Prediction of Pulmonary Tuberculosis with Hemoptysis based on Deep Learning

From a medical perspective, tuberculosis itself is not a severe disease. However, its complication, hemoptysis, is an extremely dangerous one and can even be life-threatening when it progresses to massive hemoptysis. Specifically, cavitary tuberculosis caused by tuberculosis and the presence of calcified granulomas in the lungs are highly likely to cause hemoptysis. Timely detection and management of these lesions in the early stages can effectively prevent hemoptysis caused by tuberculosis. Due to the utilization of multi-scale training and the use of finer-grained feature maps for examination, YOLOV7 is more suitable for detecting small target lesions mentioned in the previous text. Therefore, in this study, we employed the YOLOV7 object detection model to detect these two types of lesions. To minimize information loss during feature extraction, we designed one residual branch in the backbone network to better integrate shallow and deep information. Additionally, we combined it with the Contextualtransformer module to improve feature extraction for small target lesions. Finally, we proposed an improved model based on YOLOV7, named RC-COT-YOLOv7. Experimental results showed that under the same conditions, the improved RC-COT-YOLOV7 model achieved approximately a 2% upper mean average precision (MAP) compared to the original model.