Analysis of Relevant Features from Photoplethysmographic Signals for Atrial Fibrillation Classification

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial Fibrillation (AF) is a prevalent cardiac arrhythmia affecting millions globally.
• Electrocardiogram (ECG) is the standard diagnostic tool, but photoplethysmography (PPG) offers a portable alternative.
• Identifying critical PPG signal features for AF detection remains an area of research.

Purpose of the Study:

• To systematically review and identify features of PPG signals used for AF detection.
• To experimentally evaluate the relevance of these features using machine learning.
• To determine the most influential PPG features for accurate AF diagnosis.

Main Methods:

• A systematic review adhering to PRISMA-DTA guidelines was conducted across six databases.
• Forty-four potential PPG signal features (time, frequency, time-frequency domains) were identified.
• Machine learning was employed to evaluate feature relevance, implementing 27 of the identified features.

Main Results:

• A total of 44 PPG signal features were identified in the literature for AF detection.
• Machine learning analysis revealed that only 11 of these features are truly relevant for AF detection.
• An AF detection algorithm utilizing these 11 features achieved high performance: 98.43% sensitivity, 99.52% specificity, and 98.97% accuracy.

Conclusions:

• A reduced set of 11 key features significantly enhances PPG-based atrial fibrillation detection.
• Machine learning effectively identifies the most impactful features for PPG signal analysis in AF diagnosis.
• The developed algorithm demonstrates the potential of PPG for efficient and accurate AF screening.