Author + information
- Received December 24, 1996
- Revision received October 15, 1997
- Accepted November 18, 1997
- Published online March 1, 1998.
- Otto Muzik, PhDA,* (, )
- Claire Duvernoy, MDA,
- Rob S.B. Beanlands, MD, FRCP(C)A,
- Steve Sawada, MD, FACCA,
- Firat Dayanikli, MDA,
- Edwin R. Wolfe Jr., MSA and
- Markus Schwaiger, MD, FACCA
- ↵*Dr. Otto Muzik, Positron Emission Tomography Center, Children’s Hospital of Michigan, 3901, Beaubien Boulevard, Detroit, Michigan 48201-2196.
Objectives. Regional myocardial blood flow (MBF) and flow reserve measurements using nitrogen-13 (N-13) ammonia positron emission tomography (PET) were compared with quantitative coronary angiography to determine their utility in the detection of significant coronary artery disease (CAD).
Background. Dynamic PET protocols using N-13 ammonia allow regional quantification of MBF and flow reserve. To establish the diagnostic performance of this method, the sensitivity and specificity must be known for varying decision thresholds.
Methods. MBF and flow reserve for three coronary territories were determined in 20 normal subjects and 31 patients with angiographically documented CAD by means of dynamic PET and a three-compartment model for N-13 ammonia kinetics. Ten normal subjects defined the normal mean and SD of MBF and flow reserve, and 10 normal subjects were compared with patients. PET flow obtained in the territory with the most severe stenosis in each patient was correlated with the angiographic assessment of the stenosis (severity ≥50%, ≥70%, ≥90%). Receiver operating characteristic (ROC) curve analysis was performed for 1.5, 2.0, 2.5, 3.0 and 4.0 SD of flow abnormalities.
Results. MBF and flow reserve values from the normal subjects and from territories with documented stenoses ≥50% were significantly different (p < 0.05). A significant difference was found between normal subjects and angiographically normal territories of patients with CAD. High diagnostic accuracy and sensitivity, with moderately high specificity, were demonstrated for detection of all stenoses.
Conclusions. Quantification of myocardial perfusion using dynamic PET and N-13 ammonia provides a high performance level for the detection and localization of CAD. The specificity of dynamic PET was excellent in patients with a low likelihood of CAD, whereas an abnormal flow reserve in angiographically normal territories was postulated to represent early functional abnormalities of vascular reactivity.
The demand for an accurate and clinically applicable noninvasive method for the quantification of myocardial blood flow (MBF) has increased in parallel with the proliferation of therapeutic devices and recognition of the inadequacy of coronary angiography for the physiologic assessment of coronary artery disease (CAD). Experimental studies have shown that positron emission tomography (PET) allows accurate quantification of MBF under rest and stress conditions, indicating the potential utility of this technique for the evaluation of patients with known or suspected CAD. Analysis of rest/stress PET images is generally carried out visually and semiquantitatively, using data acquired in static fashion. Dynamic data acquisition allows the quantification of myocardial tracer kinetics, which can then be translated into MBF using the appropriate model. The clinical relevance of this method remains to be determined. It is hypothesized that quantification of coronary flow and flow reserve will permit accurate noninvasive definition of the functional significance of coronary stenoses and will allow for early delineation of abnormalities in vascular reserve that may be used as sensitive markers for atherosclerosis risk before the clinical appearance of disease [1, 2].
This analysis uses a three-compartment model and automatically defined myocardial regions of interest. Corrections for blood–tissue cross contamination and partial volume effects are incorporated within the equation used for the estimation of MBF . The method also incorporates a semiautomated correction for movement.
The feasibility of quantification of regional blood flow using this method has been demonstrated in normal volunteers . More recently, the potential value of this technique for the detection of CAD has been shown in a study correlating regional measurements of MBF with the results of quantitative coronary angiography.
The purpose of the present study was to examine, using receiver operating characteristic (ROC) curve analysis, the diagnostic utility of nitrogen-13 (N-13) ammonia PET using quantitative coronary angiography as the reference standard. The use of ROC curve analysis provides more complete information about the performance of a diagnostic test over a range of decision-making thresholds than does the determination of sensitivity and specificity using a single angiographic definition of significant disease.
