Research On The Pulmonary Nodules Recognition In CT Images Based On Deep Learning

According to the China Cancer Annual Report issued by the National Cancer Center in2017,the incidence and mortality of lung cancer in China are the highest among all cancers,and lung cancer has become the greatest threat to human health.Pulmonary nodules are the early manifestations of lung cancer.If Pulmonary nodules can be identified to diagnose lung cancer,lung cancer patients can be treated as soon as possible,and the survival rate can be improved.Aiming at the problem of pulmonary nodule recognition,this paper explores the application of depth learning in pulmonary nodule recognition of CT images.The main research work of this paper is as follows:Building a lung nodule data set.The first step is to segment the lung parenchyma from the original CT image and extract the suspected pulmonary nodules.First,median filtering is performed on the original CT image to remove noise and filter out some lung tissues.Second,aiming at the long time consuming of Otsu,this paper improves it,and uses the improved Otsu to pre-segmented CT images.Third,the morphological processing and the hole filling operation are performed on the pre-segmented images to remove the tracheobronchial and obtain the lung parenchyma mask.Next,the lung parenchyma was extracted from the filtered CT images using a mask.Finally,suspected pulmonary nodules are extracted from the lung parenchyma using modified otsu and etching operations.Study convolutional neural networks to identify pulmonary nodules.First build a convolutional neural network model,and design its parameters,and select the best parameters to train the network;Then the trained convolutional network model was used to classify and identify the extracted suspected pulmonary nodules.The recognition accuracy of the experiment reached 93.52%,and the missed diagnosis rate was 8.04%.The experiment proved that the convolutional neural network can effectively identify the pulmonary nodules;In order to reduce the rate of missed diagnosis,this paper integrates the convolutional neural network,uses two different convolutional neural networks to train and test the same samples in parallel,and calculates the final result through the OR operation in the logic operation.The rate of missed diagnosis was reduced to 3.35%.In this paper,a large number of lung nodule samples are identified by convolutional neural network.Experiments show that convolution neural network can effectively identifypulmonary nodules with high accuracy and low missed diagnosis rate.It provides a more objective and accurate auxiliary diagnosis for doctors and has a positive effect on the early diagnosis and treatment of lung cancer.