Radiomics Study For Predicting Methylation Status Of MGMT Promoter In Glioblastomas

Objective:To evaluate the utility of radiomics analysis for predicting the methylation status of the MGMT promoter in glioblastoma based on multi-parameter(T1WI,T2 WI,ADC)MRI radiomics model and construct a classification model,so as to provide significant informations for clinical diagnosis and treatment.Methods:The T1 WI,CE-T1 WI,T2WI and FLAIR images of 164 glioblastoma patients with pathologically confirmed from TCIA database were included,including 62 patients with MGMT promoter methylation and 102 patients without methylation.164 patients were divided into training group and testing group(115 and 49)randomly in the ratio of 7:3.3DSlicer software was used to manually draw the region of interest(ROI),and extract the radiomics features form ROI.1820 features in T1 WI,CE-T1 WI,T2WI and FLAIR images and 3640 combined features were standardized by Z-Core and screened by least-absolute shrinkage and selection operator regression(LASSO).The prediction model was constructed by using the logistic regression(LR)classifier.In the meanwhile the prediction efficiency of the model was also evaluated.Results:The predictive model for predicting the methylation status of the MGMT promoter in glioblastoma: T1 WI model,CE-T1 WI model,T2 WI model,FLAIR model and T1WI+CE-T1 WI model,T2WI+FLAIR model obtained 48、33、54、55、61 and 82 most valuable features after feature selection.In the training data set,the AUCs of T1 WI,CE-T1 WI,T2WI,FLAIR and T1WI+CE-T1 WI,T2WI+FLAIR models established by LR algorithm are 0.88,0.65,0.71,1.00,0.83 and 0.88 respectively;The sensitivity was0.72,0.13,0.49,0.97,0.76 and 0.78 respectively;The specificity was 0.87,0.93,0.81,1.00,0.86,0.83;The accuracy rates were 0.82,0.64,0.68,0.97,0.76 and 0.81 respectively.In the validation set the AUCs of T1 WI,CE-T1 WI,T2WI,FLAIR and T1WI+CE-T1 WI,T2WI+FLAIR models established by LR algorithm are 0.66,0.64,0.65,0.97,0.83 and 0.86 respectively;The sensitivity was 0.39,0.29,0.48,0.86,0.76 and 0.73 respectively;The specificity was 0.69,0.75,0.90,0.92,0.77 and 0.94respectively;The accuracy rates were 0.56,0.56,0.64,0.89,0.77 and 0.87 respectively.Conclusion:The FLAIR,T2WI+FLAIR,T1WI+CE-T1 WI models established by LR classifier have good diagnostic efficacy in predicting the methylation status of MGMT promoter in glioblastomas.