Noninvasive Assessment of Oxygen Extraction Fraction in Chronic Ischemia Using Quantitative Susceptibility Mapping at 7 Tesla
Background and Purpose—The oxygen extraction fraction (OEF) is an effective metric to evaluate metabolic reserve in chronic ischemia. However, OEF is considered to be accurately measured only when using positron emission tomography (PET). Thus, we investigated whether OEF maps generated by magnetic resonance quantitative susceptibility mapping (QSM) at 7 Tesla enabled detection of OEF changes when compared with those obtained with PET.
Methods—Forty-one patients with chronic stenosis/occlusion of the unilateral internal carotid artery or middle cerebral artery were examined using 7 Tesla-MRI and PET scanners. QSM images were obtained from 3-dimensional T2*-weighted images, using a multiple dipole-inversion algorithm. OEF maps were generated based on susceptibility differences between venous structures and brain tissues on QSM images. OEF ratios of the ipsilateral middle cerebral artery territory against the contralateral side were calculated on the QSM-OEF and PET-OEF images, using an anatomic template.
Results—The OEF ratio in the middle cerebral artery territory showed significant correlations between QSM-OEF and PET-OEF maps (r=0.69; P<0.001), especially in patients with a substantial increase in the PET-OEF ratio of 1.09 (r=0.79; P=0.004), although showing significant systematic biases for the agreements. An increased QSM-OEF ratio of >1.09, as determined by receiver operating characteristic analysis, showed a sensitivity and specificity of 0.82 and 0.86, respectively, for the substantial increase in the PET-OEF ratio. Absolute QSM-OEF values were significantly correlated with PET-OEF values in the patients with increased PET-OEF.
Conclusions—OEF ratios on QSM-OEF images at 7 Tesla showed a good correlation with those on PET-OEF images in patients with unilateral steno-occlusive internal carotid artery/middle cerebral artery lesions, suggesting that noninvasive OEF measurement by MRI can be a substitute for PET.
- carotid artery, internal
- cerebrovascular disorders
- magnetic resonance imaging
- middle cerebral artery
- positron-emission tomography
Patients with severe hemodynamic ischemia termed as misery perfusion, which can be identified by an increased oxygen extraction fraction (OEF),1 have a high risk of stroke recurrence and embolic complications during surgery.2,3 OEF can be directly measured only by 15O2-positron emission tomography (PET), which is considered the gold standard.4 However, PET involves several disadvantages, such as radiation exposure, invasiveness, including arterial blood sampling, long examination times, a limited number of clinically available scanners, and lower spatial and temporal resolutions.
Recently, several approaches have attempted to measure OEF using MRI techniques.5 In general, these techniques used blood oxygen level–dependent (BOLD) effects induced by differences in magnetic susceptibility between oxy- and deoxy-hemoglobin to quantify oxygenation in venous structures and brain parenchyma. Some of these attempted to obtain OEF values and changes by cerebrovascular challenges, such as O2/CO2 inhalation, hyperventilation, caffeine, acetazolamide, and sedatives.6–11 To calculate OEF values in a resting state, however, these techniques need complex paradigms and experimental procedures that may be unavailable in clinical practice, particularly for patients with stroke or respiratory disorders. Another approach is OEF measurement without any challenges, which includes T2* or T2’ relaxation measurements, termed as the quantitative BOLD,12,13 T2 relaxation measurement using a spin-tagging technique,14,15 and phase difference measurement.16 These methods enabled noninvasive estimation of OEF values in healthy subjects, as well as in patients with chronic ischemia and other neurological disorders.17–21 However, the techniques are not commonly available in clinical practice in many institutes because novel scanning sequences dedicated for this purpose are needed. Moreover, the aforementioned studies performed no direct comparisons with superior O2-PET, which is considered the gold standard to measure OEF. Hence, the availabilities, reliabilities, and accuracies of the MRI-based OEF measurement methods remain unclear, particularly in patients with cerebrovascular diseases.
