Pulmonary Nodule Detection Method Based On Convolutional Neural Network

In recent years,the continuous occurrence of haze seriously endangers people's health.Under the influence of air pollution,the incidence of lung cancer worldwide increases year by year.Lung cancer has become one of the cancers with the highest morbidity and mortality in the world.Lung nodules are the most important manifestation in the early stage of lung cancer.However,pulmonary nodules are usually small in size and difficult to be detected due to the interference of blood vessels,air bubbles,trachea and other tissues,resulting in a high rate of clinical misdiagnosis.The survival rate of patients with early lung cancer is very high after treatment,but when the clinical symptoms appear,most patients have reached the middle and late stage of lung cancer,that is,pulmonary nodules have worsened.Therefore,accurate detection of pulmonary nodules is the key to early screening of lung cancer.The task of traditional pulmonary nodules detection is mainly completed by doctors,who discover and make diagnosis of pulmonary nodules by reviewing patients' CT images.However,with the rapid growth of lung CT image data and the increasing difficulty for doctors to read the film,the phenomenon of misdiagnosis and missed diagnosis gradually increases.With the development of deep learning algorithm,computer image processing and other technologies,computer aided diagnosis(CAD)system,which is built based on it,has gradually become a diagnostic tool for imaging doctors.Pulmonary nodule auxiliary diagnosis system is usually composed of three parts: image preprocessing,nodule candidate set detection and false positive reduction.CAD system aims to make the interpretation of pulmonary CT images faster and more accurate,and thus improve the screening efficiency of pulmonary nodules.Region of interest(ROI)extraction is the basis for the establishment of pulmonary nodule detection model.The purpose of extracting the region of interest is to process the CT images of the lungs and obtain the images containing the detailed features of pulmonary nodules and their surrounding tissues,and then use the classification model to classify the images,so as to achieve the purpose of detecting pulmonary nodules.In this paper,Lung Nodule Analysis 2016(LUNA 16)Lung CT image data set was taken as the research object to establish a pulmonary Nodule detection method based on convolutional neural network.This method can effectively improve the accuracy of pulmonary Nodule detection and reduce the false positive rate.The main work includes three parts:In the first part,LUNA16 image data set was selected as the sample to propose a set of lung CT image preprocessing process.First,accurate and complete pulmonary nodule image samples were obtained through normalization.After that,the data reduction and data amplification method was added to solve the problem of uneven positive and negative samples in the data set,and the extraction of lung ROI region was completed.Finally,by means of information merging,the preprocessed data are classified into datasets and converted into formats.Through this series of CT image preprocessing operations,accurate,standard and complete data samples can be provided for the subsequent classification model of pulmonary nodule detection,so as to complete the training and evaluation of the model.In the second part,build based on convolutional neural network(convolutional neural network,CNN)of pulmonary nodules classification model.To address the problems existing in the early original CNN model,this paper adjusted the CNN network architecture by experiment and comparison,selected the Re LU activation function to speed up the convergence,and introduced Dropout strategy in the eight-layer network model to prevent the occurrence of overfitting.Then,the data set samples obtained after the pretreatment of the first part were used for training and comprehensive and effective evaluation of the model by using the method of half-fold cross validation.The accuracy,sensitivity,specificity and the AUC values of the ROC curve reached 92.3%,92.1%,92.6% and 0.97.The model performance was excellent and stable.In the third part,a pulmonary nodule detection software based on convolutional neural network is generated by integrating the first part of data sample set building method and the second part of CNN network model building method.The software includes image data preprocessing,model training,model application and other functions to automate the detection of pulmonary nodules.In this paper,LUNA16 data were taken as samples to propose a set of lung CT image preprocessing scheme,to construct the data set required by the model,and to introduce the adjusted convolutional neural network,so as to finally realize the detection and classification of pulmonary nodules.Experiments show that the method designed in this paper can improve the accuracy of model classification and reduce the false positive rate,and the model is stable and reliable,which can provide effective help for the early screening of lung cancer.