Research Of Lung Cancer CT Image Recognition Method Based On Convolutional Neural Network

Lung cancer is one of the most harmful malignant tumors to human health,it has the characteristics of high morbidity and high mortality.In addition,the cost of treatment is relatively high,and the cure rate is low,which brings a huge economic burden and psychological pressure to patients and their families.Therefore,there is an urgent need for refined and efficient lung cancer diagnosis and treatment methods.In recent years,deep neural networks have made important contributions in the field of smart medical care,but this kind of big data-driven learning algorithm requires a large amount of medical data support.Due to the particularity of medical field,such as privacy,source,labeling cost and other factors,it is difficult to form a large dataset for the training of complex neural network.However,artificial intelligence models trained with public dataset cannot meet actual needs.Therefore,this paper is committed to solving the above problems.The main research contents can be summarized as follows:1.According to the characteristics of the original lung cancer CT image dataset,this paper uses the histogram equalization method to grayscale the lung cancer CT image.Changing the histogram of the original image from a relatively concentrated grayscale range to uniform throughout the grayscale range distribution,the feature pixels of the lesions and important organs in the image are enhanced,the image contrast is effectively enhanced,and the visual quality of the lung cancer CT medical image is improved,and the image is clearer.At the same time,because the dataset is small and the images of each type of lung cancer extracted from the dataset are not balanced,the data enhancement method is used to solve the limited dataset,by using the methods of translation,rotation,transform from the existing data to create a batch of "new" data,to increase the amount of training data,improve the generalization ability and the robustness of the model,enlarge the diversity of the data,avoid the classification results from being biased towards a large number of samples,and improved the efficiency of the convolutional neural network learning characteristic information.2.Combined with the Ada Boost ensemble method,a novel lung cancer subtype classification algorithm based on convolutional neural network is proposed.Firstly,densenet convolutional neural network was trained based on the collected lung cancer CT image data,and the CT images were divided into three types: squamous cell carcinoma,adenocarcinoma and small cell carcinoma.Then use Ada Boost algorithm to aggregate multiple classification results to improve classification performance.Experimental results show that this method can achieve higher recognition accuracy,and has better performance than Dense Net,Res Net,VGG16 and Alex Net networks.It provides an effective non-invasive detection tool for the pathological diagnosis of lung cancer types.3.In view of the relatively small amount of lung cancer CT medical image data,the transfer learning method is introduced,and a feature extraction and lesion detection model based on convolutional neural network is proposed.Using transfer learning and convolutional neural network to detect the location information of lung cancer from CT images,it solves the problem of overfitting in the training process due to insufficient image data.First,train the pretrained network based on the PASCAL VOC2007 large-scale dataset,then use the lung cancer CT image data to adjust the network parameters,so as to obtain a network model that can accurately detect lung cancer location information from CT images.It can assist doctors in diagnosis and treatment,promote communication between doctors and patients,and improve diagnosis efficiency in the clinical field of lung cancer medicine.