Multimodal Neuroimaging Analysis for Uncovering Brain Subnetwork Associations with Psychiatric Disorders

Internalizing psychopathologies (IPs), such as depression and anxiety, are highly prevalent, costly and disabling psychiatric disorders. Findings about potential neuropathological sub- strates are more often overlapping between the IPs than distinct: many studies have concluded that similar structural and functional brain networks, commonly implicated in the regulation of emotion processing, are dysfunctional in these disorders. However, much of the previous neuroimaging work conducted on IPs focuses on a single disorder and imaging modality. Ad- ditionally, there is increasing evidence that pathological associations with psychiatric disor- ders extend beyond the central nervous system, ranging from immunological changes, cytokine signaling perturbations and alterations in the microbial communities of the gut. Given the complexity of psychiatric disorders, there is an unmet need for research methodology in which multimodal perspectives are integrated into the analytical approach. Here, we present applications of network-based analyses to better understand the distributed physiological perturbations that underlie the complex cognitive and behavioral disruption found in patients suffering from these conditions. To this end, we primarily study the functional and diffusion MRI data of a transdiagnostic cohort of IP patients and healthy controls. We first use a graph theoretical approach to quantify the properties of white matter networks that constitute the structural basis for neural communication in the brain. To incorporate functional brain networks into our analysis, we next develop a novel framework for identifying subnetworks with impaired information diffusion dynamics. We extend this methodology to demonstrate that models of neuromodulation that target subnetwork impairments normalize IP patient information diffusion dynamics towards those of healthy controls. An alternative approach to multimodal MRI analysis of brain networks, based on graph signal processing, is also applied to this dataset. We conclude by investigating the temporal dynamics of brain and microbiome networks of patients with comorbid depression and obesity using a data fusion method based on the coupled factorization of network tensors. Our findings demonstrate that methodological approaches incorporating fusion of heterogeneous data have the potential to elucidate features of complex multi-system disorders that may not be found with traditional techniques.