A qBOLD-based clinical radiomics-integrated model for predicting isocitrate dehydrogenase-1 mutation in gliomas

Abstract

Background

Quantitative blood oxygenation level–dependent (qBOLD) technique can be applied to detect tissue damage and changes in hemodynamic in gliomas. It is not known whether qBOLD-based radiomics approaches can improve the prediction of isocitrate dehydrogenase-1 (IDH-1) mutation.

Purpose

To establish a qBOLD-based clinical radiomics-integrated model for predicting IDH-1 mutation in gliomas.

Methods

A total of 125 patients of grade II–IV glioma (IDH1 mutation: IDH1 wild-type = 50:75) were divided into a training group (n = 87) and a validation group (n = 38). Contrast enhanced T1-weighted (CE-T1W), T2-weighted (T2W), and 3D multi-gradient-recalled-echo (MGRE) images were acquired. Radiomics features were extracted from the region of interests of each image. The feature selection and support vector machine radiomics models were established for each sequence. A clinical radiomics–integrated model was finally constructed combining the best radiomics model with age. The predictive effectiveness of the models was evaluated by area under the receiver operating characteristic curve (AUC). Brier score was used to assess overall predictive performance. Decision curve analysis and calibration curve were also conducted.

Results

The best radiomics model was CE-T1W + T2W + qBOLD with AUCs of 0.823 (95% confidence interval [CI]: 0.743–0.831) in the training group and 0.751 (95% CI: 0.655–0.794) in the validation group, respectively. The clinical radiomics–integrated model, incorporating the best radiomics model with age, showed the best predictive effectiveness with AUCs of 0.851 (95% CI 0.759–0.918) in the training group and 0.786 (95% CI 0.622–0.902) in the validation group.

Conclusion

A clinical radiomics-integrated model that combined qBOLD parametric maps, CE-T1W, and T2W images with age achieved promising performance for predicting IDH1 mutation in glioma patients.