Modified U-Net block network for lung nodule detection

Abstract

Since early detection of lung cancer is the most effective way to improve the survival rate of lung cancer patients, accurate detection of lung nodules is a key step in the diagnosis of lung cancer from CT images. In this paper, a deep learning-based approach is develop to detect lung nodules. First, a segmentation network is used to extract the suspected lung nodules, and then a classification network is used to perform the precise detection of the suspected lung nodules. On the one hand, according to the residual network and shortcut connection in ResNet, the gradient disappearance phenomenon and the more accurate acquisition of features can be effectively solved. We proposes a modified U-Net Block network, by applying the ResNet network to the U-Net network, improves the accuracy of segmentation of suspected lung nodules. On the other hand, multi-classification networks are combined to detect the suspected lung nodule areas, which effectively reduces the problem of false detection in single-class network, especially improving the accuracy of small nodule detection. Finally, superior performance was obtained in the evaluation of the lung nodule detection system using the scoring function, and the FROC score was 0.73.