AI-Powered X-ray System for Detecting Tuberculosis, Pneumonia and COVID-19

Abstract

In rural areas, the availability of radiologists is limited, and the diagnostic process for them can be challenging due to the high volume of patients at the same clinic. To address this issue and reduce the time required for diagnosis, a robust technology for chest X-ray image analysis has been introduced. This paper presents two convolutional neural network (CNN) models designed for multi-class classification of diseases such as Tuberculosis, Pneumonia, and COVID-19. The authors performed image pre-processing, data augmentation, and employed deep learning techniques to classify chest X-ray (CXR) images, using a dataset of 18,781 images obtained from Kaggle. The data was split into training, validation, and test sets in an 80:15:5 ratio. The system is designed to assign a probability to each diagnosis, generate a simple report, and eliminate the delay in report analysis. CNNs and a medical AI library were used to develop the models. Testing results showed that Model 1 achieved accuracy rates of 94.5%, 99.3%, 99.5%, and 94.9% for Normal, Tuberculosis, Pneumonia, and COVID-19 classes, respectively, with sensitivity values of 96.7%, 82.9%, 100%, and 86.9%. Model 2 achieved accuracy rates of 95.1%, 99.2%, 99.3%, and 94.8% for the same classes. The proposed models have the potential for automated and expedited diagnosis of these diseases.