Author + information
- Received September 9, 2009
- Revision received December 1, 2009
- Accepted December 14, 2009
- Published online April 1, 2010.
- Cosima Jahnke, MD⁎,⁎ (, )
- Rolf Gebker, MD⁎,
- Robert Manka, MD⁎,
- Bernhard Schnackenburg, PhD†,
- Eckart Fleck, MD⁎ and
- Ingo Paetsch, MD⁎
- ↵⁎Reprint requests and correspondence:
Dr. Cosima Jahnke, Department of Internal Medicine/Cardiology, German Heart Institute Berlin, Augustenburger Platz 1, 13353 Berlin, Germany
Objectives This study determined the value of navigator-gated 3-dimensional blood oxygen level–dependent (BOLD) cardiac magnetic resonance (CMR) at 3.0-T for the detection of stress-induced myocardial ischemic reactions.
Background Although BOLD CMR has been introduced for characterization of myocardial oxygenation status, previously reported CMR approaches suffered from a low signal-to-noise ratio and motion-related artifacts with impaired image quality and a limited diagnostic value in initial patient studies.
Methods Fifty patients with suspected or known coronary artery disease underwent CMR at 3.0-T followed by invasive X-ray angiography within 48 h. Three-dimensional BOLD images were acquired during free breathing with full coverage of the left ventricle in a short-axis orientation. The BOLD imaging was performed at rest and under adenosine stress, followed by stress and rest first-pass perfusion and delayed enhancement imaging. Quantitative coronary X-ray angiography (QCA) was used for coronary stenosis definition (diameter reduction ≥50%). The BOLD and first-pass perfusion images were semiquantitatively evaluated (for BOLD imaging, signal intensity differences between stress and rest [ΔSI]; for perfusion imaging, myocardial perfusion reserve index [MPRI]).
Results The image quality of BOLD CMR at rest and during adenosine stress was considered good to excellent in 90% and 84% of the patients, respectively. The ΔSI measurements differed significantly between normal myocardium, myocardium supplied by a stenotic coronary artery, and infarcted myocardium (p < 0.001). The receiver-operator characteristic analysis identified a cutoff value of ΔSI = 2.7% for the detection of coronary stenosis, resulting in a sensitivity and specificity of 85.0% and 80.5%, respectively. An MPRI cutoff value of 1.35 yielded a sensitivity and specificity of 89.5% and 85.8%, respectively. The ΔSI significantly correlated with the degree of coronary stenosis (r = −0.65, p < 0.001). Additionally, ΔSI and MPRI showed substantial agreement (kappa value 0.66).
Conclusions Navigator-gated 3-dimensional BOLD imaging at 3.0-T reliably detected stress-induced myocardial ischemic reactions and may be considered a valid alternative to first-pass exogenous contrast-enhancement studies.
In patients with suspected or known coronary artery disease, noninvasive assessment of stress-induced myocardial ischemic reactions is essential for optimal planning of therapy and provides useful prognostic information (1,2). The evaluation of the hemodynamic consequences of coronary artery luminal narrowing, rather than mere detection of its presence, is clinically relevant. In general, severe epicardial coronary stenoses lead to post-stenotic microvascular dilation in a compensatory effort to maintain sufficient oxygen supply at least under resting conditions. Thus, the direct characterization of myocardial microcirculation reflecting myocardial tissue oxygen supply is highly desirable.
Blood oxygen level–dependent (BOLD) cardiac magnetic resonance (CMR) has recently been introduced for measurement of capillary reserve (3,4). The basic principle of BOLD imaging relies on the intravascular confinement and paramagnetic properties of deoxyhemoglobin used as an endogenous contrast agent (5). An increased oxyhemoglobin and a decreased deoxyhemoglobin tissue content result in higher T2* or T2 values, leading to a corresponding signal enhancement on T2*- or T2-weighted imaging (6). Thus, BOLD CMR holds the promise to sensitively determine changes in the myocardial oxygenation status without the need for exogenous contrast agent application as a surrogate marker of myocardial perfusion.
