Unraveling the major role of vascular (R2') contributions to R2* signal relaxation at ultra‐high‐field MRI: A comprehensive analysis with quantitative gradient recalled echo in mouse brain

Abstract

Purpose

Ultra‐high‐field (UHF) R2* relaxometry is often used for in vivo analysis of biological tissue microstructure without accounting for vascular contributions to R2* signal, that is, the BOLD signal component, and magnetic field inhomogeneities. These effects are especially important at UHF as their contribution to R2* scales linearly with magnetic field. Our study aims to report on the results of separate contributions of R2t* (tissue‐specific sub‐component) and R2' (vascular BOLD sub‐component), corrected for the adverse effects of magnetic field inhomogeneities, to the total R2* signal at in vivo UHF MRI of mouse brain.

Methods

Four healthy, 8‐week‐old C57BL/6J mice were imaged in vivo with multi‐gradient echo MRI at 9.4 T and analyzed using the quantitative gradient recalled echo (qGRE) approach. A segmentation protocol was established using the Dorr Mouse Brain Atlas and ANTs Syn registration to warp template brain region labels to subject qGRE maps.

Results

By separating R2' contribution from R2* signal, we have established normative R2t* data in mouse brain. Our findings revealed significant contributions of R2' to R2*, with approximately 42% of the R2* signal arising from vascular contributions, thus suggesting the R2t* as a more accurate metric for quantifying tissue microstructural information and its changes in neurodegenerative diseases.

Conclusion

qGRE approach allows efficient separation of tissue microstructure‐specific (R2t*), vascular BOLD (R2'), and background gradients contributions to the total R2* relaxation at UHF MRI. Due to low concentration of non‐heme iron in mouse brain, major contribution to R2t* results from tissue cellular components.

1. INTRODUCTION

In vivo ultra‐high‐field (UHF) MRI, utilizing magnetic fields 7T and higher, has a proven potential for providing unique information related to diagnostic imaging, including neurodegenerative diseases. ¹ This is due to the higher spatiotemporal resolution and higher SNR of UHF MRI as compared to traditionally utilized 1.5T and 3T magnetic field strengths. ² One UHF MRI application involves R2* imaging which proved useful for discriminating structure of superficial white matter (WM) ³ and as a surrogate of brain tissue cytoarchitecture patterns, ⁴ , ⁵ as also reported with 3T MRI. ⁶

However, it is also understood that the BOLD signal ⁷ has a significantly higher influence on MRI parameters such as R2* (1/T2*) with increased field strength. ⁸ Evidence suggests that higher magnetic fields result in larger magnetic susceptibility artifacts. ⁹ Hence, accounting for the contributions of BOLD effect and magnetic susceptibility artifacts is crucial in analysis of R2*‐based results at UHF MRI.

To address the contributions of BOLD effect and magnetic field inhomogeneities to R2* imaging at UHF MRI, herein we use qGRE MRI, an imaging technique providing quantitative microstructural information distinguishing biological tissue cellular and vascular properties. ¹⁰ It is based on a GRE MRI sequence with multiple gradient‐recalled echoes, a biophysical model of GRE signal decay ⁸ , ¹¹ and algorithms for data analysis that allow generating images sensitive to tissue cellular (R2t* metric of qGRE signal) and hemodynamic (R2' metric of qGRE signal) properties corrected for macroscopic field inhomogeneities ¹² and physiological fluctuations. ¹³ It emerges as a potential biomarker to quantify neuronal brain structure ⁶ and functional connectivity. ¹⁴ qGRE R2t* metric also reflects changes in healthy aging, ¹⁵ as well as the progression of neurodegenerative diseases, including but not restricted to Alzheimer's Disease, ¹⁶ multiple sclerosis, ¹⁷ psychiatric diseases, ¹⁸ and traumatic brain injury. ¹⁹

The biophysical qGRE model separates R2* qGRE signal into three distinct components: R2t*, a tissue‐cellular‐specific component, R2', derived from the presence of deoxygenated blood in veins (BOLD effect), and the MACRO component, that is, adversative signal arising from macroscopic field inhomogeneities (background gradients). The R2' and MACRO components are proportional to magnetic field strength, ⁸ , ¹² thus affecting R2*‐based tissue microstructural correlates unless they are accounted for.

