Explainable AI Based TB Prognosis Using X-ray Images

Tuberculosis (TB) is a multi-drug-resistant disease which continues being a health security threat. The World Health Organization has declared that there could be a possibility to unravel speedy progress in the field of fighting tuberculosis due to the COVID-19 pandemic. This study includes the implementation, comparison and analysis of various Deep Learning for TB prognosis. Experimental results have demonstrated that the YOLOv10 model is not fit for this particular use case although it performs well in general object detection tasks. For TB screening, the Convolutional Neural Network (CNN) based models are a better approach as it has achieved a satisfactory accuracy of 92% using our custom-designed CNN model with the combination of filters like CLAHE, Gabor and Canny-Edge. ResNet50 and VGG19 models achieved the accuracy of 87% and 91% respectively with the same filter combination. Explainable AI tools like LIME and SHAP were also incorporated so as to build trust between the models and the users. The results show the areas that the model focuses on while making the prediction, so it is not just a black-box approach but also helps in knowing the reasoning of the model and the areas of interest while giving the results. An interface for the populace has been developed to scan their CXRs and provide them with a diagnosis.