Quantitative susceptibility mapping (QSM) is a post-processing technique for quantifying magnetic susceptibility of venous structures and brain parenchyma from T2*-weighted magnitude/phase images that can be easily obtained by commercial scanners. A recent study introduced a noninvasive OEF measurement method based on the QSM technique and demonstrated that OEF changes in patients with unilateral chronic steno-occlusive disease showed good correlations with those obtained by PET.22 However, absolute OEF values and changes in the affected cerebral hemisphere tended to be underestimated presumably because of insufficient BOLD-related signal changes in minute venous structures at 3 Tesla (3T) and suboptimal post-processing algorithms. Thus, in this study, we attempted to investigate whether QSM-OEF maps obtained with a 7 Tesla (7T) scanner, which has profound susceptibility effects, and optimized post-processing techniques can readily estimate OEF changes in patients with major cerebrovascular steno-occlusive disease and can accurately detect misery perfusion when compared with OEF maps obtained by PET.
From July 2012 to January 2016, 41 patients with chronic steno-occlusive disease of unilateral major cerebral arteries, which were evaluated by pre-operative imaging diagnosis using digital subtraction angiography or magnetic resonance angiography in our hospital, were prospectively recruited according to the following criteria: those who have unilateral occlusion or stenosis of the internal carotid artery (ICA) or middle cerebral artery (MCA; ICA stenosis, ≥70% based on The North American Symptomatic Carotid Endarterectomy Trial criteria23; MCA stenosis, ≥50% based on the Trial of Cilostazol in Symptomatic Intracranial Arterial Stenosis study criteria24) with no apparent cortical infarct in the MCA territory and those who were eligible for examinations with both 7T-MRI and PET. The details of the patient characteristics were as follows: 28 men and 13 women; age range, 29 to 82 years (median, 64 years); 19 patients with ICA occlusion, 3 with ICA stenosis, 8 with MCA occlusion, and 11 with MCA stenosis; 37 symptomatic and 4 asymptomatic patients. These patients underwent both MRI and PET scans with an interval of 2 to 24 days (mean, 4.6 days).
All examinations were performed after obtaining the approval of the institutional ethics committee (H23-45), and a written informed consent was obtained from all participants.
We used a 7T-MRI scanner (Discovery MR950, GE Healthcare, Milwaukee, WI) with quadrature transmission and 32-channel receive head coils. Source data of QSM were obtained using a 3-dimensional spoiled gradient recalled acquisition technique with the following scanning parameters: repetition time, 30 ms; echo time, 15 ms; flip angle, 20°; field of view, 256 mm; acquisition matrix size, 512×256; slice thickness, 2 mm; number of slices, 160; reconstruction voxel size after zero-fill interpolation, 0.5 mm3; and scan time, 3 minutes 25 seconds. The sections were set in the orthogonal axial plane from the level of the superior cerebellar peduncle to high convexity. Magnitude as well as real/imaginary phase images were regenerated from this acquisition. Structural images including T2-weighted images and magnetic resonance angiography were also obtained.
A PET study was performed using a PET/computed tomographic scanner (SET-3000GCT/M, Shimadzu Corp, Kyoto, Japan) with the inhalation technique of 15O2. The scanner was operated in a static scan mode with dual-energy window acquisition for scatter correction, the coincidence time window of 10 ns, axial field of view of 256 mm, and the full width at half maximum for in-plane and axial spatial resolutions of 3.5 mm and 4.2 mm, respectively. The subjects continuously inhaled superior O2 (1480 MBq) through a mask for 5 minutes, followed by a single breath of C15O (444 MBq). The image was reconstructed using the ordered subset expectation maximization algorithm. OEF maps were calculated using the steady-state method with correction by cerebral blood volume.25
QSM images were generated from the source images using an in-house program with a multiple dipole-inversion combination with k-space segmentation26 and regularization enabled sophisticated harmonic artifact reduction for phase data27 methods. We then applied a 2-dimensional Gaussian low-pass filter with a kernel size of 60% of the total image power in each section to extract iron deposition in deep nuclei, hemosiderin deposition, dural sinuses, and large venous structures, as well as a 2-dimensional Gaussian high-pass filter of 2% to extract small venous structures. Subsequently, small venous structures were determined by multiplying the Gaussian high-pass filter–processed binary images and the logical negations of Gaussian low-pass filter–processed binary images under the threshold for binarization of > +2 SDs.
