4D Endocardial Segmentation Using Spatio-temporal Appearance Models and Level Sets

In this paper a framework for the segmentation of cardiac MR image sequences using spatio-temporal appearance models is presented. The method splits the 4D space into 2 separate subspaces, one for changes in appearance and one subspace for changes in motion. Using the 4D appearance models in combination with a level set framework combines the robustness of model based segmentation with the flexibility of level sets. The method is tested on the first two time frames of 10 cardiac MR sequences leading to promising results. Further tests using a larger training set for the segmentation of the whole cardiac cycle shall be performed in the near future.