Artificial Intelligence-based Automatic Detection and Classification of Mitral Regurgitation

Mitral regurgitation is among the most prevalent valvular pathologies, with incidence rising with age. If left untreated, severe mitral regurgitation carries a substantial risk of cardiovascular complications and mortality, including heart failure, arrhythmias, and pulmonary hypertension. Transthoracic echocardiography remains the primary noninvasive diagnostic tool; however, its interpretation relies heavily on expert experience and requires considerable time investment.

This study aims to develop a fully automated system utilizing deep learning techniques for the detection and severity assessment of mitral regurgitation in color Doppler echocardiography images. Echocardiographic data collected between 2017 and 2025, comprising over 3,600 videos, were used to obtain 761 annotated images.

The model was trained in the MATLAB environment utilizing the R-CNN architecture. Detection and severity assessment of mitral regurgitation were conducted using deep learning methods implemented in MATLAB, with a focus on computer vision techniques.

Test results demonstrated greater than 95% accuracy in mitral regurgitation detection and over 80% accuracy in classification. These findings suggest that integrating artificial intelligence substantially reduces analysis time and enhances clinical decision-making.