Research On The Staging Prediction Model Of Colorectal Cancer Based On The Feature Pyramid

With the increasingly aging population in China as well as the changes in diet and lifestyle,morbidity of colorectal cancer is increasing rapidly,making it the malignant tumor whose mortality rate has ranked fourth.Although the staging theory of colorectal cancer has matured,the staging is still very difficult.Therefore,improving the accuracy of the staging has become an urgent task.However,the data set of the existing staging prediction model is based on the colonoscopy,which requires more stringent detection conditions and has the risk of penetrating the intestinal wall,so there is a greater medical preference for magnetic resonance imaging,a non-invasive trauma-free detection method.However,current researches mainly apply the traditional machine learning method,far from the standard of accurate staging prediction.In the light of medical theories,a staging prediction model of colorectal cancer based on feature pyramid is proposed in this thesis.The thesis is divided into three major parts.Firstly,the MRI image is preprocessed,and the data enhancement method based on moving least squares is designed to expand the dataset.Secondly,the staging prediction model of colorectal cancer based on feature pyramid is constructed,and the prediction model is divided into three stages by function: in the feature extraction stage,the three feature extraction networks VGG,ResNet and GoogLeNet are used to extract features,and the feature pyramid network is constructed to collect the multi-scale features of each network respectively;in the stage of feature fusion and selection,a feature fusion network based on deep learning is designed,which fuses pyramid features into heterogeneous features and uses feature selection algorithm based on information gain to extract key features;in the diagnostic classification stage,SVM classifier is used to classify the key features and integrate with three feature extraction networks to finally obtain the stage of colorectal cancer.Thirdly,the prediction indexes of each stage of the model are compared,the comparison test is set up,and 5-folder cross validation is used to evaluate the comprehensive index of the staging prediction model.