- Email: [email protected]
Mapping of cerebrovascular reactivity using bold magnetic resonance imaging
Magnetic Resonance Imaging, Vol. 17, No. 4, pp. 495–502, 1999 © 1999 Elsevier Science Inc. All rights reserved. Printed in the USA. 0730-725X/99 $–see front matter
PII S0730-725X(98)00211-2
● Original Contribution
MAPPING OF CEREBROVASCULAR REACTIVITY USING BOLD MAGNETIC RESONANCE IMAGING DAVID J. LYTHGOE,* STEVE C.R. WILLIAMS,*† MARISA CULLINANE,*
AND
HUGH S. MARKUS*
*Department of Clinical Neurosciences and †Neuroimaging, King’s College School of Medicine and Dentistry and the Institute of Psychiatry, London, UK Blood oxygen level-dependent (BOLD) contrast MRI is a simple non-invasive method of estimating “perfusion,” and combined with a vasodilatory stimulus, may allow estimation of cerebral vascular reserve. We compared BOLD carbon dioxide (CO2) reactivity in the middle cerebral artery (MCA) perfusion territory to MCA flow velocity reactivity determined using transcranial Doppler ultrasound (TCD) in 16 patients with unilateral carotid artery stenosis or occlusion. Both BOLD and TCD reactivities were calculated from measurements acquired when the subjects were breathing air, and again when breathing a 6% CO2/air mixture, and were normalized by dividing by the difference in end tidal (ET) CO2. There was a significant correlation between interhemispheric MCA reactivity difference (contralateral–ipsilateral to the stenosis or occlusion) determined by BOLD MRI and TCD (r 5 0.75, p < 0.001). In contrast, treating each hemisphere individually, there was no correlation between the absolute BOLD and TCD MCA CO2 reactivities (r 5 0.08, p 5 0.670). This appeared to be due to a variable BOLD signal change in the non-stenosed hemisphere between subjects, with little change in the normal hemisphere of a few subjects. In one patient, focal regions of reduced reactivity were seen in non-infarcted regions of the stenosed hemisphere, in the borderzones between arterial territories. BOLD reactivity maps provide information on the whole MCA territory reactivity, and may identify small regions of impaired reactivity which are not detected using TCD. However, BOLD reactivity maps only appear to provide semi-quantitative rather than quantitative data. © 1999 Elsevier Science Inc. Keywords: Carotid artery diseases; Cerebrovascular circulation; BOLD.
INTRODUCTION
determined and therefore impaired reactivity in small regions of the MCA territory cannot be identified. Alternative techniques of estimating regional cerebral blood flow (CBF) provide the necessary spatial resolution to determine small areas of reduced reactivity. Positron emission tomography provides information on hemodynamic compromise either by the determination of oxygen extraction fraction, or by carbon dioxide reactivity with regional CBF measurements before and after CO2 administration. However it is expensive and not widely available, limiting its clinical use. Xenon-inhalation computed tomography (Xe-CT) gives CBF measurements with good spatial resolution. Single photon emission tomography using the tracer Tc-HMPAO is more widely available and can be used to determine CO2 or acetazolamide reactivity but provides only a semi-quan-
Measurement of the cerebrovascular reactivity to hypercapnia or acetazolamide can be used to identify a subgroup of patients with either carotid artery stenosis (CAS) or occlusion who may be at increased risk of stroke.1,2 The technique has a number of potential clinical applications including identifying a subgroup of patients with asymptomatic CAS for carotid endarterectomy, and selecting a subgroup of patients with carotid occlusion who may benefit from revascularization procedures. Most studies are currently performed using transcranial Doppler ultrasound (TCD) to measure the flow velocities in the middle cerebral arteries (MCAs).3 This has the advantage of being cheap, portable, noninvasive and reproducible, but suffers from poor spatial resolution. Reactivity for the whole MCA territory is RECEIVED 8/15/98; ACCEPTED 11/13/98. Address correspondence to Dr. Steve C.R. Williams, Neuroimaging Research Group, Institute of Psychiatry, De Cre-
spigny Park, Denmark Hill, London, SE5 8AF. E-mail: [email protected] 495
496
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
titative estimate of perfusion. It can therefore be used to compare side-to-side difference in patients, but not to determine differences in quantitative reactivity between patients. Positron emission tomography, single photon emission tomography and Xe-CT suffer from the disadvantage that the patient is exposed to ionizing radiation, a particular problem when performing serial studies. Magnetic resonance imaging (MRI) offers a number of potential methods of estimating cerebral vascular reserve. One potential non-invasive method of indirectly estimating cerebral vascular reserve is the measurement of blood oxygen level dependent (BOLD) contrast4 with and without vasodilatory stimulus. This can be performed in less than five minutes using echo-planar imaging techniques. Both acetazolamide and hypercapnia increase CBF while having little effect on cerebral metabolism. Because of this, extraction of oxygen from the blood remains relatively constant, increasing the concentration of oxyhaemoglobin, thus reducing the concentration of deoxyhaemoglobin in the blood. This decrease in paramagnetic deoxyhaemoglobin leads to a rise in signal in T*2 -weighted MR images. This mechanism is also responsible for the increased signal seen during gradient echo based functional MR imaging of brain activation. Changes in BOLD signal have been used to detect ischemia and reperfusion in an animal model of stroke.5 BOLD signal increases due to hypercapnia and acetazolamide have also been reported in animal models of cerebrovascular disease.6,7 A previous pilot study in four patients with unilateral CAS indicated that BOLD MRI can demonstrate exhaustion of cerebrovascular reserve capacity; this study used acetazolamide as the vasodilatory stress.8 However there have been few studies of its use in man, and no comparisons of its use with other methods of determining cerebrovascular reactivity. Therefore, in this study we evaluated the use of the BOLD MRI technique to determine CO2 reactivity in a larger number of patients with carotid artery stenosis or occlusion, and compared the results with those obtained with the same subjects using TCD. In addition we determined whether the greater spatial resolution provided by BOLD MRI allowed detection of small regions of impaired reactivity in the MCA territory, which were not apparent using TCD. MATERIALS AND METHODS Subjects To validate the technique, and optimize the image acquisition parameters, we performed single-slice BOLD imaging with one normal volunteer (male, aged 30) during the administration of three cycles of air followed by a 6% CO2/air mixture. Robust changes in BOLD
signal were determined in a single region of interest (ROI) placed in the right MCA territory. We then studied sixteen patients (thirteen male, three female) with unilateral CAS greater than 70% or internal carotid artery occlusion. Carotid artery duplex was used to determine the degree of CAS, based on internal and common carotid artery peak systolic velocities. The degree of stenosis was 70 – 89% in five subjects, 90 –99% in seven, and 100% in four subjects. The CAS was symptomatic (ipsilateral stroke or TIA or amaurosis fugax within the last six months) in 10 patients, and asymptomatic in six patients. Seven patients had an ipsilateral carotid territory cerebral infarct on structural MR imaging. The mean age of the subjects was 67.3 years (SD 8.4 years, range 51– 81 years). Before the study, each subject gave informed written consent, and the experimental protocol was approved by the local hospital ethics committee. Imaging Protocol In all patients reactivity to 6% CO2 was determined using both BOLD MRI and TCD on separate occasions. For both studies end-tidal (ET) CO2 levels were continuously measured using the same infrared CO2 meter (Normocap 100, Datex Instrumentarium Corp., Helsinki, Finland). MRI BOLD imaging was performed on a 1.5-T GE Signa (GE Medical Systems, Milwaukee, WI), fitted with Advanced NMR Systems (Wilmington, MA, USA) hardware and software, giving echo-planar imaging capability. All patients were provided with a panic button, so that if they were anxious they could terminate the experiment at any time. An air or 6% CO2/air mixture was delivered to the subject via a mouthpiece. A nose clip was fitted to ensure the inhaled gas was derived solely from the mouthpiece. A one-way valve was used to separate inspired and expired gases during the whole procedure, and to allow sampling for ET CO2 levels. While breathing air, 10 consecutive non-contiguous T*2weighted images were acquired from each of 14 axial planes (TR 5 3s, TE 5 40 ms, u 5 90°, in-plane resolution 5 1.5 mm, slice thickness 5 5.0 mm and slice skip 5 0.5 mm). During image acquisition, ET CO2 measurements were recorded. When the acquisition was finished the subject inhaled the 6% CO2/air mixture, and 60 s after steady-state ET CO2 was attained (which usually took 1–2 min), the image acquisition sequence was repeated. MRI Analysis Following acquisition, the two data sets (breathing air, and breathing the 6% CO2/air mixture) were concate-
