Nimg-29. Radiomic Analysis on Apparent Diffusion Coefficient (Adc) Maps Predicts Platelet-Derived Growth Factor Receptor Alpha (Pdgfra) Gene Amplification for Newly Diagnosed Glioblastoma Patients

Abstract

NEURO-ONCOLOGY • NOVEMBER 2017 for hypermutated gliomas using a radiomics-based approach. MATERIALS AND METHODS: We retrospectively analyzed a total of 97 patients with primary gliomas from the University of Texas MD Anderson Cancer Center. Mutation status was determined post-biopsy (stereotactic/excisional) using Next Generation Sequencing (T200 and Foundation 1), and patients were grouped based on their mutation burden; 77 hypomutated (<30 mutations) and 20 hypermutated (>=30 mutations or <30 with MMR gene mutations). Radiomic analysis was performed on the conventional MR images (FLAIR and T1 post-contrast) obtained prior to tumor tissue surgical sampling; and a total of 2480 rotation-invariant radiomic features were extracted using: (i) the first-order histogram and (ii) grey level co-occurrence matrix. The Maximum Relevance Minimum Redundancy technique was used to select the most relevant radiomic features. ROC analysis and leave-one-out cross-validation (LOOCV) were used to assess the performance of the Support Vector Machine (SVM) classifier and AUC, Sensitivity, Specificity, and p-value were obtained. RESULTS: We found 100 radiomic features that can discriminate between hypermutated versus hypomutated gliomas, AUC 100% (CI: 96%-100%), Sensitivity 100%, Specificity 100%, p-value=1.8e-10. CONCLUSION: Increase in mutation burden seen in post-treatment glioma samples is associated with a unique radiomic texture signature on pre-sampling MRI and can be used to predict mutation burden status non-invasively.