New Zealand Statistical Association 2024 Conference

Khan Buchwald

Auckland University of Technology

Comparisons between functional brain networks for schizophrenia and healthy case controls: A Bayesian network approach

This is joint work with Matthieu Vignes, Richard Siegert, Ajit Narayanan, Margaret Sandham

The dysconnection hypothesis of schizophrenia has been advanced to explain its symptomology from a neurophysiological perspective. Network statistical methods have been used to assess the dysconnection hypothesis but not when varying network complexity. This paper leverages network statistical complexity analysis to evaluate the dysconnection hypothesis in schizophrenia. This study obtained fMRI data from the University of California Los Angles Consortium for Neuropsychiatric Phenomics LA5c Study for people diagnosed with schizophrenia (PDS) and healthy case-controls (HC). We used one hundred bootstrap samples of 45 PDS and 45 HC for encoding and retrieval memory trials and fitted a dynamic Bayesian network to each sample. The network properties were assessed at various significance thresholds, and an optimal significance threshold for an edge’s inclusion was obtained. After edge pruning, 52.8% of edges were shared between memory encoding in PDS and memory encoding in HC, and 52.7% between retrieval in PDS and retrieval in HC. This was considerably lower than the shared connections within treatment groups. PDS had a similar number of functional connections as HC. The network metrics (Clustering, shortest path length) were higher or lower in PDS and varied according to the threshold selected. Dynamic Bayesian networks can be used to support the dysconnection hypothesis in PDS, and this research suggests that the absence of shared functional connections between PDS and HC is pivotal rather than a loss or a gain in the number of functional connections.