Automated quantification of myocardial infarction using a hidden Markov random field model and the EM algorithm

Abstract

Infarct size has been recognized as a good indicator of the functional status of the ischemic heart and to evaluate the impact of myocardial infarction therapies. Its assessment can be performed from late gadolinium enhancement magnetic resonance images. A number of methods have been proposed for the semi-automatic and automatic quantification of necrosis. We developed an automatic method based on a Markov random field framework and a region growing approach within an EM optimization, which enables segmentation of both necrosis and microvascular obstructions. The method has been evaluated on both synthetic data and 10 clinical cases in 3D and lead to the best results as compared to other conventional approaches and expertise.