Lung X-Ray Image Classification via Feature-Level Fusion of AlexNet and NASNet Models Using SExpCSA Optimizer

Lung diseases continue to pose a significant global health burden, necessitating rapid and reliable diagnostic frameworks. Chest X-ray (CXR) imaging remains the most widely utilized screening modality due to its cost-effectiveness and accessibility; however, manual interpretation is time-consuming and prone to inter-observer variability, particularly in multi-class scenarios with overlapping radiographic features. To address these limitations, a hybrid deep learning framework is proposed that integrates the spatial feature-extraction capability of AlexNet with the hierarchical semantic representation strength of NASNet-Large via feature-level concatenation. The fused representations are optimized using the Serial Exponential Chameleon Swarm Algorithm (SExpCSA), a metaheuristic optimizer that balances global exploration and local exploitation through exponential decay–guided swarm dynamics. The model was trained and evaluated on 3,295 CXR images across five classes (COVID-19, pneumonia, tuberculosis, pulmonary nodules, and normal) using a stratified training–validation–testing protocol. Experimental results demonstrate an overall accuracy of 98.03%, precision of 98.26%, recall of 98.21%, F1-score of 98.20%, and strong multi-class discrimination performance. Comparative analysis indicates superior performance relative to several conventional CNN-based architectures. The findings highlight that feature-level fusion combined with adaptive swarm-based optimization significantly enhances classification robustness in high-dimensional medical imaging tasks. Although computationally intensive, the proposed framework demonstrates promising performance for computer-aided pulmonary disease classification and screening.