Serial quantification of brain oxygenation in acute stroke using streamlined-qBOLD

It has been proposed that metabolic markers of baseline brain oxygenation have a role to play in the early identification of the ischemic penumbra. Streamlined-qBOLD is a magnetic resonance imaging technique that does not require exogenous contrast. It is a refinement of the quantitative BOLD methodology that provides a simplified approach to mapping and quantifying baseline brain oxygenation related parameters (reversible transverse relaxation rate (R2′), deoxygenated blood volume (DBV) and deoxyhaemoglobin concentration ([dHb])) in a clinically relevant manner. Streamlined-qBOLD was applied to an exploratory cohort of acute stroke patients in a serial imaging study. Detailed voxel-level analysis was used to quantify the metabolic profile of ischaemic tissue on presentation and investigate these metrics in relation to tissue outcome. Individual patient examples illustrate the appropriate interpretation of R2′, DBV and [dHb] in acute stroke and demonstrate the ability of this method to deliver regional information related to oxygen metabolism in the ischaemic tissue. Regional analysis confirms that R2′, DBV and [dHb] vary between regions of ischaemia with different tissue outcomes.

The aim of this study is to demonstrate the utility of sqBOLD for identifying regional 22 changes in brain oxygenation during the acute phases of stroke. Here, sqBOLD is 23 applied in a prospective cohort of patients with acute ischaemic stroke. Using 24 detailed voxel-level analysis, the metabolic profile of ischaemic tissue is quantified 25 on presentation and regional measures of sqBOLD parameters (R2′, DBV and [dHb]) 1 are investigated in relation to tissue outcome. Patients with ischaemic stroke were recruited into a prospective observational cohort 3 study regardless of age or stroke severity under research protocols agreed by the 4 UK National Research Ethics Service committees (ref: 13/SC/0362). MRI was 5 performed at presentation, 2 hours, 24 hours, 1 week and 1 month whenever 6 possible. Nine consecutive patients were scanned on presentation. One patient was 7 excluded from further analysis because the final lesion ROI could not be defined (no 8 follow-up scan) and one patient was excluded from further analysis due to 9 haemorrhagic transformation at the time of the presenting MRI leading to no relevant 10 presenting data being acquired. As such, seven patients were included in the final 11 analysis (Table 1). 12 13 Image acquisition 14 Scanning was performed on a Siemens 3T Verio scanner for all time points. 15 Scanning protocols included diffusion weighted imaging (DWI) (three directions, 1.8 16 x 1.8 x 2.0 mm, field of view = 240 mm 2 , four averages, b = 0 and 1000 s/mm 2 , TR / 17 TE = 9000 / 98 ms, 50 slices, 2 min 53 s) with apparent diffusion coefficient (ADC) 18 calculation; T1-weighted MP-RAGE for structural imaging (1. 