Morphologic Features on MR Imaging Classify Multifocal Glioblastomas in Different Prognostic Groups

AJNR Am J Neuroradiol. 2019 Apr;40(4):634-640. doi: 10.3174/ajnr.A6019. Epub 2019 Mar 28.

Abstract

Background and purpose: Multifocal glioblastomas (ie, glioblastomas with multiple foci, unconnected in postcontrast pretreatment T1-weighted images) represent a challenge in clinical practice due to their poor prognosis. We wished to obtain imaging biomarkers with prognostic value that have not been found previously.

Materials and methods: A retrospective review of 1155 patients with glioblastomas from 10 local institutions during 2006-2017 provided 97 patients satisfying the inclusion criteria of the study and classified as having multifocal glioblastomas. Tumors were segmented and morphologic features were computed using different methodologies: 1) measured on the largest focus, 2) aggregating the different foci as a whole, and 3) recording the extreme value obtained for each focus. Kaplan-Meier, Cox proportional hazards, correlations, and Harrell concordance indices (c-indices) were used for the statistical analysis.

Results: Age (P < .001, hazard ratio = 2.11, c-index = 0.705), surgery (P < .001, hazard ratio = 2.04, c-index = 0.712), contrast-enhancing rim width (P < .001, hazard ratio = 2.15, c-index = 0.704), and surface regularity (P = .021, hazard ratio = 1.66, c-index = 0.639) measured on the largest focus were significant independent predictors of survival. Maximum contrast-enhancing rim width (P = .002, hazard ratio = 2.05, c-index = 0.668) and minimal surface regularity (P = .036, hazard ratio = 1.64, c-index = 0.600) were also significant. A multivariate model using age, surgery, and contrast-enhancing rim width measured on the largest foci classified multifocal glioblastomas into groups with different outcomes (P < .001, hazard ratio = 3.00, c-index = 0.853, median survival difference = 10.55 months). Moreover, quartiles with the highest and lowest individual prognostic scores based on the focus with the largest volume and surgery were identified as extreme groups in terms of survival (P < .001, hazard ratio = 18.67, c-index = 0.967).

Conclusions: A prognostic model incorporating imaging findings on pretreatment postcontrast T1-weighted MRI classified patients with glioblastoma into different prognostic groups.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adult
  • Aged
  • Brain Neoplasms / classification*
  • Brain Neoplasms / diagnostic imaging
  • Brain Neoplasms / pathology*
  • Female
  • Glioblastoma / classification*
  • Glioblastoma / diagnostic imaging
  • Glioblastoma / pathology*
  • Humans
  • Magnetic Resonance Imaging / methods
  • Male
  • Middle Aged
  • Prognosis
  • Proportional Hazards Models
  • Retrospective Studies
  • Survival Analysis