Neural network and wavelet average framing percentage energy for atrial fibrillation classification

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computational Cardiology

Background:

• Electrocardiogram (ECG) signals are crucial for diagnosing atrial conduction pathologies.
• Visual inspection of ECGs for arrhythmias like atrial fibrillation is challenging and subjective.
• Automated diagnostic tools are needed to improve accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a novel wavelet feature extraction method for automated atrial fibrillation detection.
• To assess the effectiveness of the proposed method using a probabilistic neural network (PNN) classifier.
• To compare the performance of the new method against existing techniques.

Main Methods:

• Utilized terminal wavelet packet transform (WPT) to extract features from ECG signals.
• Introduced the average framing percentage energy (AFE) as a key feature for classification.
• Employed a probabilistic neural network (PNN) for classifying normal ECGs, arrhythmias, and atrial fibrillation.
• Validated the method on the MIT-BIH ECG database.

Main Results:

• The proposed AFE feature extraction method demonstrated high potential for automated diagnosis.
• Achieved a classification accuracy of 97.92% for distinguishing various arrhythmias and normal ECGs.
• Performance was evaluated in an additive white Gaussian noise (AWGN) environment, yielding 55.14% at 0dB and 92.53% at 5dB.
• The AFE method outperformed several published comparative techniques.

Conclusions:

• The novel wavelet-based feature extraction method using AFE is a promising approach for automated ECG analysis.
• The PNN classifier effectively utilizes AFE for accurate arrhythmia detection.
• Further validation with larger datasets is recommended to extend the study's findings.