The Value Of Enhanced Mr Radiomics In Preoperative Grading Of Glioma And In Predicting IDH1 Genotype

Objective to explore the value of radiomics predictive model based on preoperative enhanced MR images in preoperative grading of glioma and in the prediction of IDH1 genotype in high-level glioma.Methods Retrospective analysis of 425 patients(154 cases of WHO Ⅱ grade,128 cases of WHO ⅡI grade,143 cases of WHO Ⅳ grade)with glioma confirmed by surgical pathology from December 2012 to January 2018 in the affiliated hospital of Qingdao university with complete preoperative brain-enhanced MR images,randomly divided into training group and verification group according to the ratio of 7:3.And 182 high-grade gliomas patients undergoing immunohistochemical analysis were selected,including 79 cases of IDH1 mutation(45 cases of WHO Ⅲ grade,34 cases of WHO Ⅳ grade),103 cases of IDH1 wild type(33 cases of WHO Ⅲ grade,70 cases of WHO Ⅳ grade),randomly divided into the primary dataset and validation dataset according to the ratio of 7:3.Lasso-Logistic regression analysis and Random Forest was used to construct the glioma diagnostic grading model and high-grade glioma IDH1 mutant prediction model and the two models was tested with cross-validation method.The area under ROC curve(AUC)was used to evaluate the efficacy of the two models.Results The glioma diagnostic grading modelhad a good diagnostic performance in both training setand validation set.The glioma diagnostic grading modelaccurately diagnosed grade II gliomas in grade II-IV gliomas with an AUC value in the training set was 0.92 with 95% confidence interval(CI): 0.89-0.95,and the sensitivity and specificity were 90.7% and 74.6%,and the positive predictive value and negative predictive value were 0.685 and 0.930.The glioma diagnostic gradingmodel accurately diagnosed grade II gliomas in grade II-IV gliomas with an AUC value in the validation set was 0.84(95% CI: 0.77-0.91),the sensitivity and specificity were 89.1% and 69.3%,and the positive predictive value and negative predictive value were 0.640 and 0.912.The glioma diagnostic grading model accurately diagnosed grade Ⅲ gliomas in grade II-IV gliomas with an AUC value in the training set was 0.88(95% CI: 0.84-0.92),the sensitivity and specificity were 45.6% and 96.6%,and the positive predictive value and negative predictive value were 0.837 and 0.822.The glioma diagnostic grading model accurately diagnosed grade Ⅲ gliomas in grade II-IV gliomas with an AUC value in the validation set was 0.58(95% CI: 0.47-0.70),the sensitivity and specificity were 24.4% and 87.5%,and the positive predictive value and negative predictive value were 0.421 and 0.755.T The glioma diagnostic grading model accurately diagnosed grade Ⅳ gliomas in grade II-IV gliomas with an AUC value in the training set was 0.90(95% CI: 0.86-0.94),the sensitivity and specificity were 77.6% and 87.7%,and the positive predictive value and negative predictive value were 0.768 and 0.882.The glioma diagnostic grading modelaccurately diagnosed grade Ⅳ glioma in grade II-IV gliomas with an AUC value in the validation set was was 0.80(95% CI: 0.71-0.89),the sensitivity and specificity were 61.9% and 84.8%,and the positive predictive value and negative predictive value were 0.684 and 0.807.The accuracy of the glioma diagnostic grading model in predicting predictive power in the training group was 0.74(95% CI: 0.68-0.79),and the accuracy in the validation group was 0.62 95%CI: 0.53-0.71).The high-grade glioma IDH1 mutant prediction model showed good discrimination for predicting high-grade glioma IDH1 mutants in both training set and validation set.The AUC in the training group was 0.87,with a 95% CI of 0.754 to 0.855,an accuracy of 0.798,and a sensitivity of 85.5% and specificity of 75.4%,positive predictive value and negative predictive value were 0.734、0.867.The AUC in the validation set was 0.86,with a 95% CI of 0.690-0.913,an accuracy of 0.789,and a sensitivity of 91.3% and specificity of 69.0%,positive predictive value and negative predictive value were 0.700 and 0.909.Conclusion The glioma diagnostic grading model based on preoperative enhanced MR images has a good predictive power for identifying different grades of gliomas.The classification model is the best at identifying grade Ⅱ gliomas,the second effective in the identification of grade Ⅳ gliomas,and the least effective in identifying grade Ⅲ gliomas.The high-grade glioma IDH1 mutant prediction model based on preoperative enhanced MR can effectively predict the IDH1 genotype preoperatively in high-grade glioma before operation.