Detection and Analysis of Tuberculosis Disease from Chest X-Ray (CXR) Images using Machine Learning and Deep Learning Techniques

Tuberculosis (TB), a significant global health concern, is a chronic lung disease resulting from bacterial infection and ranks among the top 10 leading causes of mortality worldwide. This research aims to propose a model to tackle Tuberculosis's early diagnosis problem. Two publicly available benchmarked datasets from Kaggle namely, Tuberculosis (TB) CXR Database and Shenzhen Hospital CXR Set are used in this work. In this pursuit, a hybrid framework for Tuberculosis classification is proposed. Initially, a deep CNN, DenseNet201 is employed for feature extraction. Following this, the robust eXtreme Gradient Boosting (XGBoost) classifier is used to classify images into TB and Normal classes. This ensemble approach aids in the automated detection of Tuberculosis from CXR images. A comprehensive set of metrics, including accuracy, precision, sensitivity, specificity and F1-scores are used to evaluate the results of the proposed model. DenseNet201-XGBoost performs exceptionally well on both datasets giving F1-scores of 99.83% and 80% respectively. Various other CNNs and Transformer based models such as ResNet101, VGG19, Inception and EfficientNet in combination with the XGBoost classifier are implemented for a comparative analysis. Thus, by amalgamating deep feature extraction models with advanced classification models, this study forges a path toward enhanced Tuberculosis detection.