Atrial fibrillation detection on compressed sensed ECG

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Compressive sensing (CS) offers efficient real-time electrocardiogram (ECG) encoding for wearable devices.
• Assessing the impact of CS compression on downstream signal processing and classification algorithms is crucial.

Purpose of the Study:

• To evaluate the effect of CS compression on atrial fibrillation (AF) detection accuracy.
• To compare different CS reconstruction models and their suitability for AF detection.

Main Methods:

• Compared wavelet and Gaussian reconstruction models for CS ECG data.
• Utilized beat-to-beat (RR) interval identification and an AF detector on the MIT-BIH atrial fibrillation database.
• Investigated a novel beat detection method operating directly in the compressed domain.

Main Results:

• AF detection accuracy remained comparable to uncompressed signals up to 30% compression.
• A 2% accuracy drop was observed at 60% compression.
• Gaussian reconstruction showed superior AF detection accuracy, with <75% compression yielding negligible performance drop compared to wavelet methods.

Conclusions:

• CS is a viable method for real-time ECG compression at moderate rates.
• CS is suitable for offline ECG analysis at high compression rates.
• Gaussian-based reconstruction enhances AF detection performance in compressed ECG signals.