Author + information
- Received October 22, 2011
- Revision received December 20, 2011
- Accepted January 3, 2012
- Published online May 29, 2012.
- Jayanth R. Arnold, BMBCh, MA, DPhil⁎,
- Theodoros D. Karamitsos, MD, PhD⁎,
- Paul Bhamra-Ariza, MBBS, BSc†,
- Jane M. Francis, DCRR, DNM⁎,
- Nick Searle, DCRR‡,
- Matthew D. Robson, PhD⁎,
- Ruairidh K. Howells, MEng⁎,
- Robin P. Choudhury, DM⁎,
- Ornella E. Rimoldi, MD§,
- Paolo G. Camici, MD§,
- Adrian P. Banning, MBBS, MD‡,
- Stefan Neubauer, MD, FMedSci⁎,
- Michael Jerosch-Herold, PhD∥ and
- Joseph B. Selvanayagam, MBBS, DPhil¶,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Joseph B. Selvanayagam, Department of Medicine, Flinders University, Flinders Medical Centre, Bedford Park, Adelaide 5042, Australia
Objectives The purpose of this study was to assess the diagnostic accuracy of blood oxygen-level dependent (BOLD) MRI in suspected coronary artery disease (CAD).
Background By exploiting the paramagnetic properties of deoxyhemoglobin, BOLD magnetic resonance imaging can detect myocardial ischemia. We applied BOLD imaging and first-pass perfusion techniques to: 1) examine the pathophysiological relationship between coronary stenosis, perfusion, ventricular scar, and myocardial oxygenation; and 2) evaluate the diagnostic performance of BOLD imaging in the clinical setting.
Methods BOLD and first-pass perfusion images were acquired at rest and stress (4 to 5 min intravenous adenosine, 140 μg/kg/min) and assessed quantitatively (using a BOLD signal intensity index [stress/resting signal intensity], and absolute quantification of perfusion by model-independent deconvolution). A BOLD signal intensity index threshold to identify ischemic myocardium was first determined in a derivation arm (25 CAD patients and 20 healthy volunteers). To determine diagnostic performance, this was then applied in a separate group comprising 60 patients with suspected CAD referred for diagnostic angiography.
Results Prospective evaluation of BOLD imaging yielded an accuracy of 84%, a sensitivity of 92%, and a specificity of 72% for detecting myocardial ischemia and 86%, 92%, and 72%, respectively, for identifying significant coronary stenosis. Segment-based analysis revealed evidence of dissociation between oxygenation and perfusion (r = −0.26), with a weaker correlation of quantitative coronary angiography with myocardial oxygenation (r = −0.20) than with perfusion (r = −0.40; p = 0.005 for difference). Hypertension increased the odds of an abnormal BOLD response, but diabetes mellitus, hypercholesterolemia, and the presence of ventricular scar were not associated with significant deoxygenation.
Conclusions BOLD imaging provides valuable insights into the pathophysiology of CAD; myocardial hypoperfusion is not necessarily commensurate with deoxygenation. In the clinical setting, BOLD imaging achieves favorable accuracy for identifying the anatomic and functional significance of CAD.
Determining the presence and severity of myocardial ischemia is a key goal in the effective management of coronary artery disease (CAD). Untreated ischemia is an important determinant of adverse future outcome; conversely, the benefits of ischemia-driven revascularization are well recognized (1,2).
It is now well established that the anatomic appearances of epicardial coronary arteries are poorly predictive of myocardial ischemia (3). Therefore, current guidelines recommend concurrent assessment of the functional severity of coronary stenosis to guide revascularization (4). In the clinical setting, a number of imaging methods are available, including nuclear imaging techniques, echocardiography, and cardiovascular magnetic resonance (CMR). Such modalities assess perfusion and contractile abnormalities, but these serve as surrogates of myocardial ischemia: ischemia per se is not measured.
