Predictive value of a prognostic model for isocitrate dehydrogenase wild-type glioblastoma based on multi-parameter MRI
Objective To construct and validate a prognostic model for predicting the prognosis of isocitrate dehydrogenase(IDH)wild-type glioblastoma(GBM)patients based on the radiomic features of multiparametric MRI.Methods A retrospective analysis was conducted on clinical information and MRI data from patients pathologically diagnosed with adult-type IDH wild-type GBM patients who underwent tumor resection surgery at the Department of Neurosurgery,the First Affiliated Hospital of Zhengzhou University(dataset 1,n=172 cases,from October 2018 to December 2020),and at the Department of Neurosurgery,Henan Provincial People's Hospital(dataset 2,n=89 cases,from January 2011 to September 2021).After preprocessing multiple MRI sequences,including T1-weighted imaging,gadolinium-enhanced T1-weighted imaging,T2-weighted imaging,fluid-attenuated inversion recovery sequences,diffusion-weighted imaging,and apparent diffusion coefficient maps,dataset 1 was divided into a training set and an internal validation set at a ratio of 1∶1.Dataset 2 was used as an external validation set.A univariate Cox proportional hazards regression model combined with Lasso-cox analysis was employed to select radiomic features.These radiomic features were integrated with clinical risk factors to develop a clinic-radiomic model for predicting the prognosis of GBM patients.The predictive performance of the model was evaluated using the concordance index,Akaike information criterion(AIC),integrated discrimination improvement(IDI),Kaplan-Meier survival curves,nomograms,calibration curves,and decision curves and then was compared with that of the model established using clinical risk factors.Results Eighteen radiomic features significantly associated with the prognosis of IDH wild-type GBM patients were selected to construct a radiomic model(Radscore).Kaplan-Meier survival curves demonstrated statistically significant prognostic differences between high-risk and low-risk groups stratified by Radscore in the training set,internal validation set,and external validation set(all P＜0.05).Univariate and multivariate Cox proportional hazards regression analyses indicated that the Radscore was an independent prognostic factor in the training set(P＜0.05).Compared with the clinical model alone,the addition of the Radscore improved the concordance index from 0.682,0.682,and 0.791 to 0.785,0.694,and 0.823,respectively,in the training set,internal validation set,and external validation set.The values of AIC decreased from 727.872,703.796,and 666.732 to 683.771,697.790,and 654.837,respectively.The IDI values for the clinic-radiomic model in the training set,internal validation set,and external validation set were 0.268,0.051,and 0.100,respectively(all P＜0.05).The clinic-radiomic model consistently demonstrated superior predictive performance on nomograms and calibration curves across the three subsets.Decision curves demonstrated that the incorporation of Radscore improved the decision-making ability of the clinical model.Conclusion The radiomic model established based on multiparametric MRI images using machine learning techniques can achieve risk stratification of IDH wild-type GBM patients and enhance the prognostic predictive ability of traditional clinical models to some extent.
NETL
NSTL
万方数据
