Author + information
- Received September 20, 2012
- Revision received January 17, 2013
- Accepted January 24, 2013
- Published online October 1, 2013.
- Andrew J. Swift, PhD∗,†∗ (, )
- Smitha Rajaram, MBChB∗,
- Judith Hurdman, MBChB‡,
- Catherine Hill, MBChB§,
- Christine Davies, MBChB§,
- Tom W. Sproson, BMedSci∗,
- Allison C. Morton, PhD†,‡,
- Dave Capener, MSc∗,
- Charlie Elliot, MBChB†,‡,
- Robin Condliffe, MD†,‡,
- Jim M. Wild, PhD∗,‡ and
- David G. Kiely, MD†,‡
- ∗Unit of Academic Radiology, University of Sheffield, Sheffield, England
- †National Institute of Health Research, Cardiovascular Biomedical Research Unit, Sheffield, England
- ‡Sheffield Pulmonary Vascular Disease Unit, Royal Hallamshire Hospital, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield, England
- §Radiology Department, Sheffield Teaching Hospitals Trust, Sheffield, England
- ↵∗Reprint requests and correspondence:
Dr. Andrew Swift, University of Sheffield, Academic Unit of Radiology, C Floor, Royal Hallamshire Hospital, Glossop Road, Sheffield S10 2J, England.
Objectives The aim of this study was to develop a composite numerical model based on parameters from cardiac magnetic resonance (CMR) imaging for noninvasive estimation of the key hemodynamic measurements made at right heart catheterization (RHC).
Background Diagnosis and assessment of disease severity in patients with pulmonary hypertension is reliant on hemodynamic measurements at RHC. A robust noninvasive approach that can estimate key RHC measurements is desirable.
Methods A derivation cohort of 64 successive, unselected, treatment naive patients with suspected pulmonary hypertension from the ASPIRE (Assessing the Spectrum of Pulmonary Hypertension Identified at a Referral Centre) Registry, underwent RHC and CMR within 12 h. Predicted mean pulmonary arterial pressure (mPAP) was derived using multivariate regression analysis of CMR measurements. The model was tested in an independent prospective validation cohort of 64 patients with suspected pulmonary hypertension. Surrogate measures of pulmonary capillary wedge pressure (PCWP) and cardiac output (CO) were estimated by left atrial volumetry and pulmonary arterial phase contrast imaging, respectively. Noninvasive pulmonary vascular resistance (PVR) was calculated from the CMR-derived measurements, defined as: (CMR-predicted mPAP – CMR-predicted PCWP)/CMR phase contrast CO.
Results The following composite statistical model of mPAP was derived: CMR-predicted mPAP = –4.6 + (interventricular septal angle × 0.23) + (ventricular mass index × 16.3). In the validation cohort a strong correlation between mPAP and MR estimated mPAP was demonstrated (R2 = 0.67). For detection of the presence of pulmonary hypertension the area under the receiver-operating characteristic (ROC) curve was 0.96 (0.92 to 1.00; p < 0.0001). CMR-estimated PVR reliably identified invasive PVR ≥3 Wood units (WU) with a high degree of accuracy, the area under the ROC curve was 0.94 (0.88 to 0.99; p < 0.0001).
Conclusions CMR imaging can accurately estimate mean pulmonary artery pressure in patients with suspected pulmonary hypertension and calculate PVR by estimating all major pulmonary hemodynamic metrics measured at RHC.
- cardiac output
- cardiovascular magnetic resonance imaging
- pulmonary artery pressure
- pulmonary capillary wedge pressure
- pulmonary hypertension
- pulmonary vascular resistance
Pulmonary hypertension (PH) is a condition of varied etiology defined by mean pulmonary artery pressure (mPAP) ≥25 mm Hg measured at right heart catheterization (RHC) (1). Mild elevations of pulmonary artery pressure are seen in the context of chronic respiratory and cardiac disease, whereas severe elevations of pulmonary artery pressure are seen in pulmonary arterial hypertension and chronic thromboembolic PH where specific therapies are available.
