Detection of Atrial Fibrillation from Single Lead ECG Signal Using Multirate Cosine Filter Bank and Deep Neural...

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Atrial fibrillation (AF) is a common cardiac arrhythmia characterized by irregular atrial activity detected in ECG signals.
• Early diagnosis of AF is vital to mitigate risks of stroke, heart disorders, and coronary artery disease.

Purpose of the Study:

• To propose a novel, accurate, and effective method for detecting atrial fibrillation (AF) from ECG signals.
• To evaluate the efficacy of a multi-rate cosine filter bank and Fractional Norm (FN) features combined with deep learning for AF detection.

Main Methods:

• A multi-rate cosine filter bank was used to extract coefficients from ECG signals across different subbands.
• Fractional Norm (FN) features were computed from these coefficients.
• A deep learning model, Hierarchical Extreme Learning Machine (H-ELM), was employed for AF detection using the extracted FN features.

Main Results:

• The proposed method achieved high diagnostic performance with an accuracy of 99.40%, sensitivity of 98.77%, and specificity of 100% for AF detection.
• Low-frequency subband FN features demonstrated significant diagnostic value, showing a mean difference of 0.69 between normal and AF classes.
• The performance was validated using ECG signals from public databases.

Conclusions:

• The developed multi-rate cosine filter bank and FN feature-based approach, utilizing H-ELM, is highly effective for accurate AF detection.
• FN features derived from low-frequency subbands are particularly significant for distinguishing between normal and AF ECG signals.
• This novel method offers a promising tool for the early and reliable diagnosis of atrial fibrillation.