Multi-Label Classification for Diagnosis of Tuberculosis from Chest X-Ray Images

The study topic's main goal is to apply science to early tuberculosis screening and diagnosis, which plays an important role in tuberculosis infection control and treatment. A computer-aided approach based on deep learning and machine learning is presented for the diagnosis of different kinds of TB lesions in chest radiographs. The YOLOV5 algorithm and the fully convolutional neural network approach are utilised in this system to separate the lung region from the complete chest radiograph for pulmonary TB identification. The suggested study, unlike prior studies that looked at the full chest radiograph, concentrates on particular TB lesion regions and provides the first multicategory tuberculosis lesion identification system. By mining indistinguishable samples during the training phase and utilising reinforcement learning to decrease false-positive lesions, the Faster Region-based Convolutional Network (Faster RCNN) enhances the identification of small-area lesions. The proposed computer-aided system outperforms current systems for assisting radiologists in diagnosing tuberculosis and public health providers in tuberculosis screening in tuberculosis-endemic areas.