Author + information
- Received May 30, 2005
- Revision received October 26, 2005
- Accepted October 31, 2005
- Published online April 18, 2006.
- Igor Klem, MD⁎,†,
- John F. Heitner, MD⁎,
- Dipan J. Shah, MD⁎,
- Michael H. Sketch Jr, MD⁎,
- Victor Behar, MD⁎,
- Jonathan Weinsaft, MD⁎,
- Peter Cawley, MD⁎,
- Michele Parker, RN, MS⁎,
- Michael Elliott, MD⁎,
- Robert M. Judd, PhD⁎ and
- Raymond J. Kim, MD⁎,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Raymond J. Kim, Duke Cardiovascular MRI Center, DUMC-3934, Durham, North Carolina 27710.
Objectives We tested a pre-defined visual interpretation algorithm that combines cardiovascular magnetic resonance (CMR) data from perfusion and infarction imaging for the diagnosis of coronary artery disease (CAD).
Background Cardiovascular magnetic resonance can assess both myocardial perfusion and infarction with independent techniques in a single session.
Methods We prospectively enrolled 100 consecutive patients with suspected CAD scheduled for X-ray coronary angiography. Patients had comprehensive clinical evaluation, including Rose angina questionnaire, 12-lead electrocardiography, C-reactive protein, and calculation of Framingham risk. Cardiovascular magnetic resonance included cine, adenosine-stress and rest perfusion-CMR, and delayed enhancement-CMR (DE-CMR) for infarction imaging. Matched stress-rest perfusion defects in the absence of infarction by DE-CMR were considered artifactual. All patients underwent X-ray angiography within 24 h of CMR.
Results Ninety-two patients had complete CMR examinations. Significant CAD (≥70% stenosis) was found in 37 patients (40%). The combination of perfusion and DE-CMR had a sensitivity, specificity, and accuracy of 89%, 87%, and 88%, respectively, for CAD diagnosis, compared with 84%, 58%, and 68%, respectively, for perfusion-CMR alone. The combination had higher specificity and accuracy (p < 0.0001), owing to incorporating the exceptionally high specificity (98%) of DE-CMR. Receiver operating characteristic curve analysis demonstrated the combination provided better performance than cine, perfusion, or DE-CMR alone. The accuracy was high in single-vessel and multivessel disease and independent of CAD location. Multivariable analysis including standard clinical parameters demonstrated the combination was the strongest independent CAD predictor.
Conclusions A combined perfusion and infarction CMR examination with a visual interpretation algorithm can accurately diagnose CAD in the clinical setting. The combination is superior to perfusion-CMR alone.
Although contemporary stress testing modalities are established methods for the diagnosis of coronary artery disease (CAD), the diagnostic performance of these tests is limited, even with the addition of myocardial imaging. For instance, the diagnostic accuracy of myocardial perfusion imaging with single-photon emission computed tomography (SPECT) might be as low as 65% to 70%, once the effect of referral bias has been taken into account (1,2). Given that over 10 million stress imaging studies are performed each year in the U.S. alone (3,4), an accurate noninvasive method of diagnosing obstructive CAD is of great importance to healthcare.
Cardiovascular magnetic resonance (CMR) can assess myocardial perfusion in a manner analogous to SPECT but with improved spatial resolution. Preliminary studies have shown promising results (5–9); however, there are still problems regarding image artifacts that can reduce specificity (8,9). Additionally, there are few data concerning the utility of perfusion-CMR for routine clinical practice, because most studies tested a time-consuming quantitative approach requiring extensive post-processing (5–8). In this context, it is noteworthy that a separate CMR technique—delayed enhancement-cardiovascular magnetic resonance imaging (DE-CMR)—is highly accurate for the diagnosis of myocardial infarction (MI) (10,11). Accordingly, we hypothesized that the addition of DE-CMR (scan time is increased only 5 to 10 min) to the stress-rest perfusion examination would help distinguish true perfusion defects from artifact and improve test reliability to the point that rapid visual interpretation could be performed with high accuracy.
