Segmentation of Left Ventricle in Cardiac Magnetic Resonance Imaging via Region-Dependent Motion Estimation

초록

We propose a variational method for the delineation of left ventricle in cardiac MRI using the tracking framework where an initial estimation for the region of interest is propagated forward in time. The propagation of the region state is performed by computing forward motion in consideration of the physical property of regions. The motion leading to segmentation is computed within the region of interest based on the assumption that the physical property of the region of interest is different from the one of its background. We follow a simple, yet effective bi-partitioning image model where the motion is computed within the region of interest independently from its background. We propose an energy functional for computing region-dependent motion with a regularization using total variation in a variational framework. The optimization of the proposed convex energy is performed using the accelerated version of the proximal gradient method that is efficient and effective for non-differentiable total variation regularization term. In the experimental results on the benchmark dataset, we demonstrate the advantage of our method that yields more accurate boundary compared to the conventional global regularity. The computational efficiency of our method can be also achieved by restricting the computation of motion within the region of interest without considering its background.

키워드