Enhancing ECG Signal Data through Denoising Features with Transformer Generative Adversarial Networks for Model Classification 1D-CNN

Abstract

Abstract An important component of telemedicine's remote cardiac health monitoring of patients is the use of artificial intelligence (AI) technology to detect electrocardiograph (ECG) signals. Failure to properly diagnose and treat abnormal ECG patterns caused by arrhythmia symptoms can result in a fatal outcome. Given that arrhythmia symptoms contribute significantly to noncommunicable cardiovascular disease (CVD), which is responsible for approximately 32% of global mortality, this concern becomes even more significant. The high sensitivity of ECG signals to both external and internal electrical disturbances makes accurate interpretation of these signals for arrhythmia detection challenging. An effective denoising technique is presented in this method as a substitute approach to reduce noise disturbances in ECG signal data and enhance the quality of the training data for AI detection models. This pre-processing technique combines a synthesis approach with Gaussian filtering, an auto-encoder-decoder (transformer), and generative adversarial networks (GANs). The MIT-BIH dataset is the subject of research for this study, which has been categorized into Normal, Atrial Premature, Premature Ventricular Contraction, Fusion of Ventricular and Normal, and Fusion of Paced and Normal. The research findings show that the quality of the synthesized data is almost identical to that of the original data. It is advised to use a deep neural network (DNN) model instead of the previous prediction model for this enhanced dataset, specifically a one-dimensional convolutional neural network (1D-CNN), which is well suited for training this reconstruction data through this experiment.