Cardiologists know mitral valve prolapse affects 2-3% of people and carries risks of heart failure and sudden death, yet spotting it on standard transthoracic echocardiograms demands skilled eyes and significant time during busy clinic days. 
In the study published in the JACC: Cardiovascular Imaging, the researchers at Massachusetts General Hospital developed DROID-MVP, a deep learning model that scans digital echocardiogram videos to classify MVP automatically, promising faster workflows and better risk stratification. 
They trained and validated the model using 1,043,893 echo videos from 48,829 studies of 16,902 cardiology patients, then tested it externally in 8,888 MGH primary care patients and 257 at Brigham and Women’s Hospital. The derivation cohort included 6,391 women (38%) with average age 61 years, where 783 patients (4.6%) had confirmed MVP. Outcomes focused on model accuracy via AUROC and average precision, plus links to mitral regurgitation severity and future mitral valve repair or replacement.
Model Delivers High Accuracy Across Settings
DROID-MVP scored outputs from 0 to 1, with excellent performance in the MGH cardiology validation set showing AUROC of 0.947 (95% CI 0.910-0.984) and average precision of 0.682 (95% CI 0.565-0.784) at 3.6% prevalence. External validation proved even stronger in MGH primary care patients with AUROC 0.964 (95% CI 0.951-0.977) and AP 0.651 (95% CI 0.578-0.716) at 2.2% prevalence, and in BWH primary care with AUROC 0.968 (95% CI 0.946-0.989) and AP 0.774 (95% CI 0.666-0.797) at 11.3% prevalence. These results mean the AI reliably flags MVP even in lower-risk primary care populations where expert reading proves challenging.
Predictions Tie to Real Clinical Risks
High DROID-MVP scores above 0.67 versus below 0.33 strongly predicted moderate or severe mitral regurgitation with adjusted odds ratio 2.0 (95% CI 1.1-3.8, P=0.030), and future mitral valve repair or replacement with adjusted hazard ratio 3.7 (95% CI 1.5-8.9, P=0.004). This positions the model not just as a screener but as a prognostic tool, helping prioritize patients for closer follow-up or surgical referral.
Workflow Boost for Echo Labs
Imagine the echo lab processing routine studies without manual MVP hunts—DROID-MVP automates detection, freeing sonographers and physicians for complex cases while catching high-risk prolapse early. Integrate it into PACS systems for real-time flagging, especially in high-volume centers serving diverse populations.
Path to Routine Clinical Deployment
Validated across hospitals and settings, this AI opens doors to digital risk markers for MVP complications, with potential FDA clearance paving widespread adoption to reduce missed diagnoses and optimize resource use.