Segmentation of Rat Cardiac Ultrasound Images with Large Dropout Regions

Abstract

Short-axis rat cardiac ultrasound images contain especially large regions of dropout which make it very difficult to segment the endocardium. Previous strategies, such as using shape priors, are not effective with such large dropout regions. This paper proposes a dropout modeling strategy, which can bridge large dropout regions and segment the endocardium when used along with shape priors. The segmentation is formulated as an active contour in a Maximum-APosteriori (M.A.P.) framework with explicit priors for the dropout function and shape. Further, the active contour is evolved by a strategy called tunneling descent. Tunneling descent is a deterministic evolution strategy which can escape from local minima. The combination of dropout modeling and tunneling descent gives active contours which can successfully segment rat cardiac ultrasound images. Experimental results comparing the performance of the new algorithm with manual segmentation and classical active contours are provided.