Probabilistic Radiographic Atlas of Glioblastoma Phenotypes

BACKGROUND AND PURPOSE: Tumor location is a significant prognostic factor in glioblastoma, which may reflect the genetic profile of tumor precursor cells. The purpose of the current study was to construct and analyze probabilistic radiographic atlases reflecting preoperative tumor locations and corresponding demographic, "-omic," and interventional phenotypes to provide insight into potential niche locations of glioblastoma cells of origin. MATERIALS AND METHODS: Preoperative anatomic MR images in 507 patients with de novo glioblastoma were analyzed. Images were registered to stereotactic space, tumors were segmented, and the stereospecific frequency of tumor occurrence was analyzed statistically by age, extent of resection, MGMT methylation, IDH1 mutation, gene expression subclassification, PTEN loss, PTEN deficiency, EGFR amplification, EGFR variant 3 expression, progression-free survival from the start of radiochemotherapy, and overall survival from initial diagnosis. RESULTS: Most glioblastomas grow into the periventricular white matter regions adjacent to the subventricular zone. MGMT promoter methylated tumors occur more frequently in the left temporal lobe, in young patients with glioblastoma, in IDH1 mutant tumors, in tumors having the proneural gene expression subtype, and in tumors lacking loss of PTEN occurring most frequently in the frontal lobe. MGMT methylated tumors with the IDH1 mutation tended to occur in the left frontal lobe. EGFR amplified and EGFR variant 3-expressing tumors occurred most frequently in the left temporal lobe. A similar region in the left temporal lobe was associated with favorable response to radiochemotherapy and increased survival. CONCLUSIONS: Radiographic atlases for specific phenotypes provide insight into overlap between prognostic variables and may help to identify niche locations for cancer cells of origin.