Convolutional Neural Networks Model for Screening Tuberculosis Disease
Abdulfattah E. Ba Alawi, Amer Al-basser, Amer A. Sallam, Amr Al-sabaeei, Hesham Al-khateeb
Abstract
Tuberculosis disease has a big concern and it is spreading quickly across the world. The secret for managing the condition is an accurate diagnosis. Acid quick staining, conventional approaches such as tuberculin skin test (TST), yield findings are unreliable or require more time to detect. This paper presents an automated solution that uses chest radiographs to diagnose tuberculosis. Chest radiographic images are used for tuberculosis diagnosis. Tuberculosis in chest radiographs is difficult to investigate under the current system of cavity identification, ribs, and diaphragm removal. By using a CNN-based model, the lung area is separated to resolve the problems. The proposed technique can classify chest x-ray (CXR) images as Tuberculosis (TB) infected or not. We analyzed 3500 CXR cases and 3500 normal cases with exposure to tuberculosis. Then, we built and trained our own CNN and found that the features map or heat-map generated from this network performed a slightly better job. The implementation was done in Tensorflow and Keras library. An accuracy of 98.71%, a sensitivity of 98.86%, and a specificity of 98.57% were achieved.