8 24 x 96 matrix, field of view = 220 mm 2 , nine 5 mm slabs consisting of four 1.25 mm 25 sub-slices, 100% partition oversampling, 1 mm slice gap, TR / TE = 3000 / 82 ms, 1 TIFLAIR = 1210 ms, ASE-sampling scheme τstart / τfinish / Δτ = -16 / 64 / 8 ms, scan 2 duration 4 min 30 s). FLAIR-GASE consists of three separate components, nulling of 3 CSF partial volumes using FLuid Attenuated Inversion Recovery (FLAIR) (Hajnal et 4 al., 1992), minimisation of MFI using Gradient Echo Slice Excitation Profile Imaging 5 (GESEPI) (Yang et al., 1998) and direct measurement of R2′ using an Asymmetric 6 Spin Echo (ASE) (Wismer et al., 1988). This FLAIR-GESEPI-ASE (FLAIR-GASE) 7 (Blockley and Stone, 2016) acquisition reduces confounding effects and when 8 combined with quantitative modelling offers a streamlined qBOLD approach. For the 9 patient presented in Figure 3, perfusion information was acquired on presentation 10 using vessel encoded pseudo-continuous arterial spin labelling (VEPCASL) (EPI 11 readout, 3.4 x 3.4 x 4.5 mm, field of view = 220 x 220 mm, 24 slices, TR / TE = 4080 12 / 14 ms, Labelling Duration = 1.4 s, Post-labelling Delays = 0.25, 0.5, 0.75, 1, 1.25 13 and 1.5 s, scan duration 5 min 55 s). Post-processing details of VEPCASL data to 14 produce cerebral blood flow (CBF) maps have previously been described (Harston et   parameter maps from the FLAIR-GASE data have previously been described (Stone 22 and Blockley, 2017). In brief, the τ-series were motion corrected using the FSL linear 23 motion correction tool (MCFLIRT) (Jenkinson et al., 2002) to the spin-echo image. 24 The spin-echo image was brain extracted using the FSL brain extraction tool (BET) 25 (Smith, 2002) to create a binary mask of brain tissue and all remaining τ-weighted 1 volumes were brain extracted using this mask. This data was then fit on a voxel-wise 2 basis to obtain parameter maps of R2′ by using a weighted log-linear fit to the mono-3 exponential regime (τ ≥ 16 ms (Yablonskiy and Haacke, 1994)). The intercept of this 4 fit is in effect the log of the ASE signal at τ = 0 extrapolated from the mono-5 exponential regime (ln(S(τ = 0)extrap)). By subtracting the log of the measured spin- 6 echo signal (ln(S(τ = 0 ms)) from this value, parameter maps of DBV can be 7 produced, as previously described (Yablonskiy, 1998) Parameter maps of [dHb] were calculated using Equation 2, where DBV and R2′ 10 were measured as above and other parameters are known or assumed constants 11 (Δ 0 = 0.264 x 10 -6 , κ = 0.03 (He and Yablonskiy, 2007)). 12 [ ] = 3. 2 ′ . 4. . . ∆ 0 . . 0 (2) 13 Regions of interest 14 The presenting infarct region of interest (ROI) was defined using the presenting 15 apparent diffusion coefficient (ADC) parameter map and a previously described 16 clustering method (Harston et al., 2015), as follows. Binary masks of presenting ADC 17 lesions were automatically generated using a threshold-defined (620 x 10 -6 mm 2 /s) 18 (Purushotham et al., 2015) cluster-based analysis of the ADC data. The ROI cluster 19 was identified and smoothed (Gaussian kernel of standard deviation 1 mm) and 20 followed by repeat cluster analysis using the FSL Cluster tool 21 (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Cluster). These automated ADC masks were 22 inspected by a clinician to ensure their accuracy and manually corrected when 23 necessary. The final infarct ROI was manually defined by an independent observer.   1   This was done preferentially using the 1-week T2-FLAIR image or, if not available,   2 the 24 hour b = 1000 s/mm 2 DWI image (Harston et al., 2017a). 3 The following tissue outcomes were used in the analysis and were defined from the 4 infarct ROIs in the native space of the sqBOLD parameter maps. 5 • The ischaemic core is tissue common to both the presenting infarct and final 6 infarct. 7 • The infarct growth is tissue present in the final infarct that is not present in the 8 presenting infarct.

