Research On Methods For Lung Parenchymal Segmentation And Pulmonary Nodules Detection Based On Convolutional Neural Network

Lung cancer is one of the most common cancer killers in the world today and poses a great threat to human health.Pulmonary nodules are early manifestations of lung cancer.Accurate screening of lung nodules has important positive effects on lung cancer prevention and treatment.Current screening for pulmonary nodules is primarily aided by computed tomography(CT).It is time-consuming and laborious for doctors to screen pulmonary nodules by observing CT images with eyes.A computer-aided diagnosis system that introduces pulmonary nodules detection can effectively improve the efficiency of lung nodules detection and reduce the rate of missed detection.The traditional lung nodules detection method has a complicated detection process,and the lung nodules feature extraction is cumbersome,the degree of automation is low,and the detection effect is general.In recent years,deep learning has developed rapidly,and convolutional neural network has made a big splash in the field of computer vision,which has brought new research directions to the detection of pulmonary nodules.The goal of this paper is to achieve automated lung parenchymal segmentation and pulmonary nodules detection based on convolutional neural network and other related methods,and to improve the accuracy of lung nodules detection as much as possible.Firstly,a lung segmentation method based on improved U-Net is proposed.Shortcut operation in residual network is introduced into U-Net network structure and the depth of the whole network is deepened.The process of improving U-Net network structure and the details of the improved network are described in this paper.At the same time,in order to improve the segmentation effect of the network on the edge of lung parenchyma,a new boundary penalty term is designed in the loss function of lung parenchyma segmentation.Secondly,this paper evaluates the advantages and disadvantages of the current mainstream target detection algorithms.After analysis and comparison,Faster RCNN is chosen as the main algorithm of lung nodule detection.In addition,this paper improves the backbone network of Faster RCNN,using two different scale residual feature extraction networks instead of VGG-16 feature extraction network in Faster RCNN source code.Finally,the K-means clustering analysis algorithm is used to search the appropriate anchor parameters in the training set,which makes the target detection algorithm more suitable for detecting small targets such as pulmonary nodules.Experiments show that the proposed lung parenchyma segmentation method and lung nodules detection method both have high accuracy and efficiency.