ECG-based cardiac arrhythmia classification using fuzzy encoded features and deep neural networks

Abstract

Cardiac arrhythmia, characterized by an irregular heart rhythm, is a leading cause of sudden and unexpected deaths among patients with cardiovascular diseases. The electrocardiogram (ECG) is a widely utilized non-invasive tool for detecting cardiac arrhythmias. This study investigates the effectiveness of ECG signals in diagnosing various irregular heart rhythms and proposes a novel framework integrating a fuzzy system with deep neural networks. Our approach combines Fourier–Bessel Series Expansion (FBSE)-Tunable Q Wavelet Transform (TQWT) and Principal Component Analysis (PCA) for automatic arrhythmia classification. Compared to conventional deep learning models that rely on raw ECG signals, our method enhances interpretability and feature extraction by incorporating time–frequency analysis and fuzzy feature encoding. Experimental validation using the MIT-BIH dataset demonstrated that our approach outperforms state-of-the-art models in classifying five arrhythmia categories (N, SVEB, VEB, Q, and F) based on the Association for the Advancement of Medical Instrumentation (AAMI) standards. Our model achieved precision scores of 0.98 (N), 0.95 (F), 0.98 (VEB), 0.90 (SVEB), and 0.99 (Q), with corresponding recall values of 1.00 (N), 0.74 (F), 0.93 (VEB), 0.72 (SVEB), and 0.98 (Q). The integration of FBSE-TQWT with a fuzzy deep neural network represents a substantial advancement in ECG-based arrhythmia detection, offering improved accuracy, robustness, and clinical applicability, particularly in distinguishing minority classes such as supraventricular ectopic beats (SVEB) and fusion beats (F).