Atypical joint density distribution of resting state dynamic spatial brain networks in Schizophrenia

2353 

Abstract Submission 

Krishna Pusuluri¹, Armin Iraji², Vince Calhoun³

¹GEORGIA STATE UNIVERSITY, Dunwoody, GA, ²Georgia State University, Atlanta, GA, ³GSU/GATech/Emory, Decatur, GA

Krishna Pusuluri  
GEORGIA STATE UNIVERSITY
Dunwoody, GA

Armin Iraji  
Georgia State University
Atlanta, GA

Vince Calhoun  
GSU/GATech/Emory
Decatur, GA

Resting state functional magnetic resonance imaging (rsfMRI) studies in schizophrenia (SZ) usually focus on spatially static networks over the course of a scan. Recent studies addressed the spatial expansion or shrinkage (Iraji et al. 2020) and variations in coupling between spatially dynamic networks via a model of the spatial chronnectome (Iraji et al. 2019). In this work, we investigated spatially dynamic brain networks, their voxel-wise changes over time, and the joint density distributions of pairs of networks using 2D histograms, clustered across time windows. 2D histograms allow for the comparison of two networks, counting the occurrence of various combinations of voxel-level intensities/activities. We found that several clusters of 2D histograms across brain network pairs show significant group differences in occupancy time between healthy controls (CN) and SZ group.

rsfMRI data for 508 subjects with 315 controls (CN) and 193 SZ patients were obtained from three datasets – FBIRN (Damaraju et al. 2014), COBRE (Aine et al. 2017), and MPRC (Adhikari et al. 2019) – and preprocessed using SPM12 toolbox as described in (Iraji et al. 2022). We performed spatial independent component analysis (sICA) at the group level using the GIFT software package (Iraji et al. 2021) and identified 14 large-scale brain networks. The prior networks from group-level analysis were used as a reference for a spatially constrained ICA (integrated in GIFT as Multivariate Objective Optimization ICA with Reference) (Du et al. 2013) for each subject across sliding windows to ensure the correspondence of brain networks across subjects and time windows (window size is 30 times TR, the repetition time). For each brain network pair (see Fig.1), 2D histograms were computed for z-scored voxel-level activity at each window for each subject and clustered across all the subjects and windows using k-means algorithm (after subtracting the overall mean 2D histogram from the window-level data).

Clustering results reveal changes in the joint density heat maps for several network pairs across subjects and time windows. We identified several clusters of 2D histograms that show significant group differences (with 2-sample t-tests) in subject-wise cluster occupancy time (defined as the ratio given by the number of windows per subject that fall within the cluster, divided by the total number of windows), revealing atypical joint density distributions of dynamic spatial brain network pairs in Schizophrenia. For 91 possible network pairs chosen across 14 brain networks, with 10 k-means clusters per network pair, we find that 290 of the clusters (out of a maximum of 91*10=910) show significant group differences after a 5% false discovery rate correction. A few of those clusters for the network pair cerebellar (CER) vs. secondary somatomotor (MTR-S) are shown in Fig.1.

This work underscores the importance of studying spatiotemporally dynamic behaviors within/across brain networks and could lead to the development of novel biomarkers for brain function and dysfunction.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ²

Activation (eg. BOLD task-fMRI)

fMRI Connectivity and Network Modeling

BOLD fMRI ¹

FUNCTIONAL MRI

Psychiatric Disorders

Schizophrenia