Deep Learning-Derived Right Ventricular Ejection Fraction Predicts Mortality in Patients Undergoing Transcatheter...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Transcatheter tricuspid valve intervention (TTVI) is a key treatment for severe tricuspid regurgitation.
• Assessing right ventricular (RV) function post-TTVI is crucial but challenging with traditional echocardiography.

Purpose of the Study:

• To evaluate RV functional changes after TTVI using a deep learning (DL) model.
• To assess RV ejection fraction (RVEF) trajectories using DL-based echocardiography.

Main Methods:

• A DL model predicted RVEF from 2D echocardiographic videos in 373 patients undergoing TTVI.
• RVEF was measured at baseline and 1-3 days post-procedure.
• One-year all-cause mortality was the primary endpoint.

Main Results:

• Median RVEF was similar pre- and post-TTVI at the cohort level, but individual trajectories varied.
• A post-TTVI RVEF threshold of 38% predicted 1-year mortality (58.4% vs. 85.1% survival).
• Patients with RVEF < 38% showed a significant drop in RVEF post-procedure.

Conclusions:

• DL-based echocardiography provides an unbiased assessment of RV function post-TTVI.
• A post-procedural RVEF < 38% identifies patients at high risk for mortality.
• DL-enhanced echocardiography may improve risk stratification and personalize follow-up for TTVI patients.