Intravoxel incoherent motion (IVIM) has been used to study perfusion in abdominal organs. However, its application within the brain is controversial.^1,2,3 Results can vary significantly depending on the fitting algorithm, the experimental parameters, and the range of b-values. Concerns have also been raised about biases introduced by fitting low SNR data, and contaminating effects of CSF due to bulk motion. The aim of this work was to investigate IVIM as a potential method for acquiring reliable blood volume and/or perfusion information in the brain. A direct comparison was performed between IVIM and pCASL, a method capable of quantifying blood flow within grey matter (GM) which has been validated against PET,⁴ but which suffers from low SNR and cannot be applied in areas with long arterial arrival times.

10 subjects were scanned on a 3T Siemens Verio scanner with a 32-channel head coil. A 6:28-minute multi-PLD pCASL scan was run (6 delay times ranging 250–1500ms) with 3.4×3.4×6mm voxels. Images were preprocessed using FSL,⁵ and analysed using BASIL.⁶ For IVIM, a diffusion sequence was run with b-values of 0, 10, 20, 40, 80, 110, 140, 170, 200, 300, 400, 500, 600, 700, 800 and 900s/m², acquired in three orthogonal directions with a twice-refocussed spin echo and an EPI readout. Images were acquired at the same resolution as pCASL. To match the pCASL imaging time, 2 repetitions were acquired and averaged, with an imaging time of 6:12 minutes.

A 1mm isotropic MPRAGE scan was acquired, and segmented using FAST⁵ and registered to both ASL and IVIM space using FLIRT.⁵ GM and white matter (WM) masks were created from voxels with partial volume estimates of >50%. IVIM data were corrected for motion and eddy-current distortions. A reversed phase-encoding b=0 scan was used to correct for susceptibility-induced distortions using the TOPUP tool.⁵ A biexponential model was fitted to each voxel in two steps⁷ using least-squares fitting in MATLAB, to estimate the diffusion coefficient $$$D$$$, the pseudo-diffusion coefficient $$$D^*$$$ and the fraction of fast-diffusing spins $$$f_v$$$. This aims to mitigate overfitting, assuming $$$D^∗$$$ is substantially larger than $$$D$$$ and may be neglected for large b-values. First, a monoexponential model was fit to b-values >200 to estimate $$$D$$$:

$$S=S_0e^{-bD}$$

Then the full biexponential model was fit to all b-values, with fixed $$$D$$$, to estimate $$$f_v$$$ and $$$D^∗$$$:

$$S=S_0\{(1-f_v)e^{-bD}+f_ve^{-bD^*}\}$$

S₀ was unconstrained in both cases. IVIM perfusion was defined as $$$f_vD^*$$$.

GM masks were applied to pCASL and IVIM images. Voxels with $$$f_v>0.3$$$ were excluded from masks as this is unphysiological. Correlations between pCASL perfusion and IVIM $$$f_v$$$ and perfusion were hypothesised to be linear and were assessed by computing the Pearson product-moment correlation coefficients (r) and the corresponding p-values, where p<0.05 was deemed to be significant.

Fig.1 shows average GM results. WM perfusion cannot be reliably measured using pCASL because of the long arterial arrival time. $$$f_vD^∗$$$ in WM was found to be 0.81±0.20×10−3m²/s, leading to a GM/WM perfusion ratio of 1.87±0.88.

Fig.2 shows pCASL and IVIM maps from one subject, along with the GM mask. Although the maps for pCASL and IVIM perfusion show similar contrasts, there is increased signal in IVIM from both WM and CSF, and quantitatively no significant correlation was observed between GM pCASL and IVIM perfusion (Fig.3(a), r=0.090, p=0.805) or pCASL and IVIM $$$f_v$$$ (Fig.3(b), r=0.107, p=0.768)

IVIM claims to be a quantitative model, yet conclusive validation of the method within the brain is still lacking, with past studies reporting both positive⁸ and negative⁹ correlations between $$$f_v$$$ and DSC or ASL MRI methods, respectively. Although qualitative similarities were observed between IVIM parameters and pCASL perfusion maps, this study found no significant correlation between average GM values.

The lack of correlation in Fig.3 could be caused by different vessel properties between individuals. Although the ability to compare absolute perfusion between subjects is very valuable, particularly in clinical cases, many neuroscience applications require only spatial or functional contrast. Thus the question is whether the contrast observed in $$$f_v$$$ and $$$f_vD^∗$$$ for a single subject is a true reflection of underlying blood volume or perfusion. Given concerns over the contaminating effects of CSF, it is difficult to conclude whether the spatial similarity between ASL and IVIM images supports this hypothesis, especially as the majority of GM voxels will also include a sizeable fraction of CSF at this relatively low resolution. Thus IVIM within the brain should not be considered a fully quantitative method. Although the contrasts may provide clinically valuable images, they should not be interpreted as corresponding directly to any specific physiological parameters.