Lung Cancer CT Image Recognition Based On Three-dimensional Convolutional Neural Network Model

As we all know,lung cancer has threatened human health and life safety for a very long time.Its morbidity and mortality rank among the top of many diseases.In our country and even in many other countries in the world,it has become a malignant disease that affects human's lung health.In the diagnosis of lung cancer,computer aided diagnosis(CAD)has been widely used in the early screening of lung cancer?Its general steps usually include the image pretreatment process,feature extraction and image processing,etc.,and its main performance mainly depends on the image pretreatment process,which separates the suspicious lesion organization from complex anatomical background,this process is very complex and takes a lot of work.In order to overcome the complexity problem of traditional method of image preprocessing,this paper mainly studies the convolutional neural network classification model of lung CT images,this method has several advantages as follows:(1)It can omit the complicated pretreatment process of image data,and directly input the original image into the convolution neural network for classification and identification;(2)We can build a 3D convolution neural network for feature extraction,which is more in line with the characteristics of lung CT image and better able to grab the related features of images in the third dimension;(3)The network platform of Tensorflow provides a variety of algorithms and related network interfaces to the implementation of the convolutional neural network and released the related network functions of 3d convolution and 3d pooling in 2017.In this paper,the three-dimensional convolutional neural network model is applied to the diagnosis and detection of lung tumor images.The main research work is as follows:(1)This paper discusses the basic hierarchical structure and all kinds of algorithms of the convolutional neural network,namely,convolution layer,activation layer,pooling layer,the whole connection layer,and classifier design,model optimization,etc.(2)A three-dimensional convolution neural network model is constructed to classify the image data of lung cancer.The good operation of a convolutional neural network depends on the interaction of various parameters and weights,which mainly through the following adjustments to several factors to do a promotion on the performance of the convolutional neural network,such as the number of convolution layer,pooling layer,the convolution kernel size,the size of the activation function in the model,using different methods of pooling,transform the optimizer.The experimental results show that this network model is effective in analyzing lung cancer images,and the reasonable selection of parameters can greatly improve the performance of the model and improve the classification accuracy.(3)Build a two-dimensional convolutional neural network model,select the appropriate layer number of convolution,the size of the convolution kernel,pooling function,activation function,the optimizer,and then the two-dirnensional convolution neural network model is applied to image classification of lung cancer.The results are compared with the classification results of the three-dimensional convolutional neural network designed in this paper.It is found that the two-dimensional convolutional neural network model can degrade the important features of the 3d image while extracting the features,resulting in the loss of key information.By using the three-dimensional convolution neural network model,the characteristics of CT images of lung cancer can be better extracted,and the results are well recognized.