Statistical methods for structural and functional integration in multi-modal neuroimaging data

点击翻译按钮获取中文摘要

Abstract Neuropsychiatric disorders, such as autism and schizophrenia, affect millions of people worldwide and place a considerable burden on both patients and family members. Existing treatments for these disorders have limited efficacy, in part due the varied clinical manifestations, and to our narrow understanding of the impacted neural processes, particularly at the system (i.e., network) level. Two key elements of networks are the underlying infrastructure or physical connections between elements and the functional signaling between entities that rides on top of this infrastructure. Recent advancements in noninvasive imaging have given us the ability to quantify structural and functional relationships in the brain via diffusion MRI, resting-state functional MRI, respectively. The size and scope of datasets measuring network structure and function are increasing in neuroimaging, and other domains, which heightens the need for new statistical frameworks that make full use of the data. Our goal is to develop frameworks for the analysis of structure-function integration in large-scale and complex networks, applied to neuroimaging studies, but also broadly applicable. This proposal will introduce three analytic paradigms: Bayesian network modeling that uses a priori structure-function knowledge for simultaneous network anomaly detection and clinical severity prediction; density regression using optimal transport theory; and end-to-end prediction using deep neural networks. In our application, infrastructure will be measured via dMRI, while function will be measured rs-fMRI. Each of our frameworks will provide a unique means to integrate these distinct imaging modalities, while also respecting the unique information provided by each data type. We also propose a unique software development effort that creates an application program interface to core software and implementations as software as as a service hosted on cloud platforms.