|
Abstract:
|
Connectomics, the comprehensive study of connections in the brain, is crucial for charting normal brain development and for understanding how etiology changes the structural and functional network landscapes. While much effort has centered on statistical methods for connectomic data, most network analyses first decompose the brain into subnetworks and then test for disease or outcome-related differences. Increasingly, kernel and distance-based methods are being employed; however, these methods provide only global and subnetwork-specific tests. In this work, we propose and develop a new framework for kernel testing that allows for optimally powerful testing across brain subnetworks while also identifying which networks or regions are associated with the outcome of interest. We illustrate this methodology using extensive simulations and the analysis of a large cohort of subjects who underwent an extensive neuroimaging protocol.
|