This study, conducted as 9.4T magnetic field strength with in vivo scans of mouse brains, aims to accurately interpret information from R2* measurements in multiple brain regions covering gray matter (GM) and WM, with particular attention to the relative contributions of R2t* and R2'. To segment brain regions, Dorr Brain Atlas ²⁰ labels were consolidated using hierarchical brain region organization provided by Allen Brain Atlas materials (ISH Data: Allen Brain Atlas: Mouse Brain (brain‐map.org)). Segmentation labels were then warped via inverse transformation and registered to individual subject brains using the Advanced Normalization Tools (ANTs) toolkit. ²¹

Our findings revealed that in the mouse brain at 9.4T the vascular BOLD contribution (R2') to R2* is comparable to the tissue specific R2t* contribution, with approximately 42% of the R2* signal in the GM arising from the BOLD effect. This suggests the R2t* subcomponent as a more accurate metric for quantifying tissue microstructural information and it changes in neurodegenerative diseases.

2. METHODS

All studies were approved by the local IACUC of Washington University.

2.1. Biophysical model of qGRE MRI signal formation

To identify brain tissue‐cellular‐specific and hemodynamic contributions to R2* signal, we use the quantitative gradient recalled echo (qGRE) MRI method. ¹⁰ The qGRE method is based on a 3D gradient recalled echo MRI sequence with multiple gradient echoes and a theoretical model that describes GRE signal relaxation properties. In the framework of this approach, the GRE signal dependence on the echo time TE is presented in the following equation ¹⁰ :

$$S(\mathit{TE})=S_0\cdot {\mathrm{e}}^{-R2t*\cdot \mathit{TE}+i\cdot 2\pi \cdot \mathrm{\Delta }f\cdot \mathit{TE}}\cdot F_{\mathit{\text{BOLD}}}(\mathit{TE})\cdot F_{\mathit{\text{macro}}}(\mathit{TE})$$ (1)

where TE is the gradient echo time, S ₀ is the signal amplitude, R2t* is the tissue‐specific (t‐ stands for tissue) transverse relaxation rate constant (describing GRE signal decay that would exist in the absence of the BOLD effect and the magnetic field inhomogeneities), Δf is the frequency shift (dependent on tissue structure and also macroscopic magnetic field created mostly by tissue/air interfaces). Function F _BOLD (TE) describes GRE signal decay due to the presence of blood vessel networks with deoxygenated blood (veins and the part of capillaries adjacent to veins and venules). ⁸ Function F _macro (TE) accounts for the adverse effects of macroscopic magnetic field inhomogeneities. ¹¹ In this paper for the BOLD model we use the following expression ⁸ , ¹¹ :

$$F_{\mathit{\text{BOLD}}}(\mathit{TE})=\exp \left[-\mathit{\text{dCBV}}\cdot f_s(\delta \omega \cdot \mathit{TE})\right].$$ (2)

In Eq. (2), dCBV is the deoxygenated cerebral blood volume fraction (deoxyhemoglobin‐containing part of the total blood volume, i.e., veins, venules, and adjacent to them portion of the capillary bed); the characteristic frequency δω is ⁸ :

$$\delta \omega =\frac{4}{3}\pi \cdot \gamma \cdot B_0\cdot \mathit{Hct}\cdot \mathrm{\Delta }{\chi }_0\cdot (1-Y).$$ (3)

In Eq. (3), B ₀ is the MRI scanner magnetic field (9.4 T in this study), $\gamma =2.675\cdot {10}^8\mathit{rad}\cdot s^{-1}\cdot T^{-1}$ is the gyromagnetic ratio, Hct is the blood hematocrit level, $\mathrm{\Delta }{\chi }_0=0.27\mathit{ppm}$ ²² is the susceptibility difference between fully oxygenated and fully deoxygenated blood, and Y is the blood oxygenation level (Y = 0 corresponds to fully deoxygenated blood and Y = 1 corresponds to a fully oxygenated blood). Function $f_s$ describes the signal decay due to the presence of the blood vessel network which was defined in. ⁸ Herein we use expression for the function $f_s$ defined in terms of hypergeometric function ${}_{1}F_2$ ²³ :

$$f_s(\delta \omega \cdot \mathit{TE})={}_{1}F_2\left(\left[-\frac{1}{2}\right];\left[\frac{3}{4};\frac{5}{4}\right];-\frac{9}{16}{(\delta \omega \cdot \mathit{TE})}^2\right)-1.$$ (4)

