基于胃组织病理图像数据集的卷积神经网络模型对胃癌的早期预测价值OACSTPCD

Objective:To explore the early prediction value of convolutional neural network(CNN)model for gastric histologi-cal image dataset for gastric cancer(GC),and the development and validation of GC early prediction model.Methods:154 patients with GC were selected and divided into early stage group(n=87)and middle stage group(n=67)according to different stages.Logis-tic regression was used to analyze the clinical covariates,and using the CNN feature extraction model,the CNN prediction model was built.The ROC evaluates the degree of differentiation and the accuracy of the calibration curve evaluation.Results:Age,underlying dis-ease,helicobacter pylori infection,red blood cell count(RBC)and white blood cell count(WBC)were independent risk factors for GC.The optimal CNN feature extraction model consists of 3 convolution layers,2 pooling layers and 1 fully connected layer.The index of CNN was better than other models.Calibration curve analysis showed that the fitting effect of CNN model was remarkable.Conclusion:The CNN model based on gastric histopathological image dataset has good predictive performance and good clinical feasibility.