R₂* (=1/T₂*) is influenced by various sources such as magnetic susceptibility, chemical shift, and microstructural anisotropy. In white matter of the brain, it has been demonstrated that R₂* is affected by fiber orientation relative to B₀ [1, 2] and myelin concentration [3]. However, the portions of R₂* that the two sources explain (i.e. the effect size) have not been explored simultaneously. In this study, we measured the contribution of fiber orientation and myelin concentration to R₂* quantitatively to demonstrate that the two sources are primary determinant for R₂*. Additionally, we proposed novel R₂* maps that are corrected for fiber orientation and/or myelin concentration biases. The bias-corrected R₂* maps and histograms may serve as reliable data that discriminate patients from healthy controls.

Three healthy volunteers were scanned at 3T. [R₂* map] 3D multi-echo GRE was acquired with following parameters: FOV=256×256×104mm³, voxel size=2×2×2mm³, TR=67ms, TE=1.5:2.03:32ms (16 echoes). For R₂* mapping, a mono-exponential function was fitted to three different echo periods: early echoes (first 8 echoes), late echoes (last 8 echoes) and all echoes. [Myelin concentration map] The multi-echo GRE data were reprocessed for GRE myelin water imaging (MWI), generating a myelin water fraction (MWF) map as a marker for myelin concentration. The process utilized a three pool complex model as summarized in [4]. [Fiber orientation map] To estimate fiber orientation, DTI was acquired. The resulting color DTI FA map was processed to generate a map of relative angles between fibers and B0. [R₂* models] The R₂* signals were modeled:

Myelin concentration model: R₂*(MWF) = a₀ + a₁·MWF

Fiber orientation model: R₂*(θ) = b₀ + b₁·cos2θ+b₂·cos4θ

Complete model: R₂*(θ,MWF) = c₀ + c₁·MWF+c₂·MWF·cos2θ+c₃·MWF·cos4θ

where θ is the angle in the fiber orientation map and MWF is from MWI. A white matter mask was generated using thresholds in FA (>0.4). To improve reliability, the fiber orientation was quantized with a step size of 10° and MWF with a step size of 2% (3.6% for the complete model). The MWF values larger than 18% were discarded due to a small number of voxels. The resulting data were fitted to each model, generating the coefficients. This approach may cause a bias due to non-uniform distribution of MWF over the fiber orientations but does not require multiple scans with different head orientations [5]. After the model fitting, R₂* plots, R₂* maps and histograms that removed the effects of each model were displayed.

Both MWF and fiber orientation showed significant effects on R₂* (Fig. 1), confirming the previous findings [1-3]. When quantified, the MWF change from 0% to 18% showed 5.09±0.06Hz change whereas the orientation dependent R₂* showed 3.29±0.92Hz change. These results suggest that the myelin concentration has 1.54±0.29 times larger effect than the fiber orientation. When the effect of fiber orientation was corrected (Fig. 1C), the linear fit of MWF improved (R² from 0.973 to 0.981). The resulting map showed a less biased R₂* map, revealing reduced R₂* contrasts for different directional fibers (white circles; Fig. 3B vs. 3C). When MWF was removed, the resulting plot revealed abnormally small R₂* in perpendicular orientation (Fig. 1D). This change is not physically and can be explained by a large number of perpendicular fibers with high myelination (which was confirmed). Hence the resulting MWF corrected map (Fig. 3D) may contain overcorrection of MWF. When the full model was fitted (Fig. 1G), the resulting R₂* showed no fiber orientation and no MWF dependency (Fig. 1E, 1F). The resulting map (Fig. 3E) revealed relatively uniform R₂* across white matter compared to the other maps. This map may reflect other susceptibility sources (e.g. iron). The histogram also demonstrated the effects of the bias corrections in each model, generating a more Gaussian distribution after compensating for both effects (Fig. 5).

In this work, the R₂* estimation used a single exponential decay fitting and is sensitive to the choice of echo times. The early echoes had larger weighting of the myelin signal and showed larger orientation effect (Fig. 4). The opposite was true for the late echoes. Hence, one may choose echo times based on the interests of the study. Similar results were observed in the R₂* histograms, showing a larger bias toward high R₂* values in the early echoes (Fig.5 black lines).

In this work, we measured the effect size of myelin concentration and fiber orientation in R₂*, demonstrating myelin concentration has 1.54 times larger effect than fiber orientation. Additionally, we generated the myelin concentration and/or fiber orientation bias free R₂* maps which may have important applications.