The important feature of function $f_s$ is a non‐linear behavior characterized by a quadratic TE dependence for short TE <1.5/δω. ⁸ Such a non‐linear behavior allows separate evaluation of parameters δω and dCBV in Eq. (1) if sampling of experimental data includes short TE interval. ¹¹ For longer TE intervals, the function $f_s$ reveals a linear behavior with respect to TE, resulting in a Lorentzian signal decay:

$$\begin{array}{ll}& F_{\mathit{\text{BOLD}}}(\mathit{TE})=\exp \left[-R{2}^{\prime }\cdot \mathit{TE}\right],R{2}^{\prime }=\mathit{\text{dCBV}}\cdot \frac{4}{3}\pi \cdot \gamma \cdot \\ & B_0\cdot \mathit{Hct}\cdot \mathrm{\Delta }{\chi }_0\cdot (1-Y)\end{array}$$ (5)

The $R{2}^{\prime }$ in Eq. (5) is related to the GRE signal loss due to the presence of paramagnetic deoxygenated blood in veins and the pre‐venous part of the capillary bed, thus connecting GRE signal with blood hemodynamic properties. The BOLD model in Eqs. ((2), (3), (4), (5))–((2), (3), (4), (5)) was previously validated in phantom ¹¹ and small animal ²⁴ models.

As mentioned above, function F _macro (TE) accounts for the adverse effects of macroscopic magnetic field inhomogeneities, calculated by means of a voxel spread function (VSF) method. ¹² It is important to emphasize that in the framework of the VSF approach calculating the function F _macro (TE) in the voxel n requires calculating the voxel spread matrix ${\mathrm{\Psi }}_{n,m}(\mathit{TE})$ that depends on magnetic fields and their gradients in the n‐th and the neighboring voxels, m, calculated from the phase information in the GRE data. Details of the VSF theory can be found in. ¹²

Eq. (1) suggests that the tissue‐related relaxation factor in GM can be described by a single relaxation rate constant $R2t^{*}$. The latter has contributions from the cellular components $R2t_{\mathit{\text{cell}}}^{*}$, and non‐heme iron $R2t_{\mathit{nhi}}^{*}$:

$$R2t^{*}=R2t_{\mathit{\text{cell}}}^{*}+R2t_{\mathit{nhi}}^{*}$$ (6)

The cellular component $R2t_{\mathit{\text{cell}}}^{*}$ is mostly proportional to cell density with major contribution from neurons and myelin. ⁶ The $R2t_{\mathit{nhi}}^{*}$ component is proportional to the concentration of non‐heme iron but also depends on a specific composition of iron particulates. A detail theory of non‐heme iron $R2t_{\mathit{nhi}}^{*}$ contribution to the GRE signal formation was developed in. ²⁵ Here we note that magnetic susceptibility effects resulting from the presence of iron also cause GRE signal relaxation deviation from a linear regime. ⁸ However, while for the blood vessel network such a deviation is present in a millisecond time range, whereas for the non‐heme iron, due to its very high magnetic susceptibility (about 500 ppm ²⁶ ), the deviation from a linear behavior generally exists only in a microsecond range. ²⁵ Hence, in a millisecond time range (where our experiments are conducted), the presence of non‐heme iron results in a linear GRE signal behavior adequately described by the $R2t_{\mathit{nhi}}^{*}$ relaxation rate parameter. Deviations from this regime can occur in cases when non‐heme iron is present in large complexes with low iron concentration (such as reported in superficial WM ³ and substantia nigra ²⁷ ). This leads to low mean magnetic susceptibility of these complexes. Consequently, the GRE signal from water molecules external to these complexes could maintain quadratic behavior over extended time ranges.