Thirty-one patients (mean [±SD] age 62 ± 12 years) with angiographically documented CAD and 20 normal volunteers (mean age 44 ± 11 years) were studied using N-13 ammonia as a blood flow agent. Fourteen patients had one-vessel disease, 10 had two-vessel disease, and 7 had three-vessel disease. Eighteen patients had previously undergone cardiac catheterization, and 12 had undergone previous balloon angioplasty. No patient had undergone previous coronary artery bypass graft surgery. All patients were clinically stable at the time of PET studies and were offered the examination in addition to thallium-201 single-photon emission computed tomographic studies, which were used for clinical decision making. There was no evidence of cardiovascular disease in any of the volunteers, as assessed by history, physical and electrocardiographic examination. Before patient enrollment, the protocol was approved by the Institutional Review Board and the Radiation Safety Committee of the University of Michigan Medical Center. All patients gave written informed consent. The subjects underwent dynamic PET imaging at rest and during pharmacologic vasodilation.
1.2 PET Studies
N-13 ammonia was synthesized at the University of Michigan cyclotron facility by the oxygen-16(p, α)N-13 reaction, as described by Gelbard et al. . Dynamic PET measurements were performed using a whole-body PET scanner (Siemens 931) that allows simultaneous acquisition of 15 contiguous transaxial images with a slice thickness of 6.75 mm. The reconstructed image resolution obtained in this study was 8.5 ± 0.35 mm at full-width half-maximum in-plane and 6.6 ± 0.49-mm full-width half-maximum in the axial direction. A 15-min transmission scan was acquired before the emission study to correct for photon attenuation. After completion of the transmission scan, 20 mCi of N-13 ammonia diluted in 10 ml of normal saline was administered as a slow bolus over 30 s using a volumetric pump (model 975, Harvard Apparatus) through a peripheral intravenous line. At study onset, dynamic scan acquisition was initiated with varying frame duration (12 × 10 s, 4 × 15 s, 4 × 30 s, 3 × 300 s). The total scanning time was 20 min.
After acquisition of the baseline N-13 ammonia study, at least 50 min was allowed for N-13 decay(physical half-life 9.9 min). Subsequently, MBF was increased by intravenous adenosine infused at 0.14 mg/kg body weight per min over 6 min using a Harvard pump. Heart rate and blood pressure were continuously monitored beginning 3 min before the adenosine infusion and continuing for 10 min thereafter. Three minutes after the onset of the adenosine infusion, N-13 ammonia was injected. The PET data acquisition protocol was identical to the baseline study.
1.3 Regional Data Analysis
1.3.1 Three-Dimensional Reorientation
PET data analysis was performed by one experienced observer (O.M.) who evaluated all baseline rest/stress studies. The positron emission tomograph provides 15 reconstructed transverse images oriented perpendicular to the sagittal and coronal planes of the body. Using a SUN workstation (SUN Microsystems Inc.), 12 transaxial images were created in the short-axis view of the heart. The vertical and horizontal cardiac long-axis angles were defined using the last frame of the N-13 ammonia dynamic sequence featuring the best contrast between blood and tissue and were subsequently used for the reorientation of all 23 time frames.
1.3.2 Myocardial Tissue and Blood Pool Regions
We previously described a method for automated region definition that facilitates kinetic analysis of the acquired dynamic data set. Based on radial activity profiles, the algorithm automatically defines myocardial regions containing a blood volume fraction of 50% to 60%. The algorithm incorporates corrections for partial volume effects and blood–tissue cross contamination into the model equation .
Using the described algorithm, 12 myocardial regions/plane (Fig. 1) were defined in six to eight planes of the last time frame featuring a high contrast between blood and tissue. These regions were subsequently copied to all remaining time frames of the dynamic sequence. Before time–activity curves were generated, the dynamic image sequence was corrected for patient motion, as described previously . The dynamic image set was sampled and time–activity curves were generated in 12 myocardial sectors for each plane of the left ventricle.
To assess the left ventricular blood pool time–activity curve, circular regions (diameter 3 pixels) were drawn in the two most basal planes of the resliced image set. These regions have previously been found to be free of resolution distortions [1, 7].
