Research On Lung Cancer Immune Biomarkers Based On Deep Learning

Lung cancer is currently one of the most common malignant tumors in the world.As an emerging treatment with almost no side effects and good prognosis,lung cancer immunotherapy has become a research hotspot in many fields.In 2018,immunotherapy officially entered China and attracted wide attention from all walks of life.At present,related drugs approved internationally for lung cancer immunotherapy are all through the immunosuppressive checkpoint(ICI)of PD-1/PD-L1 to enhance anti-cancer cells Immune response,but there are still many uncertainties in its specific efficacy in the clinical treatment process.Since then,tumor mutation burden(TMB)has been recognized as a predictive marker for immunotherapy,but it is invasive and expensive.Therefore,discovering more biomarkers that have a guiding role in lung cancer immunotherapy is a crucial step in the development of immunotherapy.The method of machine learning plays an increasingly important role in the diagnosis of clinicians,the most common of which is the deep learning method.This paper proposes a method based on deep convolutional neural network(CNN),which can predict and evaluate potential biomarkers in lung cancer immunotherapy through efficient analysis of immunologically stained pathological images of lung cancer tissue.The research in this article is different from the previous research on a single lung cancer immune-related biomarker.Instead,it predicts and analyzes the potential biomarkers of multiple immunotherapy processes in lung cancer,and demonstrates the deep learning method in the field of medical imaging.The importance of application.In the study,180 full-slice images(WSI)of lung cancer from TCGA were processed for tumor area selection,image screening,noise removal,image color normalization,etc.The processed images were annotated with immunotherapy biomarkers.The 1:1 ratio is integrated,and two-fold cross-validation is performed under the deep neural network model combined with CNN and the residual network(Res Net)to predict potential immunotherapy biomarkers in lung cancer immunohistochemical staining(H&E)images.The results showed that the area under the ROC curve(AUC)of TP53 reached 0.87,and EGFR,DNMT3 A,PBRM1,and STK11 also reached 0.71 to 0.84,respectively,showing a good model prediction effect,and the addition of residual blocks effectively overcomes the previous commonly used The over-fitting problem easily exists in the method,and the calculation speed of the model is greatly improved.This paper selects the VGG19 network similar to the research model to reconstruct the model and compares it at the TP53 site.The results show that the VGG19 model only reaches0.57 and the running speed is slow,indicating the relatively superior prediction performance of this research model.Then,according to the results of TP53,EGFR,DNMT3 A,PBRM1,STK11 and their biological functions and mechanisms,this article analyzes their possible guiding role in lung cancer immunotherapy.The research in this article shows that for medical professionals,the application of deep learning to assist in the development of targeted drugs and immunotherapies and to improve the survival rate of patents is essential.Finally,this article summarizes and looks forward to the model.