Simultaneous Segmentation and Motion Recovery in 3D Cardiac Image Analysis

Abstract

Accurate and robust estimation of the three-dimensional left ventricular geometry and deformation has important clinical implications for better diagnosis and understanding of ischemic heart diseases. So far, most image analysis efforts have performed the shape recovery and the motion tracking tasks in separate steps, typically in sequential fashion. In this paper, we present a continuum biomechanics model based framework that performs simultaneous segmentation and motion analysis of the left ventricle from 3D image sequences, achieved through the tracking of the spatiotemporal evolution of a 3D active region model driven by imaging data, statistical priors of the left ventricular boundaries, and cyclic motion models of the myocardial tissue elements. Experiments on 3D canine and human MR image sequences have shown the superiority of the strategy.