Purpose: The aim of this study was the development and external validation of a logistic regression model to differentiate gliosarcoma (GSC) and glioblastoma multiforme (GBM) on standard MR imaging.
Methods: A univariate and multivariate analysis was carried out of a logistic regression model to discriminate patients histologically diagnosed with primary GSC and an age and sex-matched group of patients with primary GBM on presurgical MRI with external validation.
Results: In total, 56 patients with GSC and 56 patients with GBM were included. Evidence of haemorrhage suggested the diagnosis of GSC, whereas cystic components and pial as well as ependymal invasion were more commonly observed in GBM patients. The logistic regression model yielded a mean area under the curve (AUC) of 0.919 on the training dataset and of 0.746 on the validation dataset. The accuracy in the validation dataset was 0.67 with a sensitivity of 0.85 and a specificity of 0.5.
Conclusions: Although some imaging criteria suggest the diagnosis of GSC or GBM, differentiation between these two tumour entities on standard MRI alone is not feasible.
Keywords: Gliosarcoma; MRI; glioblastoma; logistic regression model; multivariate analysis.