The Value Of Radiomics Models Based On Contrast-Enhanced T1WI And T2WI Images In Differentiating High-Grade Gliomas From Solitary Brain Metastases

Objective:To extract the tumor radiomics features from contrast enhanced T1WI(CE-T1WI)and T2WI images,and establish the radiomics models to explore the value of pre-operative radiomics models for differentiating high-grade gliomas(HGG)from solitary brain metastases(SBM).Materials and methods:The clinical data and brain magnetic resonance images of patients with HGG(n=55)and SBM(n=40)diagnosed by postoperative pathology were collected from January 1,2017 to September 30,2021 in our hospital.The region of interest(ROI)was delineated on CE-T1WI and T2WI images by 3D slicer software.After the images were standardized,the radiomics features were extracted to generate the corresponding data sets.All cases were divided into training group(n=66)and test group(n=29)by the ratio of 7:3.After normalizing the features data,the intraclass correlation coefficient(ICC)tests between groups and within groups were carried out to screen the stable features.Student t-test,Mann Whitney U test and least absolute shrinkage and selection operator(LASSO)were used to further filtrate features.After extracting the best features,CE-T1WI features set and T2WI features set were generated.Univariate analysis and multivariate logistic regression analysis were used to select the clinical independent predictors.Then,five predictive models are developed using logistic regression,including CE-T1WI model,T2WI model,clinical model(containing independent clinical predictors),model T based on CE-T1WI and T2WI images,model M based on independent clinical predictors and radiomics features.The receiver operating characteristic curve(ROC)was drawn,and the performance of models were evaluated by the sensitivity,specificity,accuracy and area under curve(AUC).Delong’s test was used to compare the difference of AUCs between radiomics models and clinical model,and calibration curve was drawn to evaluate the reliability of models.Results:Four radiomics features were screened from CE-T1WI images,and 3 radiomics features were screened from T2WI images;Enhancement was an independent predictor.Among the single sequence radiomics models,the prediction ability of CE-T1WI model is similar to that of T2WI model.The model "T" based on CE-T1WI and T2WI images performed better than the single sequence radiomics models.Among the five models,model "M"had the best prediction performance.The AUCs of CE-T1WI model,T2WI model,model "T",model "M" and clinical model in training groups were 0.824(95%CI:0.724-0.924)、0.780(95%CI:0.665-0.895)、0.833(95%CI:0.729-0.936)、0.870(95%CI:0.778-0.963)和 0.665(95%CI:0.571-0.759),验证组的 AUC 分别为 0.824(95%CI:0.657-0.990)、0.824(95%CI:0.671-0.976)、0.907(95%CI:0.798-1)、0.922(95%CI:0.826-1)和0.542(95%CI:0.460-0.623)。Conclusions:Radiomicsmodel based on MRIimagesare valuable for differentiating HGG from SBM;Radiomics model based on MRI multi-sequence can provide more help for clinical decision making.