In this paper we only focus on separation of R2' and R2t* effects in GM. For WM such a separation would require a more complicated model accounting for a multicompartment structure of WM (i.e., intra‐cellular, extra‐cellular, and myelin water). Since these compartments have different frequences, ²⁸ , ²⁹ , ³⁰ , ³¹ the separation of BOLD effect from the total WM signal would require accounting for these effects, which is beyond the scope of current paper. Hence, for WM we only report measurements of R2* without separation in R2t* and R2'.

2.2. Image acquisitions and biophysical model parameters estimation

Four healthy, 8‐week‐old C57BL/6J mice were used in this study. All scans were performed on Bruker (Bruker Biospin; Ettlingen, Germany), 9.4T MRI scanner, using a four‐channel mouse brain CryoProbe. 3D GRE sequence was utilized with 11 gradient echoes (TE1 = 2.63 ms, delta TE = 4 ms), and a TR of 50 ms. Resulting scans yielded images with 125 μm (isotropic) resolution and [128 × 128 × 128] matrix size. During scanning, all subject mice were under isoflurane anesthesia at approximately 1.3%–1.5% to maintain a respiration rate of 90–100 breaths per minute.

After the data acquisition, the raw k‐space data were read into MATLAB (The MathWorks, Inc., Natick, MA) for data analysis. Multi‐channel data were Fourier‐transformed into the image domain and then combined using a previously developed algorithm ³² accounting for phase offset and noise variation in the individual channels.

Biophysical model parameters estimation is accomplished in several steps. First, the function F_macro(TE) is precomputed using the VSF method ¹² relying on the phase and magnitude information in the GRE data. This allows separation of the adverse macroscopic field inhomogeneity effects (background gradients) on GRE signal from the magnetic field inhomogeneities caused by the vascular‐specific BOLD effect (the sum of these two constitutes the general R2'). Then, the F_macro(TE) function is used in Eq. (1) for obtaining the model parameters S ₀ , $R2t^{*}$, dCBV and δω by fitting Eq. (1) to experimental data on a voxel‐by‐voxel basis using the non‐linear least square (NLLS) algorithm. To reduce the influence of noise on the estimated fitting parameters, the values of dCBV and δω in each voxel, obtained at the first fitting step, are averaged with corresponding values of five neighbors on each side using sliding window. These averaged values are used then to calculate $R{2}^{\prime }=\mathit{\text{dCBV}}\cdot \delta \omega$, and to evaluate $R2t^{*}$ from a second run of fitting keeping these averaged values fixed in each voxel.

2.3. Segmentation protocol

A segmentation protocol was devised to warp template structure labels from the Dorr Brain Atlas ²⁰ to the qGRE maps. qGRE images were first combined to obtain “T2‐like” images to register with the Dorr Brain Atlas T2‐weighted template brain images. This process included dividing the mean of the first five gradient echo qGRE brain images in each voxel by their SD. The first five echoes were empirically selected as they were less affected by magnetic field inhomogeneities, thus allowing for more accurate conversion. Because regions containing cerebrospinal fluid have much longer T2* relaxation time, hence, slower decay as function of TE in comparison to WM and GM areas, they appear bright on T2‐like images, thus emulating T2‐weighted images. Example of mouse brain registration procedure is shown in Figure 1. The ANTs registration toolkit ²¹ was utilized to warp template segmentation labels onto subject scans.

FIGURE 1.

Example of mouse registration: T2‐like image (left), overlayed transformed Template labels (center), and Dorr Template Brain (right). Note that dark regions in lower part of T2‐like image are caused by strong field inhomogeneities.

Basic 356 Dorr brain region labels were consolidated into 84 regions of interest (ROIs), based on the hierarchical organization of brain regions available on Allen Mouse Brain Atlas to allow for analysis of larger brain regions. Consolidation was performed based on availability of corresponding labels between the two atlases at varying levels of hierarchical organization. Example is shown in the on‐line Figure S1, while details are provided in the on‐line Tables S1 and S2. Mean R2* and R2t* values in ROIs were obtained via ITK‐SNAP software.

