Brain oxygen extraction fraction mapping in patients with multiple sclerosis

Abstract

We aimed to demonstrate the feasibility of whole brain oxygen extraction fraction (OEF) mapping for measuring lesion specific and regional OEF abnormalities in multiple sclerosis (MS) patients. In 22 MS patients and 11 healthy controls (HC), OEF and neural tissue susceptibility ( ${\chi }_n$ ) maps were computed from MRI multi-echo gradient echo data. In MS patients, 80 chronic active lesions with hyperintense rim on quantitative susceptibility mapping were identified, and the mean OEF and ${\chi }_n$ within the rim and core were compared using linear mixed-effect model analysis. The rim showed higher OEF and ${\chi }_n$ than the core: relative to their adjacent normal appearing white matter, OEF contrast = −6.6 ± 7.0% vs. −9.8 ± 7.8% (p < 0.001) and ${\chi }_n$ contrast = 33.9 ± 20.3 ppb vs. 25.7 ± 20.5 ppb (p = 0.017). Between MS and HC, OEF and ${\chi }_n$ were compared using a linear regression model in subject-based regions of interest. In the whole brain, compared to HC, MS had lower OEF, 30.4 ± 3.3% vs. 21.4 ± 4.4% (p < 0.001), and higher ${\chi }_n$ , −23.7 ± 7.0 ppb vs. −11.3 ± 7.7 ppb (p = 0.018). Our feasibility study suggests that OEF may serve as a useful quantitative marker of tissue oxygen utilization in MS.

Introduction

Multiple sclerosis (MS) is an inflammatory demyelinating disease of the white matter (WM) with progressive neurodegeneration and is a leading cause of neurological disability in young adults. ¹ MRI has been the standard in vivo imaging technique for diagnosing MS and monitoring disease progression.^2,3 New acute MS lesions develop near the disrupted brain blood barrier (BBB) and appear hyperintense on Gadolinium-enhanced T1-weighted image (T1w+Gd)^3,4 as a consequence of contrast agent leakage. Following BBB repair, lesions are characterized by a non-enhancing status on T1w+Gd; these lesions are typically regarded as chronic and appear hyperintense on T2-weighted fluid-attenuated inversion recovery (T2FLAIR) image. This is attributable to demyelination and axonal loss.^2,3

Recently, an important subset of chronic MS lesions has been identified which exhibit low-grade sustained microglial activity and expanding demyelination at the lesion rim.^5–7 These so-called chronic active lesions have been associated with increased tissue damage, neuroinflammation, and disability, ⁸ and may be a driving factor in progressive disease.^9,10 Quantitative susceptibility mapping (QSM) ¹¹ has been used to detect these lesions based on their characteristic paramagnetic rim appearance (QSM rim+) owing to its sensitivity to the iron-rich pro-inflammatory microglia and macrophages in the rim region.^9,12–18 However, the metabolic activity of these immune cells has yet to be measured with current imaging techniques such as PET (due to limited spatial resolution) and conventional MRI (due to lack of sensitivity).

The primary objective of this study was to demonstrate the feasibility of whole brain mapping of oxygen extraction fraction (OEF) in MS patients using a widely available multi-echo gradient echo (mGRE) sequence and a novel integrated model of QSM phase signal and quantitative blood oxygenation level dependent magnitude signal (qBOLD) (QSM+qBOLD, or QQ).^19,20 QQ separates the effect of deoxyheme iron in a cylindrical micron-scale venule (OEF effect) from that of neural tissue susceptibility ( ${\chi }_n$ ) contributed for example by the diffuse nano-scale ferritin iron in tissue, based upon their unique contributions to tissue susceptibility determined on QSM^21–23 and to the magnitude signal decay as modeled by qBOLD.^24–27 QQ has been shown to provide OEF measurements that are in good agreement with dual-gas calibrated BOLD ²⁸ and the gold standard ¹⁵O PET-OEF. ²⁹

In our study, we aim to demonstrate the potential of QQ to provide a quantitative marker of oxygen metabolism in the rim area of QSM rim+ lesions, which may be useful in the development of new therapeutic targets to reduce inflammation in chronic active lesions. As a secondary objective, QQ-based OEF mapping is used to investigate regional oxygen metabolism differences in cortical gray matter (CGM) and deep gray matter (DGM) between MS patients and healthy controls (HCs), which have been implicated in neurodegenerative processes and are regarded as major indicators of tissue injury and neuronal loss in MS.^30–36

