Active shape models with invariant optimal features (IOF-ASM) application to cardiac MRI segmentation

Abstract

This paper addresses the problem of accurate segmentation of the endocardial contours of both ventricles and the epicardium of the entire heart. It reports on an extension of the active shape model framework making use of a non-linear appearance model as in the optimal features ASMs (OF-ASM) approach by van Ginneken et al. and incorporating a reduced set of differential Cartesian invariant features as local image descriptors. The new method is coined invariant optimal features ASM (IOF- ASM). Validation results on a vast cardiac MR data-set have shown that the new approach outperforms the original ASM (p < 0.001) and that both optimal features algorithms have barely the same performance, yielding reliable segmentations.