BOLD and carotid stenosis ● D.J. LYTHGOE
nated to form one image series of 20 consecutive images for each axial slice, with the 10 images obtained with the subject breathing air preceding the 10 images obtained with the subject breathing the 6% CO2/air mixture. Misregistration of the images produced during normocapnia, and those obtained during hypercapnia, can produce enhancements along the sulci, which may be confused with increased BOLD response associated with large blood vessels. In order to minimize this effect, we co-registered all the images prior to production of reactivity images. The time series images for each axial plane were combined into a 4D image consisting of 3D brain volume images at 20 consecutive time points. The sequential 3D volumes were co-aligned using software developed primarily for the analysis of fMRI data.9 Following realignment, calculated BOLD reactivity maps were produced, voxel-by-voxel, depicting the fractional increase in image intensity due to CO2 inhalation, normalized to the increase in ET CO2. This is described by Eq. 1, R5
H
JY
Iair/6% CO2 2 Iair Iair
@ET CO2 air/6% CO2 2 ET CO2 air#
(1)
where R is the reactivity, Iair and Iair/6%CO2 are the mean intensities at each voxel for the images acquired when the subjects were breathing air and the air/6% CO2 mixture respectively, and ET CO2 air and ET CO2 air/6%CO2 are the mean ET CO2 values recorded while the corresponding images were acquired. Mean T*2-weighted anatomic images were calculated from the ten co-aligned sequential images obtained when the subject was breathing air. Three anatomic slices (T*2-weighted images) centered at the level of the most superior aspect of the third ventricle, depicting the mean intensity at each location were displayed. ROIs were defined manually in each slice for the left and right middle cerebral artery territories, by an experienced operator. The regions of interest were then transferred to the calculated reactivity maps, and the mean reactivity and area of each region noted. Weighted mean reactivities (Rterr) for the three slices were calculated using Eq. 2. R terr 5
( slice R slice51. . .3 z A slice51. . .3 (slice51. . .3 Aslice51. . .3
(2)
Here Rslice is the reactivity for a particular slice, and Aslice is its area. In addition the same slices were visually analyzed by an experienced observer blinded to the side of stenosis to identify small areas of impaired reactivity within noninfarcted regions of the internal carotid artery territory, as determined by the absence of high signal on the corresponding reactivity image. In those images where
ET AL.
497
there was an infarct on the corresponding T*2-weighted image, a ROI was placed in the infarcted tissue on each of three slices in which it was present, and in the corresponding regions of the non-infarcted contralateral hemisphere. The mean BOLD reactivities were calculated for the infarcted and non-infarcted regions as described above for the MCA territories. Transcranial Doppler Analysis TCD measurements were performed using a commercially available TCD machine (Multi Dop 34, DWL Electronische GMbH) with a 2MHz probe. Continuous simultaneous bilateral MCA insonation was performed via the transtemporal windows. During the TCD reactivity tests, the subject was placed in a supine position. A mask was placed over the face of the subject to allow delivery of air or the CO2/air mixture, and sampling of ET CO2. The mask was held in place to allow the subject to acclimatize while breathing air, and then the left and right MCA flow velocities, and ET CO2 were recorded for approximately 30 s to obtain baseline readings. The 6% CO2/air mixture was then administered for 2–3 min, until steady state was reached for both the ET CO2 and MCA flow velocities, after which MCA velocity and ET CO2 were measured over a 30 s period. A reactivity index (R) was calculated, using Eq. 3, R5
H
JY
vair/6% CO2 2 vair vair
@ET CO2 air/6% CO2 2 ET CO2 air#
(3)
where nair and nair/6%CO2 are the blood velocities measured when the subjects were breathing air and the air/6% CO2 mixture respectively, and ET CO2 air and ET CO2 air/6%CO2 are the mean ET CO2 values recorded during the corresponding TCD blood velocity measurements. Statistical Analysis Using Student’s paired t-test, we compared both BOLD and TCD reactivities in the whole MCA territories ipsilateral and contralateral to the CAS. Comparisons of the MRI data with the TCD data were performed in two ways. First, data from both stenosed and nonstenosed hemispheres were pooled and the correlation between BOLD and TCD estimates of reactivity for each MCA territory determined using Pearson’s test. Secondly since different subjects might exhibit differing inherent BOLD reactivity to increased CO2 levels, the difference between reactivities in the non-stenosed and stenosed hemispheres was calculated for each subject. The interhemispheric reactivity difference for each subject estimated using BOLD was then correlated with that obtained using TCD. In addition BOLD reactivities in the
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
498
Fig. 1. BOLD imaging in the normal volunteer. Images were acquired while breathing air or while breathing the 6% CO2 in air. The paradigm was repeated three times. A reproducible rise in BOLD signal intensity is seen following 6% CO2 administration. Signal change is scaled relative to the mean signal in the first baseline period.
infarcted tissue were compared with BOLD reactivities in corresponding regions of non-infarcted tissue in the contra-lateral hemisphere using Student’s paired t-test. RESULTS In the normal volunteer, there was a reproducible rise in BOLD signal during each period of inspiration of the 6% CO2/air mixture, as shown in Fig. 1. In patients CO2 reactivity measured by TCD was significantly reduced in the stenosed compared with the contralateral MCA territory (mean 6 SD 5 13.9 6 7.1 vs. 19.9 6 7.3% kPa21; p 5 0.046). BOLD reactivity was also lower in the ipsilateral MCA territory but the difference was not significant (0.14 6 0.37% kPa21 vs. 0.17 6 0.41% kPa21, p 5 0.379). This was partly due to little increase in BOLD signal following CO2 administration in the normal hemisphere of some individuals; in three cases the BOLD signal change was , 0.1% kPa21. The rise in ET CO2 during 6% CO2 inspiration in these patients was similar to that seen in other patients during MR imaging (mean 6 SD, 1.43 6 0.77% kPa21 vs. 1.40 6 0.30% kPa21). When these non-responders were removed from the analysis the difference in BOLD signal change between the two hemispheres was: mean 6 SD stenosed 0.51 6 0.36% kPa21, non-stenosed 0.69 6 0.25% kPa21, p 5 0.1. There was a highly significant correlation (r 5 0.75, p , 0.001) between the inter-hemispheric MCA reactivity difference measured with BOLD and TCD as shown in Fig. 2. The data were best fitted by a linear relation-
ship. In contrast there was no correlation between BOLD and TCD reactivities as determined for individual hemispheres (r 5 0.08, p 5 0.67), as shown in Fig. 3. This appeared to be due to a variable BOLD signal change in the non-stenosed hemisphere between subjects. BOLD reactivity in infarcted tissue in the seven patients with infarcts was significantly reduced compared with the corresponding regions of non-infarcted tissue in the contralateral hemisphere (mean 6 SD 20.31 6 0.37% kPa21 vs. 0.35 6 0.41% kPa21, p 5 0.024). In all subjects visual analysis of the BOLD reactivity maps was performed. Regions of reduced reactivity were seen in areas of cerebral infarction as determined from the corresponding T*2-weighted image. In addition, in one patient focal regions of reduced BOLD signal intensity change were seen in the borderzones between arterial territories in the side with CAS, while no abnormalities were present in corresponding T*2-weighted images. This is illustrated in Fig. 4 along with the corresponding mean T*2-weighted image. The reduced BOLD response (Fig. 4b) appears to be in the watershed zones between middle and anterior cerebral artery (arrow 1) and middle and posterior cerebral artery (arrow 2) territories as indicated. Figure 4c shows another T*2-weighted image slice from the same subject indicating infarction in the white matter of the left hemisphere. In the corresponding BOLD reactivity image (Fig. 4d), this region shows reduced BOLD reactivity (arrow 3). The reduction in reactivity in a region of infarcted tissue in another patient is shown in Fig. 5.