Data on BOLD CMR have mainly been reported in experimental settings (4,7–9) or highly selected and small patient populations (3,5,10). These studies dealt with the capability of BOLD CMR to assess the presence of significant coronary stenosis during vasodilator stress (i.e., dipyridamole or adenosine), although the diagnostic value was reported to be limited (3). Difficulties of BOLD CMR related to a relatively low signal-to-noise ratio in combination with impaired image quality due to artifacts resulting from magnetic field inhomogeneities and blood flow as well as breathing and cardiac motion. Most importantly, the effect size of stress-induced signal intensity changes (i.e., normal vs. ischemic myocardium) at 1.5-T is quite small, thereby necessitating quantitative evaluation rather than visual image interpretation (3,11). Thus, a clinically applicable BOLD CMR approach needs to ensure: 1) a consistently high signal level to differentiate normal from ischemic myocardium; and 2) a precise coregistration of rest and stress BOLD images as the basis for subtraction of identical slice geometries and calculation of signal intensity changes.
An increased signal-to-noise ratio can be achieved with the use of higher field strengths and can further be increased using 3-dimensional (3D) imaging. The latter yields the additional advantage of full left ventricular coverage in contrast to previously described single-slice imaging (3,5). Improved coregistration can be achieved with real-time navigator techniques restricting image acquisition to the end-expiratory level, which has been proven advantageous in comparison to multiple breath-hold maneuvers.
Thus, the present study sought to determine the feasibility and diagnostic value of free-breathing, navigator-gated 3D BOLD CMR at 3.0-T for the detection of stress-induced myocardial ischemic reactions in comparison to first-pass perfusion CMR and invasive angiography in an unselected patient population with suspected or known coronary artery disease.
Fifty consecutive patients (31 men, 19 women; mean age 61.3 ± 7.6 years; range 44 to 76 years) referred for cardiac catheterization were prospectively enrolled after written informed consent was obtained. Patients with suspected or known coronary artery disease with or without prior percutaneous revascularization were included. Patients were not considered for study inclusion if they had prior coronary artery bypass grafting or typical contraindications for CMR or the administration of adenosine. All patients were instructed to withdraw any antianginal medication and to refrain from cigarette smoking and tea or coffee intake for at least 24 h before the CMR examination. The study was approved by the Charité Institutional Review Board.
CMR was performed with the patient in the supine position using a 3.0-T CMR scanner (Philips Achieva, Best, the Netherlands) equipped with a Quasar Dual gradient system (40 mT/m; 200 mT/m/ms) based on Philips software release 1.2. A 6-element cardiac synergy coil was used for signal reception, and cardiac synchronization was performed with a vector electrocardiograph. After acquisition of standard cine scans for the assessment of left ventricular function, a 3D BOLD sequence was performed at rest. Subsequently, adenosine infusion (140 μg/kg/min) was started and the identical BOLD sequence was repeated after 2 to 3 min of adenosine infusion. Stress first-pass perfusion imaging was then performed (intravenous bolus application of 0.025 mmol/kg of gadolinium-diethylenetriaminepentaacetic acid [DTPA] [Magnevist, Schering, Berlin, Germany]; injection rate 4.0 ml/s followed by a 20-ml saline flush). After termination of adenosine infusion and a 10-min waiting period for equilibration of the contrast agent within the myocardium, rest perfusion images were acquired with the identical contrast injection scheme. Finally, a bolus of 0.15 mmol/kg of gadolinium-DTPA was administered and delayed enhancement images were acquired an additional 10 min later (Fig. 1).
For cine imaging, balanced turbo field echo sequences with retrospective gating (repetition time [TR]/echo time [TE]/flip angle: 3.3 ms/1.6 ms/40°; 30 phases per cardiac cycle; spatial resolution: 1.3 × 1.3 × 8.0 mm) during end-expiratory breath-holds of 6 to 8 s were used. Cine scans were acquired in 3 short-axis (apical-, mid-, and basal short-axis view) and 3 long-axis geometries (4-, 2-, and 3-chamber view) according to standard definitions.
The BOLD images were acquired in the short-axis orientation with full coverage of the left ventricle. A 3D T2-prepared segmented gradient echo sequence (TR/TE/flip angle: 5.0 ms/1.5 ms/30°; bandwidth: 290 Hz/pixel) with a T2 preparation time of 40 ms and a fat suppression pre-pulse was performed during free breathing with navigator gating used for respiratory motion compensation (end-expiratory gating window 6 mm). Navigator efficiency was defined as the number of accepted navigator-gated acquisitions divided by the total number of navigator acquisitions (values are given in percents). Data were acquired during systole (acquisition duration: 100 ms per heartbeat), with a nominal scan duration of 104 heartbeats for acquisition of the complete 3D data set. Measured in-plane spatial resolution was 0.9 × 1.3 mm with a slice thickness of 10 mm (reconstructed voxel size: 0.7 × 0.7 × 5.0 mm).
