Abstract

Late diagnosis or identification is the cause for the severe complications in heart valve disease. Though echocardiography reveals valve structure and function quite well, valve anatomy interpretation along with Doppler flow and clinical measurements would need expert opinion. In this paper, we review AI, ML, DL technologies used for valve disease diagnosis from echo and cardiac scans. Discussions have gone on from view recognition and valve segmentation, through the estimation of severity, analysis of Doppler video, and the merging of images with clinicians' reports. Main issues that recur across the studies are limited dataset diversity, performance inconsistency between hospitals and devices, class imbalance, prediction explanations difficult to grasp, narrow disease spectrum covered, and insufficient real-world testing. We propose a multi-stage conceptual model to differentiate between normal valve, stenosis, and regurgitation as well as provide confidence level in addition to an explanation that clinicians may consider.

Keywords

How to Cite This Article

Rights