Myocardial Segmentation in MR images using Convolutional Neural Networks

Abstract: A convolutional neural network for automatic myocardial segmentation of MR images is described and implemented, based on the readily available architecture SegNet. A general network trained on both end-systolic and end-diastolic images is determined to be superior to the networks trained on the separate data. The evaluation of myocardial segmentation is discussed, as well as the importance of manual, visual inspection.