# A novel vascular model yields increased MR perfusion metrics compared to conventional dynamic susceptibility contrast algorithms in untreated glioblastoma

## TL;DR

A new MRI model shows better detection of blood flow in brain tumors compared to standard methods, which could improve diagnosis and treatment.

## Contribution

A novel Bayesian-based vascular model improves perfusion metric estimation in untreated glioblastoma compared to conventional DSC algorithms.

## Key findings

- NVM produced higher median normalized perfusion values than vendor algorithms (P < .001).
- NVM-derived rCBV was significantly higher at 3.0 T than at 1.5 T (P = .008).
- NVM showed enhanced hotspot delineation and excellent inter-rater reliability (κ and ICC ≥0.86).

## Abstract

Malignant gliomas are heterogeneous brain tumors with extensive neovascularization. Conventional gradient-echo dynamic susceptibility contrast (GRE-DSC) perfusion MRI may underestimate microvascular alterations. We hypothesized that a novel vascular model (NVM), based on Bayesian voxel-wise transit time distribution analysis, could yield higher perfusion metrics in untreated isocitrate dehydrogenase (IDH)-wild-type glioblastoma compared to standard vendor GRE-DSC algorithms.

In this retrospective, single-center study, 89 patients with neuropathologically confirmed glioblastoma underwent pretherapeutic GRE-DSC perfusion MRI at 1.5 or 3.0 T. Perfusion maps were generated using both the NVM and default vendor algorithms. Using co-registered T1-post-contrast and T2/FLAIR images, two neuroradiologists independently assessed perfusion conspicuity of color-coded maps for each algorithm and manually performed region-of-interest analyses within visually identified tumor hotspots for quantification. Relative values of cerebral blood flow (rCBF), cerebral blood volume (rCBV), and mean transit time (rMTT) were normalized to contralateral normal-appearing white matter. Nonparametric tests evaluated group differences.

The NVM yielded enhanced hotspot delineation and significantly higher median normalized perfusion values than vendor algorithms (all P < .001), with excellent inter-rater reliability (Cohen’s κ and intraclass correlation coefficients ≥0.86). At 3.0 T, NVM-derived rCBV was significantly higher than at 1.5 T (P = .008).

NVM post-processing yielded higher normalized CBF, CBV, and MTT values within tumor hotspots than vendor pipelines, suggesting that Bayesian model-based perfusion analysis may enhance the detection of microvascular changes in glioblastoma. As validation against a gold standard is missing, prospective multicenter studies are warranted to confirm our findings, particularly with regard to treatment monitoring and clinical decision-making.