A novel CMR application, blood oxygen level–dependent (BOLD) imaging, now offers the possibility of detecting ischemia more directly. This technique exploits the paramagnetic properties of the body's intrinsic contrast agent, hemoglobin (5). The transition from diamagnetic oxyhemoglobin to paramagnetic deoxyhemoglobin induces magnetic susceptibility differences, resulting in a change in magnetic resonance signal intensity and thereby generating oxygen-dependent contrast (6). Because ischemia is the initiator of the ischemic cascade, theoretically BOLD imaging could prove sensitive for detecting CAD. Previous studies have validated BOLD imaging against invasive measurement of perfusion and microcirculatory oxygenation in animal models (7–15). The potential benefits of BOLD imaging were also demonstrated in recent human studies (16–20). However, many early studies were hampered by limitations inherent in the BOLD sequences, namely, low signal-to-noise ratios, long acquisition times, and the occurrence of imaging artifacts that may mimic regional deoxygenation (21,22). The recently developed steady-state free precession (SSFP) sequence circumvents many of these difficulties and has been validated in animal models with promising initial results (12,14,15,23). Currently, there are limited data on human subjects, and the potential clinical utility of this technique in subjects with suspected CAD remains uncertain.
Therefore, the first objective of our study was to investigate the diagnostic performance of BOLD imaging in the clinical setting. The second objective was to capitalize on the multiparametric capability of CMR and apply first-pass perfusion and late gadolinium enhancement (LGE) imaging in conjunction with BOLD assessment to examine the pathophysiological relationship of myocardial tissue oxygenation with myocardial perfusion, coronary stenosis, ventricular scar, and cardiac risk factors.
To determine an appropriate BOLD threshold, 25 patients with known CAD (single- or 2-vessel disease) and 20 normal volunteers (with minimal coronary risk factors and no history of CAD) were first studied. For the prospective application of BOLD imaging, we recruited an additional 60 subjects who were scheduled for elective diagnostic angiography for investigation of exertional chest pain as part of routine clinical care.
Exclusion criteria were recent myocardial infarction (within 7 days), unstable coronary syndromes, contraindications to CMR (severe claustrophobia, metallic implants including pacemakers, defibrillators, cerebral aneurysm clips, ocular metallic deposits), contraindications to adenosine (second- or third-degree atrioventricular block, obstructive pulmonary disease, dipyridamole use) and contraindications to gadolinium (anaphylaxis, estimated glomerular filtration rate <60 ml/min). Patients underwent CMR imaging in the 2 weeks before undergoing coronary angiography.
Subjects were asked to avoid agents that could antagonize the effects of adenosine (e.g., caffeine, methylxanthines) for at least 24 h before the CMR scan. All participants gave written informed consent, and the study was approved by the regional ethics committee.
Each patient underwent CMR examination at 3-T (TIM Trio, Siemens Medical Solutions, Erlangen, Germany) using an anterior 4-element phased-array coil and a posterior phased-array surface coil. From standard piloting, short-axis cine images covering the left ventricle were acquired using a retrospective electrocardiography (ECG)-gated SSFP sequence (echo time, 1.5 ms; repetition time, 3 ms; flip angle, 60°).
For BOLD imaging, patients were monitored by ECG, sphygmomanometry, and pulse oximetry. Six consecutive mid ventricular images were acquired from the same short-axis slice using a T2-prepared ECG-gated SSFP sequence (echo time, 1.43 ms; repetition time, 2.86 ms; T2 preparation time, 40 ms; matrix, 168 × 192; slice thickness, 8 mm; flip angle, 44°) at rest. If required, frequency scouting and shim adjustments were performed to minimize off-resonance artifacts. For stress imaging, after 2 min of intravenous adenosine (140 μg/kg/min), 4 to 6 stress BOLD images were consecutively acquired in the same short-axis plane. Immediately after stress BOLD imaging (4 to 5 min after commencing the adenosine infusion), first-pass perfusion imaging was performed in the same short-axis plane as for BOLD imaging, using an ECG-gated T1-weighted fast gradient echo sequence (echo time, 1.04 ms; repetition time, 2 ms; voxel size, 2.1 × 2.6 × 8 mm3), and a peripheral bolus injection of a gadolinium-based contrast agent (0.04 mmol/kg; Gadodiamide, Omniscan, GE Healthcare Amersham, United Kingdom), followed by a 15-ml bolus of normal saline (rate 6 ml/s). After 20 min, the same sequence was repeated without adenosine to obtain resting perfusion images.
