Abstract
The detection rate of mitral valve prolapse (MVP) has been increased by using 3D echocardiography. It would be clinically meaningful for accurate assessment of prolapse histology with advanced deep learning algorithms. However, recent studies mainly focus on automatic measurement and reconstruction of cardiac structures using voxel data, while seriously neglecting the automatic diagnosis and visualization using cine data. To address the gap, we propose an artificial intelligence agent to study diagnosis and heatmap for MVP. We include 481 cardiac cycles (8422 frames) from 151 subjects. Our agent achieves the AUC, accuracy and F1 of MVP diagnosis are 99.54%, 95.08%, and 94.12% in patient-levels. The Dice and Iou of heatmap of MVP are 70.35% and 56.39%, which is about 5% higher than that between senior and junior physicians. Our agent would provide standardizing diagnosis and intuitive visualization for the surgical management and treatment plans of MVP.