A convenient method for semiautomatic atrial body segmentation

This paper introduces an improved level set algorithm based on Chan-Vese (C-V) models via reaction diffusion for atrial body segmentation. Before atrial body segmentation, a set of methods are used to extract left atrium and left ventricular as a preprocessing work. This set of methods consist of the following steps. First, a bilateral filtering operation is applied to remove random noise caused by uneven contrast agent followed by an adaptive thresholding method to extract left atrium and the connected issues. Second, regional growth combined with active contour models are applied to extract left atrium including left atrial appendage, pulmonary veins and left ventricular. Clinical validation has been performed on 6 Multi-slice computed tomography (MSCT) datasets. The proposed algorithm achieved an average overlap rate of 98.3% compared with the software Amira (Visage Imaging, Australia). It takes about less than 3s for a 370×458 slice. The convergence speed of contour evolution with the proposed improved level set algorithm via reaction diffusion is nearly 5 times of the original C-V models. Besides, the proposed improved level set algorithm can solve the singularity problem of the convergence process effectively.