For delayed enhancement imaging, an additional bolus of Gadodiamide (0.05 mmol/kg) was injected, and after 5 min, images were acquired in the 3 long axes and in the short axis plane to obtain coverage of the entire left ventricle using an ECG-gated T1-weighted segmented inversion recovery turbo fast low-angle shot sequence (echo time, 4.8 ms; voxel size, 1.4 × 2.4 × 8 mm; flip angle, 20°). The inversion time was adjusted to achieve optimal nulling of noninfarcted myocardium.
In the 6 weeks before CMR examination for known CAD patients, and in the 2 weeks after CMR examination of subjects with suspected CAD, all subjects underwent coronary angiography using standard techniques. Images of the coronary arteries were obtained in multiple projections, with avoidance of overlap of side branches and foreshortening of relevant coronary stenoses.
For analysis of myocardial perfusion, signal intensity (SI) curves were generated by tracing endocardial and epicardial contours (QMass, version 6.2.3, Medis Medical Imaging Solutions, Leiden, the Netherlands), which were manually corrected for cardiac displacement. The myocardium was divided into 6 equiangular segments in accordance with the American Heart Association 17-segment model (24). A region of interest was placed within the cavity of the left ventricle, excluding the myocardium and papillary muscles. Quantitative perfusion analysis was performed as previously described (25,26). Absolute myocardial blood flow (MBF) (in milliliters per minute per gram) was calculated for each myocardial segment by model-independent deconvolution of myocardial and arterial input SI curves. Reproducibility data from our group using this technique have previously been published (27). Perfusion analysis was performed by 2 operators (J.R.A. and M.J.H.), blinded to clinical information, angiographic data. and BOLD analysis.
For BOLD assessment, endocardial and epicardial contours were traced manually, with correction of cardiac motion (QMass version 6.2.3, Medis Medical Imaging Solutions). The myocardium was then subdivided into 6 equiangular segments (inferior septum, anterior septum, anterior, anterolateral, inferolateral, and inferior), corresponding to the mid ventricular slice in the American Heart Association 17-segment model. The mean SI within each segment was obtained, both at rest and stress, and corrected for variations in heart rate, as previously described (28). The BOLD SI index was derived by dividing the change in SI from rest to stress by rest SI ([stress SI − rest SI]/rest SI). No segments were excluded from analysis. Reproducibility data from our group using this technique were previously published (19). BOLD analysis was performed by a single operator (J.R.A., blinded to clinical information, angiographic data, and CMR perfusion, LGE, and functional analyses).
Analyses of left ventricular function (J.R.A.) and LGE (T.D.K.) were performed blinded to clinical information, angiographic data, and analysis of BOLD and perfusion data. For analysis of left ventricular function, the short-axis SSFP images were analyzed using customized software (CMRTools, Cardiovascular Imaging Solutions Ltd., London, United Kingdom). For analysis of LGE images, areas of hyperenhancement were assessed visually as present or absent, and with estimation of the transmural extent of LGE.
