Assessment of 3D myocardium segmentation in MRI using endocardium layer expansion model

In this work the authors present an assessment of 3D Left Ventricle Myocardium (LV) segmentation based on the Layer Expansion model. The evaluation was carried out using simulated myocardium with noise and MR images of the heart. The segmentation approach consists of first determining the LV cavity using competitive fuzzy connectedness between cavity and myocardium region. Myocardium is segmented using a layer expansion model after the determination of endocardium surface. Connected voxels with insertion cost below a threshold are incorporated in an orderly fashion, emulating a balloon expansion. The cost function encompasses local texture, local gradient and global shape information. The segmentation error (percentage of misclassified voxels) for the phantoms were below 11.1% for cavity and 19.7% for myocardium. The proposed approach can be used for higher dimensional data, such as dynamic 3D scenes