First-Pass Perfusion Imaging
For perfusion imaging, the identical geometries of the 3 cine short-axis views were used. A saturation-prepared single-shot spoiled gradient echo sequence (TR/TE/flip angle: 2.8 ms/0.9 ms/18°) was used, with 1 saturation pre-pulse per slice before data readout (pre-pulse delay: 95 ms). The 3 short-axis views were acquired every heartbeat (acquisition duration: 165 ms/slice) over 60 consecutive cardiac cycles. Measured in-plane spatial resolution was 2.9 × 2.9 mm (reconstructed voxel size: 1.5 × 1.5 × 8 mm).
Delayed-enhancement imaging was done in the short-axis orientation with full coverage of the left ventricle. A 3D inversion-prepared spoiled-gradient echo sequence (TR/TE/flip angle: 3.6 ms/1.8 ms/15°) was used (reconstructed voxel size 0.7 × 0.7 × 5.0 mm). The inversion recovery pre-pulse delay was determined from a Look-Locker sequence and adjusted accordingly (range 190 to 240 ms).
Quantitative coronary angiography (QCA)
All invasive coronary X-ray angiographies were performed within 48 h after the CMR examination. Quantitative coronary angiography (Philips Inturis Suite software) was used to define a significant coronary stenosis ≥50% luminal diameter narrowing in a coronary artery ≥2.0 mm diameter.
A visual score was used to grade the image quality of the 3D BOLD data sets at rest and during stress on a 4-point scale as excellent (4 = clear delineation of the left ventricular myocardium with sharp endocardial contours, no motion artifacts), good (3 = clear delineation of the left ventricular myocardium but mildly blurred endocardial contours, no motion artifacts), moderate (2 = moderately blurred endocardial contours, occurrence of respiratory/cardiac motion artifacts), or poor (1 = severely reduced demarcation of left ventricular endocardial contours and severe motion artifacts).
The BOLD images were analyzed semiquantitatively: the BOLD image data sets of stress and rest were subtracted pixel-wise and divided by rest values directly on the CMR scanner for determination of relative signal intensity changes (ΔSI) in percent (Fig. 2). The BOLD data sets of each patient were evaluated on a per-segment basis (standard 16-segment model) (12) using the ViewForum software (ViewForum Release 5.1, Philips Medical Systems, Best, the Netherlands). Myocardial segments were assigned to the supplying coronary artery based on a consensus read of the interventionalist and the CMR imager, taking the respective coronary dominance type into account.
For the assessment of intrareader variability, BOLD image analysis of 3 pre-defined myocardial segments per patient was repeated by the same CMR reader >3 months later; for the assessment of inter-reader variability, a second CMR reader performed the analysis independently.
In addition, first-pass perfusion images were semiquantitatively analyzed: endocardial and epicardial contours were traced manually and corrected for breathing-related motion. The myocardium was divided into 6 equiangular segments per slice according to standard definitions, and the regional signal intensity–time curves were measured. For baseline correction, mean signal intensity before contrast agent injection was subtracted from all post-contrast data. An index for myocardial perfusion reserve (MPRI) was calculated as previously published (13). Delayed-enhancement images were analyzed visually per segment regarding the presence of myocardial scar with a transmurality ≥50%.
Statistical analysis was performed using the SPSS software package, version 17.0.0 (SPSS, Inc., Chicago, Illinois). The paired Student t test was used to test for differences within groups. The normality of the distributions was tested with the Kolmogorov-Smirnov test. Error bars were used to compare the 95% confidence intervals of normal myocardial segments, segments supplied by a stenotic coronary artery, and infarcted segments. Differences between groups were tested by 1-way analysis of variance including pairwise multiple comparison procedures (Holm-Sidak method). A linear mixed-effect model was used to account for potential correlation within the patient. To determine the relationship between changes in BOLD signal intensity values/MPRI and the presence of coronary stenosis ≥50%, receiver-operator characteristic curve (ROC) analysis was performed and the area under the curve was calculated. Sensitivity, specificity, diagnostic accuracy, and positive and negative predictive values were calculated according to standard definitions and compared with the McNemar test. Pearson correlation was used to assess statistical correlation between BOLD imaging, MPRI, and QCA. Cohen's kappa was applied to measure inter-reader and intrareader agreement of BOLD imaging and the agreement between BOLD and MPRI using the following grading: 0 to 0.2 (poor), 0.21 to 0.4 (fair), 0.41 to 0.6 (moderate), 0.61 to 0.8 (substantial), and 0.81 to 1.0 (nearly perfect). In addition, Bland-Altman analysis was carried out to assess the inter-reader and intrareader reproducibility of BOLD measurements. All tests were 2-tailed; p < 0.05 was considered statistically significant.
