AI Deep Learning Model Automates Right Ventricle Evaluation From Echocardiograms
Clinical investigators from the Mass General Brigham Heart and Vascular Institute have developed a deep learning computer vision model to automate the assessment of right ventricular dimensions and functional performance from standard echocardiographic recordings. The artificial intelligence platform addresses longstanding diagnostic bottlenecks in routine cardiac ultrasound evaluation.
Mostafa Al-Alusi and Patrick Ellinor of the Mass General Brigham Heart and Vascular Institute are lead and senior authors of a paper published in Nature Cardiovascular Research titled 'Deep Learning-Enabled Assessment of Right Ventricular Size and Function is Associated with Cardiovascular Outcomes.'
Investigators developed the deep learning model DROID to automate standard measurements of left atrial and left ventricular structure and function from echocardiograms.
The right ventricle, which pumps deoxygenated blood into pulmonary circulation, remains difficult to quantify on echocardiograms due to its complex crescent geometry and anatomical position behind the breastbone. While cardiac magnetic resonance imaging offers precise volumetric measurements, high equipment expenses and limited availability make widely accessible ultrasound automation a clinical priority.
Architecture And Multi-Cohort Validation
Led by investigators Mostafa Al-Alusi and Patrick Ellinor, the team partnered with computational researchers at the Broad Institute to engineer DROID-RV. The neural network was trained on more than 380,000 echocardiogram video clips gathered from approximately 17,000 outpatient cardiology attendees.
The investigators validated the algorithm across roughly 9,100 primary care patients within their regional health system and an external cohort of 831 patients treated at a hospital in Hungary. The automated tool estimated right ventricular diameter to within approximately three millimeters of expert human measurements while identifying functional impairment with high diagnostic accuracy.
Prognostic Association And Clinical Workflow
When linked to longitudinal medical histories, model forecasts indicating right ventricular enlargement or functional deterioration strongly correlated with elevated risks of congestive heart failure, atrial fibrillation, and overall patient mortality. Notably, patients flagged with dilated right chambers demonstrated double the mortality hazard over a five-year follow-up period.
In clinical practice, the software serves as an automated triage tool by supplying rapid baseline measurements and highlighting anomalous scans that require prioritized physician review. The technology offers immediate evaluation in acute care environments and resource-constrained clinics where specialized imaging experts or MRI scanners remain unavailable.