BOLD and carotid stenosis ● D.J. LYTHGOE
ET AL.
499
Fig. 2. The relationship between MCA inter-hemispheric reactivity, normalized for change in ET CO2, as determined using BOLD MRI and TCD. There was a significant correlation; r 5 0.75, p , 0.001.
During inspiration of 6% CO2, end-tidal CO2 rose in all individuals. However, the mean (6 SD) rise in ET CO2 was greater during TCD than during BOLD MRI (2.42 6 0.64% kPa vs. 1.40 6 0.30% kPa, p 5 0.01). DISCUSSION Our study in the normal volunteer demonstrates that 6% CO2 administration results in a detectable and reproducible rise in BOLD MR signal. Our results in patients
with CAS demonstrate that BOLD MRI can be used to produce high spatial resolution CO2 reactivity maps. These may be used to provide information on whole MCA territory reactivity, and also to identify small areas of impaired reactivity which are not detectable using TCD. Consistent with this we found a highly significant correlation between the inter-hemispheric MCA reactivity difference as determined using TCD and BOLD MR imaging. However the information provided by the
Fig. 3. The relationship between reactivity, normalized for change in ET CO2, in individual MCA territories as determined by BOLD MRI and TCD. Stenosed and non-stenosed hemispheres are represented by different symbols for comparison. There was no significant correlation; r 5 0.08, p 5 0.67.
500
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
Fig. 4. Images from a patient with left carotid stenosis showing impaired reactivity in the regions of the left internal carotid artery territory. a, mean T*2-weighted image calculated from the 10 image series when the subject was breathing air. b, the corresponding normalised reactivity image. Region 1 and 2 show reduced reactivity in the border zones between the left MCA and anterior cerebral artery and between the left MCA and posterior cerebral artery territories, respectively. c, mean T*2weighted image from a more superior plane in the same individual, showing an infarcted region in the white matter of the left hemisphere. d, the corresponding reactivity map. Region 3 indicates a region of reduced activity in the infarcted area.
BOLD reactivity maps presented in this study appears to be semi-quantitative rather than providing true quantitative data. This is reflected in the lack of any correlation
Fig. 5. Mean T*2-weighted image from a patient with left carotid stenosis and an old MCA territory infarct in the gray and white matter of the left hemisphere. b, the corresponding normalised BOLD reactivity image with a region of reduced reactivity in a location (arrow) corresponding to the infarct.
between absolute BOLD MCA territory reactivities and absolute TCD MCA reactivities. Analysis of the BOLD response to hypercapnia in the normal MCA territory ipsilateral to the non-stenosed carotid artery demonstrated that this was variable. In addition in three individuals there was a very small, or non-significant BOLD signal change in the contralateral hemisphere. In some subjects, the TCD and BOLD interhemispheric reactivity differences were negative. This is probably a reflection of the fact that although in normal subjects the reactivity is similar in both MCA territories, some degree of interhemispheric difference can be found. BOLD changes provide a much greater degree of spatial resolution than do TCD measurements, reflecting change on a pixel to pixel basis in comparison with flow in the whole middle cerebral artery territory. To account for this we determined the mean BOLD signal change over the whole of the middle cerebral artery territory on each of the brain slices analyzed. In regions of established infarction one would not expect to see a BOLD response to a CO2 challenge. Consistent with this, we found reduced BOLD CO2 reactivity in areas of established infarction. In addition in one subject we found regions of impaired reactivity in non-infarcted tissue ipsilateral to the CAS. The pattern of abnormality was striking with impaired reactivity both in the internal watershed areas, and in the watershed areas between the MCA and both the anterior cerebral and the posterior cerebral arteries. This lack of signal change in non-infarcted but hemodynamically compromised areas may be due to a combination of factors. First, a lack of increase in blood flow in response to a rise in inspired CO2 will not give an increase in blood oxygenation, with its corresponding signal change in BOLD images. Second, previous studies have suggested a trend toward increased oxygen extraction fraction associated with decreased hemodynamic reserve capacity in the borderzones between the MCA and ACA perfusion territories.10 This would lead to an increase in paramagnetic deoxyhaemoglobin, and a corresponding reduction in signal intensity. This regional information cannot be provided using TCD and is a major advantage of a technique with high spatial resolution such as BOLD MRI. Single photon emission tomography can provide similar information but with a lower spatial resolution, and in addition it involves longer acquisition times and the administration of radioactive substances. The latter limits its use particularly for serial studies such as administration before and during CO2 inhalation. Application of a short CO2 challenge allows us to determine the extent of BOLD reactivity throughout the brain. This may have important implications in functional MRI studies of brain activation, in pathologic cases where muted BOLD response is observed despite
BOLD and carotid stenosis ● D.J. LYTHGOE
satisfactory task performance. In such cases a lack of significant BOLD contrast due to exhausted cerebrovascular reserve may be misinterpreted as neuronal dysfunction. There are several possible reasons for the information provided by BOLD imaging being semi-quantitative rather than truly quantitative. First, problems in CO2 administration, such as poorly fitting mouth-pieces or masks, can result in inadequate hypercapnic stimulus and could account for the lack of BOLD signal change in three individuals, and the variable response in other individuals. We excluded this by continuously monitoring ET CO2; in all studies there was a rise in ET CO2. However, the mean rise in ET CO2 during inspiration of the same 6% CO2/air mixture was lower during the MR study. This may reflect a increased compensatory hyperventilation while patients were receiving CO2 in the MR scanner due to an increased level of anxiety secondary to the noise of echo-planar imaging, or the confined space. Nevertheless, in all individuals there was a rise in ET CO2 while breathing the CO2/air mixture and all results were expressed as change in signal or flow velocity per kPa rise in ET CO2. Over the part of the cerebral blood flow/CO2 curve utilized in this study the relationship is near linear,3 and therefore, this should not account for the semi-quantitative nature of the results. Second, BOLDbased MR imaging itself provides an indirect rather than a direct measure of cerebral perfusion, as provided by newer MR techniques such as arterial spin labeling perfusion methods. Therefore, it is possible that under hypercapnic stimulus the relationship between BOLD signal change and CBF may differ in different individuals. A variable signal decrease of 1.2–2.6% kPa21 during hyperventilation has been demonstrated in other studies.11 This may be due to varying responses of CBF to a fall in PaCO2, or different degrees of BOLD signal change between individuals to the same change in CBF. Thirdly, technical considerations could also reduce the quantitative nature of the data. These include both changes in system gain due to drift from experiment to experiment, and also problems associated with image co-registration. The image registration method used relies on minimizing the difference between images in the echo-planar time series, and a mean image is produced from the time series. Although visual inspection, by viewing the images in a movie loop, showed the image to be better aligned following registration, it is possible that the minimization procedure reduces sensitivity to the BOLD response. System drift is the most likely of these to be the cause of the problem. In quality control studies of the MR scanner used, drift generally leads to small changes from one echo-planar run to the next. Such technical considerations would not affect region-to-region differences, such as the inter-hemispheric ratio,
ET AL.
