Comparison of neural network architectures for segmentation of the left ventricle on EchoCG images

This paper compares the quality of segmentation of echocardiographic images of the left ventricle of the heart using 5 architectures and 38 pre-trained encoders. As part of the study, we trained 1140 neural networks. On the test dataset, the accuracy was 93.18% according to the Dice metric, which is...

Full description

Saved in:
Bibliographic Details
Main Authors: Samun, V. S., Sheka, A. S., Chumarnaya, T. V., Solovyova, O. E.
Format: Conference Proceeding
Language:English
Subjects:
Coders
Computer architecture
Image quality
Image segmentation
Neural networks
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper compares the quality of segmentation of echocardiographic images of the left ventricle of the heart using 5 architectures and 38 pre-trained encoders. As part of the study, we trained 1140 neural networks. On the test dataset, the accuracy was 93.18% according to the Dice metric, which is more than our previous result at 92.78%. On cross-validation, the accuracy was 98.79%, which is higher than the previous result of 90.15%.
ISSN:0094-243X
1551-7616
DOI:10.1063/5.0032165