A method for inclusion of high-motion underrepresented youth for robust brain-behaviour relationship

1699 

Abstract Submission 

Jivesh Ramduny¹, Lucina Uddin^2,3,4, Tamara Vanderwal^5,6, Eric Feczko^7,8, Damien Fair^7,8,9, Clare Kelly^10,11,12, Arielle Baskin-Sommers^1,13,14

¹Department of Psychology, Yale University, New Haven, CT, USA, ²Department of Psychiatry and Biobehavioral Sciences, University of California Los Angeles, Los Angeles, CA, USA, ³Department of Psychology, University of California Los Angeles, Los Angeles, CA, USA, ⁴Semel Institute for Neuroscience and Human Behavior, University of California Los Angeles, Los Angeles, CA, USA, ⁵Department of Psychiatry, University of British Columbia, Vancouver, BC, Canada, ⁶BC Children’s Hospital Research Institute, Vancouver, BC, Canada, ⁷Masonic Institute for the Developing Brain, University of Minnesota Medical School, Minneapolis, MN, USA, ⁸Department of Pediatrics, University of Minnesota Medical School, Minneapolis, MN, USA, ⁹Institute of Child Development, University of Minnesota Medical School, Minneapolis, MN, USA, ¹⁰School of Psychology, Trinity College Dublin, Dublin, Ireland, ¹¹Department of Psychiatry, School of Medicine, Trinity College Dublin, Dublin, Ireland, ¹²Trinity College Institute of Neuroscience, Trinity College Dublin, Dublin, Ireland, ¹³Department of Psychiatry, Yale University, New Haven, CT, USA, ¹⁴Wu Tsai Institute, Yale University, New Haven, CT, USA

Jivesh Ramduny  
Department of Psychology, Yale University
New Haven, CT, USA

Lucina Uddin  
Department of Psychiatry and Biobehavioral Sciences, University of California Los Angeles|Department of Psychology, University of California Los Angeles|Semel Institute for Neuroscience and Human Behavior, University of California Los Angeles
Los Angeles, CA, USA|Los Angeles, CA, USA|Los Angeles, CA, USA

Tamara Vanderwal  
Department of Psychiatry, University of British Columbia|BC Children’s Hospital Research Institute
Vancouver, BC, Canada|Vancouver, BC, Canada

Eric Feczko  
Masonic Institute for the Developing Brain, University of Minnesota Medical School|Department of Pediatrics, University of Minnesota Medical School
Minneapolis, MN, USA|Minneapolis, MN, USA

Damien Fair  
Masonic Institute for the Developing Brain, University of Minnesota Medical School|Department of Pediatrics, University of Minnesota Medical School|Institute of Child Development, University of Minnesota Medical School
Minneapolis, MN, USA|Minneapolis, MN, USA|Minneapolis, MN, USA

Clare Kelly  
School of Psychology, Trinity College Dublin|Department of Psychiatry, School of Medicine, Trinity College Dublin|Trinity College Institute of Neuroscience, Trinity College Dublin
Dublin, Ireland|Dublin, Ireland|Dublin, Ireland

Arielle Baskin-Sommers  
Department of Psychology, Yale University|Department of Psychiatry, Yale University|Wu Tsai Institute, Yale University
New Haven, CT, USA|New Haven, CT, USA|New Haven, CT, USA

Consortia datasets provide an unprecedented opportunity for the neuroimaging community to estimate the true effect sizes of brain-behaviour associations with sufficient statistical power. However, in part due to the variation in data acquisition across consortia sites, researchers tend to apply strict data quality control to mitigate the impact of head motion. This practice typically excludes a disproportionate number of participants who belong to minioritised backgrounds and from lower socioeconomic status [1,2]. Here, we proposed a resampling strategy known as bagging [3] to rescue high-motion underrepresented youths to maximise sample inclusivity for establishing robust and inclusive brain-behaviour associations from the Adolescent Brain Cognitive Development (ABCD) Study [4].

Structural and resting-state fMRI data were obtained from the ABCD Study Baseline Release (9-10 years). Participants were categorised into 3 racial/ethnic groups: White (N=3,864), Black (N=910), and Hispanic (N=1,342). Asian participants were not included due to their limited sample size. Standard data preprocessing was performed using the ABCD-BIDS Community Collection [5]. For each youth, the fMRI data was parcellated [5,6] and functional connectivity (FC) was computed between each pair of ROIs to construct a 352x352 FC matrix. Partial Spearman's Rank correlation (R) was computed at the edge level to assess the relationships between FC and externalising behaviours [CBCL] and cognitive performance [NIH Toolbox] while adjusting for participant sex and head motion. For each behaviour, the edge showing the strongest R was selected after correcting for multiple comparisons (~61K edges; P < 0.05 FDR).

To establish the robustness of R across sample sizes [7], N was bootstrapped across 500 iterations, and the mean R and 95% CI was plotted as a function of N. Scrubbing was applied to estimate the FC matrices based on the participants' least motion-corrupted timepoints whose framewise displacement (FD) < 0.20 mm [8,9] (Fig.1A). Bagging was then performed to select subsets of the participants' least motion-corrupted timepoints with replacement over 500 iterations and compute their FC matrices [3,10] (Fig.1A). Finally, bagging was repeated without discarding the high-motion participants based on their motion-limited timepoints [3]. To include as many youth as possible, a minimum of 100 least motion-corrupted timepoints were bootstrapped for bagging.

Underrepresented youth exhibited significantly greater head motion relative to White youth (Fig.1B&C). Low-motion youth were retained (mean FD < 0.20 mm). This disproportionally reduced the size of the Black and Hispanic groups by 49.1% and 52.5%, respectively. The variability in R decreased with increasing sample sizes using a standard procedure (full timeseries; 33K correlations; Fig.2A) and bagging (16.5 million correlations; Fig.2B). The differences in CIs (computed by AUC) between the two approaches were 0.2-21% for NIH Toolbox and 2-5% for CBCL, with bagging producing tighter CIs around R in most cases. Bagging retained 99.97% of all White, 99.78% Black, and 100% Hispanic youth for brain-behaviour relationships (Fig.2C). When the high-motion youth were included, the differences in CIs between the two approaches were 2-5% for NIH Toolbox and 0.2-2% for CBCL, with bagging producing tighter CIs in almost all correlations.

Bagging shows merit in (1) maximising sample inclusivity by permitting inclusion of fMRI data from high-motion underrepresented youth, and (2) generating robust and inclusive brain-behaviour relationships from those who typically are excluded using traditional low-motion criteria. Bagging enhances the representation of individuals and serves to fulfil the promise of consortia data to produce generalisable effect sizes across races and ethnicities.

Normal Brain Development: Fetus to Adolescence

fMRI Connectivity and Network Modeling ¹

Motion Correction and Preprocessing

Univariate Modeling ²

BOLD fMRI

Development

FUNCTIONAL MRI

Univariate

Other - head motion, brain-behaviour, inclusivity