NIMG-53. MRI-BASED RADIOMICS FOR DIFFERENTIATION BETWEEN PERIFOCAL EDEMA AND NON-CONTRAST-ENHANCING GLIOBLASTOMA FOCI

Abstract OBJECTIVE Various intracranial tumors can lead to the development of perifocal edema. Differentiating edema from tumor on MRI might be easy in metastases and meningiomas but can be challenging in patients with glioblastoma. We set out to compare and characterize perifocal edema and non-contrast-enhancing glioblastoma foci by means of radiomic feature analysis. METHODS In this single-center, retrospective study, patients with intracranial tumors and perifocal edema on preoperative T2/FLAIR-sequences were included. Patients diagnosed with brain metastasis, meningioma or glioblastoma were eligible. Tumor and edema volumes were semi-automatically segmented and extracted. After image registration, contrast-enhancing tumor and necrosis on T1-weighted sequences were subtracted from T2/FLAIR-hyperintense tumor and perifocal edema. Radiomic features were computed. Distribution of radiomic features were tested using the Shapiro-Wilk test. ANOVA or Kruskal-Wallis tests were applied to compare the features. Significant features were identified based on p-values (<0.05) and post-hoc Tukey’s Honest Significant Difference (HSD) tests were conducted. RESULTS The cohort comprised n=60 patients and was split in three groups according to histopathology (metastases, meningioma, glioblastoma). Each subgroup comprised n=20 patients (33.3%). 107 radiomic features of the T2-hyperintensity, after subtraction of contrast-enhancing tumor and necrosis, were tested for each patient. 18 radiomic features (17%) that mainly comprised measures of shape, texture and intensity distribution, e.g., sphericity (p<0.01), surface/volume (p<0.01), entropy (p<0.01), skewness (p<0.01) and kurtosis (p<0.01), showed no significant difference between metastases and meningiomas but differed significantly from the T2-hyperintensity of glioblastoma. The textual complexity of glioblastoma was significantly higher (p<0.01). CONCLUSION Radiomic feature analysis unraveled many features that might be helpful in differentiating perifocal edema from tumor infiltration. Future studies should aim at validating our results and further exploring the T2-hyperintensity of glioblastoma patients, e.g., suspected edema versus suspected tumor infiltration zone.