501
since each part of the image will be affected equally. Administering the hypercapnic stimulus in a different way might avoid the problem of drift, for example, a number of alternate periods of normo- and hypercapnia could be administered. This would allow the effect of system drift to be removed from the data, as in fMRI studies. Although the signal intensity changes seen in some of our reactivity images appears to be largest in the vicinity of major blood vessels, the signal change appears to be global, and therefore associated with the microvasculature. This has previously been observed by others.11,12 This suggests the emphasis on larger vessels in functional MRI studies13 may not be a problem with the acquisition parameters and field strength used for this study. It should be noted, we have fitted a linear relationship between the BOLD and TCD reactivities, since scatter in the data preclude any other type of fit. Comparisons between the BOLD response and PET CBF measurements in functional activation studies where stimulus rate is increased, show that while CBF increases linearly with stimulus rate, the BOLD response is nonlinear, and tends toward saturation.14 Similarly, the total blood flow to the brain due to CO2 inhalation increases approximately linearly with PaCO2, whereas the BOLD signal change appears to be non-linear in response to PaCO2.12 Studies using other imaging modalities such as TCD show a sigmoidal relationship between CBF velocity reactivity and PaCO2, with an approximately linear relationship when additional inspired CO2 is administered in the range 2– 6% CO2 in air. In this study we used CO2, rather than acetazolamide, as the vasodilatory stimulus for three reasons. First, it provides a more reproducible increase in CBF, as estimated by TCD studies.15 Second, the reactivity index can be expressed as the change per unit rise in ET CO2, which allows controlling for the varying levels of hypercapnia induced by 6% CO2 in different individuals. This enabled us to control for the differing rise in ET CO2 occurring during MRI and TCD studies. ET CO2 has been shown to provide a good approximation of PaCO2 if steady state has been achieved, as was the case in our study. Previous work has shown a variation in PaCO2 in response to administration of a fixed concentration of CO2 in air. Reactivity measurements using sub-maximal vasodilatory concentrations of CO2 have shown that inter-individual comparisons are improved when variations in PaCO2 between individuals are corrected for.16 Consistent with this we found a weaker relationship between BOLD and TCD reactivities when inter-individual differences in ET CO2 were not corrected for. Third, maximal effect on CBF is reached more rapidly. While the degree of response induced by the two vasodilatory stimuli is similar as determined by methods determining true
502
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
CBF or perfusion, or by TCD, this may not be the case for BOLD imaging. A previous study in animals suggest that the BOLD signal response to acetazolamide may be larger than that due to CO2 inhalation, when both are given in tolerable doses, possibly due to different effects on the relationship between CBF and cerebral blood volume.6 There is no direct comparative data available from studies in man although the signal change reported in the pilot study by Kleinschmidt et al.,8 which used acetazolamide, was of a similar order to that found in our study. A potential advantage of acetazolamide is that there may be less movement artifact, as breathing increased levels of CO2 may be associated with a compensatory hyperventilation and some head movement, and this may have reduced the signal change on the composite co-registered image. CONCLUSION In conclusion, BOLD imaging gives semi-quantitative information on CO2-reactivity. It provides high spatial resolution, which allows identification of smaller regions of impaired reactivity. Such areas are also detectable using other methods of determining rCBF but BOLD based MR imaging offers an advantage over these in being truly non-invasive, quick, and not involving the administration of radioactive substances. The basic method we describe should be applicable on conventional MR hardware capable of performing rapid T*2weighted imaging. However, in a minority of cases there was little or no change in BOLD signal in the normal hemisphere in response to hypercapnia. It remains to be determined whether this problem can be resolved by the use of acetazolamide as the vasodilatory stimulus, or by modifying the presentation of the hypercapnic stimulus to reduce the effect of any systematic errors. Acknowledgments—This work was supported by grants from the Stroke Association and National Health Service South Thames Research and Development. We are grateful to Caroline Andrews and Amanda Glover for assistance with MR imaging and to Dr. John Mellers for help with the study in the normal volunteer.
REFERENCES 1. Kleiser, B.; Widder, B. Course of carotid artery occlusions with impaired cerebrovascular reactivity. Stroke 23:171– 174; 1992. 2. Yonas, H.; Smith, H.A.; Durham, S.R.; Pentheny, S.L.; Johnson, D.W. Increased stroke risk predicted by compromised cerebral blood flow reactivity. J. Neurosurg. 79: 483– 489; 1993. 3. Ringelstein, E.B.; Otis, S.M. Physiological testing of vasomotor reserve. In: D.W. Newell, R. Aaslid (Eds). Transcranial Doppler. New York: Raven Press, 1992: pp. 83–99.
4. Ogawa, S.; Lee, T.M.; Kay, A.R.; Tank, D.W. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc. Natl. Acad. Sci. USA 87:9868 –9872; 1990. 5. Jones, R.A.; Mu¨ller, T.B.; Haraldesh, O.; Baptista, A.M.; Øksendal, A.N. Cerebrovascular changes in rats during ischemia and reperfusion: A comparison of BOLD and first pass bolus tracking techniques. Magn. Reson. Med. 35: 489 – 496; 1996. 6. Graham, G.D.; Zhong, J.; Petroff, O.A.C.; Constable, R.T.; Prichard, J.W.; Gore J.C. BOLD MRI monitoring changes in cerebral perfusion induced by acetazolamide and hypercarbia in the rat. Magn. Reson. Med. 31:557–560; 1994. 7. Ono, Y.; Morikawa, S.; Inubushi, T.; Shimuzu, H.; Yoshimoto, T. T*2-weighted magnetic resonance imaging of cerebrovascular reactivity in rat reversible focal cerebral ischemia. Brain Res. 744:207–215; 1997. 8. Kleinschmidt, A.; Steinmetz, H.; Sitzer, M.; Merboldt, K.-D.; Frahm, J. Magnetic resonance imaging of regional cerebral blood oxygenation changes under acetazolamide in carotid occlusive disease. Stroke 26:106 –110; 1995. 9. Brammer, M.J.; Bullmore, E.T.; Simmons, A.; Williams, S.C.R.; Grasby, P.M.; Howard, R.J.; Woodruff, P.W.R.; Rabe–Hesketh, S. Generic brain activation mapping in functional magnetic resonance imaging: A nonparametric approach. Magn. Reson. Imaging 15:763–770; 1997. 10. Leblanc, R., Yamamoto, Y.L.; Tyler, J.L.; Diksin, M.; Hakim, A. Borderzone ischemia. Ann. Neurol. 22:707– 713; 1987. 11. Toft, P.B.; Leth, H.; Lou, H.C.; Pryds, O.; Peitersen, B.; Henriksen, O. Local vascular CO2 reactivity in the infant brain assessed by functional MRI. Pediatr. Radiol. 25:420 – 424; 1995. 12. Rostrup, E.; Larsson, H.B.W.; Toft, P.B.; Garde, K.; Thomsen, C.; Ring, P.; Søndergaard, L.; Henriksen, O. Functional MRI of CO2 induced increase in cerebral perfusion. NMR Biomed. 7:29 –34; 1994. 13. Frahm, J; Merboldt, K.-D.; Ha¨nicke, W.; Kleinschmidt, A.; Boecker, H. Brain or vein— oxygenation or flow? On signal physiology in functional MRI of human brain activation. NMR Biomed. 7:45–53; 1994. 14. Rees, G.; Houseman, A.; Josephs, O.; Frith, C.D.; Friston, K.J.; Frackowick, R.S.; Turner, R. Characterizing the relationship between BOLD contrast and regional cerebral blood flow measurements by varying stimulus presentation rate. Neuroimage 6:270 –278; 1997. 15. Kazumata, K.; Tanaka, N.; Ishikawa, T.; Kuroda, S.; Houkin, K.; Mitsumori, K. Dissociation of vasoreactivity to acetazolamide and hypercapnia. Comparative study in patients with chronic occlusive major cerebral artery disease. Stroke 27:2052–2058; 1996. 16. Markwalder, T.-M.; Grolimund, P.; Seiler, R.W.; Roth, F.; Aaslid, R. Dependency of blood flow velocity in the middle cerebral artery on end-tidal carbon dioxide partial pressure—a transcranial ultrasound Doppler study. J. Cerebr. Blood Flow Metab. 4:368 –372; 1984.
