Deep learning methods for tuberculosis diagnosis from chest radiographs

According to the World Health Organization, Every year, 10 million people fall ill with tuberculosis (TB). Despite being a preventable and curable disease, 1.5 million people die from TB each year – making it the world's top infectious killer. Despite the fact that it is important to diagnose this disease in the early stages in order to treat and prevent its further spreading, the use of traditional diagnostic tools proves to be inadequate because of a lack of tools and specialists needed for these procedures, particularly in the Third World countries. This paper depicts an autonomous analysis model of the tuberculosis that uses chest radiographs, based on deep Convolutional Neural Networks (CNNs). The model is trained on a large data set of X-ray pictures of the chest and to increase the picture quality, state of the art preprocessing techniques were adopted. The CNN design Model was identified and capable of identifying the small detailed structures that could signify TB. They presented good results after several trials- they were used until the convergence point. The model provided higher than 90% accuracy in the detection of the disease in contrast to other approaches, which are not as efficient as the implemented method, this is supported by the experiment. The findings thus elucidate the ability of deep learning to completely revolutionize TB diagnosis thus offer cheaper and easy-to-scale approaches to early identification. Therefore, the study establishes how crucial it is to integrate deep learning and medical imaging approaches in tackling critical health care issues. As a result, the collection of more data and the fine tuning of the algorithm with its implementation in a real hospital setting are proposed in the future work.