Artificial Intelligence-Enabled Cardiac Function Estimation from Phone Videos of Echocardiograms

The study evaluated whether artificial intelligence models for estimating left ventricular ejection fraction (LVEF) work effectively on echocardiogram videos recorded with mobile phones. The research included 6209 videos from 2648 studies and 2611 patients (mean age 68.4 years) from three medical centers. Without model fine-tuning, the mean absolute error was 7.00 percentage points, but after fine-tuning it decreased to 6.96 percentage points. Models were tested on both mobile phone-recorded videos and native DICOM medical formats. The results suggest that AI-assisted LVEF estimation from mobile phone echocardiogram videos may be feasible when native image export is unavailable. However, the authors emphasize that prospective evaluation is needed before clinical deployment.