Angular Central Gaussian and Watson Mixture Models for Assessing Dynamic Functional Brain Connectivity During a Motor Task

Abstract

The development of appropriate models for dynamic functional connectivity is imperative to gain a better understanding of the brain both during rest and while performing a task. Leading eigenvector dynamics analysis is among the favored methods for assessing frame-wise connectivity, but eigenvectors are distributed on the sign-symmetric unit hypersphere, which is typically disregarded during modeling. Here we develop both mixture model and Hidden Markov model formulations for two sign-symmetric spherical statistical distributions and display their performance on synthetic data and task-fMRI data involving a finger-tapping task.