Matrix Tri-Factorization for {BOLD}-f{MRI}

Abstract

We present Matrix Tri-Factorization as a means to obtain an individual's BOLD-fMRI decomposition into unique neuronal activation patterns, spatial network maps and their corresponding hemodynamic response functions. We validate our proposed method on the motor cortex localization task of the Human Connectome Project 1200 Subject Release and show that neural activation patterns from our proposed Unsupervised Machine Learning technique resemble the given motor task profiles.