In this study, we devised an interpretation algorithm that incorporates information from DE-CMR and perfusion-CMR in a proscribed manner for the diagnosis of obstructive CAD. We prospectively evaluated this algorithm by comparing the results to that of invasive X-ray angiography. To provide the best estimates of sensitivity and specificity, we studied a population with intermediate probability of CAD and with reduced pretest and post-test referral bias (12,13). Additionally, enrolled patients had comprehensive clinical evaluation including Rose chest pain questionnaire (14), 12-lead electrocardiography, blood tests for lipids and high-sensitivity C-reactive protein (hsCRP), and calculation of Framingham risk score (15) to provide context regarding the additional clinical value of the CMR examination.
The algorithm is on the basis of two principles. First, with perfusion-CMR and DE-CMR, we have independent methods to obtain information regarding the presence or absence of MI. Thus, one method could be used to confirm the results of the other. Second, DE-CMR image quality (e.g., signal-to-noise ratio) is far better than perfusion-CMR because it is less demanding in terms of scanner hardware (DE-CMR images can be built up over several seconds rather than in 0.1 s as is required for first-pass perfusion) (16). Thus, DE-CMR should be more accurate for the diagnosis of MI (16). Conceptually, it then follows that perfusion defects that have similar intensity and extent during both stress and rest (“matched defect”) but do not have infarction on DE-CMR are artifactual and should not be considered positive for CAD. Conversely, the presence of infarction on DE-CMR favors the diagnosis of CAD even if the results of perfusion imaging are equivocal. The algorithm is displayed in Figure 1.
Consecutive patients with suspected CAD referred to Duke Medical Center for elective coronary angiography were screened for study enrollment 3 days/week starting January 2003 and ending January 2004. Patients were contacted the day before scheduled angiography, and the first patient meeting study criteria who agreed to participate was recruited. To study the most relevant patient population and to reduce pretest referral or “spectrum” bias (12,13,17), we excluded all patients with known CAD, including those with prior MI or revascularization procedures. The only other exclusion criteria were contraindications to MRI (e.g., pacemaker) or adenosine (e.g., high-grade atrioventricular block). Written informed consent was obtained from all 100 enrolled patients. To avoid post-test referral bias (2,12,13,17), all patients underwent X-ray angiography within 24 h of CMR.
On the CMR procedure day, a complete medical history including responses to a Rose chest pain questionnaire (14) was obtained. Blood samples were drawn after an overnight fast for glucose, lipid profile, and hsCRP. C-reactive protein levels were measured with a rate immunonephelometric assay (Dade/Behring BNII, Dade/Behring, Deerfield, Illinois). Coronary artery disease risk factors were defined with Framingham heart study definitions (15) and the risk for CAD was calculated with the Framingham prediction algorithm (15); 12-lead electrocardiography was performed and scored for Q waves and bundle-branch block with Minnesota codes (18).
The CMR procedure consisted of four protocols that were performed in the following order: 1) cine imaging at rest for assessment of left ventricular (LV) function; 2) adenosine gadolinium first-pass imaging for assessment of stress perfusion; 3) repeated first-pass imaging without adenosine 15 min later for assessment of rest perfusion; and 4) DE-CMR for assessment of MI. Typically, the procedure was completed in 45 min.
