Cardiac 3D + time segmentation and motion estimation are recognized as difficult prerequisite tasks for any quantitative analysis of cardiac images. Some recent algorithms aim to consider a temporal constraint to increase the accuracy of results. To improve the temporal consistency, prior knowledge about cardiac dynamics can be used. In this paper, we propose to build a new Statistical Dynamic Model (SDM) of the heart by learning through a population of healthy individuals. This SDM is composed by a set of semi-landmarks which describe the heart surfaces. For each of them, a mean trajectory and variability around it are derived. The SDM provides a reasonable constraint for a temporally regularized segmentation and motion tracking algorithm.
展开▼
