A Deep Learning Model to Identify Mitral Valve Prolapse From the Echocardiogram

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Mitral valve prolapse (MVP) affects 2-3% of the population and increases the risk of heart failure and sudden death.
• Traditional diagnosis via transthoracic echocardiography is time-consuming and requires specialized expertise.

Purpose of the Study:

• To develop and validate a deep learning model, DROID-MVP, for automated MVP classification from digital echocardiogram videos.
• To assess the association between DROID-MVP predictions and clinical outcomes such as mitral regurgitation and mitral valve repair/replacement.

Main Methods:

• Trained and validated DROID-MVP on over 1 million echocardiogram videos from 16,902 cardiology patients.
• Externally validated the model on primary care patient cohorts from two major hospitals (MGH and BWH).
• Assessed correlations between DROID-MVP scores, mitral regurgitation severity, and subsequent mitral valve surgery.

Main Results:

• DROID-MVP demonstrated high accuracy in identifying MVP across internal and external validation sets (AUROC ranging from 0.947 to 0.968).
• Higher DROID-MVP scores were significantly associated with moderate-to-severe mitral regurgitation (OR: 2.0) and future mitral valve repair or replacement (HR: 3.7).

Conclusions:

• Deep learning model DROID-MVP effectively identifies MVP from echocardiogram videos.
• Model predictions serve as digital markers associated with clinically significant MVP, potentially aiding in diagnosis and risk stratification.
• This technology can automate MVP diagnosis and improve patient management.