDS Pipeline	Integrated structural and functional MRI preprocessing	Utilizes HCP minimal preprocessing + DCAN Labs tools; outputs volume and surface data [5]
Framewise Displacement (FD)	Quantitative motion metric	Derived from translational/rotational head motion parameters; critical for censoring decisions [3]
Global Signal Regression (GSR)	Denoising technique	Reduces motion effects; combined with motion censoring represents best practice for motion mitigation [5]
Respiratory Motion Filter	Artifact correction	Filters respiratory frequencies (18.582-25.726 bpm) from motion realignment data; improves FD estimates [5] [24]
DCANBOLDProcessing (DBP)	Signal processing software	Performs nuisance regression, motion censoring, and band-pass filtering; generates parcellated timeseries [5]
SHAMAN Algorithm	Motion impact quantification	Provides trait-specific motion impact scores with directionality (overestimation/underestimation) [3]
Permutation Testing Framework	Statistical inference	Non-parametric approach for significance testing of motion impact scores [3]

The SHAMAN methodology represents a significant advancement in quantifying and characterizing motion-related biases in functional connectivity analyses, specifically addressing the critical distinction between overestimation and underestimation effects [3]. Applied within the ABCD-BIDS preprocessing framework, this approach demonstrates that residual motion artifact persists even after comprehensive denoising and systematically impacts a substantial proportion of trait-FC relationships [3]. The differential effects of motion censoring on overestimation versus underestimation biases reveal the complex nature of motion artifacts and underscore the importance of trait-specific motion impact assessment in neuroimaging research.

Based on the empirical findings from the ABCD Study, the following recommendations emerge for researchers utilizing the ABCD-BIDS pipeline or similar preprocessing frameworks:

• Implement Motion Impact Assessment: Incorporate SHAMAN or comparable methodologies as a standard quality control procedure, particularly for studies examining traits potentially correlated with motion propensity.

• Adopt Stringent Censoring for Overestimation Control: Utilize framewise displacement thresholds of FD < 0.2 mm when concerned about false positive findings due to motion-induced overestimation of effects.

• Recognize Censoring Limitations: Acknowledge that motion censoring effectively addresses overestimation but may not resolve underestimation biases, potentially leading to false negative conclusions.

• Report Motion Impact Transparencyly: Document and report motion impact scores for key trait-FC relationships to enhance reproducibility and appropriate interpretation of findings.

• Consider Trait-Motion Relationships: Exercise particular caution when studying traits known to correlate with motion propensity (e.g., attention measures, psychiatric symptoms) and implement appropriate methodological safeguards.

As neuroimaging continues to advance through large-scale collaborative efforts like the ABCD Study, rigorous attention to motion-related biases remains essential for generating valid and reproducible brain-behavior associations. The SHAMAN framework within the ABCD-BIDS processing environment provides researchers with critical tools for identifying and addressing these pervasive methodological challenges.

Within the framework of the ABCD-BIDS preprocessing pipeline, managing noise and artifacts is paramount for generating high-quality functional and diffusion MRI data. Motion-related artifacts and thermal noise present significant challenges, particularly for developmental neuroimaging where data acquisition is often constrained. The integration of Multi-Echo (ME) acquisitions and NORDIC PCA denoising represents a methodological advance that substantially improves signal quality and reliability. These techniques are especially valuable for precision functional mapping (PFM) studies aiming to characterize individual-specific brain organization, as they enhance temporal signal-to-noise ratio (tSNR) and functional connectivity reliability without sacrificing spatial precision [60]. This application note details the protocols and quantitative benefits of incorporating these emerging methods into the ABCD-BIDS preprocessing workflow, providing researchers with practical guidelines for implementation.

