Mapping Cerebrovascular Reactivity Using Blood Oxygen Level-Dependent MRI in Patients With Arterial Steno-occlusive Disease
Comparison With Arterial Spin Labeling MRI
Background and Purpose— Blood oxygen level-dependent MRI (BOLD MRI) of hypercapnia-induced changes in cerebral blood flow is an emerging technique for mapping cerebrovascular reactivity (CVR). BOLD MRI signal reflects cerebral blood flow, but also depends on cerebral blood volume, cerebral metabolic rate, arterial oxygenation, and hematocrit. The purpose of this study was to determine whether, in patients with stenoocclusive disease, the BOLD MRI signal response to hypercapnia is directly related to changes in cerebral blood flow.
Methods— Thirty-eight patients with steno-occlusive disease underwent mapping of CVR by both BOLD MRI and arterial spin labeling MRI. The latter technique was used as a reference standard for measurement of cerebral blood flow changes.
Results— Hemispheric CVR measured by BOLD MRI was significantly correlated with that measured by arterial spin labeling MRI for both gray matter (R=0.83, P<0.0001) and white matter (R=0.80, P<0.0001). Diagnostic accuracy (area under receiver operating characteristic curve) for BOLD MRI discrimination between normal and abnormal hemispheric CVR was 0.90 (95% CI=0.81 to 0.98; P<0.001) for gray matter and 0.82 (95% CI=0.70 to 0.94; P<0.001) for white matter. Regions of paradoxical CVR on BOLD MRI had a moderate predictive value (14 of 19 hemispheres) for spatially corresponding paradoxical CVR on arterial spin labeling MRI. Complete absence of paradoxical CVR on BOLD MRI had a high predictive value (31 of 31 hemispheres) for corresponding nonparadoxical CVR on arterial spin labeling MRI.
Conclusions— Arterial spin labeling MRI confirms that, even in patients with stenoocclusive disease, the BOLD MRI signal response to hypercapnia predominantly reflects changes in cerebral blood flow.
An emerging technique for mapping cerebrovascular reactivity (CVR) is blood oxygen level-dependent MRI (BOLD MRI)1 of changes in cerebral blood flow (CBF) during manipulation of end-tidal partial pressure of carbon dioxide (Petco2).2 This method is clinically attractive because it uses a pulse sequence routinely available on MR scanners, maps the entire brain with high spatial resolution, requires less than 10 minutes to perform, and generates quantitative results rather than simply an interhemispheric difference. The major concern3,4 with this technique has been that BOLD MRI signal depends on CBF, but also (and to an unknown extent) on other factors: cerebral blood volume (CBV), cerebral metabolic rate of oxygen consumption, arterial partial pressure of oxygen (Pao2), and hematocrit.5 There is empirical evidence that, in healthy subjects, the BOLD MRI signal response to changes in Petco2 is dominated by CBF effects,6,7 but there is limited literature on this relationship in patients with cerebrovascular disease.
The purpose of this study was to determine whether, in patients with stenoocclusive cerebrovascular disease, the BOLD MRI signal response to hypercapnia is directly related to changes in CBF. We used arterial spin labeling (ASL) MRI as a reference standard for measurement of CBF changes. In its most basic form, ASL MRI uses radiofrequency pulses to magnetically label the water protons in blood flowing into the imaging plane in the brain. These modified protons then act as an endogenous contrast agent causing an MRI signal change that is proportional to CBF. Notably, ASL MRI does not have the CBV, cerebral metabolic rate of oxygen consumption, Pao2, and hematocrit dependence of BOLD MRI.8,9 Also, ASL MRI offers sufficient temporal resolution to allow direct comparison with our echoplanar imaging-based BOLD MRI protocol.
The local ethics review board approved the study protocol. A total of 38 patients with steno-occlusive cerebrovascular disease, referred from the neurology and neurosurgery services at the Toronto Western Hospital, were recruited into this study. All patients provided written informed consent.
