Identification of Patients with Potential Atrial Fibrillation during Sinus Rhythm Using Isolated P Wave...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Atrial fibrillation (AF) is frequently underdiagnosed, necessitating improved screening methods.
• Previous deep learning models for AF detection from 12-lead ECGs during sinus rhythm (SR) face challenges with data availability and reproducibility.
• Limited datasets hinder the development and validation of reliable AF screening tools.

Purpose of the Study:

• To develop an explainable feature extraction method for detecting AF during SR using limited data.
• To improve the reproducibility and accessibility of AF screening using electrocardiogram (ECG) data.
• To identify novel P-wave features indicative of underlying atrial abnormalities in patients with AF.

Main Methods:

• Collected 94,224 12-lead ECGs from 64,196 patients.
• Selected 213 patients with pre-diagnosis AF and 247 age-matched controls with SR ECGs.
• Developed the MA-UPEMD signal-processing technique to isolate P waves and extracted spatial-temporal features using PCA and inter-lead relationships.

Main Results:

• Machine learning models utilizing the extracted features achieved an Area Under the Curve (AUC) of 0.64.
• The proposed approach successfully depicted P-wave characteristics related to atrial electrical activity.
• Extracted features demonstrated superior performance compared to bandpass filter methods and deep neural networks on the limited dataset.

Conclusions:

• The study presents a physiologically explainable and reproducible approach for AF classification during SR, even with limited data.
• The MA-UPEMD technique and extracted P-wave features offer a promising alternative for AF screening.
• This method enhances the potential for early AF detection and management.