Materials and methods

Study cohort

This retrospective image analysis study was approved by the Weill Cornell Medicine local Institutional Review Board, and written informed consent was obtained from all individuals in accordance with the ethical standards of the Helsinki Declaration of 1975 and its later amendments. MS patients were selected from an ongoing imaging and clinical MS research database and were diagnosed according to the McDonald criteria. ³ A total of 22 relapsing-remitting MS patients with at least one chronic active QSM rim+ lesion was included (17 females, 5 males; mean age, 37 ± 6 years). In addition, 11 age-matched HCs (1 female, 10 males; mean age, 34 ± 12 years) were included for comparison. The demographic and clinical characteristics of the study participants, e.g. expanded disability status scale (EDSS) to evaluate the functional systems of the central nervous system (0 = normal neurologic status, 10 = death due to MS), ³⁷ are summarized in Table 1.

Table 1.

Clinical and demographic characteristics of the study cohort. Age, disease duration, and expanded disability status scale (EDSS) score are shown as mean ± standard deviation (minimum-maximum).

Variable	Healthy controls	MS patients	P-value
Number	11	22
Number of womena	1 (9%)	17 (77%)	<0.001a
Age (years)b	34 ± 12 (22–63)	37 ± 6 (28–52)	0.19b
Disease duration (years)	N.A.	16 ± 12 (1–40)
EDSS score	N.A.	1.4 ± 1.1 (0–3.5)
MS phenotype	N.A.	Relapsing-remitting
Medical treatmentc	N.A.	21 (95%)

^aχ² test.

^bWilcoxon rank sum test.

^cMedical treatment includes Gilenya (N = 3), Ocrevus (N = 9), Tecfidera (N = 3), Tysabri (N = 4), and Copaxone (N = 2).

MRI protocol

All brain MRIs were performed on a 3 T GE scanner. The typical MS imaging protocol consisted of 3 D sagittal T1w and T1w+Gd (field of view (FOV) = 24 cm, TR = 8.8 ms, TE = 3.4 ms, TI = 450 ms, flip angle (FA) = 15°, voxel size = 1.2 × 1.2 × 1.2 mm³), 2 D axial T2w (FOV = 28 cm, TR = 5917 ms, TE = 88 ms, flip FA = 90°, voxel size = 0.7 × 1.2 × 3 mm³), and 3 D T2FLAIR (FOV = 24 cm, TR = 6050 ms, TE = 149.4 ms, TI = 1820 ms, FA = 90°, voxel size = 1.2 × 1.2 × 1.2 mm³) for anatomy and lesion detection, as well as 3 D mGRE for QSM and OEF mapping (FOV = 24 cm, TR = 57.5 ms, TE1/ΔTE = 4.5/4.8 ms, number of TEs = 11, FA = 20°, voxel size = 0.7 × 0.9 × 3 mm³, axial slice acquisition parallel to B0 direction).

The imaging protocol for HCs consisted of axial 3 D mGRE (FOV = 20 cm, TR = 30.5 ms, TE1/ΔTE = 2.3/3.9 ms, number of TEs = 7, FA = 15°, voxel size = 1.2 × 1.2 × 1.2 mm³, flow-compensated in all three directions, ³⁸ axial slice acquisition parallel to B0 direction) and 3 D T1w (FOV = 20 cm, TR = 7.7 ms, TE = 2.9 ms, TI = 450 ms, FA = 15°, voxel size = 0.8 × 0.8 × 1.2 mm³).

Image processing and analysis

QSM reconstruction consisted of total field estimation from the mGRE phase data by linear fitting, ³⁹ background field removal using Projection onto Dipole Fields (PDF) algorithm, ⁴⁰ and inversion of the local field to obtain the susceptibility distribution using Morphology Enabled Dipole Inversion with automatic uniform cerebrospinal fluid zero reference (MEDI + 0) algorithm.^11,41–43 In HCs, the total field was estimated using an adaptive quadratic-fit of the mGRE phase to benefit from 3 D flow-compensation. ³⁸