A 1.5-T scanner (Siemens Sonata, Siemens, Malvern, Pennsylvania) with a phased-array receiver coil was used. Steady-state free-precession cine images were acquired in multiple short-axis (every centimeter throughout the LV) and three long-axis views. Typical parameters were: repetition time, 3.0 ms; echo time, 1.5 ms; flip angle, 60°; temporal resolution, 35 ms; voxel size, 1.7 × 1.4 × 6 mm. After cine imaging, the patient table was partially pulled outside the scanner bore to allow direct observation of the patient and full access. Adenosine (140 μg·kg−1·min−1) was infused under continuous electrocardiography and blood pressure monitoring for 2 min. The perfusion sequence was then applied, which automatically centered the patient back into the scanner and commenced imaging. Gadolinium contrast (0.065 mmol/kg gadoversetamide, Mallinckrodt, St. Louis, Missouri) followed by a saline flush (50 ml) was infused (3.5 ml/s) via an antecubital vein. On the console, the perfusion images were observed as they were acquired, with breath-holding starting from the appearance of contrast in the right ventricular cavity. Once the gadolinium bolus had transited the LV myocardium, adenosine was stopped and imaging completed 10 to 15 s later. Typically, the total imaging time was 40 to 50 s. Four to five short-axis slices (excluding most basal and apical slices) were obtained per heartbeat with a saturation-recovery, gradient-echo sequence (90° prepulse before each slice; echo time, 1.1 ms; delay time, 85 to 100 ms; temporal resolution, 110 to 125 ms; voxel size, 3.1 × 1.8-to-2.5 × 8 mm). The gap between images was set to 2 mm to have the exact same slice locations (center and plane) as cine-CMR. To speed imaging, either four k-space lines were acquired per excitation (echo-planar hybrid) or, in the last 15 patients, parallel imaging (19) with two-fold acceleration was employed. Five minutes after rest perfusion (additional 0.065 mmol/kg gadoversetamide), DE-CMR was performed with a segmented inversion-recovery technique (10,11) in the identical views as cine-CMR. Inversion delay times were typically 280 to 360 ms.
The scans were placed in random order and analyzed by the consensus of two observers who were masked to patient identity, clinical information, and the angiography results. To reduce the potential for observer bias, scans from 60 patients not enrolled in the study were included in the randomization. The individual CMR protocols and combinations were scored independently by rapid visual assessment on separate days as follows: cine alone, perfusion alone, DE-CMR alone, perfusion with DE-CMR (using the interpretation algorithm), cine with perfusion, and cine with the interpretation algorithm. For the combinations, the scoring algorithm was either explicit (i.e., interpretation algorithm) or, for those with cine-CMR, the observers were told to consider the cine findings when the perfusion results were equivocal.
Regional parameters were assessed with a 17-segment model (20). Hyperenhanced regions on DE-CMR were assumed to represent myocardial infarction (11) unless isolated midwall or subepicardial hyperenhancement was found. These latter patterns are found in nonischemic rather than ischemic disorders (21,22). For perfusion-CMR, stress and rest images were read side-by-side, and each of 16 segments (segment-17 at apex was not visualized) were scored with a four-point scale: 0, normal; 1, probably normal; 2, probably abnormal; 3, definitely abnormal. This scoring system was chosen to allow dichotomization of results into normal (≤1) and abnormal (≥2) and at the same time provide a range of scores for receiver operating characteristic (ROC) curve analysis. Because the two observers were not independent, inter-observer variability was not tested; however, only 8% of perfusion-CMR studies required a third reader to resolve disagreements.
Coronary angiography and analysis by coronary artery territory
X-ray coronary angiography was performed by standard techniques and interpreted masked to identity, clinical information, and the CMR results by the consensus of two experienced cardiologists. Luminal narrowing was estimated visually. In cases of disagreement, quantitative analysis was performed. Significant CAD was defined as ≥70% narrowing of the luminal diameter of at least one major epicardial artery or ≥50% narrowing of the left main (23).
To test the accuracy of the interpretation algorithm for individual coronary lesions, the readers also evaluated for each segment of the 17-segment model, the artery (i.e., left anterior descending coronary artery [LAD], right coronary artery [RCA], left circumflex coronary artery [LCx]) perfusing that segment, and the maximum level of stenosis. The algorithm was then applied on a segmental basis and deemed to be correct if the algorithm detected CAD in one or more of the segments perfused by a stenotic artery (true positive) or if the algorithm detected no CAD in all of the segments perfused by a non-stenotic artery (true negative). Segment-17 at the apex was interpreted as “normal” perfusion for the purposes of segmental analysis.
