Deep learning for echocardiographic assessment and risk stratification of aortic, mitral, and tricuspid regurgitation: the DELINEATE-regurgitation study

Comprehensive Summary

This study utilizes deep learning models to assess patient outcomes of valvular heart disease, including aortic, mitral, and tricuspid regurgitation. Researchers trained their AI models on over 71,000 echocardiogram color Doppler videos containing more than 1.2 million clips from two, internal and external, health systems. The deep learning models were found to accurately classify regurgitation severity based on AI comparisons to cardiologist readings, and the highest performance was accomplished upon integrating information from multiple echocardiogram videos (κ [0.73–0.81] internally, [0.64–0.76] externally, across the three valves). Additionally, a specialized model, (DELINEATE-MR-Progression), predicted the risk of mitral regurgitation progression, and AI predictions were found to go beyond currently known risk factors. Overall, this study suggests that AI-based echocardiogram video analysis can improve diagnostic agreement with cardiologists and aid in prognostic information for mitral regurgitation.

Outcomes and Implications

Valvular regurgitation is a serious condition contributing to both morbidity and mortality in patients if left untreated. The main diagnostic procedure for this condition includes echocardiography which is limited by inter-observer variability and interpretation. Deep learning models provide a consistent and comprehensive model for diagnostic and risk prediction factors which could contribute to improved clinical decision making and treatment outcomes. Echocardiography's are widely integrated into the healthcare system, making AI usage feasible to incorporate into clinical practice. This study shows promising potential for deep learning models in standardizing valvular heart disease detection and assessment to help improve treatment outcomes.