Localizing Cardiac Structures in Fetal Heart Ultrasound Video

Recently, a particle-filtering based framework was proposed to extract 'global' information from 2D ultrasound screening videos of the fetal heart, including the heart's visibility, position, orientation, view classification and cardiac phase. In this paper, we consider how to augment that framework to describe the positions and visibility of important cardiac structures, including several valves and vessels, that are key to clinical diagnoses of congenital heart conditions in the developing heart. We propose a partitioned particle filtering architecture to address the problem of the high dimensionality of the resulting state space. The state space is partitioned into several sequential stages, which enables efficient use of a small number of particles. We present experimental results for tracking structures across several view planes in a real world clinical video dataset, and compare to expert annotations.