Applications of diffusion MRI: Tensor-valued encoding, time-dependent diffusion, and histological validation

Abstract: Diffusion MRI (dMRI) sensitizes the MR signal to the diffusion of water molecules at the microscopic level and thereby non-invasively probes tissue microstructure. This is relevant when determining biological properties of tissues, for example, cancer type and its malignancy. The problem is, however, that dMRI lacks sensitivity and specificity to distinct microstructural features because an image voxel contains vast number of different features that are mapped onto relatively few dMRI observables. To tackle this issue, we aimed at solving two gaps in current knowledge—the first was related to what microstructural aspects are of most importance and the second to how adding new observables to the dMRI measurement could improve brain tumor imaging.In this work, we first investigate the biological underpinnings of dMRI observables—focusing on the degree to which larger-scale microstructural arrangements are of relevance. In Paper I, we investigated the effects of non-straight propagation of axons and found that they are indistinguishable from those originating from the diameter of a straight axon, at least for typical measurements with a clinical scanner. We propose that the use of short diffusion times could help separate them. In Paper II, in a comparison between histology and microimaging of meningioma brain tumors, we quantified to what degree the common biological interpretation of one of the most used dMRI observable holds—mean diffusivity (MD) as reflecting cell density and fractional anisotropy reflecting tissue anisotropy. We found that the local variability in MD was explained in minority of the samples whereas FA in majority by the common interpretations. We suggested additional relevant features such as tumor vascularization, psammoma bodies, microcysts or tissue cohesivity for explaining MD variability.Second, we examined whether a framework that introduces a new measurement observable brings value in intracranial tumor imaging. This new variable is termed the b-tensor shape and is derived from the tensor-valued dMRI paradigm. In Paper IV, we adjusted and shortened by 40 % (from 5 to 3 minutes) a tensor-valued dMRI protocol for clinical imaging of intracranial tumors and applied it to characterize to a wide range of different intracranial tumors. The protocol was also used in clinical studies of patients with intracranial tumors—gliomas and meningiomas—in Paper III and Paper V, respectively. In Paper III, we found that using so-called spherical b-tensor encoding leads to enhanced conspicuity of glioma hyperintensities to white matter in all patients and on average the signal-intensity-ratio increased by 28 %. In Paper V we found that it may also inform on meningiomas preoperatively. The standard deviation of isotropic kurtosis was associated with tumor grade and with and the 10th percentiles of the mean and anisotropic kurtoses with firm tumor consistency. Preoperative knowledge of the consistency is important for the neurosurgeons when choosing the optimal surgical procedure.