The OEF maps with voxels-of-interest of 25 mm3 were generated from the processed QSM images according to a previous study.22 In brief, the susceptibility difference between venous structures and surrounding brain tissues, Δχ, is expressed by the following equation:(1)
where Δχdo is the difference in the susceptibility per unit hematocrit between fully deoxygenated and fully oxygenated blood (we used 0.18 ppm [cgs]),28 Hct is hematocrit (we used 0.45), Yv is venous oxygen saturation, and Pv is a correction factor for partial volume effects that was defined as ≈7.0 according to the previous study.22 However, OEF is defined as (Ya−Yv)/Ya, where Ya is arterial oxygen saturation and can be estimated as 1–Yv under usual conditions in which Ya is ≈100%.29 Hence, OEF can be calculated with the following equation:(2)
Using Statistical Parametric Mapping 12 (Wellcome Department of Imaging Neuroscience, University College London, UK),30 PET-OEF images that were coregistered to QSM source images, as well as QSM-OEF images after Gaussian smoothing (σ=10 pixel), were warped to Montreal Neurological Institute coordinates. The OEF values were then automatically measured on the sections through the lateral ventricle body and centrum semiovale; that is, z coordinate of 30 to 75 mm, using an image analysis program (ITK-SNAP, www.itksnap.org)31 with the region of interest (ROI) of MCA territory (a combined area of pre-central, central, parietal, and angular segments) provided by 3-dimensional stereotaxic ROI template (Figure 1).32,33 Mean OEF values were calculated for each cerebral hemisphere, and OEF ratios of the affected hemisphere against the contralateral one were obtained.
The correlations of the OEF values of the MCA ROIs, as well as OEF ratios of affected/nonaffected hemispheres between QSM and PET, were examined in all the patients and those with/without the substantial increase in PET-OEF ratios (>1.09, according to a previous study33) using Pearson correlation coefficient and linear regression analysis. Bland–Altman analysis was performed to examine agreements for the OEF ratios between QSM and PET in the patients with/without increased OEF. In addition, receiver operating characteristic analysis was performed to determine the sensitivity and specificity of QSM-OEF ratios for the substantial increase in PET-OEF ratios. The cut-off value of the QSM-OEF ratio was determined using Youden index. The OEF values of the ROIs were compared between the hemispheres using a Wilcoxon signed-rank test. The OEF ratios were compared between the symptomatic and asymptomatic patients using a Mann–Whitney U test and between the patients with occlusion/stenosis of ICA/MCA using a Steel-Dwass test. Correlations between the QSM-PET difference in the OEF ratios and degree of MCA stenosis were examined using Pearson correlation coefficient. The cut-off α level used was 0.05.
Two patients were excluded because of profound metallic artifacts and severe motion artifacts. The remaining 39 patients (26 men and 13 women; age range, 29–82 years [median, 64 years]; 18 ICA occlusion, 3 ICA stenosis, 8 MCA occlusion, and 10 MCA stenosis; 35 symptomatic and 4 asymptomatic patients) were eligible for further analyses. We successfully obtained QSM-OEF images that appeared visually comparable with the corresponding PET-OEF images (Figure 1).
Both QSM-OEF and PET-OEF values in the affected hemisphere (range [median], 40.0%–64.4% [50.0%] and 32.4%–70.2% [46.8%], respectively) were significantly higher than those in the contralateral hemisphere (39.2%–62.1% [47.6%] and 30.5%–59.2% [44.2%]; P=0.004 and P<0.001, respectively). In addition, there was a significant correlation of the OEF values in both hemispheres between the QSM and PET in 11 patients with a substantial increase in PET-OEF ratio >1.09 (r=0.64; P=0.001; y=0.41x+29.2) although there was no significant correlation in the remaining 28 patients (r=−0.19; P=0.17; y=−0.12x+54.2; Figure 2).
