Prospects for investigating brain oxygenation in acute stroke: Experience with a non‐contrast quantitative BOLD based approach

Abstract Metabolic markers of baseline brain oxygenation and tissue perfusion have an important role to play in the early identification of ischaemic tissue in acute stroke. Although well established MRI techniques exist for mapping brain perfusion, quantitative imaging of brain oxygenation is poorly served. Streamlined‐qBOLD (sqBOLD) is a recently developed technique for mapping oxygenation that is well suited to the challenge of investigating acute stroke. In this study a noninvasive serial imaging protocol was implemented, incorporating sqBOLD and arterial spin labelling to map blood oxygenation and perfusion, respectively. The utility of these parameters was investigated using imaging based definitions of tissue outcome (ischaemic core, infarct growth and contralateral tissue). Voxel wise analysis revealed significant differences between all tissue outcomes using pairwise comparisons for the transverse reversible relaxation rate (R 2 ′), deoxygenated blood volume (DBV) and deoxyghaemoglobin concentration ([dHb]; p < 0.01 in all cases). At the patient level (n = 9), a significant difference was observed for [dHb] between ischaemic core and contralateral tissue. Furthermore, serial analysis at the patient level (n = 6) revealed significant changes in R 2 ′ between the presentation and 1 week scans for both ischaemic core (p < 0.01) and infarct growth (p < 0.05). In conclusion, this study presents evidence supporting the potential of sqBOLD for imaging oxygenation in stroke.

oxygenation that is well suited to the challenge of investigating acute stroke. In this study a noninvasive serial imaging protocol was implemented, incorporating sqBOLD and arterial spin labelling to map blood oxygenation and perfusion, respectively. The utility of these parameters was investigated using imaging based definitions of tissue outcome (ischaemic core, infarct growth and contralateral tissue). Voxel wise analysis revealed significant differences between all tissue outcomes using pairwise comparisons for the transverse reversible relaxation rate (R 2 0 ), deoxygenated blood volume (DBV) and deoxyghaemoglobin concentration ([dHb]; p < 0.01 in all cases). At the patient level (n = 9), a significant difference was observed for [dHb] between ischaemic core and contralateral tissue. Furthermore, serial analysis at the patient level (n = 6) revealed significant changes in R 2 0 between the presentation and 1 week scans for both ischaemic core (p < 0.01) and infarct growth (p < 0.05). In conclusion, this study presents evidence supporting the potential of sqBOLD for imaging oxygenation in stroke.

K E Y W O R D S
Ischaemia, oxygen metabolism, stroke

| INTRODUCTION
Ischaemic stroke is characterised by restricted blood supply to regions of tissue that may ultimately result in infarction. However, the brain can tolerate a limited reduction in perfusion if tissue oxygenation levels can be preserved. Therefore, techniques to map brain oxygenation in the acute phase of stroke may help to identify viable tissue that requires intervention to minimise the final infarct volume (Astrup, Siesjö, & Symon, 1981).

