Integration of MRI radiomics and germline genetics to predict the IDH mutation status of gliomas

Abstract

The molecular profiling of gliomas for isocitrate dehydrogenase (IDH) mutations currently relies on resected tumor samples, highlighting the need for non-invasive, preoperative biomarkers. We investigated the integration of glioma polygenic risk scores (PRS) and radiographic features for prediction of IDH mutation status. We used 256 radiomic features, a glioma PRS and demographic information in 158 glioma cases within elastic net and neural network models. The integration of glioma PRS with radiomics increased the area under the receiver operating characteristic curve (AUC) for distinguishing IDH-wildtype vs. IDH-mutant glioma from 0.83 to 0.88 (P_ΔAUC = 6.9 × 10^-5) in the elastic net model and from 0.91 to 0.92 (P_ΔAUC = 0.32) in the neural network model. Incorporating age at diagnosis and sex further improved the classifiers (elastic net: AUC = 0.93, neural network: AUC = 0.93). Patients predicted to have IDH-mutant vs. IDH-wildtype tumors had significantly lower mortality risk (hazard ratio (HR) = 0.18, 95% CI: 0.08-0.40, P = 2.1 × 10^-5), comparable to prognostic trajectories for biopsy-confirmed IDH status. The augmentation of imaging-based classifiers with genetic risk profiles may help delineate molecular subtypes and improve the timely, non-invasive clinical assessment of glioma patients.

Fig. 1. Overview of the study design.

Abbreviations: PRS polygenic risk score, GBM glioblastoma, IDH-WT IDH-wildtype, IDH-MT IDH-mutant.

Fig. 2. Distribution of AUC for classification of IDH mutation status from repeated nested cross-validation.

AUC estimates from 500 random iterations of nested cross-validation were obtained from IDH classification models using different combinations of features and model architectures. A, C Models that included radiomic features that were trained with an elastic net and a neural network, respectively. B, D Models that included UNet-based autoencoder (UNET) features that were trained with an elastic net and a neural network, respectively. Demographic factors include age at diagnosis and sex. Genetic factors are captured by the composite polygenic risk score (PRS).

Fig. 3. Predictive features for IDH status classification.

A Mean cumulative value of feature weight in the full integrated elastic net model across the 500 random iterations of nested cross-validation. Feature weights are normalized by the maximum value. Features with non-zero cumulative weights in >50% of the 500 random iterations are shown. B Distribution of the predictive features stratified by tumor subtype. Features were defined as predictive if the cumulative weight was non-zero across all 500 random iterations of nested cross-validation. Features are ordered (from left to right) in decreasing importance. Abbreviations: NET non-enhancing tumor, WT whole tumor, TC tumor core, ET enhancing tumor, GLCM gray-level co-occurrence matrix, GLSZM gray-level size zone matrix, GLRLM gray-level run-length matrix, SZLGE small zone low gray-level emphasis, SRE short run emphasis, NGTDM neighborhood gray-tone difference matrix, STD standard deviation, TGM tumor growth model.

Fig. 4. Overall survival for adult glioma cases with predicted or biopsy-confirmed IDH mutation status.

A, B Percent survival distributions for adults with glioma stratified by predicted or biopsy-confirmed IDH mutation status. Differences in event time distributions were assessed using the log rank test. A IDH mutation status confirmed by molecular profiling of tumor samples. B IDH mutation status predicted based on integrated elastic net model (“predicted”). C Features used for IDH status classification are examined for association with overall survival using univariate Cox proportional hazards regression. Hazard ratio (HR) and 95% confidence interval (CI) for each feature are presented. Abbreviations: NET non-enhancing tumor, WT whole tumor, TC tumor core, ET enhancing tumor, GLCM gray-level co-occurrence matrix, GLSZM gray-level size zone matrix, GLRLM gray-level run-length matrix, SZLGE small zone low gray-level emphasis, SRE short run emphasis, NGTDM neighborhood gray-tone difference matrix, STD standard deviation.