Using Minimum Redundancy Maximum Relevance Algorithm to Select Minimal Sets of Heart Rate Variability Parameters for...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Heart rate is typically irregular in atrial fibrillation (AF) but can be regular in specific conditions.
• Heart rate variability (HRV) quantifies variations in cardiac cycle durations (RR intervals).
• Automated detection of AF is crucial for timely intervention and management.

Purpose of the Study:

• To investigate the efficacy of HRV parameters for automated AF detection using machine learning (ML).
• To identify the most effective HRV features for distinguishing AF from normal sinus rhythm (SR).

Main Methods:

• Utilized a large database of 60-second electrocardiogram (ECG) segments from longer recordings (up to 24 hours).
• Employed the Minimum Redundancy Maximum Relevance (MRMR) algorithm for optimal HRV feature selection.
• Trained and validated seven common ML classifiers using 1-6 selected HRV features via 5-fold and blindfold cross-validation.

Main Results:

• The best ML classifier achieved 97.2% accuracy and a diagnostic odds ratio of 1566 in blindfold validation.
• Top discriminating HRV features included pRR50, SD2/SD1 ratio, and coefficient of variation (CV).
• Effective AF detection was achieved using minimal sets of simple HRV features, notably including pRR50.

Conclusions:

• HRV parameters, particularly pRR50, are highly effective for automated AF detection using ML.
• The proposed methodology enables the development of practical devices for AF screening with simple ECG analysis.
• Accurate differentiation of AF from SR is feasible with short ECG recordings and straightforward ML models.