Cardiac Arrhythmia Classification Using Advanced Deep Learning Techniques on Digitized ECG Datasets

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Electrocardiogram (ECG) classification is crucial for diagnosing cardiac diseases.
• Deep learning (DL) offers advanced tools for analyzing ECG signals to aid expert diagnosis.
• Digitizing ECG records into time series data enables sophisticated computational analysis.

Purpose of the Study:

• To digitize a dataset of ECG record images into time series signals.
• To apply and compare state-of-the-art deep learning techniques for ECG signal classification.
• To evaluate the performance of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Self-Supervised Learning (SSL) models for cardiac arrhythmia classification.

Main Methods:

• ECG images from Pakistani healthcare institutes were digitized.
• Lead II heartbeats were segmented from the digitized ECG signals.
• Multiple DL models, including CNN, LSTM, and an SSL-based autoencoder model, were trained and compared.

Main Results:

• The proposed CNN model achieved the highest classification accuracy of approximately 92%.
• The DL models were trained on a dataset derived from diverse patient ECG plots.
• The CNN model demonstrated fast inference capabilities for real-time ECG monitoring.

Conclusions:

• Digitized ECG signals processed by DL models offer a viable alternative to image-based analysis for arrhythmia classification.
• The developed CNN model provides accurate and efficient real-time monitoring of ECG signals.
• This approach facilitates direct utilization of DL models with ECG machine outputs for enhanced cardiac patient monitoring.