Knowledge Discovery Pipeline for the Diagnosis of Tuberculosis

Abstract

Accurate and early tuberculosis (TB) diagnosis is very important for efficient disease management. Chest X-rays are widely used for the TB diagnosis that infects a high number of people worldwide. The manual examination of chest X-rays is challenging and it is limited by the need for expert doctors. Moreover, TB chest X-ray is often misclassified to some other disease conditions of similar patterns such as COVID-19 and pneumonia. Chest X-ray misclassification can potentially lead to delayed treatment and disease spread. In recent years, deep-learning approaches have shown great promise in image analysis, particularly in medical image analysis. This study aims to focus on the automated detection of TB in chest X-rays using deep learning approaches. A comprehensive dataset consisting of 2500 TB, 2517 normal, and 2522 COVID-19 chest X-ray images was collected from different local hospitals in Pakistan and preprocessed. A convolutional neural network (CNN) has been employed for classification, and the performance of the CNN has been compared with different pre-trained deep learning models, including VGG-16, ResNet-50, InceptionV3, and DenseNet-121. Accuracy, precision, recall, and F1-Score were also calculated to evaluate the model performances. Out of all the models, CNN achieved a higher accuracy of 97.67%. The results obtained indicate the CNN model’s effectiveness in accurately classifying TB from chest X-rays. The trained CNN model was deployed into a user-friendly web application for real-time diagnosis. This study contributes to the field of image analysis by highlighting deep learning’s potential in TB diagnosis. For radiologists and other medical experts, the developed CNN is a valuable tool assisting them in TB diagnosis more accurately and efficiently.