Hybrid deep learning metaheuristic ensemble framework for enhanced COVID-19 classification using ResNet and particle swarm optimization

Abstract The rapid and accurate diagnosis of COVID-19 remains a major challenge in medical imaging and artificial intelligence. To address this issue, this study proposes an optimized deep residual learning framework that integrates Particle Swarm Optimization (PSO) with the ResNet architecture for precise classification of chest CT images into five distinct categories: COVID-19, Normal, Pneumonia, Probable COVID-19, and Possible Pneumonia. The proposed hybrid model achieved an overall accuracy of 99.02% for this 5-class task. The proposed hybrid model employs PSO to fine-tune ResNet’s hyperparameters, thereby enhancing its convergence stability and classification accuracy. Multiple ResNet variants are trained and ensembled to improve generalization and robustness against noisy or imbalanced data. Comprehensive preprocessing and data augmentation strategies are applied to improve model resilience and prevent overfitting. Experimental evaluations on a large-scale CT image dataset demonstrate that the PSO-optimized ResNet ensemble achieves superior performance compared to conventional deep learning and optimization-based approaches, with an overall accuracy of 99.02%, sensitivity of 98.57%, and specificity of 98.02%. The results highlight the potential of combining metaheuristic optimization with deep residual learning to deliver a reliable AI-assisted diagnostic framework for rapid and accurate COVID-19 detection in clinical practice. Graphical abstract