3. RESULTS

For general analysis we used mean values of R2*, R2t*, and R2' of the selected regions of interest. The T2‐like images of each mouse were utilized for registration, and resulting transformed labels were overlayed on the corresponding naturally coregistered R2*, R2t*, and R2' maps for analysis. Example of qGRE T1‐weighted image (corresponding to the first echo image), R2*, R2t*, and R2' maps are shown in Figure 2. The VSF approach successfully mitigated the influence of field inhomogeneities on R2*, R2t*, and R2' measurements except of regions with very strong field inhomogeneities where GRE signal was significantly (by about 50%) reduced at TE5 (fifth echo image). Artifacts, exemplified by such areas are seen on the inferior border of the R2* and R2t* images in Figure 2. To remove voxels in such areas from data analysis, segmentation labels for each individual mouse were compared with their corresponding F‐function masks (defined as regions with F(TE5) < 0.5) and manually edited to exclude them from the mask volume via the label editing tool on ITK‐SNAP. Data in the Table 3S show that only minor portion of voxels were excluded from the data analysis.

FIGURE 2.

Example of T1‐weighted, R2*, R2t*, and R2' qGRE images. Figure depicts (left to right): Image of the first gradient echo (T1‐weighted) image of mouse head; The corresponding R2*, R2t*, and R2' maps. The scale bar shows relaxation parameters (R2*, R2t*, and R2') values in 1/s.

Average GM R2*, R2t*, and R2' values were 29.1 ± 2.9 s⁻¹, 12.9 ± 1.5 s⁻¹ and 11.9 ± 1.0 s⁻¹, respectively, confirming that the BOLD effect has significant contributions to the R2* signal. Average WM R2* was 34.9 ± 4.1/s.

Detailed R2*, R2t*, and R2' results for selected labels in GM (corresponding to organization levels within the Allen Brain Atlas) are presented in Figure 3 and Table A1 (Appendix A), and for R2* in WM selected labels in Figure 4 and Table A2 (Appendix A).

FIGURE 3.

qGRE metrics in GM: R2* (blue), R2t* (orange), R2' (green). Bars represent mean values across all subjects and error bars represent corresponding SDs. Regions correspond to organization levels within the Allen Brain Atlas as specified in Supplementary Table 1S.

FIGURE 4.

R2* values in WM labels. Bars represent mean values across all subjects and error bars represent corresponding SDs. Regions correspond to organization levels within the Allen Brain Atlas as specified in Supplementary Table 2S.

4. DISCUSSION

Given an increased interest in using UHF MRI for in vivo R2* relaxometry as a surrogate for brain tissue microstructure(see for example, ¹ , ² , ³ , ⁴ , ⁵ ) it is important to establish and differentiate contributions of major subcomponents (cellular and vascular BOLD) to R2* signal. Our paper is devoted to this subject, using living mice for study. Given the small variation in measurements between subject mice in most brain region labels in the consolidated GM labels as shown by error bars in various brain regions (Figure 3), the method of separation of R2* signal into R2t* and R2' components and the segmentation protocol established in our study provide a consistent and effective method to analyze a battery of qGRE images and generating quantitative maps of R2*, R2t*, and R2' at UHF MRI.

Our results demonstrated that at 9.4T vascular BOLD contribution (R2') to R2* metric in mouse brain is comparable in value to tissue specific R2t* contribution, with 41.8% of the R2* signal in GM arising from the BOLD effect (mean GM R2* = 29.1/s, mean GM R2' = 11.9 s⁻¹). This value of R2' is consistent with previously developed theory of BOLD effect. ⁸ Indeed, for typical blood hemodynamic parameters in GM (dCBV = 0.03, Hct = 0.4, Y = 0.6–0.7), and $B_0=9.4T$, Eq. (5) predicts R2' = 10–12 s⁻¹.

In the human brain at 3T field strength, previous studies have found mean R2t* and R2' values to be 15.1 ±0.6 s⁻¹ and 5.1 ± 0.3 s⁻¹ for GM, respectively. ¹⁰ Hence, at 3T R2' contributes only about 25% to the total R2* signal, which is significantly smaller than its 42% contribution that we found at 9.4T.

