Introduction

Recently, it has been regarded as important to derive Myelin Water Fraction (MWF) calculated from magnetic resonance imaging (MRI) data to delineate myelin water content for diagnosing diseases such as multiple white matter sclerosis. The acquisition scan time for MWF imaging when using the multi-echo gradient-echo (mGRE) sequence can be shorter than that using the spin-echo (SE) sequence ¹. On the other hand, when applying the mGRE sequence with long echo time (TE), it often becomes a problem in which signal decrease occurs due to the effects of B₀ inhomogeneity and/or magnetic susceptibility ^{2, 3}. In our previous studies, we reported the development of a myelin weighted image derived from quantitative parameter mapping (QPM) ^{4, 5}. The purpose of this study was to determine MWF using QPM-relaxation parameters using quadratic polynomial approximation.

Methods

On a 3 Tesla MR scanner system (FUJIFILM Healthcare Corp.), QPM-MRI was performed on twelve healthy volunteers (seven men, five women; ages, 21-54 years; mean age, 35.2 years) using three-dimensional partially radio frequency-spoiled steady state gradient-echo (3D-RSSG) methods. The imaging parameters were TEs 4.6-36.8 ms (ΔTE, 4.6 ms); repetition times, 10-41.1 ms; flip angles 10 and 40 degrees. After QPM post processing [i.e., estimation of T₁, T₂, Proton Density (PD), and B₁ were generated from the 3D-RSSG dataset], the QPM-relaxation rates R₁ and R₂ were derived. After acquiring QPM datasets, myelin maps using R₁ and R₂^* based on the T₁-weighted/T₂-weighted (T₁ w/T₂w) ratio method ⁵ as
$$\frac{T_{1}w}{T_{2}w}\approx\frac{\frac{1}{T_{1}}PD}{\frac{1}{T_{2}^{*}}PD}=\frac{R_{1}}{\frac{1}{R_{2}^{*}}}=R_{1}\cdot R_{2}^{*},$$
where T₁, T₂ and are relaxation times, and PD is proton density of specific tissue. Signals of T₁w and T₂w images are derived from each relation: $$${T_{1}w}\approx{\frac{1}{T_{1}}PD},{T_{2}w}\approx{T_{2}PD}\approx{T_{2}^{*}PD}.$$$ Figure 1 shows the process of two types of myelin imaging and MWF derived from 3D-RSSG for QPM. In the calculation process of MWF, after performing B₁ correction on the multi-echo 3D-RSSG dataset (e.g., TE, 4.6-36.8 ms; TR, 41.1 ms; FA, 40 degrees) with QPM, rigid and non-rigid transformations were performed, and standardized to the shape of the MNI152 brain using FSL software. In addition, fitting by non-negative least squares (NNLS) was performed using bi-exponential fitting model (i.e., two-pool model) as$$S(t)=A_{short}\exp(-\frac{TE}{T_{2,short}^{*}})+A_{long}\exp(-\frac{TE}{T_{2,long}^{*}}),$$
where S is the signal of each TE, A is amplitude, and T₂^* is T₂^*relaxation time. After performing the fitting analysis of the measured decay signals, MWF was calculated as
$$MWF=\frac{A_{short}}{A_{short}+A_{long}}.$$
Additionally, the determined R₁·R₂^* map was standardized in the same way. Volume-Of-Interest (VOI) analysis using JHU-white matter (WM) atlas was performed on the data averaged from MWF and R₁·R₂^*. All the data in the 48 regions measured by VOI and the data in each region of anterior corona radiata (ACR), genu of corpus callosum (GCC), cerebral peduncle (CP) and posterior corona radiata (PTR) were plotted, and quadratic polynomial approximation was performed to obtain the coefficient of determination. Finally, using acquired quadratic polynomial approximation, the R₁·R₂^* map was converted to MWF, i.e., MWF_QPM.

Results

Table 1 summarizes mean measured MWF and R₁·R₂^* values for each region. Figure 2 shows scatter plots for demonstrating correlations between R₁·R₂^* and MWF values for all and some represented white matter VOIs. Their relationship in all WM regions had a strong correlation with a quadratic polynomial approximation (R₂ = 0.991). Figure 3 shows MWF, R₁·R₂^* and MWF_QPM maps standardized from twelve healthy volunteers to MNI. Figure 4 shows MWF_QPM maps of two volunteers (men, ages, 21 and 54 years) converted from R₁·R₂^* using a obtained regression curve.

Discussion

We focused only on the WM region using JHU-White matter atlas in this study. When using a quadratic polynomial equation, the relationship between R₁·R₂^* and MWF had a strong significant correlation for not only all but also each region of WM (Fig. 2). These results may be useful in future attempts to combine multiple regions.

Conclusion

We found that the relationship between R₁·R₂^* and MWF has a strong correlation when using quadratic polynomial approximation. MWF map derived from QPM can dramatically improve image quality when compared to the mGRE-MWF map.

Acknowledgements

No acknowledgement found.