PERTANIKA JOURNAL OF SCIENCE AND TECHNOLOGY

Healthcare practitioners can rapidly recognise lung problems due to the invaluable and essential role that chest X-ray imaging plays in lung diagnosis. Recently, deep learning techniques have gained popularity and demonstrated promising outcomes in automatically interpreting medical images, especially in the specialty of chest radiology. For classifying chest X-ray images into multiple classes, we propose an enhanced deep learning approach in this research. Initially, the input images are resized, normalised, and data augmented to improve the network's generalisation ability. The local and global features are extracted using the MobileNetV2 model. An Intelligent Prairie Dog Optimisation (IPDO) algorithm is employed to select the optimal features, thereby minimising the dimensionality of the features. This algorithm finds the optimal solution with a higher convergence rate. Finally, the multi-class chest x-ray images are classified using the Enhanced GoogLeNet (EGoogLeNet) model. The model parameters are optimised using the Adam optimiser. The proposed approach is evaluated using a large-scale C19RD dataset, and the findings indicate improved performance in lung multi-class diagnosis. Extensive experiments are conducted against existing state-of-the-art approaches to assess the proposed model's performance, yielding significant improvements in F1-score, recall, precision, and accuracy. Promising findings are found in the assessment of the proposed method's performance (EGoogLeNet), which achieves a multi-class classification accuracy of 99.41%. The proposed model performed better, enabling medical professionals to diagnose and treat patients more rapidly and effectively.