The similarities between measured cortical GM R2t* values at 3T (15.1 ± 0.6 s⁻¹) ¹⁰ and our result for 9.4T (12.9 ± 1.5) magnetic field strengths suggest that the tissue cellular specific part of R2* in cortical GM of mouse brain is largely field independent. At the same time, significant increases in R2t* values can be expected in iron‐rich brain regions. ²⁵ This additional, iron‐related increases in R2t*, would also lead to increased R2*, consistent with previous reports (see recent publications ³ , ³³ , ³⁴ , ³⁵ and references therein). For example, Stüber et al. ³³ estimated the iron contribution to the R2* in the human cortex to be about 18 s⁻¹ at 7T. This value is bigger than the cortical R2t* value measured herein in the mouse brain (Figure 3). A perceived discrepancy between the measurement is due to a significant (order of magnitude) difference in iron content of mouse and human brains. While in the healthy human brain iron concentration is about 30–50 μg per gram of wet tissue in the cortex and even higher – about 200 μg per gram of wet tissue in pallidum, ³⁶ in the healthy mouse brain, it is only about 3.5–4.3 μg per gram of wet tissue across all brain structures. ³⁷ Hence, the expected contribution of iron to mice cortical R2t* at 9.4T is much smaller then in human brain and is about 2 s⁻¹.

Another important issue is the difference between R2* in GM and WM. Our results show R2* values of 29.08 ± 2.9 s⁻¹ for GM, and 34.85 ± 4.1/s for WM, which is consistent with the presence of myelinated fibers in WM. A stronger variability of R2* in WM (STD = 4.1/s) as compared with GM (STD = 2.9 s⁻¹) can also be related to a previously reported effect of R2* anisotropy, ²⁸ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷

It is noteworthy to compare our R2t* measurements with existing R2 data in mice. For example, Falangola et al. ⁴⁸ reported R2 in GM about 25 s⁻¹ which is bigger than our findings for R2t* (approximately 15 s⁻¹) but smaller than R2* (approximately 30 s⁻¹). This is an interesting result that requires further theoretical development which is beyond the scope of the current paper. Herein we only mention that an often‐used assumption that R2* = R2 + R2' is only valid in extreme static dephasing regime, that is, dephasing time, τ _deph, is much shorter than the diffusion time, τ _diff (see original estimates in ⁸ ). However, detailed Monte‐Carlo simulations of BOLD‐related magnetic susceptibility effects on GRE and SE signals (see ⁴⁹ and detail discussion in ²³ ) show that the BOLD model ⁸ quite accurately estimates vascular susceptibility contributions to GRE signal even when τ _deph˜τ _diff. In this case, the contribution of magnetic susceptibility effects to GRE signal is only about two to three times bigger than the contribution of magnetic susceptibility effects to the SE signal (see Figures 5–7 in ⁴⁹ ). This consideration, at least partially, can explain differences between measurements of R2*, R2t* and R2 in mouse brain. Indeed, a 5 s⁻¹ difference between R2* and R2t* is about half of R2' (12 s⁻¹). We should also note that the interpretation of transverse relaxation of GRE signal in biological tissue in terms of the assumption R2* = R2 + R2', instead of R2* = R2t* + R2', can have additional difficulties due to the inherent complexity of biological tissue structure where imaging voxels contain hundreds to thousands of cells of diverse morphology within a network of interconnected intra‐ and extracellular spaces filled with diffusing water. Consequently, MR signal relaxation properties are influenced by this network's structure and cellular boundary permeability. Due to the distinct water movement trajectories probed by GRE and SE experiments, the assumption R2t* = R2 in biological tissues should be approached with significant caution.

The fundamental finding of significant R2' contribution to R2* at high fields creates an opportunity to further improve usage of R2* relaxometry for understanding biological tissue cellular structure and pathology. Since, R2' is dependent on blood volume and oxygenation level, changes in the rates of cerebral O₂ consumption and cerebral blood volume accompanying changes in brain physiological (e.g., awake vs. sleep) or pathological (e.g., healthy vs. diseased) status (e.g., ⁵⁰ , ⁵¹ , ⁵² ) would affect R2' and thus may significantly alter R2* values, thus tempering relationships between R2* and tissue microstructure. Importantly, qGRE approach allows for removal of the R2' contribution from total R2*, thus providing a “purer” tissue specific metric, that is R2t*. Utilizing R2t* metric could improve the ability of qGRE to identify cellular microstructure and neuropathology due to its independence from the vascular BOLD contributions that can fluctuate with subjects' physiological conditions. At the same time, the R2' measurements may provide information on the brain tissue hemodynamics and vascular pathology.

