Research On Detection And Benign And Malignant Classification Method Of Pulmonary Nodules Based On Convolutional Neural Network

Lung cancer is the leading cause of death.Radiologist often use medical imaging to screen pulmonary nodules and diagnose pulmonary nodules lesion types.Due to the limitations and influences of radiologist' experience and knowledge level,the method of manual reading is highly prone to high false negative rate and false positive rate.At present,computer-aided detection and diagnosis systems based on deep learning have achieved exciting results in lung cancer,and their emergence has brought good news to patients and doctors.The lung cancer computer-aided detection system can achieve rapid pulmonary nodule screening while ensuring high detection rate.The computer-aided diagnosis system for lung cancer can assist radiologist in diagnosing the benign and malignant nodules,and can improve the accuracy and efficiency of diagnosis while reducing the workload of radiologist.In the process of lung cancer diagnosis,the effective detection of pulmonary nodules is the basis of benign and malignant diagnosis,and the diagnosis of benign and malignant pulmonary nodules is the only way to achieve rapid cure of lung cancer.This subject deeply analyzes and explores the shortcomings of existing pulmonary nodule detection and classification techniques.Based on the convolutional neural network,the following studies were performed on pulmonary nodule detection and benign and malignant classification methods:(1)The existing computer-aided detection system for lung cancer can detect nodules with high sensitivity,but it is difficult to solve the problem of high false positives in candidate nodules.In order to solve this problem,we applied the three-dimensional convolutional neural network to the false positive reduction stage,and proposed a pulmonary nodule detection method based on multi-scale three-dimensional convolutional neural network.The method first combines three specific pulmonary nodule detection algorithms to screen candidate nodules with high sensitivity,and then designs different scales of three-dimensional convolutional neural networks for different sizes of nodules.Next,cube samples of different scales are input into the corresponding scale network to train their recognition ability.Finally the output results of the three networks are combined to obtain the final recognition result.Experiments show that the proposed method achieved high detection sensitivities of 84.9% and 90.9% at 1 and 4 false positives per scan,respectively,which demonstrate that the method is suitable for reducing false positives in detection systems.(2)The traditional computer-aided diagnosis system for lung cancer has the problem of complex extraction process of pulmonary nodule and insufficient representation ability of nodule feature.In addition,the lack of pulmonary nodule dataset restricts the accuracy of classification experiments.In order to solve these problems,we introduce deep convolutional neural network into the classification model of pulmonary nodules,and propose a classification method based on adaptive feature selection of deep convolutional neural network for benign and malignant pulmonary nodules.The method firstly uses the pre-trained deep convolutional neural network to extract features from the pre-processed pulmonary nodule images,and then uses the proposed adaptive feature selection algorithm to select feature subsets related to benign and malignant pulmonary nodules to reduce redundancy or irrelevant feature.Finally,the SVM classifier is trained to classify the selected feature vectors.Compared with the existing feature selection algorithms and pulmonary nodule classification methods,the proposed method has high feature selection efficiency,accuracy,sensitivity and AUC value,which provides a new idea for the classification of benign and malignant pulmonary nodules.