Abstract Characterizing infiltration in the non-enhancing tumor (NET) is clinically crucial as infiltration leads to progression, and a reduction in survival times. As voxel-wise biopsy of these regions is infeasible, a non-invasive MRI-based identification of infiltration would be a significant contribution. Diffusion MRI (dMRI), with its ability to model different levels of water restriction, as is caused by infiltration and edema, is well positioned to model the NET. This is achieved via multicomparment modeling (MCM) of the dMRI where infiltration, vasogenic edema and healthy tissue (with complex fibers) form the different compartments. Current MCM approaches are either not designed for NET or rely on advanced MRI scans currently not feasible in the clinic. The simplest ball-tensor models, including Hoy 2014 using multi-shell data and FERNET 2020 for single-shell data, consists of a free water compartment that models vasogenic edema and a tensor modeling underlying tissue. These single-tensor methods cannot model complex WM fibers. Our proposed model consists of two steps. First, two bundles are fit: an isotropic bundle with two balls that model isotropic free diffusivity and restricted diffusivity, and a bundle containing stick and zeppelin, corresponding to intra-cellular and extra-cellular diffusion of axons, respectively. Next, a fiber orientation distribution (FOD) is estimated from the fitted parameters. We applied our modeling to multi-shell data from eight patients with glioblastoma, and one patient with a metastatic brain tumor. Our method produces maps corresponding to free water (CSF), restricted diffusion, intra-cellular, and extra-cellular volume fractions. Preliminary results show that restricted diffusivity map of NET comprises 46% of vasogenic metastatic edema compare to 15% of infiltrated GBM edema. In conclusion, we demonstrate that our proposed model shows promise to characterize the NET region of various brain tumors, and distinguishing tumor types. These compartments will provide invaluable radiomic features that no other modality can capture.
Read full abstract