Repeatability of an MRI protocol for generating habitats based on cellularity, perfusion, and hypoxia in a murine model of glioma.

BACKGROUND: Hypoxia, cellularity, and vascularity are well-established as important drivers of tumor growth and treatment response. Diffusion weighted (DW-), dynamic contrast enhanced magnetic (DCE-), and oxygen enhanced (OE-) MRI techniques have been developed to quantitatively map these three characteristics. We have developed and implemented a multi-parametric MRI protocol to identify tumor subregions that share similar biological features or "habitats".

METHODS: We performed test-retest DW-, DCE-, and OE-MRI measurements on Wistar rats bearing a C6 glioma (n = 24). MRI data was summarized as apparent diffusion coefficient (ADC) maps from DWI-MRI, area under the curve (AUC) from DCE-MRI, and normalized change in T1 (ΔT1) from OE-MRI. We then applied a data driven and k-means clustering to the aggregated dataset to identify distinct habitats within each animal. We employed four different approaches to partition the data, with each defined by a different number of habitats. The imaging data were partitioned into one of five habitats: three normoxic habitats (low cellularity-low perfusion, low cellularity-high perfusion, and high cellularity-high perfusion), one hypoxic habitat, and one necrotic habitat.

RESULTS: Across all habitat construction approaches, Pearson and concordance correlation coefficients were both above 0.90 when comparing the test and retest habitat compositions.

CONCLUSION: The combination of the MRI protocol and our data-driven approach can reliably generate habitats that capture distinct cellularity, perfusion-related information, and hypoxia patterns with a high degree of repeatability.