Automatic segmentation of the left ventricle from pediatric echocardiography images using SegFormer architecture

Abstract

Echocardiography is the most widely used imag-

ing technique for congenital heart disease (CHD) detection, as-

sessing risk, and guiding treatment strategies in pediatric cardi-

ology. However, interpreting and analyzing these types of

images can be challenging due to their complexity, which is

some cases leads to inter-observer variability. This research

work aims to develop an automated left ventricle (LV) segmen-

tation method for pediatric echocardiography images using a

semantic transformer model known as SegFormer, for aiding in

the measurement of clinical image technique. Semantic trans-

formers have demonstrated exceptional performance in seg-

mentation tasks in recent years, making them a suitable choice

for this application. To achieve accurate LV segmentation, the

SegFormer model is trained using the EchoNet-Peds dataset,

which consists of annotated pediatric echocardiography videos.

The experimental results include segmented left ventricle im-

ages, evaluated in accuracy, mean absolute error (MAE), recall

and dice score metrics for performance comparison with other

pediatric segmentation method.