Lung Cancer Image Classification And Tumor Mutational Burden Prediction Based On Convolutional Neural Network(CNN)

As a kind of high incidence of malignant tumor,lung cancer has been the number one killer of Chinese residents.Pathological image analysis has accuracy and high authority for diagnosing cancer,however it needs effort demanding and time-consuming disadvantages,maybe along with inappropriate treatment due to the difference of doctors’ experience.In recent years,computer-aided diagnosis technology becomes an indispensable diagnostic tool for doctors.Although Computer-aided diagnosis technology is strong,pathological images’ size,abundant amount of information,effective storage,those are thorny points in research subject of pathological image analysis.In this topic,the pathological images of lung cancer marked by doctors is taken as the experimental dataset,an automatic classification method,with the method of multi-feature fusion combined,has been established based on high-efficiency network model.Otherwise,as independent bio-markers,tumor mutational burden(TMB)can be used to predict the efficacy of immunotherapy.The transfer learning method is used to predict the tumor mutational burden of lung cancer patients,and explore the association between H&E pathological images and tumor mutational burden.The main research of this paper is as follows:1.Based on Convolution Neural Network(CNN),a mufti-feature fusion model of lung cancer pathological image automatic typing is constructed.Aiming at the problem of the whole-slide image(WSI),the dataset is progressed by image block processing and color normalization.Then,in order to the deep network can also learn the shallow feature information,the preprocessed dataset is input into the Efficient Net B0 network for feature learning and the multiple layer feature extraction.The extracted GLCM features is fed into the CNN for a further extraction of the texture features of the image.Finally,the extracted local features and multi-scale features are merged and to be fed into a classifier,thus establishing a model based on lung cancer pathological image classification.The experiment shows that the accuracy of the model is improved compared with other classification methods.2.In order to predict the tumor mutational burden with whole-slide H&E image,combining transform learning and random forest algorithm base on patch is proposed,which problem of predict TMB based on patch can turn into the prediction of WSI.The feature is extracted with Xception network,then introduce pooling and random forest to integrate the results of WSI,for achieving the ability of prediction tumor mutational burden on WSI to identify patients who can benefit from immunotherapy.3.Furthermore,based on Tkinter,the visual interface of lung cancer pathological image analysis is designed and realized,which can simply and intuitively classify the pathological images and predict the tumor mutational burden.