Patients were fitted with a sequential gas delivery mask (Hi-Ox-80; Viasys HealthCare, Yorba Linda, Calif) with a rebreathing bag on the expiratory limb. To provide an airtight seal, the mask was taped to the face (3 mol/L; Tegaderm, St. Paul, Minn). Gas flow to the mask was controlled by a gas blender (RespirAct; Thornhill Research, Toronto, Canada) programmed to provide the flow and blend of o2, N2, and co2 needed to attain the target Petco2 and Peto2. The gas sequence is described in Table 1. Tidal pco2 and po2 were monitored continuously (RespirAct), digitized, and recorded (LabView; National Instruments Corporation, Austin, Texas). The apparatus and technique are described in greater detail elsewhere.10
MRI was performed on a 3.0-Tesla scanner (Signa; GE Healthcare, Milwaukee, Wis) with an 8-channel phased array head coil. T1-weighted anatomic images were acquired using a 3-dimensional spoiled gradient echo pulse sequence (whole brain coverage; matrix: 256×256; slice thickness: 2.2 mm; no interslice gap). BOLD MRI data were acquired with a T2*-weighted single-shot gradient echo pulse sequence with echoplanar readout (field of view: 24×24 cm; matrix: 64×64; TR: 2000 ms; TE: 30 ms; flip angle: 85°; slice thickness: 5.0 mm; interslice gap: 2.0 mm, number of frames: 254). ASL MRI data were then acquired using a 2-dimensional spin-echo flow-sensitive alternating inversion recovery (FAIR) sequence11 with echoplanar readout (field of view: 24×24 cm; matrix: 64×64; TR: 2000 ms; TE: 22.7 ms; TI: 1000 ms; flip angle: 85°; slice thickness: 5.0 mm; interslice gap: 2.0 mm, number of frames: 254). The FAIR sequence applies alternating slice-selective and nonselective 180° pulses to acquire alternating flow-encoded and nonflow-encoded images. The FAIR sequence has limited z-axis spatial coverage. We acquired 5 axial ASL MRI slices with the middle slice centered on the bodies of the lateral ventricles. The BOLD MRI and ASL MRI acquisitions were prescribed with matching slice locations.
In addition to the CVR data acquired for this study, all patients had previously undergone routine clinical imaging of the neck vessels and circle of Willis. This was by catheter angiography (10 of 25), MR angiography (9 of 25), CT angiography (4 of 25), or ultrasound (2 of 25).
Petco2 and Peto2 values were selected automatically (LabView) from the continuous pco2 and po2 waveforms as the highest and lowest values, respectively, during exhalation. All values were confirmed by visual inspection. MRI and Petco2 data were then imported into the software AFNI.12 A plot of head position versus time for each ASL MRI data set was used to excluded those patients (13 of 38) with head displacement of one half voxel width or greater along any of 3 orthogonal axes. Each patient’s BOLD MRI data set was temporally shifted to the point of maximum statistical correlation with the patient’s Petco2 waveform. The BOLD MRI signal time course underwent least squares fitting to the Petco2 waveform on a voxel-by-voxel basis, and 2 parameters were generated: BOLDΔ=ΔBOLD MRI signal per ΔpETco2 and BOLDBaseline. Anatomic images were manually segmented into right and left hemispheres and then automatically segmented into gray matter and white matter (SPM5; Wellcome Department of Imaging Neuroscience, Institute of Neurology, University College, London, UK). These anatomic masks were used to segment the BOLD and ASL MRI data sets. BOLDΔ and BOLDBaseline were summed for each segment, and CVR was calculated for each segment as (100*total BOLDΔ)/total BOLDBaseline. Analysis of the ASL MRI data was identical with one additional step: for each pair of slice-selective and nonslice-selective images, the signal difference between the FAIR MRI images was calculated on a voxelwise basis, and it was the resulting difference map that was then fit to the Petco2 waveform to calculate CVR.11
We generated a scatterplot of hemispheric CVR measured using BOLD MRI versus ASL MRI. To quantify the correlation between the BOLD and ASL measurements, while accounting for repeated (2) measurements on each subject, we used a random effects model. First, we fitted a model with only a random intercept for each subject and then fitted a model that also included a fixed effect for ASL. Using the residual variances from these 2 models, we computed a pseudo-R2, which represents the proportional reduction in residual variance in BOLD that can be explained by ASL.
