Multi-parametric MR habitat imaging predicts prognosis of glioblastoma

Abstract

Objective To determine the predictive value of median relative cerebral blood volume (rCBVmedian) of each habitat of glioblastoma (GBM) for overall survival (OS) by using multi-parametric MR habitat imaging analysis constructed by dynamic susceptibility contrast-enhanced perfusion weighted imaging(DSC-PWI) and diffusion weighted imaging (DWI). Methods A retrospective cohort trial was conducted on 58 patients with pathology-diagnosed GBM in Army Medical Center of PLA from January 2016 to December 2021. All patients underwent routine MRI, DSC-PWI and DWI examination within 1 week before surgery. Deep learning-based nnU-Net model and K-means clustering algorithm were applied to construct traditional habitat (TH), vascular habitat 1~4(VH1~4), and their combined habitat 1~6 (CH1~6). Spearman correlation analysis was used to determine the correlation between rCBVmedian value and OS of the GBM patients in different habitats. Kaplan-Merier survival curve was plotted for rCBVmedian-high group and rCBVmedian-low group according to their rCBVmedian value, and the difference was showed by Log-rank test. Cox analysis was performed to predict independent risk factors for OS. Results The correlation between rCBVmedian value and OS in GBM patients was stronger in CH1 and VHI than in other habitats (r=-0.404、-0.398, P=0.002). In the Whole ET, VH1 and CH1 habitats, the conversion of rCBVmedian into binary variables showed significant differences in OS between the 2 groups (P < 0.05). The rCBVmedian values of VH1 and CH1 habitats were independent risk factors for the prognosis of GBM patients (P < 0.05). Conclusion The rCBVmedian values, obtained from VH1 representing the high angiogenic-enhancing tumor habitat, and from CH1 representing the area with active tumor cell proliferation, may be significant indicators for predicting the overall postoperative survival of GBM patients.

Keywords

• glioblastoma
• tumor heterogeneity
• habitat imaging