A Lightweight and Interpretable Model to Classify Bundle Branch Blocks from ECG Signals.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Automatic classification of electrocardiogram (ECG) signals is crucial for diagnosing cardiac conditions.
• Recent advances in ECG analysis utilize complex models, often sacrificing interpretability.
• Bundle branch blocks are significant indicators of underlying heart disease, necessitating accurate classification.

Purpose of the Study:

• To develop a novel, interpretable machine learning model for the automatic classification of bundle branch blocks from ECG signals.
• To address the trade-off between model complexity and interpretability in current ECG analysis techniques.
• To identify key cross-lead dependencies in ECGs indicative of specific cardiac conditions.

Main Methods:

• Utilizing multivariate autoregressive (MAR) model coefficients as features.
• Integrating MAR coefficients with a tree-based classification model.
• Employing post-hoc interpretability techniques to analyze model decisions.

Main Results:

• Successfully classified bundle branch blocks using a lightweight, interpretable model.
• Demonstrated the effectiveness of MAR coefficients in capturing relevant ECG signal dynamics.
• Identified specific cross-lead ECG patterns associated with bundle branch blocks.

Conclusions:

• The proposed MAR and tree-based model offers an interpretable alternative for ECG signal classification.
• This approach provides valuable insights into the electrophysiological basis of bundle branch blocks.
• The lightweight nature of the model makes it suitable for clinical applications requiring explainable AI.