While Positron Emission Tomography (PET) is the current benchmark for
imaging oxygen metabolism in acute stroke (Ackerman et al., 1981;Baron, 1999), it is not widely available in acute clinical settings.
Magnetic resonance imaging (MRI) is already widely used in the assessment of patients with acute stroke and has the potential to be a viable alternative to PET in measuring oxygen metabolism. Such measurements are made possible by the inherent sensitivity of the transverse MR relaxation rate (R 2 *) to deoxyhaemoglobin. R 2 * (= R 2 + R 2 0 ) is composed of the irreversible (R 2 ) and reversible (R 2 0 ) transverse relaxation rates with respect to a spin echo. As changes in R 2 (and hence R 2 *) are known to be affected by numerous factors aside from tissue oxygenation An et al., 2014), R 2 0 is predicted to have better specificity to baseline brain oxygenation (Yablonskiy & Haacke, 1994). This sensitivity has previously been exploited to demonstrate that alterations in R 2 0 are indicative of the final outcome of ischaemic tissue (Bauer et al., 2014;Geisler et al., 2006;Seiler et al., 2012;Siemonsen et al., 2008;Zhang, Zhang, & Chen, 2011). However, R 2 0 is dependent on both deoxyhaemoglobin concentration ([dHb]) and the DBV (Yablonskiy, 1998), resulting in ambiguity regarding the physiological origin of a measured R 2 0 alteration. The ability to separate [dHb] from R 2 0 would provide a quantitative physiological metric directly related to tissue oxygenation, and to achieve this, knowledge of the underlying DBV is required.
Dynamic susceptibility contrast (DSC) MRI has been used to improve the physiological interpretation of R 2 0 by providing cerebral blood volume (CBV) and cerebral blood flow (CBF) information allowing [dHb] to be estimated (Christen, Schmiedeskamp, Straka, Bammer, & Zaharchuk, 2012). These methods have previously been employed to investigate acute stroke (Gersing et al., 2015;Seiler et al., 2017a;Seiler et al., 2017b). However, DSC requires an exogenous contrast agent, which may limit the frequency with which scanning can be repeated and be contraindicated in some patients (Gulani, Calamante, Shellock, Kanal, & Reeder, 2017). In addition, DSC provides an estimate of the blood volume across all vascular compartments, while qBOLD requires a measurement of the volume occupied specifically by deoxygenated blood.
There are noncontrast based alternatives that allow the estimation of DBV and CBF. The quantitative-BOLD (qBOLD) method is able to measure DBV by modelling the transverse MR signal decay in the presence of a vascular network (An & Lin, 2000;He & Yablonskiy, 2007). Arterial spin labelling (ASL) can be used to quantify CBF by inverting the magnetisation of arterial blood to act as a diffusible tracer (Alsop et al., 2015). These endogenous methods are particularly suitable for application in acute stroke as they can be acquired noninvasively in a clinically relevant manner Lee et al., 2003).
Recently we proposed a refinement of the qBOLD method targeted at minimising confounding effects during data acquisition, rather than by postprocessing, which we term streamlined-qBOLD (sqBOLD; Stone & Blockley, 2017). By minimising the influence of nuisance signals (macroscopic field inhomogeneities, MFIs; cerebral spinal fluid, CSF; and R 2 -weighting) during image acquisition the application of the qBOLD model is simplified, improving the robustness of the resultant oxygenation maps. While measurements in healthy young subjects look promising, so far measurements have not been performed in more challenging clinical research applications. The aim of this study is to investigate the potential of sqBOLD to monitor regional changes in brain oxygenation following ischaemic stroke, and its relation to tissue outcome over time in a cohort of patients under-  A T 2 -weighted FLAIR turbo spin echo was used to identify infarction on follow-up scans  (1.9 × 1.9 × 2.0 mm, field of view = 240 × 217.5 mm 2 , TR/TE = 9,000/96 ms, TI = 2,500 ms, 58 slices, scan duration 2 min 8 s).  Okell et al., 2013;Wang et al., 2013). This is in contrast to DSC perfusion weighted imaging (PWI) where prolonged time to peak (TTP) or mean transit time (MTT) are commonly used to define hypoperfused tissue, rather than CBF.
A FLAIR-GASE acquisition was used to measure baseline brain oxygenation using the sqBOLD approach (Stone & Blockley, 2017) (Gruetter & Boesch, 1992). This is achieved by oversampling in the slice direction using a 3D acquisition. It has recently been shown that by combining GESEPI with the ASE technique (GASE) the effect of MFGs can be minimised in the majority of the brain (Blockley & Stone, 2016). However, GASE can only compensate MFGs below a critical threshold. Hence positioning of the imaging volume was focussed on the ischaemic region and slices were angeled away from areas of severe MFI. The combined FLAIR-GESEPI-ASE (FLAIR-GASE) acquisition reduces confounding effects and when combined with quantitative modelling offers a streamlined qBOLD approach.
For sqBOLD, preprocessing and parameter map calculation from the FLAIR-GASE data is based on previously described methods (Stone & Blockley, 2017). In brief, the four 1.25 mm slices of each slab were averaged to produce a single 5 mm slice. The τ-series was motion corrected using the FSL linear motion correction tool (MCFLIRT; Jenkinson, Bannister, Brady, & Smith, 2002) to the spinecho image. As an indicator of the severity of head-motion that occurred during FLAIR-GASE acquisition, the mean relative root mean square value was output from MCFLIRT, and henceforth described as the mean motion score. The spin-echo image was brain extracted using the FSL brain extraction tool (BET; Smith, 2002) to create a binary mask of brain tissue and all remaining τ-weighted volumes were brain extracted using this mask. The data were spatially smoothed using a Gaussian kernel with a full-width half-maximum that matched the inplane resolution (2.3 mm). This smoothing was chosen to reduce the impact of noisy voxels on the model fit without unduly reducing the spatial resolution of the resulting parameter maps.
Oxygenation parameters R 2 0 , DBV and [dHb] were directly inferred from the sqBOLD data using the qBOLD model (Yablonskiy, 1998). To describe this model in simple terms, the mono-exponential part of the signal decay (τ > 15 ms) is used to measure R 2 0 and the mismatch between a measured spin echo (τ = 0 ms) and the linear intercept of the mono-exponential regime provides an estimate of the DBV. The ratio of R 2 0 and DBV is proportional to [dHb] (Equation 2).
To estimate sqBOLD oxygenation parameters for each voxel, R 2 0 and DBV were organised into a vector of unknowns (x) in a linear system ). S(τ) is the signal intensity of a given voxel for a spin-echo displacement time (τ) and τ 1,2, … n refers to the range of spin-echo displacement times acquired in the long τ regime (τ > 15 ms; Yablonskiy & Haacke, 1994). In this study τ values of 16,24,32,40,48,56, and 64 ms were used. The first row of matrix A describes where τ = 0, which is insensitive to DBV.
The weighted least squares solution was then used to produce voxel-wise estimates and standard deviation of R 2 0 and DBV. The model fit was inversely weighted for τ, with data being acquired at higher values of τ receiving less weighting in the fit, due to the increasing contribution of noise with increasing signal decay. This helps to account for the lower SNR in data acquired at longer τ-values Estimating [dHb] negates the requirement for an assumed or measured haematocrit, which is required in order to estimate the oxygen extraction fraction (OEF). The standard deviation of [dHb] was calculated by propagating the standard deviations calculated for R 2 0 and DBV.