1.3.3 MBF Calculation
The model for N-13 ammonia developed by Hutchins et al. and validated at our institution was used to calculate MBF and is described in detail elsewhere [1, 9]. In brief, we previously demonstrated that a three-compartment model allows separation of initial N-13 ammonia extraction from its retention. Based on experimental data by Schelbert et al. , the initial extraction fraction has been shown to be >90%, even for flow values up to 500 ml/100 g per min. Therefore, MBF (MBFammonia) can be expressed as (1)
The unidirectional clearance rate K1is estimated by fitting the following equation to the regional time–activity curves: (2)where Cmis the measured concentration obtained by the tomograph averaged over the acquisition period and is fitted to the averaged tissue concentration predicted by the model formulated on the right side of the equation; K1(ml/g per min) represents tracer delivery and extraction into the myocardium; k2and k3are true rate constants (1/min); Carepresents the arterial blood pool concentration; time tiis the midscan time of the ith scan; timepoints t1iand t2imark the beginning and the end of the ith data acquisition period; ρtissuerepresents myocardial tissue density in units of g tissue per ml tissue; and TBV defines the total blood volume fraction (vascular and spillover contribution) in the region studied.
As described by Hutchins et al. , the factors (1 − TBV) and TBV correct for resolution distortions caused by the finite resolution of the tomograph.
1.4 Correlation Between PET and Quantitative Coronary Angiography
To correlate the PET measurements with coronary anatomy, the starting edge of the first myocardial sector was defined at the posterior junction of the left and right ventricles. The remaining 11 sectors were then appended counterclockwise (Fig. 1). Each of the 12 sectors/plane was assigned to the distribution of one of the major coronary arteries: sectors 11, 12, 1, 2 to the right coronary artery (RCA); sectors 3, 4, 5, 6 to the left circumflex coronary artery (LCx); and sectors 7, 8, 9, 10 to the left anterior descending coronary artery (LAD). To increase the detection of apical defects, the planes were divided into basal and apical regions. Time–activity curves were derived for each individual sector and averaged to describe each of the six territories. The MBF was then estimated for each territory (apical and basal LAD, RCA and LCx) according to Eqs. (1) and (2).
All patients with CAD underwent quantitative angiography within 2 months of the PET study using standard techniques. All angiograms were analyzed by one experienced angiographer. Stenosis severity was determined using an automated quantitative arteriography computer program . Cine film images were digitized at 2,048 × 2,048-pixel resolution, and arterial geometric diameter as well as percent diameter stenosis were measured automatically. Stenoses were assigned to the LAD, LCx and RCA territories as follows: the first diagonal branch was assigned to the proximal LAD, and all subsequent diagonal vessels were assigned to the distal LAD territory. Similarly, lesions affecting the first obtuse marginal vessel were assigned to the proximal LCx, and all subsequent branches were assigned to the distal vessel. The right posterolateral and right posterior descending arteries were assigned to the distal RCA. For comparison with PET sector data, each stenosis was identified as proximal or distal in location. Proximal stenoses in the major coronary arteries were considered to affect both proximal and distal sectors, whereas distal stenoses were considered to affect distal sectors only.
Four stenosis severity groups were subsequently analyzed: angiographically normal territories (0% stenosis), and territories with stenoses ≥50%, ≥70% and ≥90%. A territory was assigned to one or more of these groups depending on the most severe stenosis in either the proximal or the distal vessel.
1.5 Statistical Analysis
Blood flow and coronary flow reserve (CFR) values were determined separately for normal subjects and patients with CAD. Results are expressed as mean value ± SD. Statistical significance was defined at p < 0.05.
To evaluate the ROC curve of PET-derived MBF and CFR estimates, sensitivity (true positives/number of patients with confirmed disease) and specificity (true negatives/number of patients without disease) were assessed. To compensate for potential correlation among anatomic territories, only the most severe stenosis for each patient was considered in the ROC curve analysis. If the stenosis severity in a patient was equal in two or more anatomic territories, these regions were combined into one territory, and the blood flows were averaged.
One group of normal volunteers (Norm 1) was used to determine the normal mean and SD of PET flow and flow reserve estimates. Using these reference values, the number of regions in each stenosis severity group (≥50%, ≥70% and ≥90% stenosis) that fell outside 1.5, 2.0, 2.5, 3.0 and 4.0 SD was determined. From these data, continuous ROC curves were generated from multiple sensitivity/specificity pairs according to the method of Metz . Furthermore, the area under the resulting ROC curve was used to rate the performance of the test for different stenosis severity groups. An area of 0.5 indicates that no diagnostic information was gained by performing the test, whereas an area of 1.0 characterizes a perfect test, with 100% sensitivity and specificity.