Regarding the OEF ratio of the affected hemisphere against the contralateral side, a good correlation was found between the QSM-OEF maps and the PET-OEF maps (r=0.69; P<0.001; y=0.53x+0.48; Figure 3). In the patients with an increased PET-OEF ratio, an excellent correlation was observed between the images (r=0.79; P=0.004; y=0.53x+0.47), whereas there was no significant correlation in other patients (r=0.21; P=0.27). Bland–Altman analysis between QSM-OEF and PET-OEF ratios showed that, although most all of the values are within mean±2SD, the constant bias was observed in the patients with increased PET-OEF ratios (mean difference, −0.11; 95% confidence interval, −0.17 to −0.06) while the proportional bias was observed in the patients with maintained PET-OEF ratios (slope of regression equation, 1.17 [P<0.001]; Figure 4), indicating the suboptimal agreements.
There were no significant differences in the QSM-OEF and PET-OEF values between symptomatic and asymptomatic patients (P=0.94 and 0.47, respectively; Mann–Whitney U test) and among the patients with occlusion/stenosis of ICA/MCA (P=0.11–1.00 and 0.08–1.00, respectively; Steel-Dwass test). The QSM-PET differences in the OEF ratios showed no significant correlations with the degree of stenosis in the patients with MCA stenosis (r=0.004; P=0.99).
The receiver operating characteristic analysis showed that the area under the curve of QSM-OEF ratios against increased PET-OEF ratios was 0.85 (95% confidence interval, 0.72–0.99) and the sensitivity and specificity were 0.82 (9 of 11; 95% confidence interval, 0.48–0.98) and 0.86 (24 of 28; 95% confidence interval, 0.67–0.96), respectively, when the cut-off value was set as 1.09 (Figure 3).
In this study, we assessed accuracies of the QSM-OEF at 7T by comparing results with PET-OEF in patients with unilateral major vessel stenosis or occlusion and revealed that the QSM-OEF ratio of the ipsilateral MCA territory against the contralateral side showed a good correlation with PET-OEF and readily discriminated patients with increased OEF from others. These results were comparable to those of QSM-OEF at 3T in a previous report,22 indicating the potential use of the QSM-based OEF estimation to evaluate OEF abnormalities, including misery perfusion as a noninvasive alternative to 15O2-PET and other invasive methods.
Various MRI approaches that share some theoretical similarities have been used for OEF estimation.5 When compared with the methods reported previously,6–16 the QSM-based method we used has several advantages, such as usage of conventional sequence; short acquisition time; no need for any challenge, contrast agent, or other invasive procedures; sufficient spatial resolution with whole-brain coverage; and robustness to low perfusion status, suggesting high availability in clinical practice and clinical studies for patients with cerebrovascular and other neurological disorders. However, this method highly depends on the algorithms for generating QSM images and for estimating OEF values. Several algorithms for QSM generation were proposed,26,34–38 and they vary in terms of preservation of small venous structures, which seems crucial to obtain accurate OEF values. The software for estimating OEF values from QSM images has remained as in-house programs that need further revisions to distribute as free software programs. Further optimization of the algorithm and parameters, as well as publication of the program, is needed for wide adoption of the QSM-OEF method.
In this study, we used a 7T scanner that yields profound susceptibility effects to improve accuracies for estimating OEF values. Against our expectations, however, the QSM-OEF at 7T in this study achieved only a slight improvement in terms of the correlation coefficient and the sensitivity/specificity for PET-OEF and included substantial systematic biases in terms of the agreements, when compared with that at 3T in the previous report.22 This issue can be mainly attributed to the relatively low spatial resolution of the source images, which were comparable to that at 3T in the previous study.22 Although the BOLD effect is much stronger at 7T than at 3T, we presumably overlooked susceptibility information of minute venous structures because of the low resolution of the images. In addition, the results can be affected by the differences in the cohorts, ROIs, and post-processing methods from the previous study.22 Moreover, regarding the absolute OEF values, we found only a fair correlation between QSM and PET. Besides the aforementioned reasons, this issue can be partly explained by the theoretical discrepancy that QSM-OEF reflects oxygenation within venous structures, whereas PET-OEF mainly reflects that within brain parenchyma. Direct comparisons between 7T and 3T images in the same patients with the identical post-processing and analysis methods are needed to clarify potential advantages of the 7T system.