| Regions of interest
Binary masks of the presenting lesion were automatically generated by thresholding maps of ADC at 620 × 10 −6 mm 2 /s (Purushotham et al., 2015). Initial clustering was performed using the FSL Cluster tool (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Cluster). The ROI cluster was identified and smoothed (Gaussian kernel of standard deviation 1 mm) and followed by repeat cluster analysis. The small amount of smoothing applied to the initial binary cluster mask was performed to remove isolated noisy voxels during repeat clustering, resulting in better specificity of the ADC ROI to the presenting lesion. These automated ADC masks were inspected by a clinician to ensure their accuracy and manually corrected when necessary (Harston et al., 2015). An independent observer manually defined the final infarct ROI. This was preferentially performed using the 1-week T 2 -FLAIR image or, if unavailable, the 24 hr b = 1,000 s/mm 2 DWI image (Harston, Minks, et al., 2017).
The following tissue outcomes were used in the analysis and were defined from the infarct ROIs in the native space of the sqBOLD and ASL parameter maps.
• The ischaemic core is tissue common to both the presenting ADC lesion and final infarct.
• Infarct growth is tissue present in the final infarct that is not present in the presenting ADC lesion.
• The contralateral tissue is defined by a composite mask of the presenting and final infarct tissue mirrored to the contralateral side of the brain.

| Registration
Registration of imaging modalities within a single time point was achieved using rigid body registration (6 degrees of freedom [DOF]) (Jenkinson & Smith, 2001). Between time point registration was performed using nonlinear registration of the T 1 -weighted structural scans to limit potential error introduced by edema (Harston, Minks, et al., 2017). To create contralateral ROIs, the infarct masks were mirrored in standard space following nonlinear registration of the T 1weighted image to a standard atlas (MNI152; Mazziotta et al., 2001).
At each time point, the FLAIR-GASE spin-echo image (τ = 0 ms) was registered (6 DOF) to the T 1 -structural using the b = 0 s/mm 2 DWI image as an intermediate registration step. For the ASL data, an unsubtracted reference volume was registered (6 DOF) directly to the T 1 -structural image.

