Cardiac arrhythmia detection from ECG signal using Siamese adversarial neural network

Heart disease is one of the most serious health issues that people faced today. Almost 50 million people worldwide suffer from cardiovascular problems. Electrocardiogram (ECG) signals are more important for diagnosing and monitoring patients with various cardiovascular diseases. This article proposes a novel SANN (Siamese Adversarial Neural Network-based Cardiac Arrhythmia Detection) CAD model for detecting different arrhythmias. First, an optimized Savitzky-Golay (OS-G) digital filter model is used to enhance and smooth the ECG signal. OS-G filtering uses a Tunicate Swarm Optimization (TSO) algorithm to adjust the polynomial order and window size to preserve the characteristics of P, Q, R, S, T, and U waves. A hybrid feature extraction process called Pathfinding Least Absolute Shrinkage and Selection Vector Regression (PLASSVR), which integrates regression analysis and pathfinding optimization, is employed to extract the significant features. The extracted features are fed to a novel Siamese neural network based on the Generative Adversarial Network (SNN-GAN) to classify the signals for arrhythmia disease detection. Furthermore, the Gannet Optimization Algorithm (GOA) is applied in the SNN-GAN model to change the hyperparameters. The proposed SANN CAD model is implemented in the Python platform via the MIT-BIH Arrhythmia Database. The performance of the SANN-CAD model is evaluated against various evaluation criteria and compared to traditional classifiers, using optimization for matching and no optimization for matching. The maximum classification accuracy that the SANN CAD model achieves is 99.74% and 98.23% for with and without optimization, and is superior to conventional classifiers.