Segmenting rodent cardiac ultrasound images using direct posterior models

Lack of an accurate generative model makes it hard to use classical MAP segmentation algorithms to jointly segment the epi- and the endocardium in ultrasound rodent cardiac images. This paper proposes an alternate methodology for such segmentation. The methodology directly models the posterior probability of segmentation using penalized logistic models. A level-set segmentation algorithm is developed using direct posterior models. Finally, experimental evaluation is provided which compares the algorithm segmentation with manual segmentation using real-world data.