Amide Proton Transfer-Weighted MR Imaging Based Radiomics And Phenotypes For Predicting IDH Mutation And Prognosis In Brain Glioma

BackgroundAccurate glioma grading and IDH mutation status prediction are critically essential for individualized preoperative treatment decisions.Lower-grade gliomas(LwGG)of histologic grades Ⅱ and Ⅲ follow heterogeneous prognosis,which necessitates risk stratification.Amide proton transfer(APT)and velocity-selective pulse based cerebral blood volume(VS-CBV)imaging,respectively,can provide unique tumor physiology information of concentration of mobile protein or peptides and angiogenesis.Objective(1)To explore the feasibility of VS-CBV imaging in evaluating glioma,and to evaluate its correlation with APTw,VS-CBF,DSC-PWI.(2)To developed and validated a radiomics model based on APTw for preoperative prediction of IDH1 mutation status in glioma.(3)To explore and validate the value of APTw imaging as an imaging biomarker in predicting progression-free survival(PFS)in LwGG.(4)To explore and validate the value of APTw imaging in predicting overall survival(OS)of LwGG.Materials and Methods(1)Forty-eight consecutive patients with newly diagnosed glioma who underwent preoperative VS-CBV imaging were retrospectively evaluated.Lesion visibility on VS-CBV was performed by evaluating the tumor signal intensity relative to surrounding.The relative tumor blood volume(rTBV),relative tumor blood flow(rTBF)and relative APTw(rAPTw)were measured from sampling intra-tumoral areas of hot-spot on the CBV,CBF or APTw map,respectively,and normalized against the contralateral normal parenchyma.Linear regression and Bland-Altman analyses were performed to evaluate the correlation and agreement of rTBV measurements between dynamic susceptibility contrast enhanced perfusion weighted imaging(DSC-PWI).Spearman correlation analysis was used to evaluate the correlation among VS-CBV,APTw,VS-CBF and DSC-PWI.(2)One hundred and twenty-six glioma patients with 3D APTw imaging and had known IDH status were retrospectively studied.Radiomics features were extracted from the APTw imaging.Logistic regression model was generated using the APTw radiomic features and compared to APTw maximum value and clinical model in differentiating IDH1 mutation status.The areas under the ROC curve(AUC),accuracy,sensitivity and specificity were calculated to illustrate diagnostic power.(3)A retrospective analysis of 179 patients with LwGG who underwent preoperative APTw imaging was performed.The maximum and mean value of the relative APTw(rAPTwmax or rAPTwmean)were extracted in the development set(n=120).The ability of rAPTwmax and rAPTwmean to predict PFS was evaluated using biomarker threshold model.Multivariate Cox proportional risk regression analysis was conducted to construct models based on clinical,conventional MR,APTw,pathological and therapeutic features,and compared with two combined models.The predictive power,calibration,and clinical usefulness of the model were validated in an independent internal cohort.(4)Based on the data grouping in Part 3,biomarker threshold models were performed and evaluated the ability of rAPTwmax and rAPTwmean to predict OS.Two APTw models(one integrating preoperative and one integrating pre-and postoperative variables)were established by multi-parameter Cox stepwise regression analysis,comparing with two semantic clinical models,and validated in an independent internal cohort.Patients were stratified and validated based on the risk scores of the two APTw models.Results(1)Lesion visibility of VS-CBV was good with overall mean score of 2.26.VS-CBV and DSC-CBV showed strong correlation and agreement in perfusion quantification(R2=0.84,P<0.001).The quantitative parameter of VS-CBV were positively correlated with that of APTw,VS-CBF and DSC-PWI(P all<0.05).(2)The APTw Radiomics signatures determined by 11 radiomics features showed a good ability to predict IDH mutation status.The combined models that incorporated radiomics signatures,clinical risk factors exhibited superior diagnostic superiority over the clinical models and APTw maximum signature in training(AUC:0.943 vs.0.739 vs.0.622,respectively)and validation sets(AUC:0.933 vs.0.778 vs.0.772 respectively).The calibration curves showed good consistency,and the decision curve analysis supported the clinical utility of APTw Radiomics signatures and the combined model.(3)Relative APTwmax and rAPTwmean parameter thresholds stratified the risk of progression for LwGG(high versus low risk,P<0.001).In the training and validation sets,the combined model based on rAPTwmax and IDH1 mutation status showed better prediction performance(C-index≥0.80),smaller prediction error(integrated Brier score,IBS≤0.15),better calibration and greater clinical benefit than the comprehensive model,the single category models and the WHO grading system.(4)APTw feature not only was an independent factor for OS but also has added prognostic value when integrated with clinical features.In the training and validation sets,the post-APTw model showed better predictive performance(C-index>0.82,P<0.05),lower predictive error(IBS<0.142)and greater net benefit,and was able to grade the OS risk of LwGG at high,medium and low risk.Conclusions(1)VS-CBV is a promising perfusion method for the evaluation of glioma,and found significant positive correlations of VS-CBV with APTw,VS-CBF and DSC-PWI.(2)APTw-derived radiomics and its combination with clinical factor could accurately predict IDH mutant status in glioma.(3)APTw has significantly increased value in predicting progression of LwGG compared to clinical prognostic factors and molecular markers.(4)APTw imaging parameters integrated with preoperative and postoperative characteristics can predict the OS of LwGG.