Quantitative coronary angiography was performed by 2 operators (N.S. and R.P.C.) blinded to clinical information and CMR data. Diameters of reference and stenotic coronary arteries were measured by a computer-assisted quantitative method (Quantcor Coronary Analysis, Siemens Medical Solutions). The contrast-filled catheter was used for image magnification calibration. Significant CAD was defined angiographically as the presence of at least 1 stenosis of ≥50% diameter in any of the main epicardial coronary arteries or their branches with a diameter of ≥2 mm. Each myocardial segment was assigned to a coronary artery territory according to standard criteria (24). For CAD patients in the derivation arm, attention was paid to coronary dominance and the location of any coronary stenoses in relation to the mid ventricular slice of the American Heart Association 17-segment model: segments were labeled as stenosed if subtended by significant coronary stenoses or remote if there were no significant stenoses proximal to the mid ventricular level.
Data analysis was performed using SPSS 15.0 for Windows (SPSS Inc., Chicago, Illinois) and Medcalc 18.104.22.168 (Mariakerke, Belgium). Normally distributed variables are presented as mean ± SD. Non-normally distributed data are presented as median and interquartile range (25th and 75th percentiles). Continuous variables were tested for normal distribution using the D'Agostino-Pearson test. Comparisons among the 3 groups of myocardial segments were performed with analysis of variance or the Kruskal-Wallis test, as appropriate.
To investigate the diagnostic performance of BOLD imaging, first, the receiver-operating characteristic (ROC) curve analysis was applied in the derivation group to identify a hyperemic MBF cutoff defined using a ≥50% threshold of coronary stenosis. The optimal cut-point was defined as that giving rise to the highest sum of sensitivity and specificity. Segments were classified as ischemic or nonischemic using this threshold, and a corresponding BOLD SI index threshold was determined using ROC curve analysis. This predetermined cutoff value for the BOLD SI index was then applied in the validation group to determine sensitivity, specificity, and diagnostic accuracy of BOLD in the identification of ischemia. The ability of BOLD imaging to identify significant coronary stenosis was also determined, as was that of first-pass perfusion imaging.
The association of an abnormal BOLD response (based on the cutoff value from the ROC curve analysis) with clinical, segmental LGE, and perfusion variables was analyzed with a mixed-effects logistic regression model, using the package lme4 in the R statistical computing environment (R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria, 2010). Within-patient correlation of measurements was modeled with a compound symmetry covariance structure for the random effects, using the patient identifier as a variable to group data. The odds ratios and their confidence intervals were calculated from the estimated regression coefficients for the fixed effects (glmer in R package lme4).
Generalized additive mixed models (gamm in the R package mgcv) were used to examine data trends between the BOLD SI index and hyperemic MBF, both measured at the segmental level. A smoothing spline was used as a nonparametric representation of the nonlinear relationship between BOLD SI index and hyperemic MBF as a predictor. The optimal amount of smoothing was determined by generalized cross-validation. The nonparametric model analysis was chosen here due to the absence of any accepted functional model for the (nonlinear) relationship between the BOLD SI index and hyperemic MBF and because standard linear regression would represent a gross oversimplification. The direction and strength of the association between 2 variables were assessed with the Spearman correlation coefficient. Permutation tests of the correlations were performed to account for correlations of repeated measures, using corrperrm in the R package. The tests are based on combining the Spearman correlations or p values for all possible combinations of repeated measures for the 2 measured variables. Correlation coefficients were compared following the Fisher Z transformation.
Table 1 outlines baseline characteristics for the study participants. In the derivation arm of the study, 20 healthy volunteers and 25 CAD patients were recruited. In the latter, 11 patients (44%) had angiographically significant left anterior descending artery disease, 8 (32%), left circumflex artery disease, and 8 (32%), disease of the right coronary artery. Based on coronary anatomy, 54 myocardial segments were supplied by stenosed vessels (stenosed segments) and 96 segments, by vessels with minimal or no disease (segments remote to ischemia). Myocardial segments (n = 120) from normal volunteers were labeled as normal segments. In all subjects in the derivation arm of the study, there was no LGE in the slice positions corresponding to perfusion or BOLD imaging.
