Klasifikasi Aritmia Berbasis Model Hybrid Convolutional Neural Network dan Transformer dengan Implementasi Berbasis Web

Abstract

Cardiac arrhythmia is a heart rhythm disorder that can trigger serious cardiovascular conditions and significantly increase the risk of sudden cardiac death. Conventional arrhythmia detection processes still rely on the manual interpretation of electrocardiogram (ECG) signals by medical experts, which necessitates high precision and is time-consuming. Advancements in artificial intelligence, particularly in Deep Learning, have paved the way for the development of faster and more consistent automated detection systems. This study proposes an arrhythmia classification model based on a hybrid architecture combining Convolutional Neural Networks (CNN) and Transformers. The CNN is utilized to extract spatial features from ECG signals, while the Transformer functions to capture temporal patterns within the signal sequences. The MIT-BIH Arrhythmia Database was employed for training and validation, encompassing five heartbeat classes: Normal, Supraventricular, Ventricular, Fusion, and Unknown beats. Experimental results demonstrate that the model achieved a validation accuracy of 98%, accompanied by balanced precision, recall, and F1-scores for the majority class. Furthermore, the model demonstrated robustness in detecting critical minority classes, achieving a sensitivity (recall) of 93.7% for Ventricular beats and 57.1% for Supraventricular beats. This model was subsequently implemented into a web-based system to enable real-time and flexible ECG signal analysis. These findings indicate that the integration of CNN and Transformer effectively enhances arrhythmia detection accuracy, not only for majority classes but also for rare pathological classes.