Abstracts

Rationale: Recent studies have revealed that IDH1 mutated brain tumors are more likely to be associated with seizures as compared to IDH1 wildtype tumors. No rigorous statistical analysis has been performed as to whether this could be due to a difference in tumor morphology or location. We used large publicly available databases to determine whether IDH1 mutated and wildtype tumor aggregated together as measured by a tumor similarity score. Methods: De-identified patient information and IDH1 mutation status were obtained through The Cancer Genome Atlas study. Corresponding MRI scans were obtained through The Cancer Imaging Archive. Tumors were delineated by the extent of gadolinium enhancement or maximum extent of abnormality on T1 weighted MRI when no enhancement was present. Each image underwent affine registration to the standard MNI 152 atlas through an automated method using FSL v5.0; registration was checked individually for fidelity. To increase statistical power, left hemispheric tumors were flipped across the XY plane to place tumors into one hemisphere. The tumor similarity score was determined by calculating the Hausdorff distance between each pair of tumors. An enrichment score (ES) was then calculated for each tumor by performing a random walk against the entire dataset, with increasing score when encountering a tumor with the same phenotype, and decreasing it when the alternate phenotype was encountered (Fig 1). The maximum ES was calculated, and its statistical significance was determined via a non-parametric permutation test in which IDH1 mutation status was randomized 50,000 times. This allows for the ranking of each tumor according to strength of similarity to tumors of the same IDH1 phenotype. Results: A total of 111 patients (26 IDH1 wild type, 85 IDH1 mutated) were included. The normalized volume of IDH1 mutated tumors (98.02 cubic cc) was larger than IDH1 wildtype tumors (56.35 cubic cc, p=0.018). Three IDH1 wild type tumors and 17 IDH1 mutated tumors (Fig 2) showed statistically significant aggregation to tumors of the same phenotype when assessed by the tumor similarity score as compared to random distribution. Conclusions: We demonstrate that IDH1 mutated and wildtype tumors segregate according to a metric that assesses for similarity in spatial distribution. This suggests that there is spatial clustering of tumors, even if not readily apparent on visual inspection. IDH1 mutated tumors also demonstrate greater tumor volume as compared to wildtype tumors. Further investigation is required to determine whether this clustering affects seizure propensity. Funding: NINDS R03 NS091864 Chinese national scientific grants 81301206; Chinese Phd students oversea joint grants