Recognition Of Stomach Cancer Pathology Based On Convolutional Neural Network

Stomach cancer has a high morbidity and mortality rate.It is a worldwide disease that makes many people living in pain.Generally speaking,stomach cancer is difficult to be find at early stage,many cases have been delayed when the best treatment time has been confirmed.According to statistics,if the patient can be find at early stage,their five-year survival rate can reach more than 80%or even 90%,but if they are discovered in the middle or even late stage,the five-year survival rate will rapidly decline.With the development of technology,people care more about their health,the check of stomach cancer has become more and more popular,but what follows is the rapid increase in the amount of pathological slice data.There is a large occupational gap in pathologists,and reading pathological slides is another time-consuming and labor-intensive task.Compared with other medical images such as MRI and CT,pathological images are more likely to be misdiagnosed or missed,and pathological diagnosis is cancer.Pathological is the most effective method in cancer diagnosis,is bound to be stricter for doctors.At present,computer-aided diagnosis in pathological diagnosis has been some progress,but most of these researches use traditional methods,and manually extracting features has higher requirements for the medical professional knowledge.And,using computer for medical diagnosis has little to study in stomach pathology.Then,the final imaging effect of pathological images is greatly influenced by various factors in the production process,which brings additional difficulties to the computer-aided identification of pathological images.For the current situation,this subject has three aspects:First,explore possible methods for processing stomach cancer pathology data,and try various image processing methods to reduce the impact of various environmental factors on the final imaging effect and enhance useful features in the data,provide more effective data for subsequent classification and segmentation work.At the same time,research and analyze the SVG file format and design algorithms to directly extract and generate usable label images from SVG files.Finally,constructing a neural network training set through certain enhancement methods.Second,train the convolutional neural network model to complete the automatic classification task of stomach cancer pathological data.First use the data set to train AlexNet and GoogLeNet separately,and optimize GoogLeNet according to the experimental results,reduce the need for computing resources and ensure the effectiveness of diagnosis.Then,put forward a model fusion idea,by designing a parallel convolutional neural network structure,fusion AlexNet and GoogLeNet,integrated depth learning model different structures to further improve the performance of stomach pathology image classification.Third,train a fully convolutional network model to achieve semantic segmentation of stomach cancer pathological data.Based on the idea of end-to-end network,a fully convolutional network model MIFNet using multiple scales for data input is proposed,and designed an effective multi-scale feature map fusion strategy to fuse feature matrices generated from multiple different inputs together.The experimental results show that MIFNet is superior to U-Net,SegNet and PSPNet in this task.This topic combines deep learning techniques with stomach cancer pathological images.By using convolutional neural network technology to classify and segment stomach cancer pathological images,and experiments show that the algorithm proposed in this paper can achieve good results and high reliability.It can help doctors to effectively diagnose stomach cancer patients,and has strong practical significance.