Patient characteristics and hemodynamic data
During adenosine infusion, 1 patient suffered from severe dyspnea necessitating premature termination of the CMR examination; this patient was excluded from further analysis due to the incomplete CMR data acquisition. All remaining 49 patients completed the CMR examination successfully without any relevant side effects, thus 784 myocardial segments were included in the analysis. Table 1 summarizes the hemodynamic data and the effective scan duration of the BOLD sequence.
Invasive angiography excluded significant coronary luminal narrowing in 23 patients (11 patients without any angiographically assessable coronary disease and 12 patients with prior diagnosis of coronary disease but currently no significant stenosis). Of the remaining 26 patients, 15 had significant coronary single-vessel disease and 11 had double-vessel disease.
Delayed-enhancement imaging identified myocardial scar in 24 myocardial segments, and these segments were excluded for the comparison with QCA only. The remaining 760 myocardial segments were classified according to the QCA results as follows: 627 normal segments (176 segments without and 451 segments with known coronary artery disease but no stenosis ≥50% on current angiography) and 133 myocardial segments supplied by a coronary artery with luminal narrowing ≥50%.
Image quality of BOLD imaging
The mean visual score of BOLD CMR was 3.5 ± 0.7 at rest and 3.2 ± 0.8 during adenosine stress (p = 0.017). The BOLD imaging at rest resulted in excellent image quality in 28 patients (57.1%) and in good image quality in 16 patients (32.7%), 4 patients showed moderate image quality (8.2%), and only 1 patient showed poor image quality (2.0%). During adenosine stress, image quality of BOLD imaging was excellent in 22 patients (44.9%), good in 19 patients (38.8%), moderate in 6 patients (12.2%), and poor in 2 patients (4.1%). All myocardial segments of all 49 patients were considered for further analysis. Imaging examples are given in Figure 3.
Semiquantitative analysis of BOLD imaging
Linear mixed modeling was applied to account for multiple within-patient measurements with the repeated effect showing no significant influence on the results of ΔSI (p = 0.56). The ΔSI measurements differed significantly between normal myocardial segments (7.1 ± 0.3), myocardial segments supplied by a stenotic coronary artery (−7.1 ± 0.9), and infarcted myocardium (−1.9 ± 0.8): normal versus QCA ≥50%, p < 0.001; normal versus infarcted, p < 0.001; QCA ≥50% versus infarcted, p = 0.013; corresponding 95% CIs are illustrated in Figure 4A.
Semiquantitative analysis of myocardial perfusion reserve index
Linear mixed modeling excluded any significant influence of multiple within-patient measurements on the results of MPRI (p = 0.49). The MPRI measurements differed significantly between normal myocardium (MPRI, 1.88 ± 0.02) and myocardium supplied by a stenotic coronary artery (MPRI, 1.03 ± 0.02) or infarcted myocardium (MPRI, 1.14 ± 0.06) as follows: normal versus QCA ≥50%, p < 0.001; normal versus infarcted, p < 0.001; QCA ≥50% versus infarcted, p = 0.597 (Fig. 4B).
Comparison with QCA
The ROC analyses were performed to identify the cutoff value of ΔSI and MPRI for the detection of significant coronary stenosis (Fig. 5). For BOLD imaging, a cutoff value of ΔSI = 2.7% yielded a sensitivity, specificity, and diagnostic accuracy of 85.0%, 80.5%, and 81.3%, respectively; positive and negative predictive values were 51.9% and 96.2%, respectively. For MPRI, the optimal cutoff value was 1.35 with a sensitivity, specificity, and diagnostic accuracy of 89.5%, 85.8%, and 86.4%, respectively; positive and negative predictive values were 57.2% and 97.5%. Diagnostic values were significantly higher for MPRI in comparison to ΔSI (p = 0.014). The ΔSI of BOLD imaging and MPRI significantly correlated with the degree of coronary artery stenosis (r = −0.65 and r = −0.70, respectively; p < 0.001) (Fig. 6).