To establish a threshold for differentiating between normal and abnormal on the reference standard (ASL MRI), we calculated mean hemispheric CVR on ASL MRI across all hemispheres with ipsilaterally normal angiography and defined abnormal as 2 or more SDs below the mean. Using this threshold, we generated receiver operating characteristic curves for BOLD MRI measurement of hemispheric reactivity in the gray matter and white matter.
Because paradoxical CVR (hypercapnia-induced decrease in CBF) has particularly strong prognostic implications,13,14 we sought to determine whether regions of paradoxical CVR on BOLD MRI correspond with paradoxical CVR on ASL MRI. Due to image noise, there are inevitably a small number of voxels with paradoxical CVR even in normal subjects. A threshold was therefore established to differentiate a small number of paradoxical voxels due to image noise from truly paradoxical CVR. For all hemispheres with ipsilaterally normal angiography, we counted the number of paradoxical voxels in each hemisphere on BOLD MRI as a proportion of the total number of voxels in the hemisphere. We then calculated the mean of this proportion across all hemispheres and defined significant paradoxical CVR as 2 or more SDs above the mean. Mean CVR on ASL MRI was calculated for the region corresponding with paradoxical CVR on BOLD MRI for each hemisphere. Similarly, mean CVR on ASL MRI was calculated for the region corresponding with nonparadoxical CVR on BOLD MRI. In addition, differences in the spatial extent of paradoxical CVR on BOLD MRI versus ASL MRI were assessed by performing a paired t test of the proportion of voxels with a paradoxical response in each hemisphere for the 2 techniques.
All Petco2 stages were attained with standard deviation of 2 mm Hg or less, and Peto2 was maintained with standard deviation of 3 mm Hg over all gas stages. There was no significant difference in end-tidal gas concentrations for the BOLD MRI versus ASL MRI acquisitions (Table 3).
Correlation Between Blood Oxygen Level- Dependent MRI and Arterial Spin Labeling MRI Measurement of Hemispheric Cerebrovascular Reactivity
Hemispheric CVR measured by BOLD MRI was significantly correlated with that measured by ASL MRI for both gray matter (R=0.83, P<0.0001) and white matter (R=0.80, P<0.0001; Figure 1). There was only a single hemisphere (case 19) with paradoxical hemispheric gray matter CVR on BOLD MRI, whereas 7 hemispheres (cases 4, 6, 12, 13, and 24) had paradoxical hemispheric gray matter CVR on ASL MRI.
Accuracy of Blood Oxygen Level-Dependent MRI Diagnosis of Impaired Hemispheric Cerebrovascular Reactivity
For the 11 hemispheres with ipsilateral normal angiography, mean hemispheric CVR on ASL MRI (%ΔCBF/mm ΔPetco2) was 5.70% (95% CI=2.67 to 8.73) for gray matter and 5.48% (95% CI=1.16 to 9.81) for white matter. Defining 2 or more SDs below the mean as abnormal resulted in 23 of 50 and 15 of 50 hemispheres categorized as having reduced CVR on ASL MRI for gray and white matter, respectively. Figure 2 shows receiver operating characteristic plots for BOLD MRI diagnosis of impaired CVR using ASL MRI CVR as a reference standard. The areas under the curves are 0.90 (95% CI=0.81 to 0.98; P<0.001) for gray matter and 0.82 (95% CI=0.70 to 0.94; P<0.001) for white matter.
Region-of-Interest Analysis of Paradoxical Cerebrovascular Reactivity
For the 11 hemispheres with ipsilateral normal angiography, the mean proportion of voxels with a paradoxical response on BOLD MRI was 2% (95% CI=0% to 6%). Defining significant paradoxical CVR as 2 or more SDs above this mean yielded 19 of 50 hemispheres containing a region of significant paradoxical CVR on BOLD MRI. For these regions of paradoxical CVR, corresponding regions on ASL MRI showed paradoxical CVR in 14 of 19 of the hemispheres. For the regions of nonparadoxical CVR on BOLD MRI in these same 19 hemispheres, corresponding regions on ASL MRI showed nonparadoxical CVR in 15 of 19 of the hemispheres. There were 31 of 50 hemispheres with no regions of paradoxical CVR on BOLD MRI, and ASL MRI was nonparadoxical in all of these 31 hemispheres.