Multi-Echo Acquisitions for Enhanced fMRI

Core Principles and Physiological Basis

Multi-Echo (ME) acquisition is an fMRI technique that captures images at multiple echo times (TEs) during a single readout of the T2* signal decay. The T2-weighted fMRI signal decays exponentially over successive echoes following the radiofrequency pulse. Crucially, T2 relaxation times vary across age, brain regions, and tissue types due to underlying neurobiological properties such as myelination, iron concentration, and proton density [60]. This biological variability means that a single echo time is suboptimal for all voxels. ME acquisition addresses this by allowing optimal combination of echoes based on the T2* of the underlying tissue, thereby maximizing the contrast-to-noise ratio for BOLD effects [60].

Technical Implementation and Protocol

Implementing ME acquisitions within the ABCD-BIDS framework requires specific sequence parameters and processing steps. The following protocol is adapted from successful implementations in developmental populations:

• Sequence Parameters: Acquire 3-5 echoes with TEs spanning the expected T2* range for the population of interest. For adult populations, common TEs range from 12-35 ms. For infant populations, longer TEs may be beneficial due to slower T2* decay related to developmental factors including reduced myelination and different water compartmentalization [60].
• Optimal Echo Combination: Use the tedana (TE-dependent analysis) pipeline to combine echoes weighted by their TE and estimated T2*. This combination improves tSNR and BOLD contrast compared to single-echo acquisition [60].
• ME-ICA Denoising: Apply Multi-Echo Independent Component Analysis (ME-ICA) to separate BOLD effects from non-BOLD artifacts based on their distinct TE-dependence. This method effectively identifies and removes non-BOLD components without requiring external physiological monitoring [60] [71].
• ABCD-BIDS Integration: Process ME data through the modified FMRIVolume stage of the ABCD-HCP pipeline, with subsequent ME-ICA denoising prior to the DCANBOLDProcessing stage for nuisance regression [5].

Table 1: Quantitative Benefits of Multi-Echo Acquisitions Across Developmental Stages

Population	tSNR Improvement	Connectivity Reliability	Optimal TE Range	Key Considerations
Adults	Significant increase	Major improvement	Standard (~12-35 ms)	Well-established protocol
Children	Significant increase	Major improvement	Intermediate	Longer T2* than adults
Infants	Moderate increase	Moderate improvement	Extended	Substantially longer, more variable T2*; requires further optimization

Developmental Considerations

The benefits of ME acquisitions are particularly relevant for developmental neuroimaging, though implementation requires special considerations. Infant brains exhibit slower T2* decay compared to older children and adults, related to ongoing myelination and changes in tissue composition [60]. This necessitates longer echo times for optimal signal capture. Furthermore, ME acquisitions show promise for longitudinal investigations where individually optimized echo combinations based on T2* times can be determined for each acquisition time point while maintaining the same acquisition sequence [60].

NORDIC PCA Denoising for Enhanced SNR

Theoretical Foundation

NOise Reduction with DIstribution Corrected (NORDIC) PCA is a advanced denoising technique that addresses the challenge of thermal noise in MRI data. Unlike conventional denoising methods that may remove structured signal along with noise, NORDIC employs low-rank modeling of g-factor-corrected complex-valued MRI data and non-asymptotic random matrix distributions to selectively remove signal components that cannot be distinguished from thermal noise [72] [73]. This approach preserves biologically relevant signal while effectively suppressing thermal noise, making it particularly valuable for high-resolution imaging where SNR is inherently limited.

Application Protocol Across Modalities

The implementation of NORDIC varies slightly depending on the imaging modality but follows the same core principles:

For Functional MRI

• Data Requirements: NORDIC requires complex-valued (magnitude and phase) data for optimal performance. If only magnitude data is available, a modified implementation can be applied with reduced efficacy [74].
• Processing Pipeline: Incorporate NORDIC after reconstruction but before main pipeline processing. For the ABCD-BIDS workflow, apply after the FMRIVolume stage but before FMRISurface and DCANBOLDProcessing stages [5] [74].
• Parameter Settings: Use default NORDIC parameters for most applications, as the method is designed to be parameter-free. The algorithm automatically determines the optimal cutoff between signal and thermal noise components [72].
• Quality Assessment: Verify denoising efficacy by comparing tSNR before and after processing. Expect substantial tSNR improvements, particularly in high-resolution acquisitions [74].