OEF and neural tissue susceptibility ( ${\chi }_n$ ) maps were computed from QSM and mGRE magnitude data using the QQ algorithm.^19,20 QQ unifies two biophysical models of mGRE data: ¹⁹ 1) a QSM-based model which distinguishes the susceptibility contribution of blood (determined by its oxygenation level) from that of neural tissue ^22,23,44 on a per voxel basis, and 2) qBOLD which models the mGRE magnitude signal decay by the field variation inside a voxel based on the susceptibility difference between blood and the surrounding tissue.^{24–27,45–47}

To improve the robustness of QQ against noise in the data fitting, the original cluster analysis of time evolution (CAT) algorithm ²⁰ was refined by incorporating a brain/CSF segmentation into clustering.^48,49 This segmentation was obtained from the echo-combined magnitude mGRE image using the FSL FAST segmentation algorithm. ⁵⁰ The CAT method groups voxels of the same tissue type with similar temporal mGRE magnitude evolutions into a cluster that are assumed to have similar model parameter values including OEF. Then, cluster-wise optimization is performed to solve for the model parameters. Total variation denoising ⁵¹ was applied to enhance the effective signal-to-noise ratio of OEF reconstruction.

For MS lesion-specific OEF analysis, two neuroradiologists (S.Z., 6 years of experience and W.H., 11 years of experience) identified lesions on T2FLAIR and manually traced QSM rim+ lesions on QSM based on their relative contrast with respect to the adjacent normal appearing white matter (NAWM).^12,13 To exclude both regional variation within a subject and inter-subject variation, the average OEF within each lesion was referenced to the average of its adjacent NAWM, OEF contrast (OEFC) =  $\overline{OE{F}_{\hspace{0.17em}}}-\overline{OE{F}_{\mathit{NAWM}}}$ . Lesion ${\chi }_n$ contrast was defined in the same way ( ${\chi }_{n}C=\overline{{{\chi }_n}_{ }}-\mathrm{ }\overline{{\chi }_{n,\mathit{NAWM}}}$ ). The average and standard deviation of $\overline{{\chi }_{n,\mathit{NAWM}}}$ among all the lesions was −21.9 ± 13.8 ppb. For each QSM rim+ lesion, the central veins were removed, and the hyperintense rim area in QSM was traced as lesion rim by three neuroradiologists (S.Z., 6 years of experience, W.H., 11 years of experience, and I.K., 23 years of experience), with the remaining part of the lesion defined as the lesion core. No significant blooming artifact were found near the lesions in the susceptibility maps (Figures 1 and 3). The blooming artifact seen on the gradient echo magnitude images is caused by the field inhomogeneity in the vicinity of a strong susceptibility source such as iron or air-tissue interface. The field can be described mathematically as the spatial convolution of the susceptibility source with a dipole kernel. QSM deconvolves the field to remove the blooming artifact and reveal the true susceptibility. ⁵² Acute T1w+Gd enhancing lesions were excluded from analysis. To compare OEF between HC and MS patients, FreeSurfer analysis of T1w anatomical images was used to obtain regions of interest (ROIs) ⁵³ including cortical gray matter (CGM) and selective DGM including thalamus, caudate, putamen, and pallidum, which have been shown to be associated with neuronal loss and atrophy in MS.^{30–33,35,54} MS lesions were excluded from these ROIs.

Figure 1.

Example of T1w+Gd image, T2w image, OEF and neural tissue susceptibility ( ${\chi }_n$ ) maps obtained from an MS patient (disease duration, 11 years) showing two chronic active lesions with paramagnetic rims (A and B). In both lesions, the rim regions (red) show higher OEF and ${\chi }_n$ than the core regions (green). The central vein within lesion A is excluded from the lesion ROI. Normal appearing white matter (NAWM) mask for each lesion is shown in cyan blue in lesion ROI map.

Figure 3.

Exemplary OEF and neural tissue susceptibility ( ${\chi }_n$ ) maps obtained in HC and an MS patient (disease duration, 2 years). Pink arrow indicates a QSM rim+ lesion with a hyperintense rim consistent with iron-rich activated microglia. Brain OEF map for both subjects appears uniform except in lesion areas.