Overall differences in group means were examined first by one-way analysis of variance, with subsequent ttests (including the Bonferroni correction for multiple comparisons) to identify the sources of these differences.
2.1 Hemodynamic Data
Hemodynamic data from the 20 normal subjects and the 31 patients with CAD were analyzed at rest and during adenosine infusion. The rest rate–pressure product was 7,308 ± 1,723 beats/min × mm Hg in normal subjects and 8,569 ± 1,977 beats/min × mm Hg in the patients with CAD. During adenosine infusion, the rate–pressure product increased to 11,399 ± 2,626 beats/min × mm Hg in normal subjects, and decreased to 8,205 ± 1,797 beats/min × mm Hg in patients with CAD.
2.2 Angiographic Findings
Twenty-three men and eight women (62 ± 12 years old) underwent quantitative coronary angiography; the results are summarized in Table 1.
2.3 MBF and CFR in Normal Subjects
The 20 normal subjects were arbitrarily classified into two equal groups (Norm 1 and Norm 2) before analysis of the PET data. Mean rest MBF was 68.0 ± 13 ml/100 g per min in Norm 1 and 66.4 ± 9 ml/100 g per min in Norm 2. Stress MBF in Norm 1 was 299 ± 59 ml/100 g per min and CFR 4.42 ± 0.56. Stress MBF in Norm 2 was 273 ± 39 ml/100 g per min and CFR 4.15 ± 0.75. There were no significant differences in the values for MBF and CFR between the two groups. Overall rest MBF (Norm 1 and Norm 2) was 67.2 ± 11.2 ml/100 g per min, with an overall stress MBF of 285 ± 49 ml/100 g per min, and an overall CFR of 4.28 ± 0.65.
2.4 Correlation Between PET-Derived Stress MBF and CFR Estimates With Coronary Angiography
Vascular territories were assigned to the stenosis severity groups as follows: 38 (41%) showed 0% stenosis; 31 (33%) showed stenoses ≥50%; 23 (25%) showed stenoses ≥70%; and 17 (18%) territories were found with stenoses ≥90%. Fig. 2shows the average stress MBF in each of the five groups as well as the corresponding CFR values. Analysis of variance showed statistically significant differences between stress MBF and CFR values derived from the normal subjects and values derived from all other territories. Furthermore, angiographically normal territories were significantly different from territories with documented stenoses ≥50%.
2.5 ROC Analysis of Stenosis Severity Versus MBF and CFR Estimates
Three stenosis severity groups were used to determine the sensitivity and specificity pairs for stress MBF and CFR estimates. The respective ROC curves using ≥50%, ≥70% and ≥90% stenosis criteria for the determination of stress MBF and CFR abnormalities in the chosen coronary flow territories are displayed in Fig. 3.
A stress MBF threshold of 152 ml/100 g per min, 2.5 SD below the mean value obtained in normal subjects, showed the best discriminatory capacity for detecting CAD according to ROC curve analysis (this point was closest to the upper left-hand corner of the plots). For CFR, the threshold value showing the best discriminatory capacity was 2.74, 3.0 SD below the mean value in normal subjects.
The areas under the ROC curves for stress MBF and CFR estimates ranged between 0.79 and 0.91. The area under the ROC curve characterizing CFR derived from a ≥90% stenosis was the highest overall of the three groups.
The sensitivity, specificity and accuracy at 2 SD for detection of CAD in the most severely affected territory of a patient’s myocardium are shown in Table 2which also presents a comparison of current data with previously published results for detection of CAD using semiquantitative static PET analysis. CFR estimates showed a higher accuracy for all stenosis severity groups compared with stress MBF estimates. Both measurements were equally specific when applied to the normal subjects. Dynamic measurements showed the greatest sensitivity and accuracy for detection of CAD at the 90% stenosis threshold, with some corresponding loss in specificity at increasingly severe stenosis levels. Overall, dynamic and static PET measurements showed equivalent accuracy, with somewhat higher sensitivities and lower specificities seen for the dynamic versus the static analysis method.
