Automatic Segmentation of Left Ventricular Endocardium and Epicardium from Cardiac Cine MRI Using Deep Learning

Abstract

This research work uses cardiac cine magnetic resonance imaging (MRI) and deep learning (DL) methods to focus on automated left ventricle epicardium and endocardium segmentation. Using left ventricle segmentation, researchers have already established novel image analysis methods. The literature review aims to contextualize fundamental concepts that, while technical, impact cardiac MRI diagnostic capabilities. Recent advancements in cardiac MR imaging have rekindled interest in MR in radiology and cardiology. The method employs multi-channel DL to segment LV endocardial & epicardial structures. To ensure segmentation accuracy and durability, multi-channel DLsegmentation contours were merged into a level-set formulation targeting annular shapes. The Dice coefficient was utilized to compare manual delineation to segmentation results. Human and machine segmentation overlap was compared using dice values. Manual and automatic segmentation agreements improve with dice value. In terms of segmentation performance as well as average Dice coefficient for LV endocardial & epicardial borders, the suggested method performs better manual segmentation in general. It also outperforms DLand level-set manual segmentation techniques.