Cardiac MRI Semantic Segmentation for Ventricles and Myocardium Using Deep Learning

Automated noninvasive medical diagnosis of the heart is necessary for early detection of cardiac disorders and cost-effective management. Automated diagnosis comprises automated image segmentation and analysis of cardiac magnetic resonance, CT scan or echocardiogram. Automated segmentation of cardiac substructures and their feature-attributes is necessary for evaluating cardiac functions, disorders, and diagnosis of cardiovascular diseases such as cardiomyopathy, valvular diseases, abnormalities caused by septum perforations, ischemia and blood-flow rate. Semantic segmentation labels an image at the pixel-level, and is used to localize various subcomponents of an object. Localization of the subcomponents facilitates the detection of abnormalities, including abnormalities in cardiac wall motions in an aging heart with muscle abnormalities, vascular abnormalities, and valvular abnormalities. In this paper, we describe a model to improve semantic segmentation of CMR images. The model extracts edge-attributes and context information during down-sampling of the U-Net and infuses this information during up-sampling to localize three major cardiac structures: left ventricle cavity (LV); right ventricle cavity (RV); LV myocardium (LMyo). We present an algorithm and performance results. A comparison of our model with previous leading models, using similarity-metrics between the actual image and the segmented image, shows that our approach improves Dice similarity coefficient (DSC) by 2%–11% and lowers Hausdorff distance (HD) by 1.6–5.7 mm.