Statistical analysis

For QSM rim+ lesions, a linear mixed-effect model with a random effect for patient was used to assess the OEFC between the rim and core regions (core = 0, rim =1), adjusting for lesion volume. The random effect for patient accounts for the correlation within a patient. For OEF comparison between HC and MS, linear regression models were fit for each ROI (whole brain, CGM, and DGM), with average OEF in the region as the dependent variable and an indicator of MS versus HC (0 = HC, 1 = MS), gender (female = 0, male = 1), and age as independent variables. Multiple comparisons correction was performed using the false discovery rate. ⁵⁵ The Jarque-Bera test confirmed the normality of each ROI data distribution (all p-values > 0.09). ⁵⁶ A p-value of less than 0.05 was considered statistically significant. The same analyses were performed for ${\chi }_n$ .

Results

In MS patients, a total of 80 chronic QSM rim+ lesions were identified. Of these, the mean lesion volume was 440.0 ± 431.0 mm³, and the mean volume of rim and core regions was 267.0 ± 267.4 mm³ and 172.9 ± 185.4 mm³, respectively.

Figure 1 shows a representative case with two QSM rim+ lesions, showing higher OEF and ${\chi }_n$ in the rim region compared to the core region. Over 80 QSM rim+ lesions, the mean OEF and ${\chi }_n$ in the rim region was found to be significantly higher than that in the core region in the linear mixed-effect model analyses (Figure 2): OEFC = −6.6 ± 7.0% vs. −9.8 ± 7.8% ( $\beta$  = 0.037, 95% CI: [0.023, 0.051], p < 0.001) and ${\chi }_n$ C = 33.9 ± 20.3 ppb vs. 25.7 ± 20.5 ppb ( $\beta$  = 0.007, 95% CI: [0.001, 0.013], p = 0.017). Between MS and HC, the clusters appear similar, consisting mainly of GM, NAWM, and WM lesions (in MS subjects) (Supporting Information Figure S1). MS lesions with rim appearance tend to have distinctive clusters corresponding to the rim and core regions.

Figure 2.

Comparison of mean OEF and neural tissue susceptibility ( ${\chi }_n$ ) contrast relative to adjacent normal appearing white matter (OEFC and ${\chi }_n$ C) between the rim and core regions of QSM rim+ lesions (n = 80). The rim region showed significantly higher mean OEFC and ${\chi }_n$ C than the core region in linear mixed-effect model analysis. Red line, blue box, black whisker, and red cross indicates median value, interquartile range, the range extending to 1.5 of the interquartile range, and outlier beyond the whisker range. Asterisk (*) indicates the significant difference (p < 0.05, linear mixed-effect model).

The lesion volume remained a significant covariate for OEFC ( $\beta$  = −0.011, 95% CI: [−0.019, −0.003], p = 0.009), but not for ${\chi }_{n}C$ (p = 0.172). The ${\chi }_n$ map was very similar to the QSM map (not shown) with an approximately 5 ppb difference (Figures 1 and 3).

Both HCs and MS patients generally show uniform OEF maps except for lesions and GM/WM contrast in ${\chi }_n$ maps (Figure 3). Compared to HCs, MS patients had significantly lower OEF values (Figure 4) as revealed by the linear regression model analyses: 30.4% ± 3.3% vs. 21.4 ± 4.4% ( $\beta$  = −0.099, 95% CI: [−0.143, −0.055], p < 0.001) in whole brain, 29.9 ± 2.9% vs. 19.9 ± 4.3% ( $\beta$  = −0.112, 95% CI: [−0.153, −0.071], p < 0.001) in CGM, 31.3 ± 3.5% vs. 23.3 ± 4.3% ( $\beta$  = −0.086, 95% CI: [−0.130, −0.042], p = 0.001) in thalamus, 31.7 ± 3.0% vs. 23.3 ± 3.6% ( $\beta$  = −0.088, 95% CI: [−0.125, −0.051], p < 0.001) in caudate, 31.5 ± 3.2% vs. 24.6 ± 3.9% ( $\beta$  = −0.072, 95% CI: [−0.111, −0.033], p = 0.002) in putamen, 31.0 ± 3.5% vs. 25.2 ± 3.4% ( $\beta$  = −0.052, 95% CI: [−0.089, −0.015], p = 0.012) in pallidum. Gender and age were not significant covariates: p-values ≥ 0.751 for gender and ≥ 0.738 for age.

Figure 4.

