Automatic Assessment of Ejection Fraction in Heart Ultrasound

Abstract

The quantification of cardiac function, including the measurement of ejection fraction (EF) is an essential assessment for heart condition. However, there is significant variability between different clinicians in evaluating EF. In addition, the manual process of selecting the keyframes and tracing the left ventricle is time-consuming. Therefore, in this paper, an automated way of ejection fraction assessment is proposed. The system composed of two sub-systems, a key-frame extractor to extract the best representative frame of the cardiac function, and a deep learning model to predict the ejection fraction based on the key-frames. The best performance for EF prediction resulted in MAE of 5.57%, RMSE of 7.64%, and R2 of 0.605, which suggests that there is a good agreement between the predicted and the experts labeled EF.

Date of Award	2021
Original language	American English
Awarding Institution	HBKU College of Science and Engineering