An Automated Detection and Classification of Lung Diseases Using a Hybrid Deep Learning Technique on Chest X-ray Images

Lung diseases, such as COVID-19, pneumonia, TB, lung cancer, and lung opacity, continue to be significant clinical difficulties requiring quick and precise diagnosis. Because of overlapping radiographic patterns, low contrast, and subjective variation, manual interpretation of chest X-rays (CXR) is highly prone to error. For reliable multi-class lung illness classification, this work proposes a novel hybrid deep learning architecture that combines sophisticated pre-processing, saliency-driven region localisation, optimised deep feature extraction, and meta-heuristic feature selection. The pipeline employs a unique 14-layer deep saliency network to isolate disease-relevant regions, bilateral filtering and CLAHE for quality enhancement, and Bayesian-optimized EfficientNet-B7 for feature extraction. A Hybrid Binary Dwarf Mongoose Optimisation with Simulated Annealing (HBDMOSA), the first such application in medical imaging, to further improve high-dimensional features while ensuring the selective retention of maximally discriminative features. KNN is used for the final classification throughout the optimised feature space. The suggested model has an external validation accuracy of 95.6% on the NIH ChestX-ray14 dataset and achieves 98.63% accuracy, 93.89% precision, 88.39% recall, and 91.6% F1-score when tested on publicly available multi-class CXR datasets. The synergistic combination of deep saliency segmentation, Bayesian optimisation, and HBDMOSA improves performance over non-optimized baselines by 3–6%, according to ablation analysis. The findings show that the proposed structure offers an extremely reliable and therapeutically flexible approach to automated lung disease classification.