Dynamic Functional Network Connectivity States, Head Motion, and Mental Health Symptoms in Children
Poster No:
1825
Submission Type:
Abstract Submission
Authors:
Donovan Roediger1, Andrea Wiglesworth1, Bonnie Klimes-Dougan1, Mark Fiecas1, Monica Luciana1, Zening Fu2, Vince Calhoun2, Bryon Mueller1, Kathryn Cullen1
Institutions:
1University of Minnesota, Minneapolis, MN, 2GSU/GATech/Emory, Atlanta, GA
First Author:
Co-Author(s):
Introduction:
Dynamic functional connectivity analysis is a promising tool for identifying fMRI biomarkers of mental health disorders but is vulnerable to artificial correlations introduced by in-scanner motion (Savaa, 2020). In pediatric imaging, head motion is a common problem and investigations may be further complicated when motion is itself correlated with other measures of interest. Here, we used a sliding window method and k-means clustering algorithm to explore fMRI-derived dynamic "states" and their relationship with head motion and clinical correlates in data from a large-scale adolescent study.
Methods:
Using the Adolescent Brain and Cognitive Development dataset, we analyzed baseline data from 11201 participants (ages 9-11) with available data from the Child Behavior Checklist (CBCL), the Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS), and at least one resting-state scan. Raw fMRI data (first run) were downloaded and preprocessed using previously described methods (Fu, 2023). We then used Neuromark, an ICA-based hybrid framework within the Group ICA of fMRI Toolbox, to derive time series for 53 IC-based ROIs (Du, 2020; Iraji, 2021). These time series were further denoised by applying 6-parameter motion regression, detrending, despiking, and bandpass filtering (0.01 - 0.15Hz). A sliding window approach (40 volumes) was used to calculate dynamic functional network connectivity (dFNC) matrices. Finally, we used a k-means clustering algorithm (k=4) to identify recurring dFNC patterns ("states") across all windows and participants.
Initial assessments correlated mean framewise displacement (FD) with the percent of volumes spent in each state to determine the extent of residual motion-related effects that remained after preprocessing and the relationship between mean FD and clinical measures of interest. GLMs were then used to assess how the percentage of scan time spent in each of the four states relates to binary lifetime suicidal ideation (SI; either caregiver or child-report) as well as internalizing, externalizing, and total problems from the CBCL (caregiver report). In addition to time spent in each state, the number of state switch events during the scan was explored as a predictor of clinical outcomes.
Initial assessments correlated mean framewise displacement (FD) with the percent of volumes spent in each state to determine the extent of residual motion-related effects that remained after preprocessing and the relationship between mean FD and clinical measures of interest. GLMs were then used to assess how the percentage of scan time spent in each of the four states relates to binary lifetime suicidal ideation (SI; either caregiver or child-report) as well as internalizing, externalizing, and total problems from the CBCL (caregiver report). In addition to time spent in each state, the number of state switch events during the scan was explored as a predictor of clinical outcomes.
Results:
Cluster centroids for the four identified dFNC states are shown in Figure 1. Despite efforts to remove motion artifacts, residual motion appeared to heavily influence the proportion of scan time spent in each state, with higher movers spending more time in States 1 and 4 and lower movers spending more time in States 2 and 3 (Figure 2). Positive correlations were found between mean FD and all clinical measures (r's ≈ 0.1). To account for the confounding effects of motion, all GLMs were adjusted for mean FD. Patterns observed across clinical outcome measures were consistent, where time spent in States 1 and 4 was weakly but consistently associated with higher CBCL scores (more symptoms) and lifetime SI while spending more time in States 2 and 3 was associated with lower CBCL scores and lifetime SI. After Bonferroni correction, only the relationships between time spent in States 1 through 4 and the CBCL "total problems" score remained significant (p's < 0.0025). The frequency of transitions between states was unrelated to any clinical measures.
Conclusions:
Motion events appeared to drive windows toward particular dFNC states despite preprocessing methods implemented to limit the impact of motion. This reinforces previous observations that dFNC analyses are especially vulnerable to motion and that steps should be taken during both preprocessing and analysis to identify and control for any motion-related effects (Abrol, 2017). Notably, we also observed significant relationships between dFNC and clinical measures that were robust to this motion, highlighting the utility of dFNC clustering analyses in exploring relationships between fMRI and mental health in children.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 2
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Motion Correction and Preprocessing
Keywords:
FUNCTIONAL MRI
PEDIATRIC
Psychiatric Disorders
1|2Indicates the priority used for review
Provide references using author date format
Du, Y. et al. (2020), 'NeuroMark: An automated and adaptive ICA based pipeline to identify reproducible fMRI markers of brain disorders', NeuroImage: Clinical, vol. 28, p. 102375.
Fu, Z. et al. (2023), 'Functional connectivity uniqueness and variability? Linkages with cognitive and psychiatric problems in children'. Nature Mental Health, pp. 1–15.
Iraji, A. et al. (2021), 'Tools of the trade: Estimating time-varying connectivity patterns from fMRI data', Social Cognitive and Affective Neuroscience, vol. 16(8), pp. 849–874.
Savva, A. D. et al. (2020), 'Effects of motion related outliers in dynamic functional connectivity using the sliding window method'. Journal of Neuroscience Methods, vol. 330, p. 108519.