Integration of MRI Radiomics and Germline Genetics to Preoperatively Predict the IDH Mutation Status of Gliomas

Introduction: Molecular profiling of gliomas relies on resected tumor samples, which can delay the accurate evaluation of treatment options and prognostication until after surgery. This highlights the need for non-invasive, preoperative predictors of clinically-relevant tumor features, such as isocitrate dehydrogenase (IDH) mutations. In this study we investigated the integration of germline genetic risk scores (GRS) with radiomic features for the preoperative prediction of IDH mutation status in gliomas.

Methods: We developed an elastic net classifier of IDH mutation status based on a panel of 256 radiomic features from preoperative MRI scans, a GRS for IDH mutation, age at diagnosis and sex. Model performance for distinguishing IDH-wildtype vs. IDH-mutant gliomas was estimated using 5-fold nested cross-validation in 159 glioma cases (82 IDH-wildtype, 77 IDH-mutant) from The Cancer Genome Atlas. Hazard ratios (HR) for predicted IDH status were estimated using Cox proportional hazards models with adjustment for patient demographics and tumor grade.

Results: The GRS-based classifier demonstrated an AUC of 0.70 (95% CI: 0.62-0.78). The integration of the germline GRS improved the performance of the radiomics-based model with a statistically significant increase in AUC from 0.82 to 0.89 (P=0.002). Classification performance was further improved with the inclusion of age at diagnosis and sex (AUC=0.92, 95% CI: 0.88-0.96). The most predictive features of the integrative model were: age at diagnosis, germline genetic susceptibility, the relative enhancing tumor volume, and the extent of frontal lobe invasion. Our multimodal classifier also predicted survival. Patients predicted to have IDH-mutant vs. IDH-wildtype tumors had significantly lower mortality risk (HR=0.27, 95% CI: 0.14-0.51, P<.001), which was comparable to the prognostic trajectories observed for biopsy-confirmed IDH mutation status (HR=0.21, 95% CI: 0.11-0.40, P<.001).

Conclusion : The augmentation of imaging-based classifiers with personalized genetic risk profiles may help preoperatively delineate molecular subtypes and improve the timely, non-invasive clinical assessment of glioma patients.