Time-varying coherence function for atrial fibrillation detection.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial fibrillation (AF) is a common arrhythmia requiring accurate detection.
• Current detection methods may have limitations in sensitivity or specificity.
• Automated detection systems are crucial for efficient diagnosis and management.

Purpose of the Study:

• To introduce and validate a novel method for automatic AF detection using time-varying coherence functions (TVCF).
• To assess the performance of TVCF, alone and combined with Shannon entropy (SE), across diverse cardiac databases.
• To evaluate the efficacy of the method using both long (128 beats) and short (12 beats) data segments.

Main Methods:

• TVCF was estimated by multiplying two time-varying transfer functions (TVTFs) derived from adjacent signal segments.
• The TVCF's ability to differentiate between normal sinus rhythm (NSR) and AF segments was analyzed.
• The combined TVCF-SE approach was tested on multiple public and clinical databases (MIT-BIH AF, MIT-BIH NSR, MIT-BIH Arrhythmia, 24-h Holter).

Main Results:

• TVCF demonstrated high coherence values (near 1) for NSR segments and significantly lower values when AF was present.
• The combined TVCF-SE method achieved high detection rates, e.g., 97.9% sensitivity and 98.2% specificity on the MIT-BIH AF database with 128 beat segments.
• Accurate AF detection was also achieved using shorter segments (12 beats), improving AF burden calculation.

Conclusions:

• The proposed TVCF-based method, especially when combined with SE, provides a robust and accurate approach for automated AF detection.
• The algorithm demonstrates high performance across various databases and segment lengths, including short segments beneficial for AF burden analysis.
• This novel method holds promise for improving the diagnosis and monitoring of atrial fibrillation.