Transformer-based ECG classification for early detection of cardiac arrhythmias

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Electrocardiogram (ECG) classification is vital for diagnosing cardiovascular diseases.
• Current methods may lack the precision needed for early and accurate detection.
• Automated analysis promises enhanced diagnostic capabilities.

Purpose of the Study:

• To develop and evaluate a Transformer-based deep learning framework for automated ECG classification.
• To integrate advanced signal preprocessing, feature selection, and dimensionality reduction for improved performance.
• To assess the model's effectiveness in classifying various cardiac arrhythmias.

Main Methods:

• Signal preprocessing: denoising, normalization, and relabeling of raw ECG data.
• Feature selection and dimensionality reduction: Principal Component Analysis (PCA) and correlation analysis.
• Transformer-based deep learning model trained with optimized loss functions and regularization.

Main Results:

• The Transformer model achieved strong performance on the MIT-BIH benchmark dataset.
• Demonstrated effective classification of Normal, Atrial Premature Contraction (APC), Ventricular Premature Contraction (VPC), and Fusion beats.
• Visualization using t-distributed stochastic neighbor embedding (t-SNE) showed clear class separability.

Conclusions:

• Transformer-based models show significant potential for automated ECG diagnostics in biomedical signal processing.
• The proposed framework offers a scalable approach for improving cardiovascular disease monitoring.
• Further optimization is needed for real-time and resource-constrained applications.