Classification Of Pancreatic Cystic Tumors Based On Radiomics Features Fusion

The most common pancreatic cystic tumors are serous pancreatic cystic tumors and mucinous pancreatic cystic tumors.The former are mostly benign tumors,while the latter are mostly malignant tumors.The two have a high similarity in clinical practice,and the accuracy of doctors’ diagnosis based on medical images based on experience is not high;although it can also be diagnosed by puncture technology,it usually causes damage,and in severe cases will cause complication.Therefore,the computer-aided diagnosis of CT images of pancreatic cystic tumors has great clinical significance.In the framework of radiomics analysis,this paper proposes a tumor image classification method based on dissimilarity multi-class imageomics feature fusion.First,multi-class radiomics features of pancreatic cystic tumor images are extracted,then a joint dissimilarity feature matrix for multi-class image omics feature fusion is formed,and finally the classification is based on the joint dissimilarity feature matrix.Methods The accuracy of CT image classification results of pancreatic cystic tumors was improved.At the same time,the method of multi-category radiomics feature de-redundant fusion is also studied in this paper,which reduces the feature redundancy and improves the accuracy of CT image classification results of pancreatic cystic tumors.The main work and results of this article are as follows:1.According to the ROI region and feature acquisition principles in the radiomics process,manually segment the ROI region of the pancreatic cystic tumor,and define and extract the radiomics features for the acquired ROI regions,which are the subsequent image-based radiomics features.Fusion and image classification provide data.2.In order to fuse different types of radiomics features to improve the effectiveness of classification,a multi-class feature fusion method based on dissimilarity is proposed and applied to pancreatic cystic tumor image classification.This method first extracts 195 radiomics features in three categories: gray,geometry and texture,and then uses a random forest algorithm to construct a dissimilarity matrix between different features in each category;then,features of different categories The averaged dissimilarity matrix is averaged to form the joint dissimilarity matrix of different categories of features,and then input to the classifier to obtain the final classification result.This method is used for the classification of pancreatic cystic tumor CT image data sets,and the classification accuracy rate is 86.23%.The experimental results show that this method effectively improves the accuracy of pancreatic cystic tumor CT image classification results.3.Aiming at the problem of the high degree of redundancy in radiomics features,a de-redundant fusion algorithm based on multi-class radiomics features was proposed,and the fused features were used to classify pancreatic cystic tumor images.The method proposed in this paper first performs a preliminary classification test on each feature,selects the features that can improve the classification accuracy from the features of radiomics,and concatenates these multiple categories of radiomics features to form a new feature vector,which is input into the classifier get classification results.In this paper,the feature selection methods of correlation coefficient method,chi-square test,mutual information,principal component analysis,and linear judgment analysis are used to perform multi-class feature fusion on the radiomics features extracted in this paper.Second,a classifier is used to explore multi-classes in different dimensions.Classification of CT images of pancreatic cystic tumors by the number of features.The experimental results show that the extracted 195 imaging omics features do have redundant information,and the accuracy of the classification results of the multi-class pancreatic cystic tumor features reduced to 32 dimensions is the highest,with 84.7%.Multi-class feature selection fusion of pancreatic cystic tumors improves the accuracy of CT image classification results of pancreatic cystic tumors.