Long-Term ECG Analysis Through Image Conversion and Deep Learning

Due to the lengthy information, irregular structure, and numerous types of distortion and disconnections, analyzing long ECG signals can be difficult. An ECG-to-image conversion is built enabling the use of image processing methods and tools for semantic segmentation that uses neural networks. That...

Full description

Saved in:
Bibliographic Details
Main Authors: Hernandez-Fernandez, Carlos, Gomez-Moreno, Hilario, Holgado-Cuadrado, Roberto, Blanco-Velasco, Manuel
Format: Conference Proceeding
Language:English
Subjects:
Electrocardiography
Heat maps
Neural networks
Residual neural networks
Semantic segmentation
Thresholding (Imaging)
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Due to the lengthy information, irregular structure, and numerous types of distortion and disconnections, analyzing long ECG signals can be difficult. An ECG-to-image conversion is built enabling the use of image processing methods and tools for semantic segmentation that uses neural networks. That segmentation is based on the ResNet50 network, which is incorporated into the DeepLabv3 architecture, and is used to identify ventricular beat fusions and premature ventricular contractions. An established database is used for training and to assess the model's efficacy. The ECG data are converted into images using thresholding and heat maps, and PyTorch is used to train the network. The obtained binary masks mark the found abnormalities with accuracy. The presented method can help to detect heart anomalies more quickly and accurately by automating the study of long-term ECG readings.
ISSN:2325-887X
DOI:10.22489/CinC.2023.091