Separating blood and water: Perfusion and free water elimination from diffusion MRI in the human brain.

The assessment of the free water fraction in the brain provides important information about extracellular processes such as atrophy and neuroinflammation in various clinical conditions as well as in normal development and aging. Free water estimates from diffusion MRI are assumed to account for freely diffusing water molecules in the extracellular space, but may be biased by other pools of molecules in rapid random motion, such as the intravoxel incoherent motion (IVIM) of blood, where water molecules perfuse in the randomly oriented capillary network. The goal of this work was to separate the signal contribution of the perfusing blood from that of free-water and of other brain diffusivities. The influence of the vascular compartment on the estimation of the free water fraction and other diffusivities was investigated by simulating perfusion in diffusion MRI data. The perfusion effect in the simulations was significant, especially for the estimation of the free water fraction, and was maintained as long as low b-value data were included in the analysis. Two approaches to reduce the perfusion effect were explored in this study: (i) increasing the minimal b-value used in the fitting, and (ii) using a three-compartment model that explicitly accounts for water molecules in the capillary blood. Estimation of the model parameters while excluding low b-values reduced the perfusion effect but was highly sensitive to noise. The three-compartment model fit was more stable and additionally, provided an estimation of the volume fraction of the capillary blood compartment. The three-compartment model thus disentangles the effects of free water diffusion and perfusion, which is of major clinical importance since changes in these components in the brain may indicate different pathologies, i.e., those originating from the extracellular space, such as neuroinflammation and atrophy, and those related to the vascular space, such as vasodilation, vasoconstriction and capillary density. Diffusion MRI data acquired from a healthy volunteer, using multiple b-shells, demonstrated an expected non-zero contribution from the blood fraction, and indicated that not accounting for the perfusion effect may explain the overestimation of the free water fraction evinced in previous studies. Finally, the applicability of the method was demonstrated with a dataset acquired using a clinically feasible protocol with shorter acquisition time and fewer b-shells.

Intravoxel incoherent motion (IVIM) imaging at different magnetic field strengths: what is feasible?

BACKGROUND: Due to limited SNR the cerebral applications of the intravoxel incoherent motion (IVIM) concept have been sparse. MRI hardware developments have resulted in improved SNR and this may justify a reassessment of IVIM imaging for non-invasive quantification of the cerebral blood volume (CBV) as a first step toward determining the optimal field strength. PURPOSE: To investigate intravoxel incoherent motion imaging for its potential to assess cerebral blood volume (CBV) at three different MRI field strengths. MATERIALS AND METHODS: Four volunteers were scanned twice at 1.5 T, 3 T as well as 7 T. By correcting for field-strength-dependent effects of relaxation, estimates of corrected CBV (cCBV) were obtained in deep gray matter (DGM), frontal gray matter (FGM) and frontal white matter (FWM), using Bayesian analysis. In addition, simulations were performed to facilitate the interpretation of experimental data. RESULTS: In DGM, FGM and FWM we obtained cCBV estimates of 2.2 ml/100 ml, 2.7 ml/100 ml, 1.4 ml/100 ml at 1.5 T; 3.7 ml/100 ml, 5.0 ml/100 ml, 3.2 ml/100 ml at 3 T and 15.5 ml/100 ml, 20.3 ml/100 ml, 7.0 ml/100 ml at 7 T. CONCLUSION: Quantitative cCBV values obtained at 1.5 T and 3 T corresponded better to physiological reference values, while 7 T showed the largest deviation from expected values. Simulations of synthetic tissue voxels indicated that the discrepancy at 7 T can partly be explained by SNR issues. Results were generally more repeatable at 7 T (intraclass correlation coefficient, ICC=0.84) than at 1.5 T (ICC=0.68) and 3 T (ICC=0.46).