Comparison of average OEF and neural tissue susceptibility ( ${\chi }_n$ ) in whole brain, cortical gray matter (CGM), and select deep gray matter (DGM, thalamus, caudate, putamen, pallidum) between healthy controls (HC, 34 ± 12 years, n = 11) and MS patients (37 ± 6 years, n = 22). Asterisk (*) indicates significant difference (p < 0.05, linear regression model). Red line, blue box, black whisker, and red cross indicates median value, interquartile range, the range extending to 1.5 of the interquartile range, and outlier beyond the whisker range, respectively.

MS patients showed significantly higher ${\chi }_n$ values in the whole brain, −23.7 ± 7.0 ppb vs. −11.3 ± 7.7 ppb ( $\beta$  = 0.010, 95% CI: [0.002, 0.018], p = 0.018) and CGM, −23.3 ± 7.4 ppb vs. −7.1 ± 8.0 ppb $(\beta$  = 0.015, 95% CI: [0.007, 0.023], p = 0.003) than HCs (Figure 4), whereas ${\chi }_n$ values were comparable in DGM: −12.6 ± 10.4 ppb vs. −12.9 ± 10.6 ppb (p = 0.599) in thalamus, 38.7 ± 10.2 ppb vs. 42.2 ± 14.0 ppb (p = 0.613) in caudate, 21.5 ± 13.5 ppb vs. 15.8 ± 16.6 ppb (p = 0.055) in putamen, and 86.9 ± 19.9 ppb vs. 84.8 ± 22.8 ppb (p = 0.201) in pallidum. Gender and age were not significant covariates: p-values ≥0.583 for gender and ≥0.494 for age.

Discussion

Our feasibility study demonstrated that high-resolution OEF maps can be extracted from mGRE data using the QQ algorithm for the study of global and regional tissue oxygen utilization in MS brains. We showed for the first time an OEF increase in the rim region compared to the core region in chronic active MS lesions with paramagnetic rim appearance. As QSM has become a reliable imaging tool for measuring susceptibility changes in the brain, the ability to further separate the contributions of blood deoxyheme iron and neural tissue to tissue susceptibility by the QQ algorithm has the potential to provide a more comprehensive picture of both tissue oxygenation status and change in tissue iron and/or myelin content in chronic active MS lesion.^9,12–17

In QSM rim+ lesions, the rim region was found to have significantly higher mean OEF and ${\chi }_n$ values than the core region (Figures 1 and 2). This difference is consistent with the histological evidence of higher density of iron-rich microglial cells in the lesion rim and PET imaging findings showing increased overall inflammatory activity in chronic active rim lesions.^8,12,16,57 The lower OEF observed in the core area of these lesions is also consistent with the irreversible tissue damage seen on histopathology.^8,10,57 These promising results suggest that OEF derived from mGRE data may serve as a sensitive marker of tissue oxygen metabolism in QSM rim+ lesions, which can be important for lesion status monitoring and development of novel MS therapeutic targets.

In both HCs and MS patients, brain OEF maps generally appear uniform except in lesions (Figure 3). This observation is consistent with OEF maps that were obtained by the reference standard ¹⁵O PET,^58,59 which supports the presence of an equilibrium between metabolic needs and blood flow in the resting state. ⁵⁹ The observed lower global OEF obtained by the QQ algorithm in MS brains as compared to HCs (Figures 3 and 4) agrees well with the previously reported reduced OEF in MRI,^34,35,60 near-infrared spectroscopy, ⁶⁰ and PET studies. ⁶¹ Lower OEF may represent a unique tissue injury mechanism underlying progressive demyelination and neurodegeneration in the MS brain. Several studies have indicated that impaired mitochondrial function via nitric oxide overproduction likely induces chronic oxygen deprivation in neuronal cells, which may result in neuronal dysfunction.^62–65

Compared to a previous MRI-based OEF study using vein susceptibility modeling, ³⁵ cortical OEF values in our study were similar for HCs, 29.9 ± 2.9 vs. 31.5 ± 3.0%, but lower for MS patients, 19.9 ± 4.3 vs. 28.1 ± 3.0%, which may be explained by the longer disease duration of our cohort (16 vs. 8.5 years). ³⁵ Significantly lower OEF in frontal, temporal, parietal, and occipital CGM in MS, compared to HC (Supporting Information Figure S2), agrees with lower OEF observed in regional cortices, e.g. sensorimotor, parietal, and prefrontal cortex, with using the vein susceptibility modeling. ³⁵ Furthermore, our global OEF was lower for both HCs (30.4 ± 3.3 vs. 38.6 ± 4.1%) and patients (21.1 ± 4.4 vs. 32.8 ± 5.2%) compared to a study using venous T2 modeling. ³⁴ This discrepancy may be explained by the complexity in estimating oxygenation from T2 modeling. For instance, performance of T2 oxygenation modeling is strongly dependent on accurate calibration of its model parameters. ⁶⁶