9
• The contralateral tissue is defined by a composite mask of the presenting and 10 final infarct tissue mirrored to the contralateral side of the brain.   applied to investigate the distributions of these parameters also. 16

1
Group characteristics 2 Nine consecutive large volume stroke patients were scanned on presentation, seven 3 of which were included in the final analysis (Table 1). One patient was excluded from 4 the analysis because the final lesion ROI could not be defined (no follow-up scan) 5 and one patient was excluded from further analysis due to haemorrhagic 6 transformation at the time of the presenting MRI leading to no presenting FLAIR- 7 GASE data being acquired. The median National Institute of Health Stroke Scale at 8 presentation was 12 (range 3 -25) and the median symptom onset to MRI was 10 9 hours 52 minutes (range 2 hours 20 minutes -1 day 4 hours 19 minutes). Eight 10 patients received intravenous thrombolysis. 11 12 Voxel-level analysis 13 To relate the presenting sqBOLD parameter maps to tissue outcome, the extracted ROIs and revealed statistically significant differences between the tissue outcome ROIs for each parameter (infarct core and contralateral tissue; infarct growth and 1 contralateral tissue; infarct core and infarct growth; p < 0.01 in all cases). In the acute phases of stroke, sqBOLD is shown to provide metabolic information 2 that is indicative of the viability of ischaemic tissue. Serial-imaging from example 3 patient cases and detailed regional analysis demonstrate that R2′, DBV and [dHb] 4 are sensitive to oxygenation related changes in ischaemic tissues with varying 5 outcomes. Significant pairwise differences in voxel distributions were observed 6 between the regional tissue ROIs using multiple comparisons analysis for R2′, DBV values for all parameters increased in the ischaemic regions (ischaemic core and 10 infarct growth) when compared to healthy tissue on the contralateral side (Figure 1). 11 The most obvious increase is seen in R2′ and this is driven by a statistically 12 significant increase in both deoxyhaemoglobin volume fraction (DBV) and 13 concentration ([dHb]).
14 15 Ischaemic penumbra 16 The definition of the infarct growth region used in this study is expected to be 17 spatially and metabolically consistent with the ischaemic penumbra. In this region, an Ischaemic core 16 From Figure 1, larger increases in all parameters were observed in the ischaemic 17 core compared to infarct growth on presentation. This trend appears surprising at 18 first, particularly if the elevated signal in the core is to be associated with the 19 presence of deoxyhaemoglobin as a by-product of ongoing metabolism. The infarct 20 growth region is expected to contain tissue that is metabolically active on 21 presentation but later recruited to the final infarct volume. This is in contrast to the 22 non-viable tissue present in the ischaemic core. However, the elevated brain fully deoxygenated as the remaining oxygen is metabolised leading to an increase in 10 the amount of deoxyhaemoglobin present. This is likely to be the main contributing 11 factor to the trend seen in Figure 1, where the ischaemic core demonstrates the  6 It is evident that the relaxometry based method used in this study is sensitive to 7 deoxyhaemoglobin regardless of the patency of the blood supply and can therefore 8 exhibit elevated R2′ in the ischaemic core. As such, knowledge of the local blood 9 supply is important to distinguish stationary deoxyhaemoglobin present in infarcted 10 tissue from active tissue with an elevated metabolism. This motivated the calculation 11 of [dHb] rather than OEF to avoid the false interpretation of a high R2′ as elevated 12 oxygen extraction in the absence of flow information. This sensitivity to stationary 13 deoxyhaemoglobin also reconciles the apparent differences between PET and BOLD 14 based measurements (Geisler et al., 2006). In PET the oxygen sensitive tracer is 15 prevented from being delivered to the ischaemic core, meaning that signal is not 16 detected there and reduced oxygen metabolism is inferred. This is in contrast to 17 BOLD based measurements, which don't rely on the arrival of a tracer and hence the 18 presence of deoxyhaemoglobin will still cause an increase in R2′. 19 20

Group heterogeneity, flow and further work
From the patient-level analysis (Figure 2), considerable heterogeneity was apparent 1 across the group, meaning regional trends in R2′, DBV and [dHb] were not 2 significantly different. The heterogeneity in regional parameters across the group can 3 be attributed, at least in part, to the differences in onset to scan time ( Table 1) and 4 differences in the perfusion and reperfusion status of the ischaemic tissue. As such, 5 it is difficult to hypothesise the expected metabolic state of this tissue on 6 presentation, but it is likely to be varied across the group, with each patient 7 potentially undergoing a different pathway to infarction (del Zoppo et al., 2011). This 8 is supported by the high coefficient of variation across subjects measured in the 9 infarct growth region, particularly for DBV and [dHb] (Figure 2). 10 11 The absence of comprehensive perfusion information is a distinct limitation of this comparisons analysis found significant differences between these distributions 8 suggesting that tissue outcome is dependent on tissue oxygenation and that the 9 parameter maps derived from sqBOLD are sensitive to identifying this information on 10 presentation. As such, sqBOLD provides complimentary information to existing 11 imaging modalities such as DWI and ASL and the combination of this information 12 may allow for earlier identification of tissue under metabolic stress during the acute 13 phases of stroke . 14 15 In addition, this study supports the further investigation of sqBOLD in a larger scale 16 study and highlights the importance of controlling for onset to scan time and tissue 17 perfusion status. The non-invasive, quantitative nature of this method also means it 18 is suitable for longitudinally monitoring stroke evolution and may provide unique 19 insight into the various pathways to infarction and recovery, as well as providing 20 valuable biomarkers with which to assess treatment and intervention. to refine the identification of the ischemic penumbra.      delineate the DWI lesion more clearly from the normal contralateral hemisphere. 10   Mazziotta, J., Toga, A., Evans, A., FOX, P., Lancaster, J., Zilles, K., Woods, R.,