Pulmonary tuberculosis detection using deep learning convolutional neural networks

Tuberculosis (TB) is classified as one of the top ten reasons for death from an infectious agent. This paper is to investigate the accuracy of two methods to detect Pulmonary Tuberculosis based on the patient chest X-ray images using Convolutional Neural Networks (CNN). Various image preprocessing methods are tested to find the combination that yields the highest accuracy. Moreover, a hybrid approach using the original statistical computer-aided detection method combined with Neural Networks was also investigated. Simulations have been carried out based on 406 normal images & 394 abnormal images. The simulations show that a cropped region of interest coupled with contrast enhancement yields excellent results. When further enhancing the images with the hybrid method even better results are achieved.