Frequency tracking of atrial fibrillation using Hidden Markov Models

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Biology

Background:

• Atrial fibrillation (AF) detection relies on accurate electrocardiogram (ECG) signal analysis.
• Noise in ECG recordings can significantly hinder the precise tracking of AF frequency.
• Robust methods are needed to improve the reliability of AF frequency estimation.

Purpose of the Study:

• To develop and evaluate a Hidden Markov Model (HMM) for robust noise-insensitive tracking of AF frequency in ECG.
• To assess the performance improvement offered by the HMM compared to traditional methods under noisy conditions.

Main Methods:

• A Hidden Markov Model (HMM) was employed, with each state representing a specific frequency interval.
• Observed states were derived from the residual ECG after QRST cancellation using the short-time Fourier transform.
• The Viterbi algorithm utilized state transition and observation matrices, incorporating AF characteristics and signal-to-noise ratios (SNRs), to determine the optimal state sequence.

Main Results:

• The HMM approach demonstrated a substantial reduction in root-mean-square (RMS) error for AF frequency tracking.
• At a 5 dB SNR, the RMS error decreased from 1.2 Hz to 0.2 Hz, indicating improved accuracy.
• The state transition and observation matrices effectively integrated knowledge of AF dynamics and estimation parameters.

Conclusions:

• Hidden Markov Models offer a robust solution for accurate AF frequency tracking in noisy ECG signals.
• The proposed HMM-based method significantly enhances the reliability of AF frequency estimation.
• This approach has the potential to improve the diagnosis and management of atrial fibrillation.