Prediction Of Survival Time Of GBM Brain Cancer Patients Based On Deep Learning

Among the various types of cancer,the survival time of brain cancer is relatively short,and the survival time of patients is often less than one year.However,glioblastoma multiforme(GBM)is a subtype of brain cancer with rapid progress and short survival time.Generally,the survival time of more than two years is called long life.Because the public data required for the study of brain cancer is far less than that of common cancers,the research on brain cancer in bioinformatics focuses on medical image segmentation,and the omics research on brain cancer is less.In this paper,we choose the research direction of predicting the survival time of GBM brain cancer patients,which has practical value.This paper collected the CNV,DNA methylation,mRNA expression data of GBM in the cbioportal database and the clinical data of corresponding patients,constructed the survival prediction model gbmnet for GBM patients,and achieved a more accurate prediction of the survival of GBM patients.This paper includes the following work:(1)In this paper,a method for predicting the survival of GBM patients based on methylation data set is proposed,which improves the learning ability of the network through attention mechanism and weighted average pooling.In this paper,the DNA methylation data and clinical data are standardized first,and then the attention mechanism is used to weight the characteristics of the data,which improves the weight of the data that are more important to the prediction results.In this paper,CNN(convolutional neural network)and cnn-lstm(long and short period memory network)are used as the basic network fitting data,and the weighted average pooling method is adopted.The average value obtained by adding all the features of the feature matrix and averaging them is used as the global variable to calculate the difference value between each feature and it,and the standardized difference value is used as the pooling layer weight.This method preserves the association between pooling windows through a global quantity,reduces the loss of information caused by the traditional pooling method,and improves the prediction accuracy.(2)In this paper,a method for predicting the survival time of GBM brain cancer patients based on multi omics data gbmnet is proposed.The multi omics data of DNA methylation data,CNV and mRNA expression data of GBM brain cancer patients are connected in series,and the survival time prediction is realized through the method of model integration.This paper uses the lightweight optimized convolutional neural network vgg16 and perceptionnetv3 for parallel training,uses the hybrid attention mechanism to improve the performance of the model,and obtains the final prediction results through the voting integration mechanism.Compared with single omics data set,the improvement of prediction accuracy reflects the advantages of multi omics data.This paper also uses the grad cam method to output the feature graph of the multiomics feature matrix as a class activation graph to analyze the role of different omics data in the classification process,and proves the effectiveness of multiomics data.In addition,the prediction accuracy of gbmnet model exceeds that of SVM,LR,CNN,Minet and other models,which reflects the superiority of the performance of the model proposed in this paper.(3)This paper designs and implements a survival prediction system for GBM patients.