Gastric Cancer Pathology Image Recognition Based On Deep Learnin

The morbidity and mortality of gastric cancer rank among the forefront of digestvive system cancer.The Early examination can effectively improve patient survival rate.Compared with CT and other examinations,pathological diagnosis is the gold standard.Accurate histopathological diagnosis is of great significance to the treatment of gastric cancer.To better assist pathologists in diagnostic analysis and improve diagnostic efficiency,this paper uses deep learning algorithms to identify gastric cancer pathological images.There are mainly two aspects of research on pathological image analysis,one is the classification of gastric cancer subtypes,and the other is the segmentation of tumor regions.Aiming at the classification of gastric cancer pathological images,a multi-classification model based on local features and global features is proposed.Although existing studies have obtained good results in the classification of gastric cancer pathological images,most studies use binary classification for gastric cancer pathological images,which has a certain gap with clinical requirements.Therefore,we propose a multi-classification method based on deep learning,which is more clinically practical.In classification research,we developed Sto His Net,a multi-scale model based on Transformer and Convolutional Neural Network(CNN),for fine-grained multi-classification tasks.Sto His Net uses Transformer to learn global features,which alleviates the perceptual field limitations of convolution operations,and uses CNN for inductive learning of local features,which accelerates the convergence of the network.The proposed model was tested on the gastric pathological image dataset and two other pathological image datasets.The results show that the model has good classification and generalization ability.Aiming at the segmentation of gastric cancer pathological images,a multi-scale segmentation model based on CNN and Transformer is proposed.In pathological images,multi-scale information such as the size of different tissues and tumor regions needs to be considered,and the pathologist’s reading is also a process of constantly zooming in and out of the microscope.Therefore,to better simulate the diagnosis process,in the study of lesion region segmentation,we proposed a multi-scale feature fusion model Gas Unet based on CNN and Transformer.For the utilization of global information and local information,Gas Unet adopts a multi-path channel module including CNN and Transformer,and uses a multi-path feature fusion module based on shift and projection operations after feature extraction to interact with multi-scale features of different channels.To reduce the calculation amount of Transformer in the channel,this paper uses channel self-attention and reduces the size of tokens to reduce the calculation amount of self-attention and improve the overall performance of the model.Compared with other models,Gas Unet achieved better segmentation results on gastric cancer and colorectal cancer pathological image datasets,which verified the performance of the model.In this paper,classification and segmentation models based on deep learning are proposed,and better classification and segmentation results are obtained on gastric cancer pathological images.In addition,this study also verifies the good generalization of the models on other pathological image datasets.The study shows that the models proposed in this paper are potential tools for analyzing gastric histopathological images,which help to better assist pathologists in diagnostic analysis.