Weighted component-based tensor distance applied to graph-based segmentation of cardiac DT-MRI

We propose a new weighted component-based tensor distance that linearly combines tensor shape (three tensor invariants) and orientation distances. Moreover the weighted component-based tensor distance allows users to easily adjust relative contributions of the distance components toward an optimal single distance for a particular application. We apply the weighted component-based tensor distance to graph-based multi-label segmentation of DT-MRI of infarcted hearts. We evaluate it using a synthetic tensor field that reflects important myocardial tensor field attributes, and three experimentally measured DT-MRI datasets from post-infarct porcine hearts.