Automatic evaluation of the ejection fraction on echocardiography images

Abstract

The ejection fraction measures the amount of
blood pumped by the heart. This study aims at automatically
evaluating the ejection fraction from echocardiography images.
First, a supervised learning model with different deep neural
network architectures was proposed for the segmentation of the
ventricle at the end of diastole and end of systole on echocardiography images. Then, a regression model was designed to study
the difference of the area between the ventricle at the systole and
the diastole, which provide an estimation of the ejection fraction. The model was evaluated on a subset of the EchoNet-Dynamic dataset. The four models described in this article study
the ejection fraction of the hearts with another approach than
the previous EchoNet-Dynamic studies. The results could be
used in future projects to predict heart diseases.