In the validation arm of the study, a total of 60 patients were recruited. Two subjects could not complete the scan due to claustrophobia, and in 1 subject, the study protocol was not carried out due to the finding of severe dilated cardiomyopathy on cine imaging. One subject requested cessation of the scan after stress BOLD imaging but before stress perfusion imaging could be performed (thus, 57 patients available for BOLD analysis, and 56 patients for perfusion analysis). No clinical cardiac events occurred in the period between the scans and coronary angiography. In total, 39 of 57 patients (68%) had angiographically defined significant CAD (stenosis of ≥50%): 60% with significant left anterior descending artery disease, 26% with left circumflex disease, and 33% with right coronary artery disease. LGE was present in 31% of patients (mean transmurality 52 ± 23%).
ROC curve analysis
In the derivation group, using quantitative coronary angiography as the gold standard, ROC curve analysis defined a cutoff value for stress MBF <1.8 ml/min/g (area under the curve 0.77 ± 0.03, p = 0.0001) to identify segments subtended by ≥50% stenosis. Using this as the threshold of ischemia, ROC curve analysis identified a BOLD SI index threshold of 2.64% (area under the curve 0.62 ± 0.04, p = 0.0009) to distinguish ischemic from nonischemic myocardium on a per-segment basis.
Relationship among tissue oxygenation, perfusion, and coronary stenosis
Segmental data from all subjects were pooled to investigate the relationship between tissue oxygenation and MBF. Figure 1 shows the relationship between the BOLD SI index and hyperemic MBF. The BOLD SI index increased with increasing MBF (r = 0.26), but with a plateau at higher MBF (>3 ml/min/g). However, there was evidence of dissociation between the BOLD SI index and MBF, with 50% of hypoperfused segments (<1.8 ml/min/g) demonstrating no evidence of deoxygenation (BOLD SI index >2.64%), and 23% of segments with MBF >1.8 ml/min/g showing some evidence of deoxygenation. Nonetheless, the likelihood of abnormal BOLD SI change decreased as hyperemic MBF increased (Fig. 2). Both the BOLD SI index and hyperemic MBF decreased with increasing stenosis, but there was a weaker correlation with myocardial oxygenation (r = −0.20, p < 0.001) than with perfusion (r = −0.40, p < 0.001; p = 0.005 for difference).
Impact of segment location, LGE, and cardiac risk factors on tissue oxygenation
Assessment of the BOLD SI index in normal volunteers and patients with normal coronary angiography revealed that the BOLD SI index varies by myocardial sector, with lower values in anterolateral and inferolateral segments (p = 0.0003 and p = 0.0033, respectively, compared with the BOLD SI index in anterior segments) (Fig. 3).
Figure 4 shows the relationships between the presence of LGE and both the BOLD SI index and MBF. Whereas the presence of LGE was associated with significantly lower MBF (p < 0.0001), there was no significant difference in the BOLD SI index between segments with and without LGE (p = 0.26). Figure 5 depicts the effects on an abnormal BOLD response of hyperemic MBF, LGE, and other cardiac risk factors, all included simultaneously as predictors in a multivariate logistic regression model. The odds of an abnormal BOLD response approximately doubled with the presence of hypertension (p = 0.025), and a quantitative coronary angiography stenosis >50% (p = 0.0048), although it was reduced by approximately one-fourth for each 1-ml/min/g increment of the hyperemic MBF response (p = 0.025). There was a trend for smoking (p = 0.088) to increase the odds of an abnormal BOLD response, whereas LGE (p = 0.97), hypercholesterolemia (p = 0.15), and diabetes mellitus (p = 0.77) did not significantly change the odds of an abnormal BOLD response.
