An Improved Deep Network of Pulmonary Tuberculosis Lesions Detection based on YOLO.

Tuberculosis (TB) has been a potentially serious infectious lung disease for decades and remains the leading cause of death worldwide. Digital radiography (DR) imaging technology can locate and display suspicious areas of tuberculosis, and improve the quality of tuberculosis diagnosis. However, the tuberculosis lesions area has high compatibility with the background, the target has serious diffusion and the edge morphology is extremely irregular, which seriously interferes with the accuracy of diagnosis. Aiming at the above problems, this paper proposes a detection framework for pulmonary tuberculosis lesions based on a deep feature enhancement network. which is mainly composed of a deep-level feature enhancement module (DFEM) to strengthen the model to extract the deep semantic information of DR images. The experimental results show that on the TBX11K data set, the Average Precision (IOU=0.5) of the proposed detection framework reaches 82.9%, and the recall rate reaches 77.6 ^% . Compared with the YOLO series network, the model in this paper effectively improves the detection accuracy of pulmonary tuberculosis lesions and achieves accurate positioning and classification results.