The Entropy of Task-Induced Centrality Dynamics in Schizophrenia: Novel Applications of Graph Theory

1816 

Abstract Submission 

Dhruval Bhatt¹, John Kopchick², Dalal Khatib², Patricia Thomas², Usha Rajan², Luay Haddad², Alireza Amirsadri², Jeffrey Stanley², Vaibhav Diwadkar¹

¹Wayne State University, Detroit, MI, ²Wayne State University, Department of Psychiatry, Detroit, MI

Dhruval Bhatt  
Wayne State University
Detroit, MI

John Kopchick  
Wayne State University, Department of Psychiatry
Detroit, MI

Dalal Khatib  
Wayne State University, Department of Psychiatry
Detroit, MI

Patricia Thomas  
Wayne State University, Department of Psychiatry
Detroit, MI

Usha Rajan  
Wayne State University, Department of Psychiatry
Detroit, MI

Luay Haddad  
Wayne State University, Department of Psychiatry
Detroit, MI

Alireza Amirsadri  
Wayne State University, Department of Psychiatry
Detroit, MI

Jeffrey Stanley  
Wayne State University, Department of Psychiatry
Detroit, MI

Vaibhav Diwadkar  
Wayne State University
Detroit, MI

Both resting and task-evoked brain network dynamics are impaired in schizophrenia (SCZ)(Meram et al., 2023). These dynamics have been explored using Dynamic Functional Connectivity (DFC) primarily applied to resting state fMRI signals (Hutchison et al., 2013). Given that task-induced changes are the most substantive modulators of brain network states (Logothetis, 2008), we introduce a novel framework for the study of centrality dynamics, thereby yoking DFC with graph theory (Rubinov and Sporns, 2010). First, we drove brain networks in SCZ and heathy controls (HC) using an established learning paradigm (Stanley et al., 2017). Next, using a moving window technique (the window width corresponded to the width of each task epoch) we computed a series of stationary functional connectivity matrices (one for each of the 280 partially overlapping windows). Each connectivity matrix (246 functionally derived nodes)(Fan et al., 2016) captures the state of the system in that time window. Next, in each window we estimated the Betweenness Centrality (BC)(Freeman, 1977) of each of the 246 nodes before ranking them (Spearman's rank) for integrative importance. Finally, a 280 point time series (based on each node's rank in each time window) was formed for each node. Each time series (tBC) captures that node's centrality dynamics across the task. After estimating cross-similarities between tBC, agglomerative clustering was used to cluster regions with similar centrality dynamics. From the analyses of the clustering solutions (and the recovered centrality dynamics in each cluster) we demonstrate a) that the clustering identified different sub-networks in SCZ and HC and b) the centrality dynamics of sub-networks in SCZ were characterized by higher entropy (irregularity).

fMRI data (Siemens Verio 3T) were collected while participants (n=88, 49 SCZ, ages:18-45) learned nine object-location associations. Learning occurred over eight iterations of a paradigm where each iteration consisted of four distinct epochs (each 27s): Encoding (objects location shown for naming) , Post-Encoding Consolidation (instruction-free fixation), Retrieval (locations cued in random order with participants required to name the associate object), and Post-Retrieval Consolidation (instruction-free fixation). fMRI data were preprocessed using typical methods (SPM12). Within each group (HC and SCZ), the 246 tBC were averaged across the respective participants before estimating the cross similarity matrix in centrality dynamics across all 246 nodes (30,135 pairs). Next, the matrix was submitted for agglomerative hierarchical clustering (Ward, 1963)(Figure 1) to cluster regions based on similarities in their centrality dynamics.

Two distinct cluster solutions were observed (Figure 1; five clusters in HC and three in individuals with SCZ) with the averaged time series in each cluster (tCluster) evincing distinct centrality dynamics. Next, we calculated the Approximate Entropy (ApE), over each tCluster to characterize the degree of stochasticity of the cluster's centrality dynamics. ApE is an index of the irregularity of the fluctuations of a time series (Bonal & Marshak, 2019). As seen in Figure 2a, the ApE of each tCluster in the three SCZ clusters were higher than in the five HC clusters. The ApE was also linked to indices relating to network flexibility (Clark & Bjørnstad, 2004) (Figure 2b) and clinical symptoms (Figure 2c).

There are many avenues for studying brain dynamics (Heitmann and Breakspear, 2018). Here we derive the concept of centrality dynamics, a new measure that encapsulates (task-driven or spontaneous) changes in the cumulative functional connectome over time. As shown, these dynamics are more stochastic in schizophrenia patients, and this increased stochasticity was linked to salient clinical symptoms. Overall, our findings further highlight the importance of studying dysfunctional network dynamics in complex psychiatric conditions like schizophrenia.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ²

Connectivity (eg. functional, effective, structural)

fMRI Connectivity and Network Modeling ¹

FUNCTIONAL MRI

Learning

Psychiatric Disorders

Schizophrenia

Other - Graph Theory and Dynamic Functional Connectivity