Quantifying the functional severity of coronary stenoses is becoming increasingly important to guide treatment interventions, such as coronary artery bypass graft surgery or percutaneous transluminal coronary angioplasty. Noninvasive quantitative PET flow measurements and invasive coronary angiography are two of several modalities providing this type of information. At present, both these modalities are considered complementary. Together, they are thought to provide both an anatomic and functional description of coronary artery narrowing [13, 14]. The present study was designed to investigate the relation between quantitative flow measurements and angiographic appearance of CAD to define the potential clinical utility for the diagnostic process. To our knowledge, the present study represents the first time that dynamic PET acquisition with quantitative assessment of MBF and CFR has been characterized as to its clinical relevance in patients with known CAD. The present results demonstrate that absolute quantification of myocardial perfusion using dynamic PET and N-13 ammonia accurately detects and localizes CAD. In addition, calculation of the ROC curves allowed the definition of the threshold values for both absolute CFR and stress MBF that could provide the best diagnostic performance. This information may be clinically relevant for decision making in the future. Furthermore, this method appears to be more sensitive than coronary angiography for early stages of CAD, as evidenced by the high incidence of abnormal flow reserve in angiographically normal vessels of patients with CAD. For this investigation, quantitative coronary angiography was used as the reference standard and was related to the physiologic assessment of myocardial perfusion using PET. It could be argued that PET, which permits the evaluation of blood flow at the level of the microcirculation, might provide a superior characterization of CAD.
3.1 Diagnostic Performance of Dynamic PET Measurements
CFR was found to be more sensitive, but less specific, than stress MBF for localization of significant coronary stenoses. For all stenosis severity groups, diagnostic accuracy was found to be higher for CFR than for stress MBF. Our group has previously described the diagnostic utility of automated analysis of semiquantitative static PET analysis for the detection and localization of coronary stenoses . Results of that study are not directly comparable at all stenosis thresholds because the severity cutoffs differed for the middle range (70% vs. 75%). However, comparison of the present results obtained for dynamic PET analysis shows that diagnostic accuracy is essentially equivalent, with somewhat higher sensitivities and correspondingly lower specificities obtained for the dynamic method. Laubenbacher et al. also performed data analysis of static PET images using both the difference and the ratio between rest and stress polar maps. Using these mathematical combinations, they were able to improve sensitivity for the detection of CAD without any loss in specificity. Accuracy remained in the 80% range, similar to the current results. In the present study, the normalcy rate was determined for the control subjects and was found to be equally high for both CFR and stress MBF (97%). Our finding that both CFR and stress MBF are significantly lower in angiographically normal regions of patients with CAD than in normal subjects may have several explanations. Angiography is a poor method for the detection of diffuse disease, where flow limitations may nonetheless be significant. Furthermore, PET may be able to detect early functional abnormalities of coronary vasculature before the angiographic appearance of discrete stenoses . The calculation of absolute values in dynamic PET analysis may allow more accurate characterization of the extent of CAD than semiquantitative analysis, which depends exclusively on relative differences in tracer uptake.
3.2 Potential Methodologic Limitations
Dynamic acquisition of N-13 ammonia PET studies with subsequent quantitative analysis of rest and stress MBF and calculation of CFR shows great promise as a research tool. However, it does not appear to be intrinsically more accurate than semiquantitative static PET analysis in the detection and localization of clinically significant coronary artery stenoses. Given the relatively complex analysis techniques required, routine clinical use of this technique does not currently appear warranted. Furthermore, one major problem of flow quantification using PET is the decreasing retention fraction of perfusion radiotracers at high flows. To avoid this error, a three-compartment model for N-13 ammonia kinetics was developed at our institution that separates the initial extraction of N-13 ammonia from its retention in the myocardium . Assuming almost complete extraction of tracer during first pass by myocardial tissue, the transport rate constant K1directly reflects MBF. Validation of this assumption in open chest dogs using microspheres has shown a linear correlation between blood flow and K1up to flows of 500 ml/100 g per min [8, 16]. In contrast to the rate constant k3, which represents the metabolic retention of N-13 ammonia in tissue, K1merely reflects the delivery of the tracer into myocardial tissue.
Measurement of tissue tracer concentration is affected by the geometry of the tissue under study. The thickness of the left ventricular wall is less than the spatial resolution of the PET scanner, which results in underestimation of the true tracer tissue concentration (partial volume effect) and significant cross contamination of activity between the myocardium and surrounding tissue. The tracer kinetic model used in the present study incorporates a correction term for both these resolution distortions . By incorporating the resolution effects in the model, the dynamic behavior of the radionuclide concentration in blood and tissue enables the determination of the true myocardial tissue concentration. This method clearly simplifies the PET blood flow measurement without the need for wall thickness measurements by independent methods, such as echocardiography , or the application of a constant partial volume correction factor for various myocardial regions derived from phantom studies .
