A Novel AI Diagnostic Assistant Tool for Echocardiographic Videos Comparable to Senior Cardiologists

Echocardiography is the common diagnostic technique for heart diseases. However, the major limitation remains to be analysis of echocardiographic videos by experts with extensive clinical experience. The recent development of handheld ultrasound devices could tremendously broaden the application scope in clinics and even at home, if artificial intelligence (AI)-assistance diagnostic tool comparable to trained cardiologists could be developed with medical videos rather than the annotated static medical images. Here, we present an AI echocardiogram diagnosis network (AIEchoDx) that differentiates four common cardiovascular diseases (Atrial Septal Defect, Dilated Cardiomyopathy, Hypertrophic Cardiomyopathy, prior Myocardial Infarction) from normal subjects with performance comparable to that of senior cardiologists and accurately recognize critical anatomic regions of interest (ROI) in each disease. Furthermore, our analysis suggested that Dilated Cardiomyopathy could be classified into two subclasses with significant differences in clinical characteristics. Finally, AIEchoDx performed efficiently for anomaly detection and disease identification of handheld device-produced echocardiographic videos.