While Diffusion Tensor Imaging (DTI) identifies anisotropic diffusion by use of a mono-exponential model, IntraVoxel Incoherent Motion (IVIM) accounts for two motion compartments, pseudo-diffusion and diffusion. However, the IVIM model does not take into account anisotropy, although it has been reported that pseudo-diffusion in the human kidney is anisotropic¹. A combined IVIM-DTI model that contains a diffusion tensor and a pseudo-diffusion fraction tensor has proven to perform well². Both renal DTI and renal IVIM showed significant dependence of their parameters on the cardiac cycle^3,4. The purpose of this study was to investigate the impact of the cardiac cycle on the combined IVIM-DTI model.

Thirteen heathy volunteers were included in the study. Images were acquired at 3 T (Magnetom Prisma, Siemens Healthcare, Erlangen, Germany). ECG triggering was applied. Coronal images of the kidney were acquired twice, once during systolic peak blood flow (trigger delay 200ms) and once during low blood flow (trigger delay $$$\geq$$$400ms).

The diffusion-weighted EPI sequence used the following parameters: Field of view $$$400\times400$$$ mm²; matrix $$$208\times208$$$; partial Fourier 5/8; slice thickness 4 mm; minimum TR 3.000 ms; TE 87 ms; b-values 0,200,250,700,750,800 s/mm²; diffusion directions 30; averages 2. All diffusion weighted images were used for analyses with the standard DTI model and the IVIM-DTI model.

For standard DTI analysis the signal was fitted to the equation

$$S(b) = S_{0}e^{-b\mathbf{D}},$$

where $$$S_{0}$$$ is the signal without diffusion weighting and $$$\mathbf{D}$$$ denotes the diffusion tensor.

The IVIM-DTI model combines the diffusion tensor $$$\mathbf{D}$$$ and the pseudo-diffusion fraction $$$f$$$. Furthermore, not only $$$\mathbf{D}$$$ is considered to be a tensor, but also $$$\mathbf{f}$$$ is described as a tensor. A more thorough explanation of this model was presented before². We assume that the fast motion regime contributes negligibly to the signal for b $$$\geq200$$$ s/mm². Hence, for the IVIM-DTI analysis the signal was fitted to

$$S(b) = S_{0}(1-\mathbf{f})e^{-b\mathbf{D}}$$

Image analysis

Mean diffusivity MD and fractional anisotropy of the $$$\mathbf{D}$$$-tensor FA_D, as well as mean pseudo-diffusion fraction Mf and fractional anisotropy of the $$$\mathbf{f}$$$-tensor FA_f where determined in renal cortex and medulla. The results were tested for statistical significant differences between peak flow and low flow using Wilcoxon’s signed rank test. MD and FA_D where also tested for differences between the DTI and IVIM-DTI model. $$$P<0.05$$$ was considered statistically significant.

Figure 1 lists MD and Mf for peak flow and low flow. DTI yields significantly higher cortical MD during peak flow compared to low flow ($$$P<0.01$$$). There is no significant difference between peak and low flow in cortical MD when applying the IVIM-DTI model. Yet, cortical Mf is higher during peak flow compared to low flow ($$$P<0.001$$$).

For both cortex and medulla MD is significantly lower in the IVIM-DTI model compared to DTI at both times.

Figure 2 shows FA_D and FA_f for peak and low flow.

FA_D as determined by DTI is significantly different in the medulla for peak and low flow ($$$P<0.01$$$). In contrast, medullary FA_D is not significantly different when using the IVIM-DTI model. Instead, medullary FA_f is significantly lower during peak flow compared to low flow. The cortex did not reveal significant differences of FA_D and FA_f between the two time points.

Figure 3 shows a T2-weighted image, as well as MD, FA_D , MF and FA_f maps of one kidney during low flow (a,b,c,d,e) and peak flow (f,g,h,i,k). The FA_f map appears less noisy during peak flow.

The results show that when using the DTI model, the time of acquisition within the cardiac cycle significantly affects MD and FA_D. This is in agreement with previously reported findings^3,4.

We found that the time-dependence of MD and FA_D is extinguished when using the IVIM-DTI model. High cortical MD during peak flow in the DTI model relates to high Mf in the IVIM-DTI model. Low medullary FA_D during peak flow relates to low FA_f in the IVIM-DTI model. FA_D is lower when using IVIM-DTI compared to DTI, because part of the anisotropy is covered by the $$$\mathbf{f}$$$-tensor.

In voxels with very low Mf the value for FA_f might be increased due to noise. However, this would hardly cause a difference between the DTI and the IVIM-DTI model, since a low Mf value will have little impact on MD and FA_D.

The time-dependence of MD and FA_D using the DTI was assumed to be caused by pseudo-diffusion. The IVIM-DTI model allows accounting for pseudo-diffusion. Our findings confirm that in the IVIM-DTI model the pseudo-diffusion related parameters Mf and FA_f depend on the cardiac cycle, while the diffusion related parameters MD and FA_D appear constant.