Continuous data were expressed as mean ± SD. Comparisons were made with two sample ttests for continuous data and chi-square tests for discrete. Fisher exact test was used when the assumptions of the chi-square test were not met. McNemar’s test was used to compare the diagnostic accuracy of techniques; pairwise comparisons were made with Bonferroni adjustment. Univariate logistic regression analysis was performed to assess the relation between predictor variables (Table 1)and the presence of CAD. Stepwise logistic regression analysis was then used to identify multivariable models. Receiver operating characteristic curve analyses were performed to compare the diagnostic performance of techniques (24). Statistical tests were two-tailed; p < 0.05 was considered significant.
Cardiovascular magnetic resonance imaging stress-testing was completed in 92 of the 100 enrolled patients. In three, imaging could not be performed or completed because of CMR-related issues: one did not fit in the scanner (body habitus); in one, the scanner electrocardiogram cable malfunctioned; and in one, a scanner software upgrade that morning prohibited opening the protocol. In five, imaging was omitted because of non–CMR-related issues: one had caffeine that morning; one withdrew consent; in one, intravenous access could not be obtained; in one, the contrast injection pump failed; and in one, there was severe adenosine-induced dyspnea, which led to early termination of the protocol. The patient with dyspnea, which quickly resolved after stopping adenosine, had the only adverse event during stress-testing. The infusion duration for adenosine was 2.9 ± 0.4 min. The heart-rate was 71 ± 13 beats/min at baseline and 93 ± 19 beats/min at peak stress. Of the 92 patients that completed imaging, 1 patient had atrial fibrillation and 2 patients had frequent ventricular ectopy chronically. All 92 patients were considered to have evaluable images and are included in the analysis.
Table 1summarizes the patient characteristics. Forty-nine percent were men; most had >1 CAD risk factor; and one-third had typical anginal chest pain. Only 40% had angiographically significant CAD, even though the majority (76%) had an abnormal nuclear, echo, or treadmill stress test before angiography. The mean LV ejection fraction was 56 ± 13% and was not significantly different between those with and without CAD (p = 0.40).
Comparison of CMR techniques for the detection of CAD
Table 2summarizes the diagnostic value of the interpretation algorithm and the individual CMR techniques for the detection of CAD. The interpretation algorithm had a five-point increase in sensitivity compared with perfusion-CMR alone (89% vs. 84%). The higher sensitivity was the result of two patients in whom infarction was observed on DE-CMR even though stress-rest perfusion-CMR was normal. Images exemplifying this situation are shown in Figure 2(top row). More commonly, however, stress perfusion imaging was more sensitive for CAD than DE-CMR or cine-CMR alone. Images from a typical patient with a reversible perfusion defect are shown in Figure 2(middle row).
The interpretation algorithm had markedly higher specificity than perfusion-CMR alone (87% vs. 58%, p < 0.0001). The higher specificity was primarily the result of changing the diagnosis from positive to negative for CAD in 13 patients in whom infarction was not observed on DE-CMR even though perfusion-CMR demonstrated matched stress-rest perfusion defects (Fig. 2, bottom row). In 12 of these patients (92%), the change in diagnosis was correct.
The interpretation algorithm provided higher diagnostic accuracy than the individual CMR methods when significant CAD was redefined with different cutoff values (Table 2) when only patients with single-vessel (n = 24) or multivessel disease (n = 13) were considered and when the analysis was performed on a “per coronary artery” rather than “per patient” basis (Table 3).The accuracies for the RCA, LAD, and LCx perfusion territories were similar at 88%, 85%, and 90%, respectively.
The addition of cine imaging did not improve the detection of CAD. The sensitivity, specificity, and accuracy of the combination of cine plus perfusion-CMR were 89%, 49%, and 65% respectively, which were not significantly different from that of perfusion-CMR alone (p = 0.26 for accuracy). Likewise, the addition of cine-CMR to the interpretation algorithm (sensitivity = 86%, specificity = 89%, accuracy = 88%) did not provide results that were significantly different from that of the interpretation algorithm alone (p = 1.00 for accuracy).
In the last 15 patients that underwent parallel imaging, the sensitivity, specificity, and accuracy of perfusion-CMR were similar to that in the first 77 patients (p = NS).
