A comparative analysis of CNNs and LSTMs for ECG-based diagnosis of arrythmia and congestive heart failure

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Cardiac arrhythmias pose a significant global health risk, necessitating early and accurate detection for effective diagnosis and management.
• Automated analysis of electrocardiogram (ECG) data holds promise for improving arrhythmia detection rates.

Purpose of the Study:

• To evaluate and compare the performance of Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks for classifying cardiac arrhythmias.
• To assess the suitability of different deep learning models for varying ECG data durations and monitoring scenarios.

Main Methods:

• Utilized three PhysioNet datasets containing ECG records, segmented into approximately 10-second intervals.
• Transformed ECG data into scalograms using Discrete Wavelet Transform (DWT) for training a VGG-16 (CNN) model.
• Employed Wavelet Transform (WTS) for feature extraction and dimensionality reduction to train an LSTM network.

Main Results:

• The VGG-16 model achieved a test accuracy of 96.44% in classifying cardiac arrhythmias.
• The LSTM network achieved a test accuracy of 92% for the same classification task.
• VGG-16 demonstrated higher effectiveness for analyzing short-duration ECG segments.

Conclusions:

• CNNs, specifically VGG-16, are highly effective for rapid, short-duration cardiac arrhythmia detection.
• LSTMs show potential for continuous, long-term monitoring of cardiac arrhythmias, particularly on edge devices for personalized healthcare applications.