Multiclass Classification of Chest X-rays based Pulmonary Disorder Using a Specialized VGG-19 Deep Neural Network

Respiratory infections such as COVID-19, tuberculosis (TB) and pneumonia, remain important global health challenges, often requiring rapid and accurate diagnosis to prevent complications. Due to the visual similarities in chest X-ray (CXR) images, distinguishing between these diseases can be complex. In this study, we proposed, a deep learning (DL)-based model utilizing a customized VGG-19 architecture for multiclass classification of lung diseases, including COVID-19, pneumonia, TB, and healthy cases. A total of 5,928 CXR images were collected from open-access platforms, comprising COVID-19, pneumonia, TB, and normal cases. The dataset was pre-processed using bilateral filtering for noise suppression and Multiscale Retinex for image enhancement. FFurthermore, data augmentation and image resizing were also applied to increase robustness. When compared with the state-of-the-art techniques, the proposed method achieved a classification accuracy of 98.48% in identifying various lung disorders, with precision at 97% and an F1-score of 96%, indicating that it is an appropriate technique for computerized lung disease diagnosis in clinical environments.