A Method for Detecting Lung Nodules Based on YOLOv8-DLung

As a serious public health problem, lung cancer has the highest morbidity and mortality among all cancer types. Accurate detection of lung nodules is essential for early intervention of lung cancer and prevention of its spread. Therefore, this paper proposes a deep learning network, YOLOv8-DLung, to improve the detection accuracy of lung nodules by using deep learning methods, reduce the misdiagnosis rate, and thus improves the survival probability of patients. First, the model refers to the overall architecture of YOLOv8 model, adding dilatative convolution in the backbone network to expand the filter region, aiming to capture a wide range of correlation information. At the same time, SENet is used after SPPF module to further filter and fuse the information extracted from the backbone network. The spatial information of the lesions and the information between the channels in CT images of pulmonary nodules are effectively used. The results in LUNA16 public data set show that the accuracy of the model is 94.1%, and the mAP is 95.5%. In addition, the average reasoning speed of each image in the test set is 25 ms, which can effectively detect the pulmonary nodule region.