Detection Of Bronchial Tuberculosis Using ASFF-Yolov5S Model

On CT images, irreversible damage caused by bronchial tuberculosis can be seen, and early observation and intervention can provide assistance for the patient’s prognosis. So in this article, it is proposed to use depth learning to detect CT images of patients. However, in the learning process of deep learning classification models, as each level continues to expand, the focus points of each layer’s features will be covered and the connection will become weaker, leading to the interruption of the connection between levels and the deterioration of the learning effect. In this article, we use an adaptive spatial feature fusion mechanism, which can automatically learn features at different spatial scales and fuse them together, thereby improving the model’s generalization ability and increasing the connectivity of hierarchical relationships in the judgment of bronchial tuberculosis, making the judgment more accurate. Specifically, based on the YOLO V5 model, various levels of information extracted from the Neck section are utilized to enhance features for prediction through an adaptive spatial feature fusion mechanism. The model proposed in this article outperforms commonly used object detection models in terms of performance and accuracy. In addition, the proposed framework may serve as a preliminary screening tool for the differential diagnosis of bronchial tuberculosis.