MBF and BOLD SI index in normal volunteers and subjects with known CAD
Table 2 shows the MBF and BOLD SI changes for the 3 groups of myocardial segments in the derivation arm of the study. In normal volunteers, MBF increased from 0.88 ± 0.28 ml/min/g at rest to 2.86 ± 0.98 ml/min/g during hyperemia (p < 0.0001). In CAD patients, there was a blunted increase in hyperemic flow in diseased segments, from 0.76 ± 0.22 ml/min/g at rest to 1.59 ± 0.68 ml/min/g during stress (p = 0.0165 for change). Segments remote from ischemia showed an intermediate stress MBF between those of normal and stenosed segments (Table 2).
Table 2 also shows the relative BOLD SI changes. In normal volunteers, the mean relative increase in the BOLD SI index was 12%. However, in diseased segments, the relative BOLD SI increase was significantly lower (mean 4%), mirroring the behavior of stress MBF. Segments remote to ischemia demonstrated an intermediate increase in BOLD SI index, as reported in previous studies (16,19,20). Pooled segmental data from all subjects (N = 102) confirmed this finding.
Prospective application of BOLD and perfusion imaging
Using quantitative coronary angiography as the reference standard for determining anatomically significant CAD, 39 of 57 subjects (68%) had significant CAD. BOLD imaging achieved an accuracy of 86%, a sensitivity of 92%, and a specificity of 72% for identifying significant coronary stenosis, compared with 91%, 92%, and 89%, respectively, for perfusion imaging. Sensitivity, specificity, and predictive values for each technique by coronary territory are listed in Table 3.Figure 6 demonstrates patient examples of perfusion and BOLD imaging.
Using CMR perfusion imaging as the reference standard for determining functionally significant CAD, 38 of 56 subjects (68%) had evidence of ischemia (as defined by hyperemic MBF <1.8 ml/min/g). BOLD imaging using the predetermined cutoff of 2.64% achieved a diagnostic accuracy of 84% for the identification of ischemic myocardium, with a sensitivity of 92% and a specificity of 72%. Sensitivity, specificity, and predictive values for BOLD imaging to identify ischemia by coronary territory are listed in Table 4.
On a per-segment basis, agreement between BOLD and perfusion imaging for the overall diagnosis of CAD was 66% (kappa = 0.29). On a per-subject basis, agreement was 84% (kappa = 0.66).
BOLD MRI provides valuable insights into the pathophysiology of CAD. In this study involving multiparametric, high-field strength CMR, we demonstrate that hypoperfusion is not necessarily commensurate with deoxygenation, notably in segments with myocardial scar, where perfusion is reduced but without associated deoxygenation. Furthermore, although hypertension increases the odds of abnormal BOLD response, diabetes mellitus and hypercholesterolemia are not associated with significant deoxygenation. Although downstream hyperemic flow decreases with increasing coronary stenosis, the correlation between quantitative coronary angiography and deoxygenation is weaker. Nonetheless, our data demonstrate that this technique can identify the presence and functional significance of CAD with favorable accuracy.
Angiographic appearances do not reliably indicate the hemodynamic significance of a lesion. As demonstrated in previous studies, we also found that the BOLD effect in remote segments in patients with CAD had a level intermediate between stenosed segments and normal myocardium in volunteers, suggesting that not only epicardial flow limitation but also microvascular function may play an important role in the BOLD response (16,19). This may also explain the lower correlation of the BOLD SI index and MBF than in previous studies involving healthy canines with coronary hydraulic occluders (12,14). Our data reveal that tissue oxygenation correlates poorly with quantitative coronary angiography and that this relationship is even weaker than that between tissue perfusion and stenosis severity. Interestingly, we also found that microvascular perfusion may itself be uncoupled from tissue oxygenation; a significant proportion of hypoperfused segments did not show evidence of deoxygenation (19). This result highlights a potential limitation of perfusion imaging techniques, namely, that hypoperfusion is not necessarily commensurate with deoxygenation or ischemia. This is best exemplified in segments with LGE, where the hyperemic response is predictably reduced; by contrast, the BOLD SI index in these segments remains unchanged (Fig. 3). This may reflect the presence of nonischemic but viable tissue in the vicinity of scar, which returns a normal BOLD response. It is also possible that to avert ischemia in the face of hypoperfusion, there may be a down-regulation of energy-using processes, as has been postulated to occur in hibernating myocardium (29). There may exist a degree of physiological reserve, whereby perfusion may decrease within a safety margin, without corresponding deoxygenation.