Of the 7 patients with paradoxical hemispheric CVR on ASL MRI, none had net paradoxical hemispheric CVR on BOLD MRI. However, 7 of 7 had subhemispheric foci of significant paradoxical CVR on BOLD MRI. Similarly, the one patient with paradoxical hemispheric CVR on BOLD MRI had nonparadoxical hemispheric CVR on ASL MRI but did have a region of significant paradoxical CVR on ASL MRI.
Spatial Extent of Paradoxical Cerebrovascular Reactivity on Blood Oxygen Level-Dependent MRI Versus Arterial Spin Labeling MRI
For the 11 hemispheres with normal angiography, the proportion of voxels with a paradoxical response on BOLD MRI (mean=8%, SD=4%) was not significantly different from that on ASL MRI (mean=11%, SD=8%; paired t test, P=0.282). For the remaining 39 hemispheres, the mean proportion of voxels with a paradoxical response on BOLD MRI (mean=23%, SD=15%) was significantly less than that on ASL MRI (mean=54%, SD=51%; paired t test, P<0.001). Figure 3 is a representative case (case 24) illustrating this difference.
Our results demonstrate a strong correlation between hemispheric CVR measured using BOLD MRI and ASL MRI. This suggests that even in patients with steno-occlusive disease, the BOLD MRI signal response to hypercapnia is directly related to changes in CBF. Previous studies did not allow this conclusion. Shiino et al7 studied 10 patients with cerebrovascular disease and found a significant correlation (R=0.698, P<0.0001) between CVR measured by BOLD MRI (with breathholding as the vasodilatory stimulus) and I-123 single photon emission CT (with acetazolamide as the vasodilatory stimulus). However, the BOLD and single photon emission CT measurements are difficult to compare because breathholding and acetazolamide injection are physiologically different stimuli, and also, the 2 measurements were performed at different times. In contrast, we have used identical vasodilatory stimuli for both BOLD and ASL MRI, and the 2 types of imaging were performed in the same session, within minutes of each other. Lythgoe et al15 studied 16 patients with unilateral carotid artery stenosis or occlusion and found no significant correlation between BOLD MRI and transcranial Doppler ultrasound responses to co2 provocation. However, Ziyeh et al6 later performed a similar comparison of BOLD MRI and transcranial Doppler in 27 patients and reported a significant correlation (R=0.71, P<0.001).
Furthermore, our findings support the hypothesis that BOLD MRI can accurately discriminate between normal and abnormal hemispheric CVR. Diagnostic accuracy was reflected in the large areas under the receiver operating characteristic curves for both gray and white matter (Figure 2). We found that regions of paradoxical CVR on BOLD MRI had a moderate positive predictive value (73%) for true paradoxical CVR. Importantly, complete lack of paradoxical CVR on BOLD MRI had a high negative predictive value (100%).
Our results might be criticized on the grounds that ASL MRI is not as well established as other techniques for measuring CBF. Our decision to use ASL MRI arose from a need to map CBF with comparable spatial and temporal resolution to the BOLD MRI technique. Although transcranial Doppler is clinically the most commonly used method for measuring CVR, it offers no spatial resolution and only assesses the middle cerebral artery territory. Also, it has limited diagnostic accuracy.16 H215O positron emission tomography is considered the most accurate method of measuring CBF in vivo. However, H215O has a 123-second half-life necessitating a 10- to 15-minute delay between Petco2 states to allow for decay of the radiopharmaceutical between measurements of CBF. This precludes direct comparison with our BOLD MRI protocol that involves relatively rapid changes in Petco2 (to minimize study duration). Single photon emission CT also requires a delay between measurements. In contrast with these other techniques, ASL MRI has comparable spatial and temporal resolution to BOLD MRI. FAIR MRI measurement of CBF is accurate when compared with positron emission tomography,17 and in the functional MRI literature, FAIR has been used as a reference standard for measurement of CBF.9 FAIR has also been validated for measuring CVR in normal subjects18 and in patients with steno-occlusive disease.19 Currently, most ASL MRI pulse sequences have limited spatial coverage, some require specialized hardware, and few are widely available. However, ASL MRI research is progressing rapidly, and this technique may eventually prove better than BOLD MRI for mapping CVR.
