AI-Driven Approaches for Real-Time Beat-by-Beat ECG Signal Classification and Cardiac Health Monitoring in Smart Healthcare Systems

Cardiovascular disease, which refers to a range of diseases that affect the heart, including vascular diseases such as coronary artery disease, arrhythmia, myocardial disease and various forms of heart disease, is a leading cause of death worldwide. Smart Health (sHealth) in future can transform the utilization of biosignals including Electrocardiogram (ECG) for monitoring and managing cardiac patients. The ECG classification system uses technologies such as artificial intelligence (AI) to automatically detect and classify complex cardiac events with high accuracy. With the operability of wearable devices and IoT in the healthcare field, continuous real-time ECG monitoring has become more feasible. If the heart rhythm is abnormal, point-of-care ECG monitoring can immediately generate alert and notify the individual or medical professional. We have explored the role of feature engineering in optimizing heartbeat classification models. We utilized a large set of ECG features, then narrowed it down to the features that have the most impact on model performance through feature ranking technique. Machine learning (ML) and Deep learning (DL) models such as Random Forest and Support Vector Machine (SVM) achieved significant accuracy by leveraging this technique. We further explored the combination of signal-specific and signal-independent features to enhance the models, achieving peak performance through fine-tuned configurations. We further examined the effects of intra- and inter-patient analysis in ECG beat-by-beat classification by using DL models and selecting features through variation analysis and principal component analysis (PCA). Our research highlights the complexity and necessity of customizing healthcare models based on individual patient characteristics and collective patient data sets. The approach supports personalized patient care by optimizing prediction accuracy and adapting models to individual and grouped patient data. Finally, we proposed a novel hybrid transformer neural network model designed for efficient and accurate ECG heartbeat classification. This study evaluates the performance and energy consumption of various models, exploring the superiority of hybrid transformer models in terms of accuracy and computational resource management. This model can be used for real time cardiac patient monitoring and can be integrated into sHealth systems to enhance patient care.