NIMG-11RADIOMIC SUBCLASSIFICATION OF GLIOBLASTOMA

Abstract

BACKGROUND: To identify clinically relevant glioblastoma (GBM) sub-classification/subtypes based on radiomic analysis. Integrated genomic analysis has already identified genomic subtypes of GBM as characterized by specific genomic aberrations and mutations. In our study, we seek to demonstrate a GBM sub-classification based on radiomic texture analysis that is both predictive and prognostic. METHODS: We retrospectively identified 80 GBM treatment naive patients from the Cancer Genome Atlas with corresponding MRI data in the Cancer Imaging Archive. We applied texture analysis (528 features) to each tumor. We extracted texture features and performed feature selection. Univariate and multivariate analysis were also performed. Clustering identified radiomic subclasses. RESULTS: Multiple texture features were associated with GBM and we were able to classify GBM into 4 subtypes. However, subclassification into 3 subclasses were associated with differences in patient progression free and overall survival. CONCLUSIONS: GBM subtypes based on radiomic texture signatures were identified that were both predictive and prognostic. These findings are important as this can help determine patient outcome based on imaging (obtained at the time of diagnosis before treatment); this can also help determine response to therapy at the pre-treatment stage to stratify patients and serve as imaging endpoints for clinical trials.