| Regional analysis
For the MRI data acquired during presentation and follow-up scanning, voxel values of R 2 0 , DBV, [dHb], and CBF were extracted from the native space of the acquired parameter maps using the ROI definitions of ischaemic core, infarct growth and contralateral tissue.
A voxel-level analysis of the presenting imaging data was con-

| Comparison of presentation scans and tissue outcome
To demonstrate the potential of sqBOLD for quantitatively mapping oxygenation in a clinical research setting, data acquired during the initial MRI scan session are displayed in Figure 1 for all patients (n = 9).
For each patient, oxygenation (sqBOLD) and blood flow (ASL) parameter maps are presented alongside diffusion (DWI, b = 1,000 s/mm 2 ) images. Core (blue), growth (orange), and contralateral (yellow) tissue outcome ROIs are displayed on the sqBOLD spin-echo image. As a measure of the error in the oxygenation parameters, maps of the standard deviation of R 2 0 , DBV and [dHb] are provided. Higher error is noticeable in patients with more severe head motion during acquisition. For example, patient P05 displays the highest mean motion score and shows the largest standard deviation in the oxygenation parameters compared with patient P09, who moved the least. The mean motion scores estimated during motion correction for the presentation scans are listed in Table 1. It can also be seen that high DBV in the ventricles coincides with high DBV standard deviation and a large  revealed differences between all ROIs for each parameter (p < 0.01 in all cases). A more detailed voxel-level investigation of CBF in a similar patient cohort has been published elsewhere (Harston, Okell, et al., 2017).

| Comparison of presentation and follow up scans
The pertinent features of the sqBOLD technique as applied to acute stroke are illustrated through four example patients (Figures 4-7).
Firstly, the use of sqBOLD to investigate brain oxygenation over multiple time points is demonstrated in two patients and, secondly, two important confounding effects are considered through further examples.
Lastly, the potential of serial imaging is investigated at the patient level by comparing the presenting data with images acquired at 1 week.
The images in Figure 4 were acquired from patient P03 on presentation (2 hr 20 min postonset) and 24 hr after presentation. This patient received IV thrombolysis at 1 hr 24 min post-onset. On presentation, a lesion is clearly evident on the b = 1,000 s/mm 2 map. The presenting R 2 0 parameter map shows a large region of elevated R 2 0 in the area surrounding the presenting DWI lesion, which corresponds to a region of reduced CBF in the presenting CBF map. At the follow  extracted from the ischaemic core ROI, which is defined based on the presenting DWI, is still greater than the infarct growth ROI ( Table 2).
The measure of R 2 0 in this region appears to decrease between the presenting and follow-up imaging time points. This patient received IV thrombolysis at 2 hr 31 min post-onset. At presentation, a well defined lesion is visible in the b = 1,000 s/mm 2 images. This is paralleled by similarly well defined regions of elevated R 2 0 and reduced CBF. An elevation in DBV can also be seen. At 1 week R 2 0 appears to have normalised, but now hyperperfusion is seen where previously hypoperfusion had been present. Finally, CBF is observed to be normalised by the 1 month scan. Figure 6 shows images acquired from patient P05 who did not receive IV thrombolysis and was scanned on presentation (28 hr 20 min postonset), with follow up scanning performed at 38 hr and 1 month after presentation. Elevated R 2 0 is observed in regions expected to contain CSF at presentation. This is consistent with a failure of the CSF nulling FLAIR preparation. This is most likely due to patient motion since this patient recorded the highest mean motion score of 2.12, which compares with the 1 month scan where CSF signal appears nulled and a mean motion score of 0.86 was recorded.
Maps of CBF demonstrate a heterogenous pattern of perfusion in the ischaemic region and appear to be relatively consistent between imaging time-points.  (Table 2).

Figure 8 compares measurements made at presentation with those
made at the follow up scan at 1 week, which was available in 6 of the 9 patients. The null hypothesis was rejected for the R 2 0 measurements in the ischaemic core (paired t-test, p < 0.05) and infarct growth (paired t-test, p < 0.01) ROIs. In both cases R 2 0 is reduced at 1 week compared with presentation, which is consistent with increased blood oxygenation.
This increase in blood oxygenation is mirrored in measurements of DBV and [dHb], which both result in a nonsignificant reduction in the amount of deoxyhaemoglobin present in the voxel.