For Diffusion MRI

• Implementation: Apply NORDIC to the complex-valued dMRI data during reconstruction. The method leverages the high dimensionality of dMRI data (multiple diffusion directions and weightings) to distinguish signal from noise [72] [75].
• Performance: NORDIC demonstrates up to 6-fold improvement in apparent SNR for 0.9mm whole-brain dMRI at 3T, substantially enhancing quantitative performance for estimating diffusion tractography and resolving crossing fibers compared to conventional denoising methods like MP-PCA [72].
• Integration with Motion Correction: When using Eddy for dMRI head motion correction, apply NORDIC denoising beforehand to improve registration target generation, though note that this may affect noise distribution assumptions in the Gaussian Process model [75].

Table 2: NORDIC Denoising Efficacy Across MRI Modalities

Modality	SNR/tSNR Improvement	Impact on Quantitative Metrics	Data Requirements	Recommended Stage in ABCD-BIDS
fMRI	>3x increase in rodents; significant in humans	Enhanced functional connectivity reliability	Complex-valued preferred	Post-FMRIVolume, pre-FMRISurface
dMRI	Up to 6x apparent SNR improvement	Improved fiber orientation estimation, crossing fiber resolution	Complex-valued required	Pre-head motion correction
Combined with ME	Additive benefit with ME-ICA	Superior to either technique alone	Multi-echo complex data	After echo combination, before ME-ICA

Integrated Implementation for ABCD-BIDS Pipeline

Synergistic Integration Framework

The combination of ME acquisitions and NORDIC denoising produces additive benefits for data quality. When implemented together within the ABCD-BIDS pipeline, these methods create a powerful preprocessing stream that maximizes signal fidelity:

• Acquire Multi-Echo Data: Collect fMRI data with multiple echo times appropriate for the target population.
• Apply NORDIC Denoising: Process the complex-valued ME data with NORDIC to reduce thermal noise while preserving signal.
• Optimal Echo Combination: Use T2*-weighted combination to generate a single time series with enhanced tSNR.
• ME-ICA Denoising: Remove non-BOLD components using the TE-dependence of signals.
• Standard ABCD-HCP Processing: Continue through FMRISurface, DCANBOLDProcessing, and subsequent stages of the established pipeline [5] [60].

This integrated approach has demonstrated superior performance compared to either technique alone, particularly for precision functional mapping where both spatial precision and reliability are crucial [60].

Complementary Denoising Strategy

For optimal results, combine ME-ICA with additional denoising methods:

• Post-ME-ICA Processing: After ME-ICA, apply anatomical Component Based Correction (aCompCor) to remove spatially diffuse noise [71].
• Alternative Option: The aggressive option of ICA-AROMA also provides effective denoising for ME data, though it may remove more signal of interest compared to the ME-ICA + aCompCor combination [71].

Research Reagent Solutions

Table 3: Essential Tools for Implementing Advanced Denoising Methods

Tool/Resource	Function	Application Context	Accessibility
tedana	Multi-echo data analysis and echo combination	Critical for ME fMRI processing	Open-source Python package
NORDIC Implementation	Thermal noise removal from complex-valued MRI data	fMRI and dMRI denoising	Code available from original developers
ABCD-HCP BIDS Pipeline	Core processing framework for structural and functional MRI	Foundation for integrating new denoising methods	Docker containers available
3dSHORE/SHORELine	Head motion correction for non-shelled dMRI schemes	Alternative to Eddy for DSI/CS-DSI data	BSD 3-Clause license
MP-PCA (dwidenoise)	Conventional dMRI denoising	Baseline comparison for NORDIC performance	Part of MRtrix3 package
ICA-AROMA	Automatic removal of motion artifacts via ICA	Comparison/combination with ME-ICA	Standalone tool compatible with major pipelines