Comparison between BOLD and MPRI
The ΔSI of BOLD imaging showed a significant correlation with MPRI per myocardial segment (r = 0.69, p < 0.001) (Fig. 7). Applying the cutoff values derived from ROC analysis, 85.6% of myocardial segments were identically classified by BOLD and MPRI resulting in a substantial agreement (kappa value 0.66). Comparing BOLD imaging with MPRI as the reference standard, ROC analysis of ΔSI resulted in a sensitivity and specificity of 85.7% and 83.0%, respectively.
Inter-reader and intrareader variability of BOLD imaging
The ΔSI measurements as assessed by 2 different CMR readers resulted in a significant correlation (r = 0.91, p < 0.001) with a substantial inter-reader agreement (kappa value 0.77). The ΔSI measurements as assessed twice by the same CMR reader resulted in a significant correlation (r = 0.96, p < 0.001) and a nearly perfect intrareader agreement (kappa value 0.85). The corresponding Bland-Altman plots illustrating the inter-reader and intrareader reproducibility of BOLD ΔSI measurements are shown in Figure 8.
This study evaluated a navigator-gated 3D BOLD CMR approach at 3.0-T for the detection of stress-induced myocardial ischemic reactions. The navigator-gated BOLD sequence had an effective scan duration of <3 min, and thus can easily be integrated in the standard adenosine infusion protocol with a maximum infusion duration of 6 min. Notably, navigator efficiency significantly decreased during adenosine stress due to stimulated breathing; however, effective scan duration did not differ between rest and stress as a result of the concomitant increase in heart rate.
Importantly, navigator gating resulted in a consistent coregistration of rest and stress image geometry in all patients enabling automatic subtraction of the data sets and subsequent direct visualization of signal intensity changes. Because quantitative analysis is mandatory for BOLD CMR, automatic subtraction may facilitate time-saving evaluation. However, at present the need for semiquantitative evaluation of BOLD images may be considered a limitation with regard to the routine applicability of the technique.
The signal intensity differences between rest and stress reliably differentiated normally perfused myocardium from myocardial segments supplied by a stenotic coronary artery; in addition, infarcted myocardium demonstrated significantly different signal intensity behavior, and can thus be identified. When comparing ΔSI with myocardial perfusion reserve index, a significant correlation and a substantial agreement were found. Notably, ΔSI significantly correlated with the degree of coronary stenosis, and a cutoff value of ΔSI = 2.7% achieved a sensitivity and specificity of 85% and 81% for the detection of coronary stenosis ≥50%. However, because the cutoff value was defined in a retrospective manner from the data set, the definitive diagnostic value of BOLD CMR needs confirmation in future prospective studies.
Noninvasive stress testing for the detection of myocardial ischemic reactions represents one of the main objectives of CMR. With technical refinements of CMR taking place in recent years, the BOLD technique has emerged as a promising alternative to first-pass perfusion imaging and may benefit from inherent methodical advantages: BOLD imaging allows for the direct determination of myocardial oxygenation status and does not depend on the application of an exogenous contrast agent as a surrogate marker of myocardial blood flow (14). Hence, BOLD CMR abolishes the need for accurate timing of dynamic perfusion imaging during bolus arrival and provides the opportunity of sequence repetition if necessary. In addition, an improved spatial resolution can be achieved by distributing data acquisition over several cardiac cycles.
Although initial experimental (4,8) and volunteer (5) studies using BOLD imaging of the heart showed promising results, the commonly reported drawback was a low signal-to-noise ratio at 1.5-T together with misregistration resulting from breathing and limited left ventricular coverage due to single-slice acquisition. In an animal study, the benefit of increasing field strength to enhance sensitivity for the detection of myocardial oxygenation abnormalities has already been examined (15). Further evaluation of BOLD imaging at higher field strengths in patient studies has therefore been mandated (16). In the present study, a free-breathing, navigator-gated 3D BOLD sequence has been implemented on a 3.0-T clinical scanner system to achieve an increased signal-to-noise ratio, improved coregistration, and reduced breathing-related artifacts. The 3D approach facilitated full coverage of the left ventricle with a submillimeter in-plane spatial resolution.