In addition to making measures of mPAP, RHC allows direct measurement of cardiac output (CO), pulmonary capillary wedge pressure (PCWP) and calculation of pulmonary vascular resistance (PVR), which are key variables in diagnosis, risk stratification, and follow-up. CO is an invasively measured variable that can be quantified noninvasively using a number of techniques including cardiac magnetic resonance (CMR) phase contrast imaging and cardiac volumetric analysis (2,3). On the contrary, use of CMR for estimation of mPAP, PCWP, and PVR remains uncertain. Doppler echocardiography is currently the modality of choice in clinical practice for noninvasive assessment of patients with suspected PH, however, the technique does not perform well for some etiologies of PH (4,5), and there is a limited role in patient follow-up.
RHC is the gold standard for establishing a diagnosis of PH and changes in PVR are thought to indicate the effectiveness of therapies for patients with pulmonary arterial hypertension. In severe PH, however, falls in mPAP may contribute to a fall in PVR in the setting of worsening of right ventricular (RV) function. Hence, some investigators consider CMR and RHC to be complementary in the assessment of patients with severe PH. An accurate and reproducible noninvasive technique capable of estimating pulmonary hemodynamic metrics in addition to morphological and functional measures of RV function would be highly desirable. Several preliminary studies have shown significant relationships between CMR metrics and invasively measured mPAP (6–10) and PVR (11,12) in PH, though the results of these studies were not validated in prospective cohort studies. The aim of this study was to derive and validate physiologically plausible composite numerical models of mPAP and PVR from CMR imaging parameters.
The cohort used for model derivation included consecutive treatment naive patients identified between July 2009 and March 2010 from the ASPIRE (Assessing the Spectrum of Pulmonary Hypertension Identified at a Referral Centre) PH registry (13). Inclusion required the patient to be referred with suspected PH and have undergone same day RHC and CMR imaging. Patients were excluded if the imaging was of nondiagnostic quality. Patients were classified following complete diagnostic work-up according to recent guidelines (1), including standard blood testing, echocardiography, lung function, exercise testing and imaging investigations including computerized tomography, perfusion lung scanning, CMR, and RHC. Ethical approval for retrospective review of routinely performed investigations in the derivation cohort was granted by the local institutional review board. Patients in the validation cohort were recruited from the Sheffield Pulmonary Hypertension Biobank and gave written informed consent to allow use of their prospectively collected data.
CMR imaging was performed on a 1.5-T whole body scanner GE HDx (GE Healthcare, Milwaukee, Wisconsin), using an 8-channel cardiac coil with the patients supine. Four-chamber and short axis (SA) cine images were acquired using a retrospective cardiac gated multislice steady-state free precession (SSFP) sequence. A stack of axial images in the SA plane with slice thickness of 8 mm with a 2 mm interslice gap or 10 mm with no interslice gap were acquired fully covering both ventricles from base to apex. The SSFP sequence parameters were repetition time (TR) 2.8 ms, echo time (TE) 1.0 ms, flip angle 50°, field of view 48 × 43.2, 256 × 256 matrix, 125 kHz bandwidth, and slice thickness of 8 to 10 mm.
Phase contrast flow imaging was performed with a compensated velocity encoded 2D gradient echo sequence with parameters were TR 5.6 ms, TE 2.7 ms, slice thickness 10 mm, field of view 48 × 28.8, bandwidth 62.5 kHz, 256 × 128 matrix, 20 reconstructed cardiac phases, and velocity encoding 150 cm/s. The plane of the phase contrast imaging was orthogonal to the pulmonary arterial trunk, 2 perpendicular SSFP planning slices were acquired in the long axis of the pulmonary artery to achieve the orthogonal position.
Image analysis was performed on a GE Advantage Workstation 4.4 and GE Advantage Workstation ReportCard, the observer was blinded to the patient clinical information and cardiac catheter parameters. Patient scans were defined as nondiagnostic when image quality significantly affected cardiac measurements or volumetric analysis could not be accurately performed. The CMR parameters were corrected where appropriate for body surface area (BSA), as previously reported in the literature (14,15).