We previously developed a semiautomated analysis program that uses an automated ROI definition scheme. This program optimizes the placement of myocardial ROIs to account for resolution distortion using the described kinetic model. The method further uses a two-dimensional motion correction scheme to correct for patient motion artifacts, which is necessary to optimally use the kinetic model in the clinical setting.
Czernin et al. previously showed that stress MBF and CFR are both slightly correlated with patient age. Their data indicated a decrease in CFR on the magnitude of 0.2 to 0.3/decade of age, primarily caused by an increase in rest blood flow. In our study, the normal subjects were statistically significantly younger than the patients (44 ± 11 vs. 62 ± 12 years). However, the main focus of the present analysis was to examine the performance of dynamic PET for the detection and localization of CAD with regard to changing diagnostic thresholds. These thresholds are valid for the patient with CAD independent of their origin. Therefore, the age difference between the normal subjects and patients with CAD should not affect the significance of our results.
An additional limitation lies in the use of quantitative coronary angiography as the reference standard for comparison. This method has obvious inadequacies for the accurate detection and characterization of lumen narrowing within vessels. It may be more accurate to use intravascular ultrasound or angioscopy to more completely characterize vessel abnormalities.
The present study demonstrates that noninvasive quantification of MBF using N-13 ammonia and dynamic PET correlates well with quantitative coronary angiography. Our analysis of dynamic PET data compared favorably with previously published values for semiquantitative static PET analysis at all stenosis thresholds. However, its most valuable application may be for the detection of early changes in coronary vasculature that precede the angiographic appearance of disease and for the definition of the true extent of vascular abnormalities. To that end, the noninvasive assessment of the effects of specific pharmacologic or lifestyle interventions on MBF and CFR may provide a valuable research tool. Our results showed that CFR and stress MBF measurements differ in normal volunteers and in angiographically normal territories of patients with CAD. This difference may explain the observed low specificity of flow measurements in patients with CAD. Quantitative assessment of CFR provides a functional assessment of CAD that may be more sensitive than angiography in the early stages of CAD, when abnormalities in vascular reactivity predominate over fixed stenoses. In addition, CFR measurements may be more sensitive to the presence of diffusely diseased vessels. However, the prognostic significance of abnormal CFR in angiographically normal territories remains to be determined before clinical application of such quantitative measurements can be recommended for the workup of patients with CAD.
☆ This work was performed during the tenure of an Established Investigatorship of the American Heart Association, Dallas, Texas (Dr. Schwaiger) and was supported in part by Grant RO1 HL41047-01 from the National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland. Dr. Duvernoy was supported by a research stipend from the Alexander von Humboldt Foundation, Bonn, Germany. Dr. Beanlands is a Research Scholar of the Medical Research Council of Canada, Ottawa, Ontario.
- coronary artery disease
- coronary flow reserve
- left anterior descending coronary artery
- left circumflex coronary artery
- myocardial blood flow
- positron emission tomography (tomographic)
- right coronary artery
- receiver operator characteristic
- region of interest
- total blood volume
- Received December 24, 1996.
- Revision received October 15, 1997.
- Accepted November 18, 1997.
- The American College of Cardiology
- Dayanikli F,
- Grambow D,
- Muzik O,
- Mosca L,
- Rubenfire M,
- Schwaiger M
- Hutchins G,
- Caraher J,
- Raylman R
- Beanlands R,
- Muzik O,
- Sutor R,
- et al.
- Muzik O,
- Beanlands R,
- Wolfe E,
- Hutchins G,
- Schwaiger M
- Weinberg I,
- Huang S,
- Hoffman E,
- et al.
- Muzik O,
- Beanlands R,
- Hutchins G,
- Mangner T,
- Nguyen N,
- Schwaiger M
- Schelbert H,
- Phelps M,
- Huang S,
- et al.
- Gould K
- Gould KL,
- Kirkeeide RL,
- Buchi M
- Laubenbacher C,
- Rothley J,
- Sitomer J,
- et al.
- Krivokapich J,
- Stevenson L,
- Kobashigawa J,
- Huang S,
- Schelbert H
- Wisenberg G,
- Schelbert H,
- Hoffman E,
- et al.
- Czernin J,
- Mueller P,
- Chan S,
- et al.