PII S0730-725X(98)00211-2
● Original Contribution
MAPPING OF CEREBROVASCULAR REACTIVITY USING BOLD MAGNETIC RESONANCE IMAGING DAVID J. LYTHGOE,* STEVE C.R. WILLIAMS,*† MARISA CULLINANE,*
AND
HUGH S. MARKUS*
*Department of Clinical Neurosciences and †Neuroimaging, King’s College School of Medicine and Dentistry and the Institute of Psychiatry, London, UK Blood oxygen level-dependent (BOLD) contrast MRI is a simple non-invasive method of estimating “perfusion,” and combined with a vasodilatory stimulus, may allow estimation of cerebral vascular reserve. We compared BOLD carbon dioxide (CO2) reactivity in the middle cerebral artery (MCA) perfusion territory to MCA flow velocity reactivity determined using transcranial Doppler ultrasound (TCD) in 16 patients with unilateral carotid artery stenosis or occlusion. Both BOLD and TCD reactivities were calculated from measurements acquired when the subjects were breathing air, and again when breathing a 6% CO2/air mixture, and were normalized by dividing by the difference in end tidal (ET) CO2. There was a significant correlation between interhemispheric MCA reactivity difference (contralateral–ipsilateral to the stenosis or occlusion) determined by BOLD MRI and TCD (r 5 0.75, p < 0.001). In contrast, treating each hemisphere individually, there was no correlation between the absolute BOLD and TCD MCA CO2 reactivities (r 5 0.08, p 5 0.670). This appeared to be due to a variable BOLD signal change in the non-stenosed hemisphere between subjects, with little change in the normal hemisphere of a few subjects. In one patient, focal regions of reduced reactivity were seen in non-infarcted regions of the stenosed hemisphere, in the borderzones between arterial territories. BOLD reactivity maps provide information on the whole MCA territory reactivity, and may identify small regions of impaired reactivity which are not detected using TCD. However, BOLD reactivity maps only appear to provide semi-quantitative rather than quantitative data. © 1999 Elsevier Science Inc. Keywords: Carotid artery diseases; Cerebrovascular circulation; BOLD.
INTRODUCTION
determined and therefore impaired reactivity in small regions of the MCA territory cannot be identified. Alternative techniques of estimating regional cerebral blood flow (CBF) provide the necessary spatial resolution to determine small areas of reduced reactivity. Positron emission tomography provides information on hemodynamic compromise either by the determination of oxygen extraction fraction, or by carbon dioxide reactivity with regional CBF measurements before and after CO2 administration. However it is expensive and not widely available, limiting its clinical use. Xenon-inhalation computed tomography (Xe-CT) gives CBF measurements with good spatial resolution. Single photon emission tomography using the tracer Tc-HMPAO is more widely available and can be used to determine CO2 or acetazolamide reactivity but provides only a semi-quan-
Measurement of the cerebrovascular reactivity to hypercapnia or acetazolamide can be used to identify a subgroup of patients with either carotid artery stenosis (CAS) or occlusion who may be at increased risk of stroke.1,2 The technique has a number of potential clinical applications including identifying a subgroup of patients with asymptomatic CAS for carotid endarterectomy, and selecting a subgroup of patients with carotid occlusion who may benefit from revascularization procedures. Most studies are currently performed using transcranial Doppler ultrasound (TCD) to measure the flow velocities in the middle cerebral arteries (MCAs).3 This has the advantage of being cheap, portable, noninvasive and reproducible, but suffers from poor spatial resolution. Reactivity for the whole MCA territory is RECEIVED 8/15/98; ACCEPTED 11/13/98. Address correspondence to Dr. Steve C.R. Williams, Neuroimaging Research Group, Institute of Psychiatry, De Cre-
spigny Park, Denmark Hill, London, SE5 8AF. E-mail: [email protected] 495
496
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
titative estimate of perfusion. It can therefore be used to compare side-to-side difference in patients, but not to determine differences in quantitative reactivity between patients. Positron emission tomography, single photon emission tomography and Xe-CT suffer from the disadvantage that the patient is exposed to ionizing radiation, a particular problem when performing serial studies. Magnetic resonance imaging (MRI) offers a number of potential methods of estimating cerebral vascular reserve. One potential non-invasive method of indirectly estimating cerebral vascular reserve is the measurement of blood oxygen level dependent (BOLD) contrast4 with and without vasodilatory stimulus. This can be performed in less than five minutes using echo-planar imaging techniques. Both acetazolamide and hypercapnia increase CBF while having little effect on cerebral metabolism. Because of this, extraction of oxygen from the blood remains relatively constant, increasing the concentration of oxyhaemoglobin, thus reducing the concentration of deoxyhaemoglobin in the blood. This decrease in paramagnetic deoxyhaemoglobin leads to a rise in signal in T*2 -weighted MR images. This mechanism is also responsible for the increased signal seen during gradient echo based functional MR imaging of brain activation. Changes in BOLD signal have been used to detect ischemia and reperfusion in an animal model of stroke.5 BOLD signal increases due to hypercapnia and acetazolamide have also been reported in animal models of cerebrovascular disease.6,7 A previous pilot study in four patients with unilateral CAS indicated that BOLD MRI can demonstrate exhaustion of cerebrovascular reserve capacity; this study used acetazolamide as the vasodilatory stress.8 However there have been few studies of its use in man, and no comparisons of its use with other methods of determining cerebrovascular reactivity. Therefore, in this study we evaluated the use of the BOLD MRI technique to determine CO2 reactivity in a larger number of patients with carotid artery stenosis or occlusion, and compared the results with those obtained with the same subjects using TCD. In addition we determined whether the greater spatial resolution provided by BOLD MRI allowed detection of small regions of impaired reactivity in the MCA territory, which were not apparent using TCD. MATERIALS AND METHODS Subjects To validate the technique, and optimize the image acquisition parameters, we performed single-slice BOLD imaging with one normal volunteer (male, aged 30) during the administration of three cycles of air followed by a 6% CO2/air mixture. Robust changes in BOLD
signal were determined in a single region of interest (ROI) placed in the right MCA territory. We then studied sixteen patients (thirteen male, three female) with unilateral CAS greater than 70% or internal carotid artery occlusion. Carotid artery duplex was used to determine the degree of CAS, based on internal and common carotid artery peak systolic velocities. The degree of stenosis was 70 – 89% in five subjects, 90 –99% in seven, and 100% in four subjects. The CAS was symptomatic (ipsilateral stroke or TIA or amaurosis fugax within the last six months) in 10 patients, and asymptomatic in six patients. Seven patients had an ipsilateral carotid territory cerebral infarct on structural MR imaging. The mean age of the subjects was 67.3 years (SD 8.4 years, range 51– 81 years). Before the study, each subject gave informed written consent, and the experimental protocol was approved by the local hospital ethics committee. Imaging Protocol In all patients reactivity to 6% CO2 was determined using both BOLD MRI and TCD on separate occasions. For both studies end-tidal (ET) CO2 levels were continuously measured using the same infrared CO2 meter (Normocap 100, Datex Instrumentarium Corp., Helsinki, Finland). MRI BOLD imaging was performed on a 1.5-T GE Signa (GE Medical Systems, Milwaukee, WI), fitted with Advanced NMR Systems (Wilmington, MA, USA) hardware and software, giving echo-planar imaging capability. All patients were provided with a panic button, so that if they were anxious they could terminate the experiment at any time. An air or 6% CO2/air mixture was delivered to the subject via a mouthpiece. A nose clip was fitted to ensure the inhaled gas was derived solely from the mouthpiece. A one-way valve was used to separate inspired and expired gases during the whole procedure, and to allow sampling for ET CO2 levels. While breathing air, 10 consecutive non-contiguous T*2weighted images were acquired from each of 14 axial planes (TR 5 3s, TE 5 40 ms, u 5 90°, in-plane resolution 5 1.5 mm, slice thickness 5 5.0 mm and slice skip 5 0.5 mm). During image acquisition, ET CO2 measurements were recorded. When the acquisition was finished the subject inhaled the 6% CO2/air mixture, and 60 s after steady-state ET CO2 was attained (which usually took 1–2 min), the image acquisition sequence was repeated. MRI Analysis Following acquisition, the two data sets (breathing air, and breathing the 6% CO2/air mixture) were concate-