Volume and mass measurements
Left and right endocardial surfaces were manually traced from the stack of SA cine images to obtain end-diastolic volume and end-systolic volumes, using MR workstation software (GE Advantage Workstation ReportCard). From end-diastolic and end-systolic volumes, the left ventricular (LV) and RV ejection fractions and right stroke volumes were calculated. LV stroke volume was multiplied by heart rate to determine CO. The RV epicardial and endocardial borders on each end-diastolic SA slice image were carefully outlined. The interventricular septum was considered as part of the LV. The myocardial volume for each slice was calculated by multiplying the area of the RV wall by the slice thickness. The product of the sum total of the myocardial slice volumes for each ventricle and the density of myocardium (1.05 g/cm3) gave an estimate of right ventricular mass (RV mass) (6,8) (Fig. 1). The LV epicardial and endocardial borders on each end-diastolic SA slice were also outlined, LV end-diastolic mass was thus derived. Ventricular mass index (VMI) was defined as RV mass divided by LV mass.
Phase contrast measurements
Phase contrast Q-flow CMR imaging was analyzed using ReportCard software, and the contours of the vessel were automatically traced, with manual correction when required. Positive pulmonary arterial flow (l/min) was recorded as the average from all pixels traced from the vessel region of interest. Average pulmonary arterial velocity (cm/s) was defined as average pulmonary arterial flow divided by average pulmonary arterial area. Maximal and minimal pulmonary arterial areas were measured, and relative area change was defined by the following equation: (maximal area – minimum area)/minimum area (10,15,16).
The interventricular septum was assessed on the mid-chamber SA cine cardiac images at the phase of maximal septal displacement. Interventricular septal angle was measured by determining the angle between the mid-point of the interventricular septum and the 2 hinge points. The angle between the 2 hinge points and the mid-point of the LV free wall was also measured (Fig. 1). Septal angle ratios were calculated from the ratio of interventricular septal angle to free wall angle.
Left atrial volume
Left atrial (LA) volume was estimated using the well-established biplane area length method (17,18). LV long axis (2-chamber view) and 4-chamber views were analyzed. LA volume was calculated from the equation: (0.85 × LA area 2-chamber view × LA area 4-chamber view)/([LA length 2-chamber + LA length 4-chamber]/2) (19,20). The LA volume was adjusted for body surface area (LA index) to estimate PCWP.
RHC was performed using a balloon-tipped 7.5-F thermodilution catheter (Becton-Dickinson, Franklin Lakes, New Jersey). PH was defined as mPAP ≥25 mm Hg at rest. Patients referred with suspected PH who were found to have a mPAP <25 mm Hg were defined as no PH. CO was measured using the thermodilution technique. PVR was determined as follows: PVR = (mPAP – PCWP)/CO. Cardiac index (CI) was corrected for the patient's BSA: CI = CO/BSA.
The derivation cohort was used to build a composite numerical parametric model of mPAP. To satisfy the assumptions of the statistical model each variable was tested for any nonlinear relation between continuous independent variables and mPAP. Scatter plots of CMR metrics versus mPAP were constructed. Regression curve fitting determined the best fit relationship using linear, quadratic, exponential, logarithmic, inverse-linear, or power analysis. To achieve linearity, variable transformations were performed where necessary. Statistically significant CMR variables at univariate analysis were entered into a multiple linear regression model of mPAP. The CMR data from the prospective validation cohort were used to test the CMR-predicted mPAP versus RHC measured mPAP. Comparisons of CMR measurements between the derivation and validation cohorts (Table 1) were analyzed using an independent t test for continuous data, and the chi-square test for categorical data.
CMR-measured LA volume index was considered a surrogate marker of PCWP. This was based on previous work showing that LA volume measurements are reproducible (20) and have a reasonable correlation with PCWP (21). Noninvasive estimation of PCWP was made using linear regression analysis to determine the relationship between LA volume and invasively measured PCWP. Transpulmonary gradient was thus determined as follows: transpulmonary gradient = CMR-predicted mPAP – CMR-estimated PCWP. CMR phase contrast imaging has been well validated as a method of estimating CO in humans (3). By dividing CMR-estimated transpulmonary gradient (predicted mPAP – predicted PCWP) by phase contrast CO, a CMR-based PVR was derived, assuming the standard formula PVR = mPAP – PCWP/CO.
