Automated Arrhythmia Detection Using Joint Duration-Bandwidth Localized Biorthogonal Wavelet Filters

As the cardiovascular system weakens with age, the risk of arrhythmia increases. Arrhythmias such as atrial fibrillation (Afib), atrial flutter (Afl), and ventricular fibrillation (Vfib) are life-threatening conditions that require accurate detection. Electrocardiography (ECG) serves as the primary diagnostic tool, but its manual interpretation is challenging due to the complexity of ECG signals. To address this, we propose a computer-aided diagnosis (CAD) system using an optimal 13/7 parametrized biorthogonal wavelet filter bank for feature extraction. The extracted features—Rényi entropy, fuzzy entropy, sample entropy, norm, and energy—are classified using a Least squares support vector machine (LS-SVM) with 10-fold cross-validation. We formulate three classification tasks: two-class (Afib vs. Normal sinus rhythm (Nsr)), three-class (Afib vs. Afl vs. Nsr), and four-class (Afib vs. Afl vs. Vfib vs. Nsr). Using two-and five-second ECG signals, our approach achieves high classification performance, with up to 99.5% accuracy for two-class, 98.0% for three-class, and 98.0% for four-class classification. These results demonstrate the effectiveness of our method in automated arrhythmia detection, assisting clinicians in objective diagnosis.