| DISCUSSION
Streamlined-qBOLD is shown to provide metabolic information that is indicative of tissue viability following acute stroke. Detailed regional revealing this effect to be driven by a significant difference between ischaemic core and both infarct growth and contralateral tissue. Furthermore, serial analysis at the patient level revealed significant changes in R 2 0 between the presentation and 1 week scans for both ischaemic core (p < 0.01) and infarct growth (p < 0.05).

| Ischaemic penumbra
The definition of the infarct growth region used in this study is expected to be spatially and metabolically consistent with the fraction of the ischaemic penumbra that does not survive. In this region, an increase in [dHb] is anticipated in order to maintain the rate of oxygen metabolism in tissue that is experiencing a reduction in CBF. The potential of sqBOLD to detect these changes was demonstrated by the statistically significant increase in [dHb] measured in the infarct growth ROIs at the voxel level (Figure 2; p < 0.01). Furthermore, [dHb] was found to be elevated, and CBF reduced, at the patientlevel, although this was not statistically significant ( Figure 3). In addition to the observed changes in CBF and [dHb], DBV was found to be elevated in the infarct growth region (Figure 3). In order to interpret this it is important to understand the definition of DBV, which represents the blood volume occupied by deoxygenated blood. In healthy tissue this is largely contained within the veins and capillaries and an elevation would therefore normally result from passive inflation during increases in CBF. Since CBF is reduced here, increases in DBV can only be achieved by decreasing blood oxygenation in normally highly oxygenated vessels, such as precapillary arterioles. These vessels have already been shown to have a higher degree of desaturation than previously thought (Vovenko, 1999) which could feasibly be further reduced during ischaemia.
The spatial correspondence of the sqBOLD parameter maps with the infarct growth ROI can also be observed at the individual patient level (Figure 4). This example case shows presenting measures of R 2 0 , DBV, and [dHb] that are elevated in regions that correspond to the infarct growth ROI. The CBF parameter map acquired on presentation demonstrates a large region of decreased flow that coincides with the elevated regions on the sqBOLD parameter maps. A restriction in flow is expected to result in an elevated OEF, which in turn causes an increase in the relative amount of deoxyhaemoglobin produced. As such, the observation of reduced CBF and elevated [dHb] in this patient is suggestive of the early identification of tissue exhibiting the physiological traits of the ischaemic penumbra. This opens up the prospect that concurrent MR based oxygenation and flow imaging can be used to identify tissue at risk of infarction (Astrup et al., 1981). In this patient, infarction occurs in this region at some point between the presenting and follow up scan times as evidenced by the infarct growth ROI (defined from the b = 1,000 s/mm 2 image). Therefore, early identification of penumbral tissue would provide a window of opportunity for interventions that might salvage this tissue.

| Ischaemic core
From Figure 2, it can be seen that increases in all of the baseline brain oxygenation parameters are observed in the ischaemic core compared to infarct growth on presentation. This trend appears surprising at first, particularly if the elevated signal in the core is to be associated with the presence of deoxyhaemoglobin as a by-product of ongoing metabolism. The infarct growth region is expected to contain tissue that is metabolically active on presentation but later recruited to the final infarct volume. This is in contrast to the nonviable tissue present in the ischaemic core. However, the elevated qBOLD signal measured in the ischaemic core can be explained either as stationary deoxyhaemoglobin in metabolically inactive regions with no blood supply or as ongoing metabolism in the diffusion lesion. This also offers two possible scenarios that explain the significant increase in core [dHb] measured at the patient-level (Figure 3).

| Stationary deoxyhaemoglobin
A similar regional trend in R 2 0 can be extrapolated from a previous study (Geisler et al., 2006) which looked at comparable tissue outcome ROIs. Here it was proposed that the elevated R 2 0 in the ischaemic core may result from stationary deoxyhaemoglobin present in vessels without blood supply. In the event of a complete occlusion of flow, stationary haemoglobin beyond the blockage will become fully deoxygenated as the remaining oxygen is metabolised leading to an increase in the amount of deoxyhaemoglobin present. This is likely to be the main contributing factor to the trend seen in Figure 2, where the ischaemic core demonstrates the largest elevation in R 2 0 , DBV, and [dHb]. In the context of the DBV elevation observed in the infarct growth region this would similarly reflect the desaturation of arteriolar blood vessels.