Workflow Visualization

Integrated ME and NORDIC Denoising Workflow

The integration of Multi-Echo acquisitions and NORDIC PCA denoising within the ABCD-BIDS preprocessing pipeline represents a significant advancement for motion denoising research. These complementary methods address distinct noise sources—ME acquisitions leverage TE-dependent BOLD effects to separate signal from artifacts, while NORDIC selectively removes thermal noise without disturbing structured signal. For researchers investigating individual differences in functional brain organization, particularly in challenging developmental populations, these techniques offer enhanced tSNR and reliability that can reduce the scanning time needed for precision functional mapping. Implementation within the established ABCD-BIDS framework ensures compatibility with existing analytic workflows while providing substantial improvements in data quality for both functional and diffusion MRI applications.

Conclusion

The ABCD-BIDS preprocessing pipeline represents a sophisticated, validated approach for addressing the pervasive challenge of motion artifact in resting-state fMRI data, particularly within large-scale studies like the ABCD cohort. Through its integrated denoising strategy—combining respiratory filtering, motion regression, and despiking—it achieves substantial reduction in motion-related variance. However, residual motion effects persist even after processing, necessitating careful implementation of supplementary strategies like motion censoring at FD < 0.2 mm and trait-specific impact assessment using frameworks like SHAMAN. Future directions include integration of emerging methodologies such as multi-echo acquisitions and NORDIC thermal noise removal, which show promise for further enhancing signal quality. For biomedical and clinical research, rigorous motion denoising is not merely a technical preprocessing step but a fundamental requirement for ensuring the validity of brain-behavior associations and advancing toward reproducible biomarkers in neuropsychiatric drug development.