5. CONCLUSIONS

In conclusion, herein we demonstrated that the qGRE approach allows successful generation of a battery of qGRE quantitative maps (R2* in WM and R2*, R2t*, R2' in GM) at UHF MRI. The method is based on the previously proposed theoretical biophysical model that accounts for signal relaxation caused by tissue cellular microstructure (R2t*), vascular BOLD effect (R2'), and the adverse effects of macroscopic B0 field inhomogeneities (background gradients). As demonstrated by the small variation in measurements between subject mice in most of the WM and GM areas of the brain, the segmentation protocol established in our study provides a consistent and effective method for separation of different brain structures in the mouse brain. Our data suggest that, at the higher fields, the BOLD contribution becomes increasingly critical to R2* (approximately 42% at 9.4T in the mouse brain, in comparison to 25% at 3T) and must be accounted for to accurately assess the tissue cellular‐microstructure‐specific R2t* contribution to the R2* signal. Our results are the first experimental validation of the qGRE approach at the UHF MRI. They also confirm the theoretical prediction of a large impact of vascular BOLD influence on the R2* signal at UHF MRI. Hence, the R2' BOLD contribution should be accounted for in the models relating R2* to tissue microstructure, especially at ultra‐high‐field MRI.

Supporting information

Figure S1. Image of Original Unconsolidated (Left) and Consolidated (Right) Dorr Atlas segmentation labels.

Table S1. Levels of label organization in Allen Brain Atlas and corresponding Dorr Brain segmentation labels for gray matter labels.

Table S2. Levels of label organization in Allen Brain Atlas and corresponding Dorr Brain segmentation labels for white matter labels.

Table S3. Total voxels, unaffected by field inhomogeneities (good) voxels, % of good voxels of analyzed segmentation labels.

APPENDIX A.

TABLE A1.

Summary of qGRE metrics (R2*, R2t*, R2') and corresponding label volumes in GM.

Region	R2* (1/s)		R2t* (1/s)		R2' (1/s)		Volume (mm3)
	Mean	STD	Mean	STD	Mean	STD	Mean	STD
Cerebral Ctx, cort. plate, isocortex	27.5	0.2	14.9	0.5	11.5	0.4	99.4	3.2
Cereb. Ctx, cort. plate, olfactory area	30.8	1.2	17.2	2.4	11.6	1.2	27.1	1.9
Cereb. Ctx, cort. plate, hippocampal formation	30.8	1.2	15.5	0.6	12.6	0.5	29.2	0.5
Cereb. Ctx, cortical subplate	29.0	0.3	15.4	1.2	12.4	0.5	9.7	0.6
Cerebral nuclei	29.5	0.8	16.5	0.3	11.9	0.7	37.3	2.2
Brainstem	32.5	0.2	18.4	0.5	12.9	0.7	32.9	1.3

Note: Data represent mean values and their STD across all mice.

TABLE A2.

Summary of R2* and corresponding label volumes in WM.

Region	R2* (1/s)		Volume (mm3)
	Mean	STD	Mean	STD
Olfactory nerve	29.9	1.1	1.96	0.17
Optic nerve	33.1	1.8	1.41	0.11
Oculomotor nerve	36.8	0.3	2.62	0.15
Cerebellar peduncles	40.5	0.8	2.95	0.06
Arbor vitae	43.2	1.1	7.94	0.36
Corpus callosum	34.3	1.5	11.35	0.48
Corticospinal tract	37.3	0.5	6.52	0.45
Rubrospinal tract	42.4	0.8	0.13	0.01
Cerebrum related	32.8	0.6	5.67	0.39
Mammillary related	30.3	1.2	0.22	0.02
Epithalamus related	31.4	1.0	0.90	0.03

Note: Data represent mean values and their STD across all mice.

Im J, Xiang B, Levasseur VA, et al. Unraveling the major role of vascular (R2') contributions to R2* signal relaxation at ultra‐high‐field MRI: A comprehensive analysis with quantitative gradient recalled echo in mouse brain. Magn Reson Med. 2025;94:761‐770. doi: 10.1002/mrm.30529