| Reperfusion
In contrast, R 2 0 and [dHb] parameter maps in Figure 4 demonstrate a decrease in the ischaemic core. This may be explained by the presence or restoration of flow to an infarcted region. In this case, metabolically inactive tissue would not produce new deoxyhaemoglobin and previously produced deoxyhaemoglobin would be removed by the restored blood flow, leading to a decrease in R 2 0 , DBV, and [dHb]. This patient received thrombolysis at 1 hr 24 min postonset. Despite this, flow is still noticeably reduced in this patient in the ischaemic core at presentation (Table 2) suggesting that recanalisation has not occurred. It is unclear whether this flow is sufficient to clear the deoxyhaemoglobin from metabolically inactive tissue.

| Ongoing metabolism in the diffusion lesion
Elevated [dHb] and DBV in the ischaemic core may also be explained by observations made using PET, which have shown that regions of ongoing oxygen metabolism are possible within the presenting diffusion lesion (Fiehler et al., 2002;Guadagno et al., 2004Guadagno et al., , 2006Kidwell et al., 2000).
As such, it is possible that deoxyhaemoglobin production may still be occurring in regions of decreased ADC and contribute towards elevated R 2 0 in the ischaemic core. While none of the patients in this study clearly demonstrate this phenomenon (ie, elevated R 2 0 in the presence of nonzero CBF), the parameter maps shown in Figure 6 show some evidence of a heterogeneous pattern of blood oxygenation within the diffusion lesion. Unfortunately, these parameter maps are of low quality due to

| Interpretability of [dHb] and OEF
It is evident that the relaxometry based method used in this study is sensitive to deoxyhaemoglobin regardless of the patency of the blood supply and can therefore exhibit elevated R 2 0 in the ischaemic core. As such, knowledge of the local blood supply is important to distinguish stationary deoxyhaemoglobin present in infarcted tissue from active tissue with an elevated metabolism. This motivated the calculation of [dHb] rather than OEF to avoid the false interpretation of a high R 2 0 as always representing elevated oxygen extraction. This sensitivity to stationary deoxyhaemoglobin also reconciles the apparent differences between PET and BOLD based measurements (Geisler et al., 2006). In PET the oxygen sensitive tracer is prevented from being delivered to the ischaemic core, meaning that signal is not detected there and reduced oxygen metabolism is inferred. This is in contrast to BOLD based measurements, which do not rely on the arrival of a tracer and hence the presence of deoxyhaemoglobin will still cause an increase in R 2 0 . However, in this study all of the patients with hypoperfused lesions had nonzero median CBF in the ischaemic core ROI at presentation (Table 2). It remains to be seen whether there is a threshold CBF below which the blood is functionally stationary with respect to the accumulation of deoxyhaemoglobin.

| Prospects for serial imaging of brain oxygenation
Streamlined-qBOLD and ASL data were successfully acquired at each of the time points listed in Table 1

| Confounds: Non-deoxyhaemoglobin related elevations and patient-motion
Streamlined-qBOLD is sensitive to other sources of magnetic susceptibility in the brain not related to deoxyhaemoglobin and care must be taken when interpreting elevations in signal. Iron and myelin are known sources of susceptibility that can confound the accurate quantification of brain oxygenation with this method and are of particular relevance as both can vary during ageing and in different pathologies. Figure 7 shows bilateral elevations in R 2 0 on both the affected and unaffected sides of the brain due to the high iron content of the deep grey matter structures. The presenting [dHb] map appears to be more highly elevated in deep grey matter on the affected side, pointing towards the importance of interpreting the R 2 0 , DBV, [dHb], and CBF parameter maps in combination, as well as being aware of non-oxygen related sources of susceptibility in the locality of the region of interest. Furthermore, a change in local haematocrit level could influence regional sqBOLD measurements (Broisat et al., 2018). For example, a local increase in Hct would increase R 2 0 even in the case of high CBF.  Figure 1). This provides some assurance of oxygenation measurements made in tissue even in cases where there is significant head motion during sqBOLD acquisition.
Through patient-wise comparison of mean motion scores (Table 1) and R 2 0 standard deviation maps (Figure 1)  From Figure 5 a noticeable increase in DBV can be seen between the presentation and 1 week follow-up scans. On closer inspection, the 1 week DBV error map (data not shown) appears uniformly elevated compared to presentation. Further inspection of this data, which was found to have a low mean motion score, indicates that patient motion was not a problem here. It is therefore possible that the observed elevation in DBV could be of physiological origin. However, further work is required to establish the accuracy and repeatability of this technique. From Figure 1 it can be seen that the error in DBV can be quite high and from previous work it has been noted that the accuracy of this measurement requires improvement (Stone & Blockley, 2017).

