Intra-voxel incoherent motion (IVIM) can probe both diffusion and pseudo-perfusion by using a range of b-values, typically 0 to <1000 (s/mm²)^1–3. Lower b-values are sensitive to capillary-level perfusion, and higher b-values are sensitive to tissue diffusion. This ability to measure diffusion and perfusion in a single scan is desirable for studies where both metrics are desired, such as alcoholism^4,5 and stroke³. This work introduces both a new IVIM fitting procedure and robust anatomically-based region-driven analysis in order to investigate test-retest variability within and between scan sessions and age effects on IVIM in five brain structures.

Equation 1 shows IVIM signal equation assuming pseudo-perfusion (fe^-bD*) and diffusion ((1-f)e^-bD) compartments. A four parameter fit of S0 (signal), f (perfusion fraction), D* (pseudo-diffusion coefficient), and D (diffusion coefficient) is performed where S(b) is the signal intensity at each of the diffusion values b. The product fD* represents a pseudo-perfusion parameter¹.

Equation 1: $$$S\left(b\right)=S0\left(fe^{-bD^{*}}\right)+\left(1-f\right) e^{-bD} )$$$

Subjects:

Healthy normal volunteers (9 men, 6 women; age 23.5 to 67.6 years, mean±std=41.7±17.4) were scanned on a GE MR750 3T (GE Medical Systems, Waukesha, WI, USA) with an 8-channel head coil (Invivo Co, Gainesville, FL, USA).

Acquisition:

All subjects received a structural, two identical echo-planar imaging (EPI) IVIM, and a reversed phase encode (PE) polarity EPI scan (Table 1). Seven subjects were scanned twice with the same protocol >24 hours apart to determine the intra-session (between scans in the same session) and inter-session (between scans on separate days) repeatability.

Pre-Processing:

All IVIM scans were EPI-distortion and eddy-current corrected using FSL5 (Topup, Eddy)^6,7. The SRI24 atlas⁸ was registered to the IVIM image via the IR-SPGR, and all IVIM images were registered together using ANTS⁹. SRI24 brain and parcellation masks were transformed to IVIM space and eroded using a 3x3x3 kernel to minimize cross-region and fluid-tissue contamination.

IVIM fitting:

Each image voxel and ROI signal vector were fit individually for S0, f, D, and D* (Equation 1) using a multi-step approach illustrated in Figure 1, which was implemented in Python2.7¹⁰.

1. Linearly fit $$$\ln\left(S\left(b>200\right)\right)$$$ for S0’ and D, where S0’ is the zero intercept of the diffusion fit at b>200.

2. Linearly fit $$$\ln\left(S\left(b\leq200\right)\right)$$$ for S0 and D*’, where D*’ is the combination of D and D* decays.

3. Calculate f: $$$f=1-\frac{S0'}{S0}$$$.

4. Fit all b-values using Equation 1 for D* keeping S0, f, and D fixed.

5. Fit S0, f, D, and D* using Equation 2 with the previous values as initial guesses and λ=0.01.

This produces a single set of IVIM parameters for each ROI and each voxel in the image (Figure 2).

Analysis:

The statistical analysis in R¹¹ focused on the thalamus, cingulum, amygdala, caudate nucleus, and pons using factors f, D, and fD*. The intra-session (for all 15 subjects and the seven young subjects with two scan sessions) and inter-session (for all subjects with two scan sessions) differences were analyzed using intraclass correlation (ICC). The age analysis used a two-tailed Mann-Whitney U test on all subjects to compare two age groups (<45 years, n=10; >45 years, n=5).

Equation 2: $$S\left(b\right)=S0\left(fe^{-bD^{*}}\right)+\left(1-f\right) e^{-bD} )+\\\lambda\left|\frac{S0-S0_{initial\_guess}}{S0_{initial\_guess}}\right|+\lambda\left|\frac{f-f_{initial\_guess}}{f_{initial\_guess}}\right|+\lambda\left|\frac{D-D_{initial\_guess}}{D_{initial\_guess}}\right|+\lambda\left|\frac{D^{*}-D^{*}_{initial\_guess}}{D^{*}_{initial\_guess}}\right|$$

The ICC results show that for most factors in all regions, the intra-session repeatability is moderate to extremely high, and the inter-session repeatability is high with a few exceptions, the caudate, amygdala, and pons (Table 2). The values for f, D, and fD* were higher in the older group, illustrated in Figure 3.

We found that reliable results are produced using the new fitting procedure combined with regional analysis. The intra-session ICC values are significant for all regions and metrics for either the full set of scans or the set of seven younger subjects scanned twice. The majority of the inter-session ICC values are also significant, indicating good inter-day repeatability. The lower inter-session relative to intra-session ICC values may be related to physiological variation across time.

The age-related correlation of greater f, D, and fD* in older adults is contrary to the expected perfusion decrease with age¹² and occurred despite atlas erosion to minimize CSF contamination. This age difference may arise from CSF-filled perivascular spaces which can influence IVIM measures¹³ and also expand with age¹⁴. These results emphasize how age is an important factor to consider when interpreting IVIM results.

IVIM metrics are significantly repeatable between scans and scan days and show significant and increasing effects with age in f, D, and fD*, which may arise from enlargement of CSF-filled perivascular space with age and highlight how age is an important consideration when using IVIM.