BOLD and carotid stenosis ● D.J. LYTHGOE
nated to form one image series of 20 consecutive images for each axial slice, with the 10 images obtained with the subject breathing air preceding the 10 images obtained with the subject breathing the 6% CO2/air mixture. Misregistration of the images produced during normocapnia, and those obtained during hypercapnia, can produce enhancements along the sulci, which may be confused with increased BOLD response associated with large blood vessels. In order to minimize this effect, we co-registered all the images prior to production of reactivity images. The time series images for each axial plane were combined into a 4D image consisting of 3D brain volume images at 20 consecutive time points. The sequential 3D volumes were co-aligned using software developed primarily for the analysis of fMRI data.9 Following realignment, calculated BOLD reactivity maps were produced, voxel-by-voxel, depicting the fractional increase in image intensity due to CO2 inhalation, normalized to the increase in ET CO2. This is described by Eq. 1, R5
H
JY
Iair/6% CO2 2 Iair Iair
@ET CO2 air/6% CO2 2 ET CO2 air#
(1)
where R is the reactivity, Iair and Iair/6%CO2 are the mean intensities at each voxel for the images acquired when the subjects were breathing air and the air/6% CO2 mixture respectively, and ET CO2 air and ET CO2 air/6%CO2 are the mean ET CO2 values recorded while the corresponding images were acquired. Mean T*2-weighted anatomic images were calculated from the ten co-aligned sequential images obtained when the subject was breathing air. Three anatomic slices (T*2-weighted images) centered at the level of the most superior aspect of the third ventricle, depicting the mean intensity at each location were displayed. ROIs were defined manually in each slice for the left and right middle cerebral artery territories, by an experienced operator. The regions of interest were then transferred to the calculated reactivity maps, and the mean reactivity and area of each region noted. Weighted mean reactivities (Rterr) for the three slices were calculated using Eq. 2. R terr 5
( slice R slice51. . .3 z A slice51. . .3 (slice51. . .3 Aslice51. . .3
(2)
Here Rslice is the reactivity for a particular slice, and Aslice is its area. In addition the same slices were visually analyzed by an experienced observer blinded to the side of stenosis to identify small areas of impaired reactivity within noninfarcted regions of the internal carotid artery territory, as determined by the absence of high signal on the corresponding reactivity image. In those images where
ET AL.
497
there was an infarct on the corresponding T*2-weighted image, a ROI was placed in the infarcted tissue on each of three slices in which it was present, and in the corresponding regions of the non-infarcted contralateral hemisphere. The mean BOLD reactivities were calculated for the infarcted and non-infarcted regions as described above for the MCA territories. Transcranial Doppler Analysis TCD measurements were performed using a commercially available TCD machine (Multi Dop 34, DWL Electronische GMbH) with a 2MHz probe. Continuous simultaneous bilateral MCA insonation was performed via the transtemporal windows. During the TCD reactivity tests, the subject was placed in a supine position. A mask was placed over the face of the subject to allow delivery of air or the CO2/air mixture, and sampling of ET CO2. The mask was held in place to allow the subject to acclimatize while breathing air, and then the left and right MCA flow velocities, and ET CO2 were recorded for approximately 30 s to obtain baseline readings. The 6% CO2/air mixture was then administered for 2–3 min, until steady state was reached for both the ET CO2 and MCA flow velocities, after which MCA velocity and ET CO2 were measured over a 30 s period. A reactivity index (R) was calculated, using Eq. 3, R5
H
JY
vair/6% CO2 2 vair vair
@ET CO2 air/6% CO2 2 ET CO2 air#
(3)
where nair and nair/6%CO2 are the blood velocities measured when the subjects were breathing air and the air/6% CO2 mixture respectively, and ET CO2 air and ET CO2 air/6%CO2 are the mean ET CO2 values recorded during the corresponding TCD blood velocity measurements. Statistical Analysis Using Student’s paired t-test, we compared both BOLD and TCD reactivities in the whole MCA territories ipsilateral and contralateral to the CAS. Comparisons of the MRI data with the TCD data were performed in two ways. First, data from both stenosed and nonstenosed hemispheres were pooled and the correlation between BOLD and TCD estimates of reactivity for each MCA territory determined using Pearson’s test. Secondly since different subjects might exhibit differing inherent BOLD reactivity to increased CO2 levels, the difference between reactivities in the non-stenosed and stenosed hemispheres was calculated for each subject. The interhemispheric reactivity difference for each subject estimated using BOLD was then correlated with that obtained using TCD. In addition BOLD reactivities in the
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
498
Fig. 1. BOLD imaging in the normal volunteer. Images were acquired while breathing air or while breathing the 6% CO2 in air. The paradigm was repeated three times. A reproducible rise in BOLD signal intensity is seen following 6% CO2 administration. Signal change is scaled relative to the mean signal in the first baseline period.
infarcted tissue were compared with BOLD reactivities in corresponding regions of non-infarcted tissue in the contra-lateral hemisphere using Student’s paired t-test. RESULTS In the normal volunteer, there was a reproducible rise in BOLD signal during each period of inspiration of the 6% CO2/air mixture, as shown in Fig. 1. In patients CO2 reactivity measured by TCD was significantly reduced in the stenosed compared with the contralateral MCA territory (mean 6 SD 5 13.9 6 7.1 vs. 19.9 6 7.3% kPa21; p 5 0.046). BOLD reactivity was also lower in the ipsilateral MCA territory but the difference was not significant (0.14 6 0.37% kPa21 vs. 0.17 6 0.41% kPa21, p 5 0.379). This was partly due to little increase in BOLD signal following CO2 administration in the normal hemisphere of some individuals; in three cases the BOLD signal change was , 0.1% kPa21. The rise in ET CO2 during 6% CO2 inspiration in these patients was similar to that seen in other patients during MR imaging (mean 6 SD, 1.43 6 0.77% kPa21 vs. 1.40 6 0.30% kPa21). When these non-responders were removed from the analysis the difference in BOLD signal change between the two hemispheres was: mean 6 SD stenosed 0.51 6 0.36% kPa21, non-stenosed 0.69 6 0.25% kPa21, p 5 0.1. There was a highly significant correlation (r 5 0.75, p , 0.001) between the inter-hemispheric MCA reactivity difference measured with BOLD and TCD as shown in Fig. 2. The data were best fitted by a linear relation-
ship. In contrast there was no correlation between BOLD and TCD reactivities as determined for individual hemispheres (r 5 0.08, p 5 0.67), as shown in Fig. 3. This appeared to be due to a variable BOLD signal change in the non-stenosed hemisphere between subjects. BOLD reactivity in infarcted tissue in the seven patients with infarcts was significantly reduced compared with the corresponding regions of non-infarcted tissue in the contralateral hemisphere (mean 6 SD 20.31 6 0.37% kPa21 vs. 0.35 6 0.41% kPa21, p 5 0.024). In all subjects visual analysis of the BOLD reactivity maps was performed. Regions of reduced reactivity were seen in areas of cerebral infarction as determined from the corresponding T*2-weighted image. In addition, in one patient focal regions of reduced BOLD signal intensity change were seen in the borderzones between arterial territories in the side with CAS, while no abnormalities were present in corresponding T*2-weighted images. This is illustrated in Fig. 4 along with the corresponding mean T*2-weighted image. The reduced BOLD response (Fig. 4b) appears to be in the watershed zones between middle and anterior cerebral artery (arrow 1) and middle and posterior cerebral artery (arrow 2) territories as indicated. Figure 4c shows another T*2-weighted image slice from the same subject indicating infarction in the white matter of the left hemisphere. In the corresponding BOLD reactivity image (Fig. 4d), this region shows reduced BOLD reactivity (arrow 3). The reduction in reactivity in a region of infarcted tissue in another patient is shown in Fig. 5.