Our study revealed several limitations of using BOLD MRI signal to map CVR. First, the predictive value of paradoxical CVR on BOLD MRI was only moderate. Second, BOLD MRI underestimated the spatial extent of paradoxical CVR. We may gain insight into these findings by considering the nature of BOLD MRI signal. BOLD MRI signal depends on the intravoxel concentration of deoxyhemoglobin. This concentration is determined by the fraction of the voxel occupied by blood (CBV), the concentration of hemoglobin in blood (hematocrit), the concentration of dissolved oxygen in blood (Pao2), the rate of inflow of fresh blood and outflow of deoxygenated blood (CBF), and the rate of oxygen use by tissues (cerebral metabolic rate of oxygen consumption).
In normal subjects, hypercapnia induces an increase in both CBF and CBV. The relationship between ΔCBF and ΔCBV was determined empirically by Grubb20: ΔCBV=ΔCBF0.38. Through inflow of fresh blood and washout of deoxygenated blood, increased CBF reduces the concentration of deoxyhemoglobin in blood, increasing BOLR MRI signal. For a given concentration of deoxyhemoglobin in blood, increased CBV increases the fraction of the imaging voxel occupied by blood, reducing BOLD MRI signal. Despite these opposing influences, hypercapnia9 and neuronal activation21 each induce an increase in BOLD MR signal that is linearly related to the induced change in CBF. This indicates that in normal subjects, the BOLD MRI signal response to hypercapnia is dominated by flow effects and not blood volume effects.
In patients with cerebrovascular disease, the relationship between hypercapnia and CBF is well known. However, there is little literature on the relationships between hypercapnia and CBV and hypercapnia and BOLD MRI signal. Sabatini et al22 found that acetazolamide-induced ΔCBV/ΔCBF was greater in those with carotid occlusion than in healthy volunteers, suggesting that Grubb’s relationship is not valid in this patient population. Okazawa et al23 reproduced these findings and, of particular relevance, also described the acetazolamide-induced ΔCBV and ΔCBF for a subset of patients (4 of 16) with paradoxical CVR. These latter patients showed an acetazolamide-induced increase in CBV and decrease in CBF. Because an increase in CBV and a decrease in CBF both reduce BOLD MR signal, these changes do not explain why BOLD MRI underestimated the spatial extent of paradoxical CVR in our study. However, the Okazawa results are derived from only 4 patients, and only mean changes in CBV and CBF for entire hemispheres were measured. We anticipate that future studies that map, with high spatial resolution, hypercapnia-induced changes in both CBF and CBV in patients with steno-occlusive disease would help resolve these issues.
We have considered changes in CBV as a possible confounder in BOLD MRI mapping of CVR. Pao2, hematocrit, and cerebral metabolic rate of oxygen consumption are additional confounders to consider. When a hypercapnic stimulus is applied, the resulting hyperventilation results in a tandem increase in arterial po2, reducing deoxyhemoglobin concentration, causing an increase in BOLD MR signal independent of any change in CBF. Our method maintains Peto2 at normoxia within very narrow limits (mean=100 mm Hg, SD 2 mm Hg),10 independent of minute ventilation and the target Petco2, and thus removes variation in arterial po2 as a confounding variable. Hematocrit is also an unlikely confounder because one would not expect a hypercapnia-induced change in hematocrit during the course of a CVR acquisition. Similarly, hypercapnia does not affect cerebral metabolic rate of oxygen consumption.24,25
Another potential confounder is the time delay between change in partial pressure of carbon dioxide in the pulmonary capillaries and arrival of this change in a given voxel in the brain. This time delay is spatially heterogeneous.26 If the relative difference in time delay between 2 regions of brain was large enough, then theoretically, the changes in arterial pco2 in one region of brain could be 180° out of phase with the changes elsewhere, mimicking a steal phenomenon.27 We have used a nonperiodic Petco2 stimulus to avoid this possibility.
