Abstract
Brain pathologies often manifest as partial or complete loss of tissue. Thegoal of many neuroimaging studies is to capture the location and amount oftissue changes with respect to a clinical variable of interest, such as diseaseprogression. Morphometric analysis approaches capture local differences in thedistribution of tissue or other quantities of interest in relation to aclinical variable. We propose to augment morphometric analysis with anadditional feature extraction step based on unbalanced optimal transport. Theoptimal transport feature extraction step increases statistical power forpathologies that cause spatially dispersed tissue loss, minimizes sensitivityto shifts due to spatial misalignment or differences in brain topology, andseparates changes due to volume differences from changes due to tissuelocation. We demonstrate the proposed optimal transport feature extraction stepin the context of a volumetric morphometric analysis of the OASIS-1 study forAlzheimer's disease. The results demonstrate that the proposed approach canidentify tissue changes and differences that are not otherwise measurable.