The theoretical basis for cardiac BOLD imaging has been described in detail elsewhere (5,17). Briefly, in healthy subjects, myocardial oxygen supply exceeds demand during vasodilator stress. Myocardium being supplied by a normal coronary artery shows an increased oxygen saturation with a consequential relative decrease in deoxyhemoglobin under stress. On the contrary, myocardium being supplied by a narrowed coronary artery shows an already-dilated capillary bed even at rest, and further provocation of vasodilatation under stress is limited. Moreover, with persistent stress the proportion of deoxyhemoglobin further increases while any compensatory hyperemic effect is abolished; the resultant decrease in signal intensity as seen on BOLD images can thus be considered indicative of a stress-induced myocardial ischemic reaction (5,18).
In this study, during adenosine stress a mean increase in signal intensity of 7% was observed in all normal myocardial segments, whereas in patients without any coronary artery disease an average increase in signal intensity of 10% was seen. These findings are consistent with previous studies showing a higher stress-induced signal intensity increase in volunteers than in healthy patients (3). This phenomenon can be explained by an already-abnormal response of the microvasculature even in myocardial segments without any angiographically assessable epicardial coronary stenosis in patients with known coronary artery disease.
Myocardial segments being supplied by significantly narrowed coronary arteries showed a mean decrease of signal intensity of 7%. In contrast to previous patient studies noting signal intensity changes only in the presence of coronary stenosis ≥75% (3), we found a significant decrease in signal intensity already occurring in myocardial segments being supplied by ≥50% stenosed arteries. In an experimental setting, a linear relationship between signal intensity differences and coronary flow was established, indicating that BOLD imaging can differentiate between varying degrees of coronary stenosis (4). The proposed BOLD imaging approach of the current study confirmed a significant correlation between signal intensity changes and the degree of coronary stenosis in an unselected patient population with a higher correlation value compared with previous patient studies (3). The relatively large variability of signal intensity values in myocardial segments being supplied by an occluded coronary artery in the absence of myocardial scarring is most likely due to angiographically evident collateral circulation.
In addition, signal intensity differences between normal and chronically infarcted myocardium were seen: scarred myocardium is scarcely capillarized, hence capillary recruitment during vasodilator stress is negligible. Consequently, scar tissue as defined by delayed-enhancement CMR showed nearly identical BOLD signal intensities at rest and under adenosine stress.
Signal intensity changes of BOLD imaging correlated well with myocardial perfusion reserve index, and a substantial agreement could be proven. This is not surprising because both techniques functionally characterize stress-induced myocardial ischemic reactions. However, analysis of first-pass perfusion imaging was performed semiquantitatively, and absolute quantification of myocardial blood flow has not been attempted, which may be considered a limitation of the present study.
BOLD CMR can be regarded as a useful alternative to first-pass perfusion imaging in patients with suspected or known coronary disease.
The authors thank Uwe Kokartis for thoroughly performing the quantitative coronary angiographic measurements.
- Abbreviations and Acronyms
- relative signal intensity changes of blood oxygen level–dependent cardiac magnetic resonance
- blood oxygen level–dependent
- cardiac magnetic resonance
- diethylenetriaminepentaacetic acid
- myocardial perfusion reserve index
- quantitative coronary angiography
- receiver-operator characteristic curve
- Received September 9, 2009.
- Revision received December 1, 2009.
- Accepted December 14, 2009.
- American College of Cardiology Foundation
- Hundley W.G.,
- Morgan T.M.,
- Neagle C.M.,
- Hamilton C.A.,
- Rerkpattanapipat P.,
- Link K.M.
- Jahnke C.,
- Nagel E.,
- Gebker R.,
- et al.
- Friedrich M.G.,
- Niendorf T.,
- Schulz-Menger J.,
- Gross C.M.,
- Dietz R.
- Wacker C.M.,
- Hartlep A.W.,
- Pfleger S.,
- Schad L.R.,
- Ertl G.,
- Bauer W.R.
- Fieno D.S.,
- Shea S.M.,
- Li Y.,
- Harris K.R.,
- Finn J.P.,
- Li D.
- Cerqueira M.D.,
- Weissman N.J.,
- Dilsizian V.,
- et al.
- Al-Saadi N.,
- Nagel E.,
- Gross M.,
- et al.
- Dharmakumar R.,
- Arumana J.M.,
- Tang R.,
- Harris K.,
- Zhang Z.,
- Li D.
- Klocke F.J.,
- Li D.