In the context of adenosine stress testing, microvascular blood volume (MBV) may also be an important factor contributing to the dissociation between perfusion and the BOLD SI index. Although increased MBV may serve as a mechanism for meeting increased oxygen demand, MBF and MBV may become decoupled during vasodilator stress, with the occurrence of capillary derecruitment (9,16,30–32). Because the BOLD SI index is determined by absolute deoxyhemoglobin content, it reflects not only oxygenation, perfusion, and the balance of myocardial oxygen supply and demand, but also changes in coronary microvasculature. Reduced MBF represents only 1 factor contributing to myocardial ischemia; as a consequence, in pharmacologically induced hyperemia or in high cardiac workloads, there may be dissociation between MBF, MBV, oxygen delivery, and high energy metabolism. Only the latter can be considered a reliable marker of cellular ischemia; thus, both MBF and BOLD measurements may have potential limitations in the detection of true myocardial ischemia.
In our study, we elected to compare BOLD imaging with CMR perfusion imaging, which has the advantage of superior spatial resolution compared with alternative imaging modalities. This approach also ensured precise spatial coregistration of BOLD and perfusion images and that both imaging modalities were performed under the same hemodynamic conditions. By contrast, previous studies (involving positron emission tomography or single-photon emission computed tomography) invariably conducted the 2 imaging protocols under differing hemodynamic states (17,19). Our study is also the first human study to compare BOLD imaging with absolute quantification of perfusion with CMR, which itself has been shown to be superior to semiquantitative methods (33).
The BOLD SI index threshold defined in our study (<2.64%) is comparable to the thresholds defined in previous human studies (Jahnke et al. , <2.7%; Karamitsos et al. , <3.74%; Friedrich et al. , <1.2% [for >75% stenoses]). Previous studies of BOLD imaging in human subjects used populations with known CAD or high prevalence of CAD, which may distort assessment of sensitivity and specificity (17,19,20). Furthermore, to evaluate diagnostic performance, BOLD SI index thresholds were applied within the same study population from which they were determined. This approach may have led to overly optimistic estimates of the sensitivity and specificity (17,19,20). In our study, we used an independent derivation group to determine the BOLD SI index threshold, and to determine diagnostic performance, this threshold was then applied prospectively in a separate population with suspected CAD. This 2-step approach enabled a more robust assessment of diagnostic performance: our results reveal that despite only imaging 1 myocardial short-axis slice, BOLD CMR achieves high sensitivity (92%), but moderate specificity (72%). The latter may relate in part to the persistence of imaging artifact. Although the use of SSFP BOLD imaging at high field strength achieves higher signal-to-noise ratios and shorter acquisition times than alternative sequences, it remains susceptible to off-resonance effects associated with local field inhomogeneities. All previous studies applied a single BOLD threshold to all segments, with the inherent assumption that the BOLD SI index is uniform in all myocardial sectors (16,17,19,20). However, in our study, segmental data from subjects without established CAD reveal that the BOLD SI index is lower in the inferolateral and anterolateral segments. Whether this represents a physiological phenomenon or a limitation of the BOLD technique (such as altered signal with varying distance from the surface coil) warrants further investigation. Against the latter, normalization of BOLD signal by normalizing stress SI by the resting SI in the corresponding segment should theoretically counter any regional variation in signal. Nonetheless, the observation of variation in the normalized BOLD SI index by sector suggests that it may be advantageous to use separate thresholds for different myocardial sectors, and this approach may improve diagnostic specificity and accuracy.
