An Ensemble CNN-Dempster Shafer based tuberculosis detection from chest x-ray images

Pulmonary Tuberculosis remains as an epidemic in the world and continues to be one of the leading causes of death according to WHO. The number of cases reported worldwide is increasing rapidly. An automated Computer Aided Diagnostic (CAD) System is required for early detection and prevention. Deep learning algorithms have emerged as popular techniques when it comes to disease detection. Though detection of tuberculosis can be done using different techniques, chest x-rays are most popular due to their availability and relatively low cost. In our proposed tuberculosis detection model, an ensemble CNN DempsterShafer based architecture is implemented. Contrast of the images are first enhanced using CLAHE method and then the training set is augmented to increase the training dataset size. The results of the experiment have shown accuracy of 94.21%. Our proposed method can detect whether an x-ray image is normal, or if it contains manifestations of tuberculosis. This model will help in timely and accurate detection of tuberculosis.