The Rlign algorithm for enhanced electrocardiogram analysis through heart rate-corrected ECG alignment for...

Area of Science:

• Biomedical Signal Processing
• Machine Learning in Healthcare
• Cardiology

Background:

• Electrocardiogram (ECG) analysis is crucial for diagnosing heart conditions.
• Deep learning, particularly Convolutional Neural Networks (CNNs), dominates current automatic ECG analysis.
• CNNs require large datasets and offer limited explainability, posing challenges for ECG interpretation.

Purpose of the Study:

• To reintroduce and enhance shallow learning methods for ECG analysis.
• To leverage the cyclic nature of ECG signals for improved diagnostic accuracy.
• To provide an interpretable and data-efficient alternative to deep learning in cardiology.

Main Methods:

• Developed an adaptive transformation to structure ECG signals for shallow learning.
• Aligned R-peaks and resampled inter-QRS segments to a reference heart rate.
• Evaluated the transformed signals using shallow learning models for classification, clustering, and explainability.

Main Results:

• The adaptive transformation significantly improved shallow learning algorithm performance.
• Shallow models trained on transformed ECGs showed superior accuracy and interpretability compared to CNNs in data-limited scenarios.
• Achieved CNN-level performance in ECG analysis using shallow learning with the proposed transformation.

Conclusions:

• Shallow machine learning, enhanced by alignment-based transformation, offers a viable alternative to deep learning for ECG analysis.
• This approach excels in classification, clustering, and explainability, particularly when training data is scarce.
• A publicly available framework for ECG signal alignment is released to support further research and adoption.