In our study, although image quality was generally good, artifacts could not always be resolved with frequency and shim adjustments, and this may have contributed to the limited specificity of BOLD imaging. Invasive angiography was not performed in the normal volunteers, although significant CAD was unlikely given the finding of normal hyperemic flows in these subjects. Our study is also limited by the lack of invasive hemodynamic measures; validation of BOLD imaging against fractional flow reserve is an important goal in future studies. The use of a single mid ventricular slice for BOLD imaging is a further weakness of our study. It is possible that distal coronary disease may have been missed. In future studies, diagnostic performance will be improved with the use of multislice imaging with parallel imaging techniques and artifact reduction methods (20,34).
The noninvasive assessment of myocardial oxygenation by BOLD imaging offers valuable insights into the pathophysiology of CAD. Our data indicate that hypoperfusion may not necessarily imply cellular ischemia because there is apparent dissociation between myocardial tissue oxygenation and microvascular blood flow. Our study also provides prospective evaluation of BOLD imaging in the clinical setting. Although the results are promising, we identify a number of shortcomings that need to be addressed before this technique can fulfill a diagnostic role in the clinical setting.
The authors thank Prof. Richard Woodman for his statistical support.
This work was supported by the British Heart Foundation, the UK Medical Research Council, and the Oxford Partnership Comprehensive Biomedical Research Centre with funding from the Department of Health's NIHR Biomedical Research Centers funding scheme. All authors have reported that they have no relationships relevant to the contents of this paper to disclose. Dr. Arnold is the senior author.
- Abbreviations and Acronyms
- blood oxygen level–dependent
- coronary artery disease
- cardiovascular magnetic resonance
- late gadolinium enhancement
- myocardial blood flow
- microvascular blood volume
- receiver-operating characteristic
- signal intensity
- steady-state free precession
- Received October 22, 2011.
- Revision received December 20, 2011.
- Accepted January 3, 2012.
- American College of Cardiology Foundation
- Shaw L.J.,
- Berman D.S.,
- Maron D.J.,
- et al.
- Kern M.J.,
- Lerman A.,
- Bech J.W.,
- et al.
- Pauling L.,
- Coryell C.D.
- McCommis K.S.,
- Zhang H.,
- Goldstein T.A.,
- et al.
- McCommis K.S.,
- Goldstein T.A.,
- Abendschein D.R.,
- et al.
- Foltz W.D.,
- Huang H.,
- Fort S.,
- Wright G.A.
- Fieno D.S.,
- Shea S.M.,
- Li Y.,
- Harris K.R.,
- Finn J.P.,
- Li D.
- Dharmakumar R.,
- Arumana J.M.,
- Tang R.,
- Harris K.,
- Zhang Z.,
- Li D.
- Wacker C.M.,
- Hartlep A.W.,
- Pfleger S.,
- Schad L.R.,
- Ertl G.,
- Bauer W.R.
- Friedrich M.G.,
- Niendorf T.,
- Schulz-Menger J.,
- Gross C.M.,
- Dietz R.
- Beache G.M.,
- Herzka D.A.,
- Boxerman J.L.,
- et al.
- Karamitsos T.D.,
- Leccisotti L.,
- Arnold J.R.,
- et al.
- Jahnke C.,
- Gebker R.,
- Manka R.,
- Schnackenburg B.,
- Fleck E.,
- Paetsch I.
- Cerqueira M.D.,
- Weissman N.J.,
- Dilsizian V.,
- et al.
- Petersen S.E.,
- Jerosch-Herold M.,
- Hudsmith L.E.,
- et al.
- Fallavollita J.A.,
- Malm B.J.,
- Canty J.M. Jr.
- Sambuceti G.,
- Marzilli M.,
- Mari A.,
- et al.
- Jayaweera A.R.,
- Wei K.,
- Coggins M.,
- Bin J.P.,
- Goodman C.,
- Kaul S.