Noninvasive Prediction of IDH1 Mutation and ATRX Expression Loss in Low-Grade Gliomas Using Multiparametric MR Radiomic Features

Background: Noninvasive detection of isocitrate dehydrogenase 1 mutation (IDH1(+)) and loss of nuclear alpha thalassemia/mental retardation syndrome X-linked expression ((ATRX(-)) are clinically meaningful for molecular stratification of low-grade gliomas (LGGs).

Purpose: To study a radiomic approach based on multiparametric MR for noninvasively determining molecular status of IDH1(+) and ATRX(-) in patients with LGG.

Study type: Retrospective, radiomics.

Population: Fifty-seven LGG patients with IDH1(+) (n = 36 with 19 ATRX(-) and 17 ATRX(+) patients) and IDH1(-) (n = 21).

Field strength/sequence: 3.0T MRI / 3D arterial spin labeling (3D-ASL), T₂ /fluid-attenuated inversion recovery (T₂ FLAIR), and diffusion-weighted imaging (DWI).

Assessment: In all, 265 high-throughput radiomic features were extracted on each tumor volume of interest from T₂ FLAIR and the other three parametric maps of ASL-derived cerebral blood flow (CBF), DWI-derived apparent diffusion coefficient (ADC), and exponential ADC (eADC). Optimal feature subsets were selected as using the support vector machine with a recursive feature elimination algorithm (SVM-RFE). Receiver operating characteristic curve (ROC) analysis was employed to assess the efficiency for identifying the IDH1(+) and ATRX(-) status.

Statistical tests: Student's t-test, chi-square test, and Fisher's exact test were applied to confirm whether intergroup significant differences exist between molecular subtypes decided by IDH1 and ATRX.

Results: Optimal SVM predictive models of IDH1(+) and ATRX(-) were established using 28 features from T₂ Flair, ADC, eADC, and CBF and six features from T₂ Flair, ADC, and CBF. The accuracies/AUCs/sensitivity/specifity/PPV/NPV of predicting IDH1(+) in LGG were 94.74%/0.931/100%/85.71%/92.31%/100%, and those of predicting ATRX(-) in LGG with IDH1(+) were 91.67%/0.926/94.74%/88.24%/90.00%/93.75%, respectively.

Data conclusion: Using the optimal texture features extracted from multiple MR sequences or parametric maps, a promising stratifying strategy was acquired for predicting molecular subtypes of IDH1 and ATRX in LGGs.

Level of evidence: 3 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019;49:808-817.