Real-Time ECG Image Classification Using GA–PSO Optimized CNN-LSTM Model Trained on PTB-XL Dataset

Objectives: To design an optimized ECG image-based classification framework where it functions in real time without compromising with diagnostic accuracy and computational complexity. Methods: The hybrid CNN–LSTM architecture for learning the spatial and temporal features from ECG image representations. The convolutional layers are used to obtain the structural features of an image from the input signal and then LSTM layers learn dependencies of signals patterns sequentially. An innovative classifier using a hybrid Genetic Algorithm and Particle Swarm Optimization based feature selection technique is used to eliminate redundant information. Training and validation are performed using the PTB-XL dataset from ECG. During training the Adam optimizer is used for a stable convergence. The whole framework is performed in MATLAB to ensure real-time classification of the ECG images into normal and abnormal (PTB-XL Dataset). Findings: The framework achieved an accuracy, precision, recall, F1-score and specificity of 98.62%, 97.92%, 97.84%, 98.04% and 98.79% respectively. Inference time per ECG image: 5.46 milliseconds. Novelty: An integrated real-time electrocardiogram (ECG) image classification framework is proposed in this study which combines the deep feature learning, hybrid metaheuristic-based feature selection and stable training optimization to satisfy practical clinical decision support needs. Keywords ECG Classification, PTB-XL Dataset, CNN-LSTM, Genetic Algorithm, Particle Swarm Optimization, Adam Optimizer