부산대학교 도서관

Lung cancer is one of the cancers with the highest incidence and mortality in the world, and early detection of lung nodules can effectively reduce the mortality rate of lung cancer. At present, most of the lung nodule detection methods based on deep learning face the problem of low sensitivity or high false positive rate. This paper proposes a loss function suitable for the clinical features of lung nodules and introduces a novel Squeeze-and-Excitation module in the depth direction, which alleviates the problem of high false positive rate in the detection process and achieves good results. In this paper, the competition performance metric (CPM) on the LUNA16 dataset reached 88.16%.