Gml-PAF: A Generalizable Machine Learning Algorithm for Paroxysmal Atrial Fibrillation Detection based on Short-Term...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Paroxysmal atrial fibrillation (PAF) detection is challenging.
• Current methods require robust and generalizable algorithms.
• Wearable screening for PAF necessitates efficient detection methods.

Purpose of the Study:

• To develop a generalizable machine learning algorithm (Gml-PAF) for paroxysmal atrial fibrillation detection.
• To evaluate the algorithm's performance across diverse electrocardiogram (ECG) databases.
• To assess the utility of Gml-PAF for wearable screening applications.

Main Methods:

• Utilized a model-agnostic framework for machine learning (ML) model selection, feature selection, and hyperparameter tuning.
• Employed 21-beat inter-beat intervals (IBI) for PAF detection.
• Trained and validated the Gml-PAF algorithm across 16 PhysioNet ECG databases.

Main Results:

• Achieved robust cross-database generalization in independent tests.
• Reported F1 scores ranging from 0.747 to 0.987.
• Demonstrated Area Under the Curve (AUC) values between 0.933 and 0.999.

Conclusions:

• The Gml-PAF algorithm demonstrates strong utility for wearable screening of paroxysmal atrial fibrillation.
• The algorithm achieves performance comparable to deep learning methods with longer inter-beat interval sequences.
• Gml-PAF surpasses conventional machine learning methods in PAF detection accuracy and generalizability.