References

• 1. Dissimilarity of functional uncovers the influence of... connectivity [https://pmc.ncbi.nlm.nih.gov/articles/PMC7814752/]
• 2. Mitigating head artifact in motion functional connectivity MRI [https://www.nature.com/articles/s41596-018-0065-y?error=cookies_not_supported&code=8f78a509-2a86-49a5-a350-7504e0f899aa]
• 3. Motion impact score for detecting spurious brain-behavior ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12479937/]
• 4. Prospective motion correction of fMRI: Improving the quality ... [https://www.sciencedirect.com/science/article/pii/S1053811920300811]
• 5. ABCD-HCP BIDS fMRI Pipeline [https://abcd-hcp-pipeline.readthedocs.io/]
• 6. Prospective Motion Correction of Functional MRI Using ... [https://archive.ismrm.org/2025/1438_W7IylzR2c.html]
• 7. Real-time multislice-to-volume motion correction for task- ... [https://pubmed.ncbi.nlm.nih.gov/41173218/]
• 8. Quality control in resting-state fMRI: the benefits of visual ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10192849/]
• 9. The past, present, and future of the brain imaging data ... - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC11415029/]
• 10. ABCD-BIDS community collection [https://docs.abcdstudy.org/v/6_0_0/documentation/imaging/abcc_start_page.html]
• 11. Brain asymmetry and its association with inattention ... [https://www.nature.com/articles/s41398-025-03327-1]
• 12. DCAN-Labs/infant-abcd-bids-pipeline [https://github.com/DCAN-Labs/infant-abcd-bids-pipeline]
• 13. Poor image quality introduces systematic bias into MRI data [https://www.thetransmitter.org/spectrum/poor-image-quality-introduces-systematic-bias-into-large-neuroimaging-datasets/]
• 14. Reproducible brain-wide association studies require ... [https://www.nature.com/articles/s41586-022-04492-9]
• 15. Motion impact score for detecting spurious brain-behavior ... [https://pubmed.ncbi.nlm.nih.gov/41022827/]
• 16. Balancing Data Quality and Bias: Investigating Functional ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11717557/]
• 17. Representing brain-behavior associations by retaining ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12322147/]
• 18. MRI Quality control [https://docs.abcdstudy.org/latest/documentation/imaging/type_qc.html]
• 19. Usage - ABCD-HCP BIDS fMRI Pipeline [https://abcd-hcp-pipeline.readthedocs.io/en/latest/usage/]
• 20. A practical guide for researchers and reviewers using the ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9156875/]
• 21. Image processing and analysis methods for the Adolescent ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC6981278/]
• 22. MRI derivatives data documentation [https://docs.abcdstudy.org/latest/documentation/imaging/data_derivatives.html]
• 23. Using FMRIPREP on (somewhat) preprocessed data (ABCD) [https://neurostars.org/t/using-fmriprep-on-somewhat-preprocessed-data-abcd/3397]
• 24. Data processing [https://docs.abcdstudy.org/latest/documentation/imaging/abcc_pipeline.html]
• 25. Quality control procedures [https://docs.abcdstudy.org/latest/documentation/imaging/abcc_qualitycontrol.html]
• 26. Data Sharing [https://abcdstudy.org/scientists/data-sharing/]
• 27. Motion artifact in studies of functional connectivity [https://pmc.ncbi.nlm.nih.gov/articles/PMC5930165/]
• 28. Motion impact score for detecting spurious brain-behavior ... [https://www.nature.com/articles/s41467-025-63661-2]
• 29. DCAN-Labs/dcan_bold_processing [https://github.com/DCAN-Labs/dcan_bold_processing]
• 30. The Global Signal in fMRI: Nuisance or Information? - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC5406229/]
• 31. Towards a consensus regarding global signal regression ... [https://www.sciencedirect.com/science/article/pii/S1053811916306711]
• 32. Global signal regression reduces connectivity patterns ... [https://www.frontiersin.org/journals/neuroimaging/articles/10.3389/fnimg.2025.1653206/full]
• 33. XCP-D: A Robust Pipeline for the post-processing of fMRI ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10690221/]
• 34. XCP-D : A Robust Postprocessing Pipeline of fMRI data [https://xcp-d.readthedocs.io/en/0.5.0/]
• 35. fMRIPrep Lifespan: Extending A Robust Pipeline for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12132292/]
• 36. Correction of respiratory artifacts in MRI head motion estimates [https://pmc.ncbi.nlm.nih.gov/articles/PMC7307712/]
• 37. Fieldmaps recognition - Software Support [https://neurostars.org/t/fieldmaps-recognition/34495]
• 38. MASILab/SynBOLD-DisCo: Synthetic BOLD images for ... [https://github.com/MASILab/SynBOLD-DisCo]
• 39. DeepPrep: an accelerated, scalable and robust pipeline for ... [https://www.nature.com/articles/s41592-025-02599-1]