BOLD and carotid stenosis ● D.J. LYTHGOE
ET AL.
499
Fig. 2. The relationship between MCA inter-hemispheric reactivity, normalized for change in ET CO2, as determined using BOLD MRI and TCD. There was a significant correlation; r 5 0.75, p , 0.001.
During inspiration of 6% CO2, end-tidal CO2 rose in all individuals. However, the mean (6 SD) rise in ET CO2 was greater during TCD than during BOLD MRI (2.42 6 0.64% kPa vs. 1.40 6 0.30% kPa, p 5 0.01). DISCUSSION Our study in the normal volunteer demonstrates that 6% CO2 administration results in a detectable and reproducible rise in BOLD MR signal. Our results in patients
with CAS demonstrate that BOLD MRI can be used to produce high spatial resolution CO2 reactivity maps. These may be used to provide information on whole MCA territory reactivity, and also to identify small areas of impaired reactivity which are not detectable using TCD. Consistent with this we found a highly significant correlation between the inter-hemispheric MCA reactivity difference as determined using TCD and BOLD MR imaging. However the information provided by the
Fig. 3. The relationship between reactivity, normalized for change in ET CO2, in individual MCA territories as determined by BOLD MRI and TCD. Stenosed and non-stenosed hemispheres are represented by different symbols for comparison. There was no significant correlation; r 5 0.08, p 5 0.67.
500
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
Fig. 4. Images from a patient with left carotid stenosis showing impaired reactivity in the regions of the left internal carotid artery territory. a, mean T*2-weighted image calculated from the 10 image series when the subject was breathing air. b, the corresponding normalised reactivity image. Region 1 and 2 show reduced reactivity in the border zones between the left MCA and anterior cerebral artery and between the left MCA and posterior cerebral artery territories, respectively. c, mean T*2weighted image from a more superior plane in the same individual, showing an infarcted region in the white matter of the left hemisphere. d, the corresponding reactivity map. Region 3 indicates a region of reduced activity in the infarcted area.
BOLD reactivity maps presented in this study appears to be semi-quantitative rather than providing true quantitative data. This is reflected in the lack of any correlation
Fig. 5. Mean T*2-weighted image from a patient with left carotid stenosis and an old MCA territory infarct in the gray and white matter of the left hemisphere. b, the corresponding normalised BOLD reactivity image with a region of reduced reactivity in a location (arrow) corresponding to the infarct.
between absolute BOLD MCA territory reactivities and absolute TCD MCA reactivities. Analysis of the BOLD response to hypercapnia in the normal MCA territory ipsilateral to the non-stenosed carotid artery demonstrated that this was variable. In addition in three individuals there was a very small, or non-significant BOLD signal change in the contralateral hemisphere. In some subjects, the TCD and BOLD interhemispheric reactivity differences were negative. This is probably a reflection of the fact that although in normal subjects the reactivity is similar in both MCA territories, some degree of interhemispheric difference can be found. BOLD changes provide a much greater degree of spatial resolution than do TCD measurements, reflecting change on a pixel to pixel basis in comparison with flow in the whole middle cerebral artery territory. To account for this we determined the mean BOLD signal change over the whole of the middle cerebral artery territory on each of the brain slices analyzed. In regions of established infarction one would not expect to see a BOLD response to a CO2 challenge. Consistent with this, we found reduced BOLD CO2 reactivity in areas of established infarction. In addition in one subject we found regions of impaired reactivity in non-infarcted tissue ipsilateral to the CAS. The pattern of abnormality was striking with impaired reactivity both in the internal watershed areas, and in the watershed areas between the MCA and both the anterior cerebral and the posterior cerebral arteries. This lack of signal change in non-infarcted but hemodynamically compromised areas may be due to a combination of factors. First, a lack of increase in blood flow in response to a rise in inspired CO2 will not give an increase in blood oxygenation, with its corresponding signal change in BOLD images. Second, previous studies have suggested a trend toward increased oxygen extraction fraction associated with decreased hemodynamic reserve capacity in the borderzones between the MCA and ACA perfusion territories.10 This would lead to an increase in paramagnetic deoxyhaemoglobin, and a corresponding reduction in signal intensity. This regional information cannot be provided using TCD and is a major advantage of a technique with high spatial resolution such as BOLD MRI. Single photon emission tomography can provide similar information but with a lower spatial resolution, and in addition it involves longer acquisition times and the administration of radioactive substances. The latter limits its use particularly for serial studies such as administration before and during CO2 inhalation. Application of a short CO2 challenge allows us to determine the extent of BOLD reactivity throughout the brain. This may have important implications in functional MRI studies of brain activation, in pathologic cases where muted BOLD response is observed despite
BOLD and carotid stenosis ● D.J. LYTHGOE
satisfactory task performance. In such cases a lack of significant BOLD contrast due to exhausted cerebrovascular reserve may be misinterpreted as neuronal dysfunction. There are several possible reasons for the information provided by BOLD imaging being semi-quantitative rather than truly quantitative. First, problems in CO2 administration, such as poorly fitting mouth-pieces or masks, can result in inadequate hypercapnic stimulus and could account for the lack of BOLD signal change in three individuals, and the variable response in other individuals. We excluded this by continuously monitoring ET CO2; in all studies there was a rise in ET CO2. However, the mean rise in ET CO2 during inspiration of the same 6% CO2/air mixture was lower during the MR study. This may reflect a increased compensatory hyperventilation while patients were receiving CO2 in the MR scanner due to an increased level of anxiety secondary to the noise of echo-planar imaging, or the confined space. Nevertheless, in all individuals there was a rise in ET CO2 while breathing the CO2/air mixture and all results were expressed as change in signal or flow velocity per kPa rise in ET CO2. Over the part of the cerebral blood flow/CO2 curve utilized in this study the relationship is near linear,3 and therefore, this should not account for the semi-quantitative nature of the results. Second, BOLDbased MR imaging itself provides an indirect rather than a direct measure of cerebral perfusion, as provided by newer MR techniques such as arterial spin labeling perfusion methods. Therefore, it is possible that under hypercapnic stimulus the relationship between BOLD signal change and CBF may differ in different individuals. A variable signal decrease of 1.2–2.6% kPa21 during hyperventilation has been demonstrated in other studies.11 This may be due to varying responses of CBF to a fall in PaCO2, or different degrees of BOLD signal change between individuals to the same change in CBF. Thirdly, technical considerations could also reduce the quantitative nature of the data. These include both changes in system gain due to drift from experiment to experiment, and also problems associated with image co-registration. The image registration method used relies on minimizing the difference between images in the echo-planar time series, and a mean image is produced from the time series. Although visual inspection, by viewing the images in a movie loop, showed the image to be better aligned following registration, it is possible that the minimization procedure reduces sensitivity to the BOLD response. System drift is the most likely of these to be the cause of the problem. In quality control studies of the MR scanner used, drift generally leads to small changes from one echo-planar run to the next. Such technical considerations would not affect region-to-region differences, such as the inter-hemispheric ratio,
ET AL.