| Group heterogeneity, flow, and further work
From the patient-level analysis (Figure 3), only the ischaemic core [dHb] demonstrated a significantly different value from infarct growth and contralateral ROIs. Although strict inclusion criteria were employed in this study to provide a uniform patient cohort, the failure to detect significant changes in R 2 0 and DBV may be partly explained by remaining heterogeneity in this group. Heterogeneity in regional perfusion status is apparent, with almost equal numbers of patients exhibiting hypo-and hyperperfusion in their presenting ischaemic core ROIs. A further source of heterogeneity across the group may be attributed to differences in onset to scan time (Table 1).
Alongside demonstrating the sensitivity of sqBOLD to changes in oxygenation, this study provides a demonstration of how complementary measurements of flow and oxygenation can be used to provide a unique insight into tissue viability. In practise, the information provided by sqBOLD could aid the interpretation of ASL CBF measurements, since low flow does not always progress to infarction in regions experiencing benign oligaemia (Kidwell, Alger, & Saver, 2003).
Although it is beyond the scope of this study, the combination of CBF and [dHb] measurements allows for the calculation of the cerebral metabolic rate of oxygen consumption (CMRO 2 ) (Blockley, Griffeth, Stone, Hare, & Bulte, 2015). This has been shown to improve tissue outcome prediction and may partly explain the variability seen in the presenting sqBOLD oxygenation measurements . The ASL acquisition used in this study utilises multiple postlabelling delays and as such offers improved accuracy of CBF quantification in the case of delayed blood arrival times. However, care should be taken when comparing these resuls with typical DSC PWI measurements where TTP and MTT are often preferred.
The identification of tissue outcomes based solely on measurements from this method does not currently appear to be possible, as evidenced by the considerable overlap between tissue outcome distributions in Figure 2. However, Kruskal-Wallis tests and post hoc pairwise comparisons found significant differences between these distributions suggesting that tissue outcome is dependent on tissue oxygenation and that the parameter maps derived from sqBOLD are sensitive to identifying this information on presentation. As such, sqBOLD provides complementary information to existing imaging modalities such as DWI and ASL and the combination of this information may allow for earlier identification of tissue under metabolic stress during the acute phases of stroke .
Future development of sqBOLD should focus on improving the robustness of the oxygenation measurements through refinement of the sqBOLD acquisition, modelling and analysis routines. This work would be aided by a study of the repeatability of the R 2 0 measurement. To date the only study to compare multiple techniques for measuring R 2 0 suggested that ASE has a marginally increased intersubject standard deviation (Ni, Christen, Zun, & Zaharchuk, 2015). However, it is difficult to compare these results with the current work, since the acquisitions differ significantly, and therefore further investigation of the GASE technique is warranted.
In addition, this study supports the further investigation of sqBOLD in a larger scale study and highlights the importance of controlling for onset to scan time and tissue perfusion status. The results presented here provide an initial estimate of the effect size that may be used to ensure follow on studies are adequately powered (Table 3).
The noninvasive, quantitative nature of this method also means it is suitable for the longitudinal monitoring of stroke evolution and may provide unique insight into the various pathways to infarction and recovery, as well as providing valuable biomarkers with which to assess treatment and intervention (Figure 8).

| CONCLUSION
Streamlined-qBOLD was used to acquire information about oxygen metabolism in a cohort of acute ischaemic stroke patients, which is complementary to conventional MRI methodologies. It was found that resting brain oxygenation related parameters (R 2 0 , DBV, and [dHb]) IGURE 8 Comparison of measurements made during presenting and 1 week stroke-onset-to-scan-times (n = 6). For each parameter (R 2 0 , DBV, [dHb], and CBF), measurements made in core and growth ROIs are normalised to contralateral tissue and group averaged (± standard deviation). Significant decreases in R 2 0 were detected in both core (p = 0.049) and growth (p = 0.006) after 1 week (paired t-test). *p < 0.05 [Color figure can be viewed at wileyonlinelibrary.com]