The heterogeneity of glioma is linked to patient prognosis and tumor invasiveness, and it can be quantified through noninvasive imaging. This study aimed to spatially validate physiological tumor heterogeneity (referred as tumor habitat) using multiparametric magnetic resonance imaging (MpMRI) on co-registered macropathology. We further investigated the potential of tumor habitat to predict microscopic tumor extension (ME) in glioma.

This prospective study recruited patients with newly diagnosed gliomas. Comprehensive MpMRI protocols were performed on all participants, encompassing T1-weighted, gadolinium-enhanced T1-weighted, T2-weighted, T2-weighted fluid-attenuated inversion recovery sequences (T2-FLAIR), diffusion-weighted imaging (DWI), and MR spectroscopy (MRS), followed by supra-total tumor resection. On MpMRI, tumor enhancing and non-enhancing regions were segmented into distinct spatial habitats using K-means clustering. These imaging-based habitats were further compared with histological and immunohistochemical (Ki-67: proliferation; CD31: microvascular density) features of sections that were co-registered using MpMRI-guided three-dimensional reconstructed specimen. Next, we measured the ME, defined as the distance between the tumor edge and the surrounding cancerous cells in the brain. A spatial habitat model (SLHM), constructed using multivariate linear regression, was developed to predict the ME of glioma. For external validation, an independent cohort of 110 patients with recurrent glioma was retrospectively included. The SLHM-derived ME was compared with regions of recurrence on postresection follow-up imaging.

Between February 2016 and July 2020, thirty patients with glioma, including 7 diffuse astrocytomas, 13 oligodendrogliomas, and 10 glioblastomas, were analyzed. MpMRI habitats demonstrated significant correlations with pathological and immunohistochemical habitat characteristics (all P < 0.05). Through integrated analysis of MpMRI and histology data, we identified three distinct spatial habitats: high-vascularity high-cellularity (HV-HC), low-vascularity high-cellularity (LV-HC), and low-vascularity low-cellularity. Multivariate analysis revealed that the increase of both HV-HC and LV-HC habitat showed significant positive associations with ME distance in glioma (all P < 0.05). The SLHM, incorporating these spatial habitat features, demonstrated superior predictive performance for ME, with an adjusted R-square of 0.858. Noteworthy, the utilization of SLHM-derived ME demonstrated a remarkable enhancement in recurrence coverage within the validation cohort compared to conventional radiation plans, achieving a mean Jaccard index of 87.6% and a mean Dice similarity coefficient of 90.2%.

MpMRI enables the identification of distinct physiological tumor habitats in gliomas. These spatially defined habitats serve as a noninvasive biomarker for precise prediction of ME distance.