Detection of arrhythmias with the improved transformer-based deep learning approach

Background Heart arrhythmias are one of the leading causes of death worldwide. Electrocardiography (ECG) is used to detect and classify these arrhythmias. ECG provides a mechanism that offers comprehensive information about the heart through its electrical activity. It does not harm the patient and is cost-effective. However, the evaluation process is time-consuming and open to observer-related errors. Improving deep learning-based approaches for detecting normal, abnormal, tachycardia, and bradycardia conditions that occur through ECG signals is a gain in reducing errors. Objective This study investigates the effectiveness of a deep learning-based architecture in automatically detecting and classifying heart rhythm disorders from ECG signals. Methods An original dataset was created with four categories (normal, abnormal, tachycardia, and bradycardia), including regular and abnormal heartbeats. The research evaluates the potential to increase the accuracy and efficiency of arrhythmia detection by hybrid use of transformer and transfer learning approaches, and the proposed normalization techniques. Results The proposed hybrid model achieved a classification accuracy of up to 98.98% on the test dataset. Conclusion These findings highlight the potential of AI-enabled systems to revolutionize ECG analysis and provide a robust, cost-effective, and scalable solution for early diagnosis and treatment in cardiology. This research is a significant contribution to the application of AI in healthcare. It is also critical for improving interdisciplinary collaboration and patient-centered diagnostic tools.