| At present,lung cancer has become the tumor disease with the highest morbidity and mortality in the world.Once patients are diagnosed with lung cancer,they are often at the advanced stage of lung cancer,and the five-year survival rate of patients with advanced lung cancer is very low.Therefore,early detection,early diagnosis and early treatment of lung cancer patients are very important.The early symptoms of lung cancer are presented in the form of lung nodules on medical images.Accurately detecting lung nodules and judging their benign and malignant is of great significance for the prevention and treatment of lung cancer.In this thesis,the public dataset of lung CT images LIDC-IDRI is taken as the research object.In order to solve the problem of lung segment nodule detection algorithm based on convolutional neural network,the lung CT images need to be segmented and pre-processed.Lung nodule detection method based on detection algorithm;Aiming at the imbalance in the classification of benign and malignant lung nodules in the data set,study benign and malignant lung nodules based on Deep Convolutional Generative Adversarial Networks The method of processing unbalanced data in this section is as follows:(1)This thesis uses the classic Faster R-CNN algorithm as the detection network based on the target detection algorithm.In order to solve the problem of parameter redundancy and low average accuracy reflected in medical image detection,the feature extraction network and candidate regions Improvements were made separately.By optimizing the parameters of the feature extraction network and introducing the Inside Net network,the parameters of the network structure were reduced,and the average accuracy of lung nodule detection was improved.Through experimental simulation and comparison,it is found that the improved Faster R-CNN algorithm has an average detection accuracy of 86.15% under the conditions of similar detection time,which is 6.05% higher than the original Faster R-CNN algorithm.(2)Aiming at the problem of unbalanced number of benign and malignant pulmonary nodules in the public data set LIDC-IDRI,a data augmentation method was used to increase the sample data volume and change the distribution of the data set.Combining traditional data enhancement methods(affine transformation)with DCGAN,and using traditional methods to expand the original data set,use DCGAN to implicitly learn the data distribution from the sample images to generate new lung nodules with similar characteristics Image data,thereby completing data enhancement and changing the uneven distribution of the data set.Classification of benign and malignant lung nodules was performed on a convolutional neural network model based on residual network.Compared with traditional data augmentation methods,the classification sensitivity and specificity of the method in this thesis have increased by 13.17% and 12.42%,respectively. |
