Bio-inspired hybrid algorithms in deep learning model for the detection of COVID-19 and lung diseases

In the field of medical image analysis, accurate and efficient lung diseases classification remains a critical challenge. This study introduces an innovative approach leveraging deep learning and metaheuristic optimisation techniques to enhance diagnostic accuracy. We investigated five pre-trained convolutional neural network models (AlexNet, VGG19, ResNet50, InceptionV3 and MobileNet) and developed hybrid optimisation strategies using cuckoo search (CS) and grey wolf optimiser (GWO) algorithms cascaded with simulated annealing (SA). The experimental results demonstrated that ResNet50 outperformed other models, serving as the foundation for further optimisation. The proposed model demonstrates exceptional performance in COVID-19 classification, achieving a remarkable 99.68% accuracy. This represents a significant improvement over the cuckoo search-simulated annealing (CS-SA) approach, which demonstrated a 99.04% accuracy rate. This research not only advances computational intelligence techniques in medical image classification but also offers a promising framework for developing more precise and reliable automated diagnostic support systems for lung diseases.