ASL MRI confirms that even in patients with steno-occlusive disease, the BOLD MRI signal response to hypercapnia predominantly reflects changes in CBF. This result contributes to the validation of BOLD MR mapping of CVR.
We thank Marat Slessarev and Alex Vesely for their contributions to the development of the breathing apparatus. We thank the Toronto Western Hospital MRI technologists, particularly Garry Detzler, Eugen Hlasny, David Johnstone, and Keith Ta, for their contributions to the data acquisition. We thank George Tomlinson for his advice on the statistical analysis. We would also like to acknowledge the support of the Radiologist Scientist Training Program, Department of Medical Imaging, University of Toronto.
Two of the study authors (JAF, DJM) contributed to the develop of the Respiract, a device used in the second experiment of the study. These authors stand to gain financially if the device is successfully commercialized by Thornhill Research Inc, a University of Toronto/University Health Network-related company.
- Received October 28, 2007.
- Revision received November 28, 2007.
- Accepted December 12, 2007.
Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci U S A. 1990; 87: 9868–9872.
Zaharchuk G, Mandeville JB, Bogdanov AA Jr, Weissleder R, Rosen BR, Marota JJ. Cerebrovascular dynamics of autoregulation and hypoperfusion. An MRI study of CBF and changes in total and microvascular cerebral blood volume during hemorrhagic hypotension. Stroke. 1999; 30: 2197–2204;discussion 2204–2195.
Ziyeh S, Rick J, Reinhard M, Hetzel A, Mader I, Speck O. Blood oxygen level-dependent MRI of cerebral co2 reactivity in severe carotid stenosis and occlusion. Stroke. 2005; 36: 751–756.
Petersen ET, Zimine I, Ho YC, Golay X, Petersen ET, Zimine I, Ho YCL, Golay X. Non-invasive measurement of perfusion: a critical review of arterial spin labelling techniques. Br J Radiol. 2006; 79: 688–701.
Webster MW, Makaroun MS, Steed DL, Smith HA, Johnson DW, Yonas H. Compromised cerebral blood flow reactivity is a predictor of stroke in patients with symptomatic carotid artery occlusive disease. J Vasc Surg. 1995; 21: 338–344;discussion 344–335.
Pindzola RR, Balzer JR, Nemoto EM, Goldstein S, Yonas H. Cerebrovascular reserve in patients with carotid occlusive disease assessed by stable xenon-enhanced CT cerebral blood flow and transcranial Doppler. Stroke. 2001; 32: 1811–1817.
Arbab AS, Aoki S, Toyama K, Miyazawa N, Kumagai H, Umeda T, Arai T, Araki T, Kabasawa H, Takahashi Y. Quantitative measurement of regional cerebral blood flow with flow-sensitive alternating inversion recovery imaging: comparison with [iodine 123]-iodoamphetamin single photon emission CT. AJNR Am J Neuroradiol. 2002; 23: 381–388.
Grubb RL Jr, Raichle ME, Eichling JO, Ter-Pogossian MM, Grubb RL Jr, Raichle ME, Eichling JO, Ter-Pogossian MM. The effects of changes in paco2 on cerebral blood volume, blood flow, and vascular mean transit time. Stroke. 1974; 5: 630–639.
Zhu XH, Kim SG, Andersen P, Ogawa S, Uurbil K, Chen W. Simultaneous oxygenation and perfusion imaging study of functional activity in primary visual cortex at different visual stimulation frequency: quantitative correlation between BOLD and CBF changes. Magn Reson Med. 1998; 40: 703–711.
Sabatini U, Celsis P, Viallard G, Rascol A, Marc-Vergnes JP. Quantitative assessment of cerebral blood volume by single-photon emission computed tomography. Stroke. 1991; 22: 324–330.
Okazawa H, Yamauchi H, Sugimoto K, Takahashi M. Differences in vasodilatory capacity and changes in cerebral blood flow induced by acetazolamide in patients with cerebrovascular disease. J Nucl Med. 2003; 44: 1371–1378.
Alberti E, Hoyer S, Hamer J, Stoeckel H, Packschiess P, Weinhardt F. The effect of carbon dioxide on cerebral blood flow and cerebral metabolism in dogs. Br J Anaesth. 1975; 47: 941–947.