Deep Learning-Assisted 17-Segment Myocardial Scar Segmentation and Analysis

Accurate myocardial scar assessment plays a crucial role in cardiovascular medicine. According to the current recommendations, localization is interpreted based on the American Heart Association's 17-segment model. Despite the recognized benefits of the 17-segment model, automated segmentation methods remain limited. Even when existing approaches are claimed to be automatic, they often require that LV contours are readily available. This study aims to develop a novel deep learning pipeline for fully automated segmentation of left ventricular myocardial scars within the 17-segment model. The approach consists of a two-stage segmentation process followed by scar quantification. Additionally, a third DL model functions as an auxiliary convolutional neural network to enhance the performance of both segmentation models. The quantification of myocardial scar volume in separate segments was successfully addressed, achieving a mean absolute error of 7.71% and a Pearson correlation coefficient of 0.88, indicating a strong correlation between predicted and actual scar volumes. This innovative approach holds promise for improving clinical decision-making and patient management in cardiovascular care.