Classification Of Liver Steatosis And Pancreas Steatosis Using Radiomics Analysis Of MRI

Different degrees of liver steatosis and pancreas steatosis could lead to the occurrence of a variety of metabolic diseases,which has been affecting a huge number of people.And the process that steatosis induces metabolic diseases is irreversible.Due to the various defects of needle biopsy,classification of steatosis is generally diagnosed by medical imaging in clinical,which can generally get high rate of accuracy for more severe steatosis.But the diagnosis of mild steatosis gets less sensitive,which affects the implementation of early intervention.Magnetic resonance imaging has more accurately on the monitor of the fat volume fraction,provides the possibility to improve the accuracy of the diagnosis of mild steatosis.As a medical imaging aid for decision-making,Radiomics has developed rapidly in recent years.The purpose of the project is to explore the application of Radiomics in the diagnosis of liver steatosis and pancreas steatosis.Firstly,based on the method of Radiomics and feature engineering,the subjects’ MRI data is preprocessed and the regions of interest of the liver and pancreas are extracted respectively.After data amplification,the Radiomics features are extracted and feature conversion is performed.Finally,the feature selection method is used to screening the features having high correlation and low redundancy with steatosis.Integrating feature engineering into Radiomics has further improved the quality of features.Subsequently,four ensemble learning principles are used to establish classification models,including random forest,adaptive boosting,gradient boosting decision tree and extreme gradient boosting.Besides,support vector machine is used as a control to compare the accuracy between ensemble learning method and traditional machine learning method.These algorithms are used to establish mild and moderate steatosis classification models based on the radiomics feature of liver and pancreas,then the classification effects of different models are evaluated to screen out the best performing model.The results show that the accuracy of the liver steatosis classification model and the pancreas steatosis classification model established by extreme gradient boosting algorithm are both highest.The correlation between clinical factors of subjects and the grade of steatosis is calculated so that the clinical factors with high correlation are input as clinical features for training the classification model so that to explore whether they can improve the performance of the model.The results show that although the accuracy of the model is not significantly improved,the clinical features have improved the detection rates of samples with moderate pancreatic steatosis to a certain extent.In addition,the importance of features in the models is calculated and ranked in order to investigation the clinical significance of the features with high importance.By analyzing the imaging performance corresponding to these features,more kinds of effective information related to steatosis can be found from the imaging data,so as to provide support for clinical diagnosis.