Comparison of clinical with CMR predictors of CAD
Univariate analysis demonstrated that male gender, elevated fasting glucose, reduced high-density lipoprotein cholesterol, elevated triglycerides, and increased Framingham score were clinical predictors of significant CAD. Each of the individual CMR methods and the interpretation algorithm were also predictors (cine-CMR, p = 0.04; DE-CMR, p = 0.0002; perfusion-CMR, p = 0.0002; interpretation algorithm, p < 0.0001). The interpretation algorithm was the strongest overall univariate predictor with an odds ratio of 56.6 (95% confidence interval, 15.3 to 208.8). Multivariable analysis including the clinical parameters and the interpretation algorithm demonstrated that the interpretation algorithm was the strongest independent predictor with an odds ratio of 54.1 (95% confidence interval, 12.7 to 229.8). The only remaining independent predictor was plasma high-density lipoprotein cholesterol with an odds ratio of 0.93 (p = 0.01).
Because the sensitivity and specificity of a diagnostic test depends on the choice of a cutoff point dichotomizing normal from abnormal, we performed ROC curve analysis to provide a metric of diagnostic performance over the range of cutoff points. For perfusion-CMR, the curve was derived from five sensitivity/specificity pairs obtained from defining a positive scan as a maximum perfusion score of at least “3,” “2,” or “1” (Methods) as well as the boundary conditions of 100% sensitivity and 100% specificity. For the interpretation algorithm, the curve was derived in a similar fashion, although “matched” perfusion defects in the absence of hyperenhancement were rescored as “0” (normal) before the decision threshold was applied. Figure 3demonstrates that the area under-the-curve for the interpretation algorithm was significantly higher than that for perfusion-CMR alone (0.94 vs. 0.84, p < 0.01). The areas under-the-curve for cine-CMR (0.62), DE-CMR (0.73), Framingham score (0.70), hsCRP (0.59), and Rose angina grade (0.65) were also lower.
The principal finding in this prospective study was that a new protocol that combines perfusion and DE-CMR using a simple interpretation algorithm could accurately diagnose CAD in a routine clinical setting. Our values for sensitivity (89%) and specificity (87%) were obtained in a cohort in whom all patients had intermediate pretest probability of obstructive CAD. We did not enroll patients with very high probability (i.e., those with already known CAD before CMR) or with very low probability (i.e., angiographic “normals” or low-risk volunteers), because it is well established that such pretest referral (“spectrum”) bias can inappropriately raise test sensitivity and/or specificity (12,13,17). Likewise, we did not enroll patients with known prior MI, because inclusion of these patients in investigations purporting to predict CAD is thought to be inappropriate (17). Moreover, our study was designed to avoid post-test referral (“work up”) bias—invasive coronary angiography was performed in all patients independent of the CMR results—because post-test referral bias can significantly affect test sensitivity and specificity (2,12,13).
To place our findings in context, we note that the current American College of Cardiology/American Heart Association/American Society for Nuclear Cardiology guidelines on radionuclide imaging list 52 published studies on exercise or vasodilator SPECT imaging for detecting CAD (see tables 5 to 7 in reference 2). Of these, only 11 excluded patients with prior MI, 5 excluded patients with known history of CAD, and 3 corrected for post-test referral bias (2 additional studies involved planar imaging). As far as we are aware, only two studies corrected for all three sources of bias. Cecil et al. (17) found a corrected sensitivity and specificity of 82% and 59%, respectively, and Miller et al. (1) found a corrected sensitivity and specificity of 65% and 67%, respectively. These results highlight two important points. First, the accuracy of stress-testing in the most appropriate patient population—those with intermediate pretest probability and after correction for post-test referral bias—is likely to be lower than accepted values from literature summaries (2). Second, in comparison with these results, the accuracy of the interpretation algorithm seems to indicate significant clinical utility.
