Unveiling Irregular Heart Rhythms: Utilizing a Pretrained ResNet50 Convolutional Neural Network Autoencoder Strategy for Detecting Anomalies in ECG Signals

This study examines the application of ResNet50 (CNN) Autoencoders for identifying abnormalities in electrocardiogram (ECG) data using the PTB Diagnostic ECG Database. The dataset comprises 14,552 samples categorized as normal heartbeats and heartbeats affected by cardiac disorders. Utilizing Transp...

Full description

Saved in:
Bibliographic Details
Published in:2024 2nd World Conference on Communication & Computing (WCONF) pp. 1 - 4
Main Authors: Agarwal, Muskan, Gill, Kanwarpartap Singh, Aggarwal, Priyanshi, Rawat, Ramesh Singh
Format: Conference Proceeding
Language:English
Published: IEEE 12.07.2024
Subjects:
Anomaly detection
Artificial Intelligence
Autoencoder
Convolution
Convolutional neural networks
Data models
Decoding
Deep Learning
Electrocardiography
Heart
Heart beat
Medical services
Model Training
Residual neural networks
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study examines the application of ResNet50 (CNN) Autoencoders for identifying abnormalities in electrocardiogram (ECG) data using the PTB Diagnostic ECG Database. The dataset comprises 14,552 samples categorized as normal heartbeats and heartbeats affected by cardiac disorders. Utilizing Transposed Convolution has significantly improved the model's performance. The study emphasizes the vital role of immediately detecting cardiac irregularities and presents a CNN Autoencoder model designed to efficiently encode and decode ECG data, aiding in the detection of aberrant patterns. The approach involves building a robust Autoencoder with encoder and decoder components that are trained to minimize errors in reconstruction. The evaluation metrics show the model's outstanding performance with an accuracy of 76.93%, precision of 55.23%, recall of 89.81%, and Fl score of 65.40%. This work highlights the need of utilizing deep learning techniques to detect abnormalities in ECG data at an early stage. This method demonstrates significant promise in increasing diagnostic abilities and enhancing patient results.
DOI:10.1109/WCONF61366.2024.10692242