Atrial repolarization wave: a new approach using spline-based feature engineering and explainable AI for atrial...

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• The atrial repolarization (Ta wave) is difficult to study due to its low amplitude and overlap with the QRS complex.
• Existing methods for analyzing atrial activity primarily focus on the P wave, leaving Ta wave characteristics underexplored.

Purpose of the Study:

• To investigate and extract features from the Ta wave, which is typically obscured by the QRS complex.
• To assess the utility of Ta wave features in classifying atrial arrhythmias using machine learning models.

Main Methods:

• Standard 12-lead ECGs from patients with Sinus Rhythm, Sinus Tachycardia, and Atrial Tachycardia were analyzed.
• A cubic spline interpolation model was employed to reconstruct and isolate the Ta wave, even when hidden within the QRS complex.
• Machine learning models, including Extra-Trees, were trained using extracted Ta wave features for multi-class classification.

Main Results:

• The cubic spline interpolation model demonstrated high accuracy (SSIM score of 0.85) in reconstructing the Ta wave.
• The Extra-Trees model achieved 99% accuracy in classifying atrial arrhythmias when utilizing combined P-Ta wave features.
• Statistically significant temporal and voltage features of the Ta wave were identified and utilized for classification.

Conclusions:

• Incorporating Ta wave features alongside P wave features enhances the classification accuracy of atrial arrhythmias.
• The proposed interpolation method is practical for clinical implementation and adaptable to various ECG datasets.
• This research highlights the potential of analyzing the Ta wave for improved diagnosis of cardiac conditions.