Our study did not remove all potential sources of referral bias, because patients were selected from those already scheduled for invasive angiography. Because, for ethical reasons, it would be difficult to obtain angiographic verification on the real population that needs to be studied (i.e., all those presenting with suspected CAD, even those with very low pretest likelihood), the primary concern is that we might have selected a “sicker” cohort with higher prevalence or severity of disease. In our study, however, the prevalence of CAD was only 40%, and the majority had single-vessel rather than multi-vessel disease. This was despite the fact that 76% of patients had an abnormal clinical stress-test (nuclear, echo, or treadmill electrocardiography) before study enrollment. Although the prevalence of CAD could be considered unusually low, we note that previous studies that excluded patients with prior MI and established CAD have observed similar rates. Cecil et al. (17) found the prevalence of obstructive CAD was 36%. Likewise, Morise and Duval (25), in a study of patients undergoing coronary angiography for the first time, found a prevalence of 38%. Indeed, the low prevalence of CAD in our cohort emphasizes the need for an accurate, noninvasive test to rule out significant disease before invasive angiography.
Practical advantages of stress perfusion CMR include the lack of ionizing radiation, a shortened examination time (30 to 45 min), and a good safety and tolerability profile. Unlike vasodilator radionuclide imaging in which adenosine is infused for 6 min (tracer injection at 3 min), we used an abbreviated protocol with an average infusion duration of 2.9 min (gadolinium injected after a 2-min stage in which the patient is observed outside the bore), because waiting for tracer uptake is not necessary. Although severe reactions to adenosine are rare, a shortened protocol is relevant, because moderate reactions that affect patient tolerability are relatively commonplace (26). A minimum 2-min infusion duration was chosen on the basis of physiological studies in humans demonstrating that maximum coronary blood flow is reached, on average, 1 min after the start of intravenous adenosine infusion and in nearly everyone by 2 min (27,28).
The results of perfusion-CMR alone demonstrated an adequate sensitivity (84%) but a relatively poor specificity (58%) for the detection of CAD. These results are consistent with previous reports. Nagel et al. (8), with a gadolinium dose of 0.025 mmol/kg, found that visual assessment had a sensitivity of 74% and specificity of 58% for the detection of CAD. Although the modest sensitivity could be blamed on the low gadolinium dose, the poor specificity was problematic, given that it was unlikely to improve with higher gadolinium doses. Not surprisingly, the same group, with a higher dose of gadolinium (0.05 mmol/kg), more recently reported an improved sensitivity (91%) but unchanged specificity (62%) (29). These results underscore the value of incorporating DE-CMR, with its exceptionally high specificity (98%), into the stress CMR examination.
Although some other studies have reported higher specificity for perfusion-CMR alone, it is important to recognize that the results of these studies are not directly comparable with the current study. For instance, most studies did not test the feasibility of perfusion-CMR for everyday clinical use, in that they required central venous catheters (5,6), imaged only one to two slices per heart beat (6), excluded patients with hypertension or diabetes (8), included patients with known CAD or prior MI (8,9,29), or required extensive post-processing after data collection (5–8). Moreover, in some studies after the data were collected, several methods of analysis were tested and different decision thresholds for test abnormality were appraised (7,8). For these studies, the reported sensitivity and specificity values are optimistic, because the end points were chosen retrospectively.
It is likely that many CMR practitioners already perform a multi-component examination, including cine and delayed-enhancement imaging, for the assessment of CAD; however, there are few data regarding the advantages or disadvantages of different approaches to performing the examination or the analysis. For instance, there is debate whether the rest-perfusion component is necessary (30), and some authors advocate a stress-only approach because it shortens both examination and analysis time (31). Likewise, with a multi-component examination, there are many possible combinations of individual test results, and currently there are no data on how to deal with discordant findings. We believe the present study is the first to address these types of issues. For example, our results indicate that rest-perfusion is an important component, because in combination with DE-CMR, it can help distinguish true defects from artifact on the stress-perfusion images. Furthermore, our results indicate that an abnormal cine component alone (without perfusion or delayed-enhancement defects) is likely negative for CAD, whereas an abnormal delayed-enhancement component alone is likely positive.
