A Deep Learning Approach for Predicting Ejection Fraction from Pediatric Echocardiogram

This study proposes a deep learning model that predicts cardiac ejection fraction (EF) from 2D echocardiogram videos of pediatric patients. The model integrates convolutional neural networks (CNN) and long short-term memory (LSTM) networks to process echocardiogram videos in two standard views: apical four-chamber (A4C) and parasternal short axis (PSAX), by using video data and patient metadata including age, weight, height, and EF percentage. The CNN sub-module extracts feature maps from the video frames to learn spatial representations, while the LSTM layers enable the learning of temporal dependencies that characterize the heart movement. This model also utilizes a user-friendly web application to allow clinical professionals to view the EF value.