Machine Learning Models for Enhanced Post-MI Survival Duration Prediction from Echocardiograms
ePoster
Authors: Bryan-Clement Tiu, University of California, Irvine School of Medicine Michelle Dai, University of Pennsylvania Abhinav Suri, National Institutes of Health Maxwell Geiger, Touro University Nevada Brian Kaw, Touro University, Nevada ,
We present evidence that machine learning models can be trained to predict post-myocardial infarction survival duration. These models harness already extant measurements from echocardiograms, a standard of care post-MI evaluation and monitoring. Our findings show potential for machine learning in predicting post-MI outcomes, with one model demonstrating remarkable accuracy in one-year survival forecasts. Simultaneously, other models demonstrate the limitations of small, unbalanced datasets. Using data augmentation as well as optimizing training set size and probability thresholds can improve model performance in midst of these limitations.