The current study is limited in that we did not compare CMR with established perfusion techniques such as SPECT. This issue is relevant because X-ray angiography is not necessarily a perfect gold standard, and a true physiologic decrease in blood flow might be seen in the absence of coronary lesions (2). In the current study, however, patients without CAD but with risk of microvascular disease, such as those with angina, diabetes, or hypertension, were no more likely to have perfusion-CMR defects than those without these risk factors (p = 0.31, p = 0.92, and p = 0.60, respectively), suggesting that this issue did not play a major role. Moreover, despite its limitations, coronary angiography remains the final arbitrator for the diagnosis of CAD in clinical practice today; and, given the established nature of radionuclide imaging in the clinical armamentarium, we believed that it would not have been possible to perform such a study free of post-test referral bias—for ethical reasons it would have been difficult to refer patients with normal radionuclide examinations to undergo invasive angiography. Therefore, in the present study, we attempted to provide clinical context by obtaining a comprehensive evaluation in each patient, including chest pain questionnaire, blood tests for lipids and hsCRP, and calculation of Framingham risk. It is well recognized that clinical indexes such as these have predictive value for CAD (15), and our goal was to determine whether the interpretation algorithm significantly adds to the information provided by these readily available tests. The finding that the interpretation algorithm has independent predictive value for CAD with a predictive power that far exceeds that of the other parameters, we believe, provides strong evidence of the added clinical value of this technique.
The authors are indebted to Francis Klocke, MD, for careful review of the manuscript.
This work was supported in part by Grants R01-HL64726 (Dr. Kim), R01-HL63268 (Dr. Judd), and K02-HL04394 (Dr. Judd).
- Abbreviations and Acronyms
- coronary artery disease
- cardiovascular magnetic resonance imaging
- delayed enhancement-cardiovascular magnetic resonance imaging
- high-sensitivity C-reactive protein
- left anterior descending coronary artery
- left circumflex coronary artery
- left ventricle/ventricular
- myocardial infarction
- right coronary artery
- receiver operating characteristic
- single-photon emission computed tomography
- Received May 30, 2005.
- Revision received October 26, 2005.
- Accepted October 31, 2005.
- American College of Cardiology Foundation
- Klocke F.J.,
- Baird M.G.,
- Bateman T.M.,
- et al.
- ↵Information Means Value (IMV) MID. 2003 Nuclear Medicine Market Summary Report. Available at: http://www.imvlimited.com. Accessed February 16, 2006.
- Information Means Value (IMV) MID. 2003 Echocardiography Census Market Summary Report. Available at: http://www.imvlimited.com. Accessed February 16, 2006.
- Al-Saadi N.,
- Nagel E.,
- Gross M.,
- et al.
- Schwitter J.,
- Nanz D.,
- Kneifel S.,
- et al.
- Nagel E.,
- Klein C.,
- Paetsch I.,
- et al.
- Wolff S.D.,
- Schwitter J.,
- Coulden R.,
- et al.
- Kim R.J.,
- Fieno D.S.,
- Parrish T.B.,
- et al.
- Wilson P.W.,
- D’Agostino R.B.,
- Levy D.,
- Belanger A.M.,
- Silbershatz H.,
- Kannel W.B.
- Fuster V.,
- Kim R.J.
- World Health Organization,
- Rose G.A.
- Choudhury L.,
- Mahrholdt H.,
- Wagner A.,
- et al.
- McCrohon J.A.,
- Moon J.C.,
- Prasad S.K.,
- et al.
- Gibbons R.J.,
- Abrams J.,
- Chatterjee K.,
- et al.
- Metz C.E.,
- Herman B.A.,
- Roe C.A.
- Cerqueira M.D.,
- Verani M.S.,
- Schwaiger M.,
- Heo J.,
- Iskandrian A.S.
- Rossen J.D.,
- Quillen J.E.,
- Lopez A.G.,
- Stenberg R.G.,
- Talman C.L.,
- Winniford M.D.
- Wilson R.F.,
- Wyche K.,
- Christensen B.V.,
- Zimmer S.,
- Laxson D.D.
- Paetsch I.,
- Jahnke C.,
- Wahl A.,
- et al.
- Pennell D.J.,
- Sechtem U.P.,
- Higgins C.B.,
- et al.
- Giang T.H.,
- Nanz D.,
- Coulden R.,
- et al.