We generated OEF maps only from the QSM images at resting state in this study. However, the QSM-based method can be applied for paradigms using cerebrovascular challenges, which were proposed by the previous studies.6–10 These methods can minimize influences by paramagnetic effects because of iron deposition of deep nuclei, hemosiderin deposition, and neighboring bone/air structures, which enabled selective assessment of susceptibility changes in response to external stimuli. By using the paradigms, we may improve accuracies of OEF estimation although the availability may be compromised because of invasiveness and long examination times.
This study had several limitations. First, we performed no comparisons with clinical outcomes of the patients because of the relatively small sample size of the heterogeneous cohort. Hence, we did not fully determine the clinical significance of the method we used although we demonstrated accurate detection of substantial increases in OEF of affected cerebral hemispheres. Second, we did not compare our method with the other methods previously reported, such as paradigms using cerebrovascular challenges, T2*/T2’ relaxation measurements, spin-tagging T2 relaxation measurements, and phase difference measurements. Therefore, whether the QSM-based OEF estimation is more accurate than the other methods remains unknown. Third, we did not compare OEF values between 7T and 3T or between QSM algorithms so that advantages of the 7T scanner and algorithm we used remain unclear. Furthermore, we did not examine the effects of spatial resolution of source images on OEF accuracies although we consider spatial resolution crucial even at 7T. To overcome the issues, further technical improvements and optimization, as well as prospective studies with larger sample sizes, are needed, some of which are ongoing.
The OEF obtained by QSM at 7T was well correlated with that obtained by 15O2-PET in terms of the ratio of affected/nonaffected sides in patients with unilateral ICA/MCA steno-occlusive lesions and enabled accurate evaluation of the substantial OEF increase, which suggests that noninvasive OEF measurements based on the QSM technique can be used as a substitute for PET for assessment of chronic ischemia and other cerebrovascular disorders.
Sources of Funding
This work was partly supported by a Grant-in-Aid for Strategic Medical Science Research (S1491001, 2014–2018) and Grants-in-Aid for Scientific Research (grant no. 16K10798) from the Ministry of Education, Culture, Sports, Science and Technology of Japan, as well as by Grain-in-Aid from Senshin Medical Research Foundation.
R. Sato is an employee of Hitachi Ltd. The other authors report no conflicts.
- Received January 12, 2017.
- Revision received April 18, 2017.
- Accepted May 30, 2017.
- © 2017 American Heart Association, Inc.
- Baron JC,
- Bousser MG,
- Rey A,
- Guillard A,
- Comar D,
- Castaigne P
- Yamauchi H,
- Higashi T,
- Kagawa S,
- Nishii R,
- Kudo T,
- Sugimoto K,
- et al
- Christen T,
- Bolar DS,
- Zaharchuk G
- Goodwin JA,
- Kudo K,
- Shinohe Y,
- Higuchi S,
- Uwano I,
- Yamashita F,
- et al
- Xie S,
- Hui LH,
- Xiao JX,
- Zhang XD,
- Peng Q
- Liu Z,
- Li Y
- 23.↵North American Symptomatic Carotid Endarterectomy Trial. Methods, patient characteristics, and progress. Stroke. 1991;22:711–720.
- Kwon SU,
- Cho YJ,
- Koo JS,
- Bae HJ,
- Lee YS,
- Hong KS,
- et al
- Sato R,
- Shirai T,
- Taniguchi Y,
- Murase T,
- Bito Y,
- Ochi H
- Chida K,
- Ogasawara K,
- Kuroda H,
- Aso K,
- Kobayashi M,
- Fujiwara S,
- et al
- Liu T,
- Spincemaille P,
- de Rochefort L,
- Kressler B,
- Wang Y