Multi-modal Magnetic Resonance Images For Predicting Glioma Genotypes Preoperativly

Background:Glioma is the most common type of primary tumor of central nervous system.Traditional pathological perspective holds that histopathological classifications of gliomas based on morphologic criteria determine the prognosis of patients.However,recent genomic-wide studies have resulted in a far more comprehensive understanding of the genomic alterations in gliomas and provide suggestions for a new molecularly based classification.Especially,isocitrate dehydrogenase?IDH?gene mutations and oxygen 6-methylguanine-DNA methyltransferase?MGMT?promoter methylation are increasingly used as prognostic or predictive biomarkers for gliomas.Mutations in IDH1/2 genes,which occur in the majority of WHO grade ? and ? gliomas,have been postulated to indicate a favorable clinical prognosis.However,IDH1/2 mutation occurs in only 5%of?Glioblastoma,GBM?patients,leaving the vast majority of GBMs be IDH1/2 wild-type.Therefore,knowledge of MGMT promoter status has great clinical significance for GBMs patients after adjuvant chemotherapy.Thus,it is very important to preoperatively identify the IDH mutation and MGMT promoter status.As popularly applied in clinical practice,the gold standard to identify genetic alterations in gliomas is surgical sampling,a necessary invasive procedure for gliomas treatment or identification,however,may induce severe complications.Furthermore,glioma heterogeneity and sampling errors increase the risk of erroneous genetic profiling.Magnetic resonance?MR?imaging phenotypes can serve as noninvasive surrogates for tumor genotypes and can provide important information for diagnosis,prognosis,and,eventually,personalized treatment.Materials and Methods:In this study,151 patients with postoperatively pathologically confirmed glioma were enrolled,including 59 cases of grade ?/? glioma and 92 cases of IDH-wild type GBMs.Conventional MR imaging,diffusion weighted imaging?DWI?,three-dimensional arterial spin labeling?3D-ASL?and amide proton transfer imaging?APT?sequences were employed to predict IDH mutation and MGMT promoter status.Part 1:Ninety-two pathologically confirmed IDH-wild type GBMs underwent preoperative structural MR imaging and the efficacy of structural image features were qualitatively analyzed using Fisher's exact test.In addition,77 of the 92 patients underwent additional advanced MRI scans including DWI and 3D-pCASL.Apparent diffusion coefficient?ADC?and relative cerebral blood flow?rCBF?values within the manually drawn region-of-interest?ROI?were calculated and compared using independent sample t test for their efficacies in predicting MGMT promoter methylation.Receiver operating characteristic curve?ROC?analysis was used to investigate the predicting efficacy with the area under the curve?AUC?and cross validations.Multiple-variable logistic regression model was employed to evaluate the predicting performance of multiple variables.Part 2:Fifty-nine patients with grade ?/? glioma were prospectively recruited.For the entire cohort,16 of 59 gliomas were IDH1-wild type,and 43 of 59 were IDH1 mutant.Conventional MRI structure image and APT imaging were performed preoperatively.After MRI acquisition,APT images were transferred to the workstation?AdvantageWorkstation 4.6;GE Medical Systems?to generate an important parametric map.The ROI was delineated on the S₀ map of APT.In this study,six types of characteristic parameters,including histogram parameters,first-order texture parameters,form factor parameters,GLCM parameters,RLM parameters and GLZSM parameters,are extracted from each ROI.Then,two data analysis strategies were compared.?1?Single MRI parameter analysis:Independent sample t test was performed to compare the difference between IDH1-wild type and IDH1 mutant gliomas.ROC analysis was used to investigate the predicting efficacy with AUC.?2?Support vector machine?SVM?based on multi-parameters:the synthetic minority oversampling technique?SMOTE?was applied to oversample for smaller sample size group.Then,5 samples were randomly selected as the test set from each group,the rest samples were used as the training set.After SVM-based recursive feature elimination?SVM-RFE?algorithm was applied to order the features.SVM-based classification strategy was established.The performances of SVM classifier was visualized by ROC curve.Results:Part 1:Ninety-two pathologically confirmed IDH1-wild type GBMs?44 MGMT promoter methylation and 48 MGMT promoter unmethylation?underwent preoperative structural MR imaging.In addition,77 of the 92 patients?37 MGMT promoter methylation and 40 MGMT promoter unmethylation?underwent additional advanced MRI scans including DWI and 3D-pCASL imaging.ADC and rCBF values within the manually drawn ROI were calculated.?1?Differences in conventional structural image features between MGMT promoter methylated and unmethylated groups:MGMT promoter methylation was associated with tumor location and necrosis?P<0.05?.?2?Performance of single advanced MRI parameter?ADC and rCBF?in predicting MGMT promoter methylation:Significantly increased ADC value?P<0.001?and decreased rCBF?P<0.001?were associated with MGMT promoter methylation in primary glioblastoma.The ADC achieved the better predicting efficacy than rCBF ?ADC:AUC,0.860;sensitivity,81.1%;specificity,82.5%;VS.rCBF:AUC,0.835;sensitivity,75.0%;specificity,78.4%;P=0.032?.?3?Efficacies of parameter combinations for predicting MGMT promoter methylation:The combination of tumor location,necrosis,ADC and rCBF resulted in the highest AUC of 0.914.Part 2:Fifty-nine patients with grade ?/? gliomas?16 IDH1-wild type,43 IDH1-mutant type?were finally recruited for this study.1033 parameters were obtained by using AK software.?1?Eighteen parameters were statistically significant between IDH1-wild type group and IDH1-mutant group.Among 18 parameters,High Grey Level Run Emphasis All Direction offset 8 SD showed the highes AUC?0.769?with 79.9%accuracy.?2?After the resampling of SMOTE and SVM-RFE,the optimal feature subsets from texture matrix?1033×59?for predicting IDH1 mutation status were consisted of top15 feature set.The accuracy and AUC of SVM-based classification strategy were 82.9%and 0.911.Conclusion:?1?ADC and rCBF are promising imaging biomarkers in clinical routine to predict the MGMT promoter methylation in IDH1-wild type GBMs.?2?The APT imaging based on machine learning method can effectively predict the IDH1 mutation status of grade ?/? gliomas,and its diagnostic efficiency is better than that of single characteristic parameter.