| Objective:1.To explore the value of MRI signs in preoperative assessment of the methylation status of the MGMT promoter in lower-grade gliomas.The value of promoter methylation status.2.Explore the value of Efficient Net-B3 convolutional neural network in predicting the methylation status of MGMT promoter before surgery,compare the performance of the established models based on different sequences,and select the best sequence combination for modeling.Methods:1.Retrospectively analyze the clinical and imaging data of 89 patients with lower-grade glioma confirmed by surgical pathology and molecular pathology from June 2016 to June 2020 in the Second Hospital of Lanzhou University.Among them,the MGMT promoter is methylated.A total of 59 cases were metabolized and30 cases were unmethylated.All patients underwent MRI scans before operation,including T1WI,T2WI,FLAIR,DWI and T1WI enhanced sequences.Perform statistical analysis on the tumor’s gender,age,location,number of lesions,tumor diameter,cystic necrosis,tumor border,hemorrhage,peritumoral edema,whether it crosses the midline,the degree of enhancement and other signs and ADCmin value,ADCmean value,n ADC value,Through the receiver operating characteristic（ROC）curve to evaluate the performance of indicators related to the methylation status of the MGMT promoter.2.The collected T2WI and T1 enhancement sequences of MRI images of 121cases of lower-grade gliomas were selected by manually selecting all the images of each patient including the lesion level,and randomly divided into training sets（n=85）and Validation set（n=36）,using Efficient Net-B3 convolutional neural network to build independent prediction models T2-net,T1C-net,TS-net based on T2WI,T1WI enhancement,T2WI combined with T1WI enhancement,and operation by subjects The ROC evaluates the predictive performance of each model separately.Results:1.Compared with the MGMT unmethylated group,the ADCmin value of the lower grade glioma MGMT methylated group[（0.97±0.30）×10-3 vs（0.84±0.23）×10-3 mm2/s]is significantly higher,and the difference is statistically significant（P=0.032）.The ADCmean value[（1.10±0.32）×10-3 vs（0.99±0.32）×10-3mm2/s]and the n ADC value[（1.48±0.46）vs（1.37±0.44）]of the MGMT methylation group and the non-methylation group.The difference was not statistically significant（all P>0.05）.When the ADCmin value is used to distinguish the lower-grade glioma MGMT methylation group from the non-methylation group,the AUC value is 0.646.When ADCmin=0.76×10-3 mm2/s,the sensitivity is 76%,the specificity is 57%.There was no statistical difference in age of onset,gender,tumor location,number of lesions,whether it crossed the midline,whether hemorrhage,tumor boundary,whether cystic transformation,tumor diameter,maximum diameter of cystic transformation,maximum diameter of edema,and degree of enhancement were not statistically different between the two groups（P>0.05）.2.The accuracy of the T2-net model on the validation set is 72.3%,the sensitivity is 64.7%,the specificity is 73.3%,and the area under the ROC curve（AUC）is 0.718;the accuracy of the T1C-net model on the validation set is The sensitivity is66.8%,the sensitivity is 68.3%,the specificity is 66.9%,the area under the ROC curve（AUC）is 0.721;the accuracy of the TS-net model on the validation set is 81.8%,the sensitivity is 63.1%,and the specificity is 85.0%,the area under the ROC curve（AUC）is 0.784.Conclusions:1.ADCminvalue has preliminary reference value in evaluating the MGMT methylation status of lower-grade gliomas before surgery;conventional MRI signs cannot be used to assess the MGMT methylation status of lower-grade gliomas.2.The Efficient Net-B3 convolutional neural network based on MRI can predict the methylation status of the MGMT promoter in lower-grade gliomas.The TS-net model based on T2WI combined with T1WI enhancement has the best prediction performance,and the model noninvasively predicts the start of MGMT before surgery Sub-methylation status has value. |
PDF Full Text Request