• 40. TOPUP - FSL - FMRIB Software Library [https://fsl.fmrib.ox.ac.uk/fsl/docs/diffusion/topup/index.html]
• 41. Frequently asked questions about TOPUP [https://fsl.fmrib.ox.ac.uk/fsl/docs/diffusion/topup/FAQ/index.html]
• 42. FSL - FMRIB Software Library [https://fsl.fmrib.ox.ac.uk/]
• 43. Using synthetic MR images for distortion correction [https://www.sciencedirect.com/science/article/pii/S1878929323000397]
• 44. The Brain Imaging Data Structure: BIDS [https://bids.neuroimaging.io/index.html]
• 45. Motion-BIDS: an extension to the brain imaging data ... [https://www.nature.com/articles/s41597-024-03559-8]
• 46. Prospective Motion Correction in Functional MRI - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC5427003/]
• 47. Motion - Brain Imaging Data Structure 1.10.1 [https://bids-specification.readthedocs.io/en/stable/modality-specific-files/motion.html]
• 48. BIDS validation - The Brain Imaging Data Structure [https://bids.neuroimaging.io/tools/validator.html]
• 49. bids-standard/bids-validator: Validator for the Brain ... [https://github.com/bids-standard/bids-validator]
• 50. Data processing [https://docs.abcdstudy.org/v/6_0_0/documentation/imaging/abcc_pipeline.html]
• 51. FMRIPREP susceptibility distortion correction-Question ... [https://neurostars.org/t/fmriprep-susceptibility-distortion-correction-question-about-field-map-voxel-and-image-dimensions/2640]
• 52. Missing information JSON files - dcm2bids [https://neurostars.org/t/missing-information-json-files/30427]
• 53. DCAN-Labs/nhp-abcd-bids-pipeline-synth [https://github.com/DCAN-Labs/nhp-abcd-bids-pipeline-synth]
• 54. mne_bids.update_sidecar_json — MNE-BIDS 0.17.0 ... [https://mne.tools/mne-bids/stable/generated/mne_bids.update_sidecar_json.html]
• 55. Usage — fmriprep version documentation [https://fmriprep.org/en/1.2.0/usage.html]
• 56. Real-time feedback reduces participant motion during task ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9882243/]
• 57. A Multi-Dataset Evaluation of Frame Censoring for Motion ... [https://www.humanbrainmapping.org/files/Aperture%20Neuro/HTML%20files/12_1070_A_Multi-Dataset_Evaluation_of_Frame_Censoring.html]
• 58. Less is more: balancing noise reduction and data retention ... [https://www.sciencedirect.com/science/article/pii/S1053811923001180]
• 59. Correction of Physiological Artifacts in Multi‐Echo fMRI Data ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12183604/]
• 60. Multi-echo Acquisition and Thermal Denoising Advances ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10634909/]
• 61. The tedana roadmap - Read the Docs [https://tedana.readthedocs.io/en/latest/roadmap.html]
• 62. Reducing Inter-Individual Differences in Task fMRI ... [https://link.springer.com/article/10.1007/s12021-025-09741-6]
• 63. Continuous evaluation of denoising strategies in resting ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10977879/]
• 64. Motion impact score for detecting spurious brain-behavior ... [https://www.semanticscholar.org/paper/Motion-impact-score-for-detecting-spurious-Kay-Montez/4e77d974b5cd25ea1684ee288936485c8a57c0e4]
• 65. The efficacy of resting-state fMRI denoising pipelines for ... [https://pubmed.ncbi.nlm.nih.gov/40800742/]
• 66. Denoising strategies of functional connectivity MRI data in ... [https://pubmed.ncbi.nlm.nih.gov/41190330/]
• 67. Increasing the representation of minoritized youth for inclusive ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11230295/]
• 68. Multi-echo acquisition and thermal denoising advances ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12319974/]
• 69. Learning residual motion correction for fast and robust 3D ... [https://www.sciencedirect.com/science/article/pii/S1361841522000391]
• 70. Residual learning for 3D motion corrected quantitative MRI [https://proceedings.mlr.press/v143/pirkl21a.html]
• 71. Comparing the efficacy of data-driven denoising methods ... [https://pubmed.ncbi.nlm.nih.gov/37669723/]
• 72. NOise reduction with DIstribution Corrected (NORDIC) PCA in dMRI with complex-valued parameter-free locally low-rank processing - ScienceDirect [https://www.sciencedirect.com/science/article/pii/S1053811920310247]
• 73. NOise reduction with DIstribution Corrected (NORDIC) PCA in dMRI with complex-valued parameter-free locally low-rank processing - PubMed [https://pubmed.ncbi.nlm.nih.gov/33186723/]
• 74. NOise Reduction with DIstribution Corrected (NORDIC) PCA improves signal-to-noise in rodent resting-state and optogenetic functional MRI - PubMed [https://pubmed.ncbi.nlm.nih.gov/36086476/]
• 75. Diffusion MRI head motion correction methods are highly ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10826632/]