An Optimized Machine Learning Model for R and T Peak Detection in ECG Signal

Detection of R and T peaks in electrocardiogram (ECG) signals is essential for monitoring and diagnosing heart-related diseases. In generic cases, these peaks suffer through different types of noise and variability, which are of important clinical significance in identifying diseases like myocardial infarction, ischemia, and electrolyte imbalance. Machine learning techniques are integrated computer-aided diagnosis systems (CIDS) [1] to reduce manual effort in peak analysis by medical professionals. Machine learning models are used for learning from ECG nonlinear patterns and tested to mark actual R and T peaks. Once the machine learning model is trained and tested. Minimizing error or loss in the implemented machine learning model is possible through tuning parameters and finding the best fitness value. This article proposed an optimized metaheuristic approach (finding best fitness) for convolutional neural network (CNN), to detect R and T peaks in captured ECG, aimed, at improving the accuracy and robustness of real-time signal data. Before training and testing optimization, different approaches of metaheuristics optimization were evaluated based on best fitness values over iteration and found optimization technique having lowest convergence. Applying Particle Swarm Optimization (PSO) with CNN achieved 99.57% accuracy, 99.13% precision, 99.28% recall, and 99.25% F1-Score on the publicly available ECG database. The paper results that the proposed approach not only detects peaks but also improves the capabilities of the model for real-time cardiac monitoring systems development.