In DGM, MS patients have lower OEF values than HCs (Figure 4), which may be associated with tissue loss in those regions in MS brains.^30,32,67 For instance, the lower OEF in the thalamus, a central link for cortical-subcortical circuits, ⁶⁸ may be related to thalamic neuronal loss ³³ and atrophy, ⁶⁹ which are predictors of long-term disability progression in MS. ⁷⁰

QSM sensitivity to tissue iron^71–73 has been applied to study MS brain tissue.^57,74 The susceptibility values of several deep gray nuclei (e.g. caudate and globus pallidus) in MS brains have been reported to be higher compared to HC.^67,74 This study furthers our understanding of QSM as it separates QSM into the contribution of oxygen metabolism (OEF effect) and neural tissue ( ${\chi }_n$ effect). While OEF is reduced in all deep gray nuclei of MS brains, ${\chi }_n$ is not significantly changed compared to HC, unlike previous literature.^67,74 This may be due to the longer disease duration of subjects in this study, as iron content of deep gray nuclei in MS brains decreases as disease duration increases, while iron content in HC brains increases or remains stable.^73,75 Additional studies may be needed as there is no agreement in the literature on QSM values in the thalamus and putamen in MS as compared to HC.^67,74

Both HC and MS showed generally higher ${\chi }_n$ in CGM compared to WM (Figure 3). This contrast is in line with the lower myelin and higher iron content in CGM. ⁷⁶ However, compared to HC, ${\chi }_n$ was significantly higher in the whole brain and CGM of the MS brains, whereas OEF was lower (Figure 4). The biological cause of higher ${\chi }_n$ in CGM remains to be elucidated, though it may be related to iron accumulation and/or demyelination.^57,77

This study has several limitations. First, the patient cohort is small and limited to relapsing-remitting MS with low EDSS. For instance, OEF was not significantly correlated with disease duration in all the ROIs (uncorrected p-values > 0.538, Spearman correlation analysis) (Supporting Information Figure S3) unlike a previously reported negative correlation. ³⁵ This may be caused by a small number of patient cohort. Also, no significant correlation was found between OEF and EDSS in this study (Supporting Information Figure S3), whereas cortical OEF negatively correlated with EDSS in the previous venous OEF study. ³⁵ This may be caused by a limited and low range of EDSS in this study (0-4).

Second, HC subjects had a different gender ratio from MS subjects which may affect the comparison result between HC and MS. To address the lack of female subjects in the HC group, we performed additional analysis to compare OEF only in the male subjects between HC (n = 10) and MS (n = 5) groups adjusting for age. We found that the MS group still showed significantly lower whole brain OEF than the HC group: 30.9% ± 3.2% vs. 18.7 ± 1.3% ( $\mathrm{\beta }$  = −0.119, 95% CI: [−0.153, −0.085], p < 0.001). Furthermore, while females generally have lower hemoglobin concentration [Hb] and higher CBF than males, ⁷⁸ prior literature in healthy adults shows that OEF in women is slightly higher ⁷⁹ or similar ⁸⁰ compared to OEF in men. These data suggest that the lower OEF observed in the MS group with more female subjects compared to the HC group is not driven by gender. Unfortunately, we could not obtain [Hb] or CBF measurements in our study.

