ECG Restitution Analysis and Machine Learning to Detect Paroxysmal Atrial Fibrillation: Insight from the Equine...

Area of Science:

• Veterinary Cardiology
• Computational Biology
• Biomedical Engineering

Background:

• Atrial fibrillation is a common arrhythmia in equine and human athletes.
• Current electrocardiogram (ECG) diagnosis lacks sensitivity for paroxysmal atrial fibrillation (PAF).
• The arrhythmogenic substrate during quiescent periods of PAF remains challenging to detect.

Purpose of the Study:

• To investigate the potential of ECG restitution analysis during sinus rhythm for detecting the arrhythmogenic substrate of PAF.
• To develop a discrimination tool using machine learning for automated PAF screening.

Main Methods:

• ECG recordings were obtained from control horses and horses with PAF during routine clinical work.
• QT, TQ, and RR intervals were extracted for ECG restitution analysis.
• A k-nearest neighbor (k-NN) algorithm was employed to classify PAF based on interval combinations.

Main Results:

• k-NN classifiers using two-interval measures achieved areas under the curve (AUC) of approximately 0.8 to 0.9.
• Simultaneous analysis of QT, TQ, and RR intervals yielded mean AUCs of 0.9 across tested k-values.
• The developed method demonstrated high accuracy in predicting PAF diagnosis.

Conclusions:

• Three-dimensional ECG restitution analysis shows promise for automated screening of PAF.
• This approach can potentially identify the arrhythmogenic substrate between PAF episodes.
• The findings suggest a novel, non-invasive method for early detection and management of equine PAF.