Bimodal classification algorithm for atrial fibrillation detection from m-health ECG recordings

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial Fibrillation (AF) is a common arrhythmia and a major stroke risk factor.
• Automated AF detection is increasingly important with the rise of m-Health devices for off-clinic monitoring.
• Current diagnostic methods for AF can be limited in continuous, real-world monitoring.

Purpose of the Study:

• To develop and evaluate a bimodal classifier for distinguishing between Atrial Fibrillation (AF) and Sinus Rhythm (SR).
• To assess the algorithm's performance for automated AF detection in a potential m-Health application.
• To establish the feasibility of using spectral and temporal ECG features for AF classification.

Main Methods:

• Collected surface ECG recordings from a handheld device and standard ECGs from 68 subjects.
• Analyzed an additional 48 subjects from the MIT-BIH Arrhythmia Database.
• Developed a bimodal algorithm computing a spectral Frequency Dispersion Metric (FDM) and temporal R-R interval variability (VRR) index from 6-second ECG segments.

Main Results:

• Scatter plots of VRR and FDM indices formed two distinct clusters, enabling a linear classification boundary.
• The algorithm correctly differentiated SR from AF in all subjects except for 3 SR subjects from the MIT-BIH dataset.
• The bimodal approach demonstrated high accuracy in classifying AF and SR waveforms.

Conclusions:

• The bimodal classification algorithm successfully acquires, analyzes, and interprets ECGs for AF detection.
• This algorithm holds significant potential for supporting m-Health diagnosis, monitoring, and therapy management in AF patients.
• The study validates the effectiveness of combining spectral and temporal metrics for robust AF detection.