Lung Nodule Detection And Benign And Malignant Diagnosis Based On Deep Convolution Neural Network

Lung cancer is a threat to people's health with the highest morbidity and mortality.If found in the early stage of lung cancer treatment,can significantly improve the survival rate of patients.CT images are the most effective technique for lung cancer detection,and pulmonary nodules are the early clinical manifestations of lung cancer,which means that the study on the detection and diagnosis of early pulmonary nodules based on CT images and the techniques and algorithms for malignant nodules has important scientific and clinical significance.At the same time,the application scope of deep learning in the field of image processing is expanding,and in recent years,it has achieved good results in medical image processing.Therefore,researchers are constantly focusing on the detection of pulmonary nodules and the diagnosis of benign and malignant pulmonary nodules by using the technology of deep learning.Based on previous research results,this paper mainly studies the detection of pulmonary nodules and the algorithm of benign and malignant diagnosis based on deep convolutional neural network.The main contents are as follows:In order to avoid the interference of other lung tissues in the detection,the lung parenchyma is segmented by the method of "threshold method + morphology",CT images of segmented lung parenchyma were used for the detection of pulmonary nodules and the diagnosis of benign and malignant pulmonary nodules.For the detection of pulmonary nodules,this paper designed an end-to-end 3D deep convolutional neural network for the automatic detection of pulmonary nodules.Using 3D multi-scale convolution neural network to extract suspected lung nodules in the network.The u-shaped dual-path network structure combines the advantages of residual learning and skip connection,extracts more abundant image features and makes full use of the features,so that the shallow network has higher semantic features and improves the detection rate of small target objects.In the extraction network of suspected pulmonary nodules,in order to ensure the high sensitivity of the model and prevent missed detection,a large number of areas that are not nodules misdiagnosed as nodules,which means false positive is too high.The network finally used a 3D CNN to detect lung nodules by false positive filtering,and further classified the suspected pulmonary nodules,so as to improve the specificity of detection.For the diagnosis of benign and malignant pulmonary nodules,3D dense convolution neural network is used to classify the detected pulmonary nodules.Due to particularity of 3D pulmonary nodule images,3D spatial information reduction of pulmonary nodules using intermediate density projection.CT images of pulmonary nodules were randomly flipped,noise added and randomly cropped to enhance the training data,In addition,positive and negative sample ratio in training set is guaranteed to be balanced.The Loss function uses Focal Loss to replace the traditional cross entropy Loss,so that the network can pay more attention to learning samples that are easy to be classified incorrectly.Finally,it optimizes the way of data storage and reduces the consumption of computing resources.