Application Of Different Radiomics Dignostic Models Based On Conventional MRI Images In The Preprotive Grading Of Brain Glioma

Objective:This study intends to use different MRI images and different machine learning models to establish a radiomics diagnostic model of preoperative grading of gliomas,confirming and exploring the feasibility of radiomics diagnostic models in preoperative grading of gliomas,and AUC,Sensitivity,and Specificity are used to evaluate the diagnostic performance of different diagnostic models.The aim is to select the MRI image and machine learning model with the highest diagnostic efficiency for preoperative grading diagnosis of glioma,providing the research basis for further research.Methods:This study retrospectively analyzed 93 patients with glioma who underwent routine MRI examination before operation in the First Affiliated Hospital of China Medical University from January 2013 to June 2018.Among them,23 were low-grade glioma(LGG)and 70 were high-grade glioma(HGG).DICOM format images were imported into GE-AK software.The regions of interest(ROI)of T2-weighted(T2WI)and T1-enhanced images were delineated by software features extraction.Two hundred and forty-three histogram and texture features were obtained,and all the features were imported into R language software package for feature preprocessing and dimensionality reduction.Twenty and twenty-two important features were obtained.Then six machine learning models were established by selecting three functions of "Support Vector Machines with Linear Kernel","Random Forest","Generalized Linear Model".All the data were divided into training sets and training sets according to the ratio of 7:3.The test set consisted of 65 training sets(49 cases of HGG,16 cases of LGG),28 testing sets(21 cases of HGG and 7 cases of LGG).The six models were trained by 10-fold cross-validation,and then the ROC curve was drawn to calculate the sensitivity,specificity and AUC of the classification of LGG and HGG.Results:The AUC of the six radiomics diagnostic models is greater than 0.8,which indicates that the diagnostic performance of the six models is higher.The AUC of the radiomics diagnostic model based on T1 enhancement image is larger than that based on T2 WI image,and the AUC of the RF model based on T1 enhancement image is the highest,reaching 0.97.Conclusion:Radiomics diagnostic model can distinguish the pathological grading of glioma well.Different image and machine learning model have certain influence on the diagnostic efficiency.In the future clinical application,appropriate MRI images machine learning model should be adopted to further optimize the radiomics diagnostic model of glioma preoperative grading.