Tuberculosis Detection based on Chest X-Rays using Ensemble Method with CNN Feature Extraction

Tuberculosis is an infectious disease which mostly attacks human lungs and has become a public health problem because the disease is easily transmitted. According to global estimation in 2019, 10 million people suffer from tuberculosis which results in 1.2 million deaths. However, this high mortality rate could be prevented by early detection. Among the several examinations performed to detect tuberculosis, chest x-rays are widely used for screening and early detection. Also, advanced technology such as machine learning and deep learning helps to improve early detection by automatically classifying the presence of disease through medical images. In this study, the combinations of Convolutional Neural Network (CNN) feature extraction and ensemble method classifiers, such as Random Forest (RF) and Extreme Gradient Boosting (XGBoost), were proposed to classify chest x-ray images into two classes, namely tuberculosis and normal. Using Kaggle Tuberculosis Chest X-Ray Database, CNN-RF model provided the best accuracy and AUC results of 98.667% and 99.933% respectively, while the CNN-XGBoost gave the best results of 98.367% and 99.866 respectively. Both models provided the best performance results, but the accuracy and AUC values of CNN-RF are better than the CNN-XGBoost in classifying tuberculosis based on chest x-ray images.