501
since each part of the image will be affected equally. Administering the hypercapnic stimulus in a different way might avoid the problem of drift, for example, a number of alternate periods of normo- and hypercapnia could be administered. This would allow the effect of system drift to be removed from the data, as in fMRI studies. Although the signal intensity changes seen in some of our reactivity images appears to be largest in the vicinity of major blood vessels, the signal change appears to be global, and therefore associated with the microvasculature. This has previously been observed by others.11,12 This suggests the emphasis on larger vessels in functional MRI studies13 may not be a problem with the acquisition parameters and field strength used for this study. It should be noted, we have fitted a linear relationship between the BOLD and TCD reactivities, since scatter in the data preclude any other type of fit. Comparisons between the BOLD response and PET CBF measurements in functional activation studies where stimulus rate is increased, show that while CBF increases linearly with stimulus rate, the BOLD response is nonlinear, and tends toward saturation.14 Similarly, the total blood flow to the brain due to CO2 inhalation increases approximately linearly with PaCO2, whereas the BOLD signal change appears to be non-linear in response to PaCO2.12 Studies using other imaging modalities such as TCD show a sigmoidal relationship between CBF velocity reactivity and PaCO2, with an approximately linear relationship when additional inspired CO2 is administered in the range 2– 6% CO2 in air. In this study we used CO2, rather than acetazolamide, as the vasodilatory stimulus for three reasons. First, it provides a more reproducible increase in CBF, as estimated by TCD studies.15 Second, the reactivity index can be expressed as the change per unit rise in ET CO2, which allows controlling for the varying levels of hypercapnia induced by 6% CO2 in different individuals. This enabled us to control for the differing rise in ET CO2 occurring during MRI and TCD studies. ET CO2 has been shown to provide a good approximation of PaCO2 if steady state has been achieved, as was the case in our study. Previous work has shown a variation in PaCO2 in response to administration of a fixed concentration of CO2 in air. Reactivity measurements using sub-maximal vasodilatory concentrations of CO2 have shown that inter-individual comparisons are improved when variations in PaCO2 between individuals are corrected for.16 Consistent with this we found a weaker relationship between BOLD and TCD reactivities when inter-individual differences in ET CO2 were not corrected for. Third, maximal effect on CBF is reached more rapidly. While the degree of response induced by the two vasodilatory stimuli is similar as determined by methods determining true
502
Magnetic Resonance Imaging ● Volume 17, Number 4, 1999
CBF or perfusion, or by TCD, this may not be the case for BOLD imaging. A previous study in animals suggest that the BOLD signal response to acetazolamide may be larger than that due to CO2 inhalation, when both are given in tolerable doses, possibly due to different effects on the relationship between CBF and cerebral blood volume.6 There is no direct comparative data available from studies in man although the signal change reported in the pilot study by Kleinschmidt et al.,8 which used acetazolamide, was of a similar order to that found in our study. A potential advantage of acetazolamide is that there may be less movement artifact, as breathing increased levels of CO2 may be associated with a compensatory hyperventilation and some head movement, and this may have reduced the signal change on the composite co-registered image. CONCLUSION In conclusion, BOLD imaging gives semi-quantitative information on CO2-reactivity. It provides high spatial resolution, which allows identification of smaller regions of impaired reactivity. Such areas are also detectable using other methods of determining rCBF but BOLD based MR imaging offers an advantage over these in being truly non-invasive, quick, and not involving the administration of radioactive substances. The basic method we describe should be applicable on conventional MR hardware capable of performing rapid T*2weighted imaging. However, in a minority of cases there was little or no change in BOLD signal in the normal hemisphere in response to hypercapnia. It remains to be determined whether this problem can be resolved by the use of acetazolamide as the vasodilatory stimulus, or by modifying the presentation of the hypercapnic stimulus to reduce the effect of any systematic errors. Acknowledgments—This work was supported by grants from the Stroke Association and National Health Service South Thames Research and Development. We are grateful to Caroline Andrews and Amanda Glover for assistance with MR imaging and to Dr. John Mellers for help with the study in the normal volunteer.
REFERENCES 1. Kleiser, B.; Widder, B. Course of carotid artery occlusions with impaired cerebrovascular reactivity. Stroke 23:171– 174; 1992. 2. Yonas, H.; Smith, H.A.; Durham, S.R.; Pentheny, S.L.; Johnson, D.W. Increased stroke risk predicted by compromised cerebral blood flow reactivity. J. Neurosurg. 79: 483– 489; 1993. 3. Ringelstein, E.B.; Otis, S.M. Physiological testing of vasomotor reserve. In: D.W. Newell, R. Aaslid (Eds). Transcranial Doppler. New York: Raven Press, 1992: pp. 83–99.
4. Ogawa, S.; Lee, T.M.; Kay, A.R.; Tank, D.W. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc. Natl. Acad. Sci. USA 87:9868 –9872; 1990. 5. Jones, R.A.; Mu¨ller, T.B.; Haraldesh, O.; Baptista, A.M.; Øksendal, A.N. Cerebrovascular changes in rats during ischemia and reperfusion: A comparison of BOLD and first pass bolus tracking techniques. Magn. Reson. Med. 35: 489 – 496; 1996. 6. Graham, G.D.; Zhong, J.; Petroff, O.A.C.; Constable, R.T.; Prichard, J.W.; Gore J.C. BOLD MRI monitoring changes in cerebral perfusion induced by acetazolamide and hypercarbia in the rat. Magn. Reson. Med. 31:557–560; 1994. 7. Ono, Y.; Morikawa, S.; Inubushi, T.; Shimuzu, H.; Yoshimoto, T. T*2-weighted magnetic resonance imaging of cerebrovascular reactivity in rat reversible focal cerebral ischemia. Brain Res. 744:207–215; 1997. 8. Kleinschmidt, A.; Steinmetz, H.; Sitzer, M.; Merboldt, K.-D.; Frahm, J. Magnetic resonance imaging of regional cerebral blood oxygenation changes under acetazolamide in carotid occlusive disease. Stroke 26:106 –110; 1995. 9. Brammer, M.J.; Bullmore, E.T.; Simmons, A.; Williams, S.C.R.; Grasby, P.M.; Howard, R.J.; Woodruff, P.W.R.; Rabe–Hesketh, S. Generic brain activation mapping in functional magnetic resonance imaging: A nonparametric approach. Magn. Reson. Imaging 15:763–770; 1997. 10. Leblanc, R., Yamamoto, Y.L.; Tyler, J.L.; Diksin, M.; Hakim, A. Borderzone ischemia. Ann. Neurol. 22:707– 713; 1987. 11. Toft, P.B.; Leth, H.; Lou, H.C.; Pryds, O.; Peitersen, B.; Henriksen, O. Local vascular CO2 reactivity in the infant brain assessed by functional MRI. Pediatr. Radiol. 25:420 – 424; 1995. 12. Rostrup, E.; Larsson, H.B.W.; Toft, P.B.; Garde, K.; Thomsen, C.; Ring, P.; Søndergaard, L.; Henriksen, O. Functional MRI of CO2 induced increase in cerebral perfusion. NMR Biomed. 7:29 –34; 1994. 13. Frahm, J; Merboldt, K.-D.; Ha¨nicke, W.; Kleinschmidt, A.; Boecker, H. Brain or vein— oxygenation or flow? On signal physiology in functional MRI of human brain activation. NMR Biomed. 7:45–53; 1994. 14. Rees, G.; Houseman, A.; Josephs, O.; Frith, C.D.; Friston, K.J.; Frackowick, R.S.; Turner, R. Characterizing the relationship between BOLD contrast and regional cerebral blood flow measurements by varying stimulus presentation rate. Neuroimage 6:270 –278; 1997. 15. Kazumata, K.; Tanaka, N.; Ishikawa, T.; Kuroda, S.; Houkin, K.; Mitsumori, K. Dissociation of vasoreactivity to acetazolamide and hypercapnia. Comparative study in patients with chronic occlusive major cerebral artery disease. Stroke 27:2052–2058; 1996. 16. Markwalder, T.-M.; Grolimund, P.; Seiler, R.W.; Roth, F.; Aaslid, R. Dependency of blood flow velocity in the middle cerebral artery on end-tidal carbon dioxide partial pressure—a transcranial ultrasound Doppler study. J. Cerebr. Blood Flow Metab. 4:368 –372; 1984.