Detection and segmentation of pulmonary nodules based on improved 3D VNet algorithm

Abstract

Lung cancer has one of the highest incidence rate and mortality rates in the world. Therefore, the recognition and diagnosis of pulmonary nodules based on CT images is an essential index for the diagnosis of lung cancer. Therefore, the detection and segmentation of pulmonary nodules are of great significance in assisting doctors in detecting lung diseases. In this paper, the luna16 data set in the LIDC-IDRI (the lung image database consortium) initiated by the National Cancer Institute is used for research. Given a large amount of data and the diversity of types and sizes of pulmonary nodules in this data set, the detection, and segmentation of various pulmonary nodules are realized based on the improved deep neural network 3D VNet. The deeper the network, the stronger the expression ability and the better the performance. However, with the increase of network depth, there will be problems such as gradient dissipation and gradient explosion. The improved 3D VNet network improves this problem. More than 200000 lung CT images in the LUNA16 data set are preprocessed, such as image denoising and interpolation sampling, and then coarse segmentation images and mask images are generated. They are trained many times in the improved 3D VNet model, and the results with high detection accuracy are obtained. The experimental results show that the dice similarity coefficient (DSC) and IOU of the improved 3D Vnet network can reach 88.29% and 87.25%, which has a good effect on the current detection methods of pulmonary nodules.