Third, QQ has some modeling assumptions: the random cylinder orientation of the venous microvasculature in qBOLD of QQ may not be valid in MS lesions and therefore induce bias in OEF estimations. Although intracranial and extracranial vascular abnormalities have not been found in MS compared to HC,^81–83 microvascular abnormalities should be further investigated and potentially incorporated into QQ. Also, QQ does not consider detailed microstructure of the brain tissue including myelin, which may lead to a bias in OEF estimation. This may be a confounding factor in the comparison between MS and HC due to the demyelination and axonal loss in MS. QQ assumes two source compartments in a voxel, deoxyhemoglobin in venous structure and medium, such as diffusively distributed non-blood tissue ferratin or protein. If other structured strong sources including myelin in WM exists, additional introvoxel field variation would be induced, which may lead to greater MRI signal decay than the sole deoxyhemoglobin effect of venous blood. Consequently, QQ-based OEF may be overestimated in WM. However, in a validation study of QQ against the reference standard ¹⁵O-PET, the OEF difference between QQ and PET was not significant in WM. ²⁹ This may suggest that the myelin effect in QQ-based OEF estimation may not be significant.

Fourth, vein voxels (blood only) were treated the same way as brain voxels (mixture of neural tissue and blood), which could influence the accuracy of OEF and ${\chi }_n$ estimations in large veins. This could be mitigated by segmenting large veins; additional studies are needed to perform accurate vein segmentation.

Fifth, different scan protocols between HC and MS patients, e.g. longer TE range, no flow compensation, and lower slice resolution in MS, may affect the OEF comparison between HC and MS patients. Additional OEF processing was performed to investigate the effects of scan parameters on OEF comparison. When MS data were processed using a similar TE range (TE₁/ΔTE/TE₆ = 4.7/4.8/28.5 ms) compared to that of HC (TE₁/ΔTE/TE₇ = 2.3/3.9/25.7 ms), we found lower OEF values in MS than HC, e.g. 23.0 ± 3.2% vs. 30.4 ± 3.5% for whole brain average OEF, in good agreement with the original findings (Supporting Information Figure S4). Lack of flow compensation may cause large veins with high blood velocity to shift due to mis-encoding. ⁸⁴ In this study, this problem was alleviated by manually excluding the central veins from the lesion ROI (identified and segmented by neuroradiologists experienced with brain QSM, S.Z., W.H, and I.K) from the MS lesion ROI. For capillary level vessels, flow-related issue is not expected as they exhibit very slow blood flow of less than 1 mm/sec. Spatial resolution, which scales linearly with SNR, could have an impact on OEF quantification. To investigate this issue, a numerical simulation was performed to compare OEF obtained at SNR =1000, 100, and 50 using CAT algorithm used in this study, and negligible difference in mean OEF value was found. ²⁰ The CAT may alleviate the OEF dependency on SNR by substantially improving the effective SNR through signal averaging in clusters. ²⁰

Sixth, the medication effect has not been considered in the statistical analysis due to the limited sample size and variation of treatment drugs.

Seventh, though the comparison of venous oxygenation in lesion draining veins and its corresponding normal veins would be valuable to study the source of decreased OEF in the lesion core, it is difficult to perform because it is hard to specify the exact draining veins and venous oxygenation estimation may be biased due to partial volume effect. For instance, in an additional comparison, a neuroradiologist (I.K., 23 years of experience) identified central veins in 17 out of 80 lesions with thickness of 1∼2 voxels and its corresponding normal veins on contralateral side. Then, venous oxygenation in the central veins was calculated using mean QSM values ⁸⁵ and compared with that in the normal veins. The venous oxygenation was similar and not significantly different between the lesion central veins and its corresponding normal veins: 88.9 ± 1.5% vs. 89.1 ± 1.1% ( $\mathrm{\beta }$  = −0.002, 95% CI: [−0.012, 0.009], p = 0.78, linear mixed-effect model analysis). This may result from that the central veins in MS lesions are not generally the draining veins of the MS lesions. ⁸⁶ Even if the draining veins are correctly identified, its oxygenation is likely to suffer from partial volume effect due to the limited resolution, e.g. 3 mm slice thickness used in this study. A higher venous oxygenation in the normal veins than previously reported venous oxygenation in large veins, 89.1% vs. 63.7%, ⁸⁵ suggests that the central vein oxygenation suffers from the partial volume effect.

Finally, this study used a cross-sectional design to investigate chronic active rim lesions. We plan to perform longitudinal studies in our future work.

In conclusion, brain OEF mapping is feasible in MS brains and can detect within-lesion difference between the rim and core regions of MS lesions with QSM hyperintense rim as well as global and regional gray matter differences between MS and HC. OEF maps may provide useful information on tissue oxygenation and viability for MS monitoring and treatment.