Linear regression was used to assess the strength of the relationships between invasive mPAP, PCWP, CO, and PVR and CMR-derived mPAP, PCWP, CO, and PVR measurements. Bland-Altman analysis was used to assess the agreement between CMR-derived hemodynamics parameters versus invasively measured values (22). The bias, standard deviation, and the 95% limits of agreement were reported. Receiver operating characteristic (ROC) curves were constructed to determine the diagnostic accuracy of CMR-derived mPAP for the detection of PH, and were used to test the diagnostic value of CMR-predicted PVR for detection of PVR >3 WU. Diagnostic accuracy was quantified by the area under the ROC curve. Area under the ROC curve ≤0.5 indicates no value. The closer the area is to 1.0, the greater the diagnostic utility and significance of the test. A p value <0.05 was considered statistically significant. To perform and display the statistics, PASW 18 (PASW, Chicago, Illinois) and GraphPad Prism 5.04 (GraphPad Software, San Diego, California) software were used.
A total of 69 patients were identified having same-day RHC and CMR from the ASPIRE registry (13), of whom 64 patients had diagnostic quality imaging. A second cohort of 66 patients were prospectively recruited, 2 of whom had nondiagnostic imaging. Thus, 64 patients were included in the validation cohort. Patient demographics and hemodynamics are shown in Table 1.
Estimation of mPAP
Linearity of CMR metrics versus mPAP was found with all tested variables except for pulmonary artery relative area change for which the best fit relationship was exponential. Table 2 presents the correlations between CMR variables and invasively measured mPAP at RHC. The multivariate linear regression model of best fit identified the measurements interventricular septal angle and VMI index. The parametric relationship was as follows: MR-predicted mPAP = –4.6 + (interventricular septal angle × 0.23) + (VMI × 16.3).
In the derivation cohort, CMR-predicted mPAP and invasive mPAP demonstrated a significant correlation (R2 = 0.75, p < 0.0001) (Fig. 2). High diagnostic accuracy for the identification of PH was identified, area under the ROC curve was 0.94 (0.87 to 1.00; p < 0.0001).
The model of predicted mPAP correlated well with invasive mPAP in the validation cohort (R2 = 0.67, p < 0.0001) (Fig. 3). For detection of the presence of PH, the area under the ROC curve was 0.96 (0.92 to 1.00; p < 0.0001), and predicted mPAP of ≥32 mm Hg was identified as the most favorable threshold with 87% sensitivity and 90% specificity (Fig. 4).
Intraobserver and interobserver agreement were measured in 20 randomly selected patients. There was good intraobserver agreement intraclass correlation coefficient 0.99 (0.97 to 2.00) and Bland-Altman analysis showed a bias of 1.1° ± 6.3° with 95% limits of agreement of –11.3 to 13.4. Good interobserver agreement was also shown intraclass correlation coefficient 0.97 (0.91 to 0.99) and Bland-Altman analysis showed a bias of 1.3° ± 6.9° and limits of agreement of –12.2 to 14.9°.
Phase contrast CO
LV SSFP and phase contrast CMR both correlated moderately with CO measured by thermodilution (R2 = 0.46 and 0.49, respectively, in the derivation cohort). When directly compared phase contrast CMR and LV SSFP also had a moderate correlation (R2 = 0.49). The correlation between CO estimated by thermodilution and phase contrast CMR was significant (R2 = 0.44, p < 0.0001) in the validation cohort. Bland-Altman agreement analysis results for CMR-derived hemodynamic including CO are presented in Table 3.
Estimation of PCWP
The regression equation of estimated PCWP from the derivation cohort was as follows: CMR-derived PCWP = 6.43 + LA volume index × 0.22. In the retrospective and prospective validation cohorts, modest correlations (R2 = 0.36 and 0.49, respectively) were found between invasively measured PCWP at RHC and CMR-estimated PCWP. See Table 3 for Bland-Altman agreement analysis results. For the noninvasive detection of elevated PCWP (≥15 mm Hg) in the validation cohort, the area under the ROC curve for the detection of elevated PCWP was 0.90 (0.78 to 1.00; p < 0.0001) with an optimal threshold value of ≥15 mm Hg identified as the optimal cutoff threshold with sensitivity of 85% and specificity of 94%.
Noninvasive CMR- and RHC-derived PVR correlated well in the derivation and validation cohorts (R2 = 0.67, p < 0.0001; R2 = 0.76, p < 0.0001, respectively). See Table 3 for Bland-Altman agreement analysis results. For detection of elevated PVR (≥3 WU at RHC) in the derivation cohort the area under the ROC was 0.95 (0.90 to 1.00; p < 0.0001). For detection of elevated PVR in the validation cohort the area under the ROC remained high at 0.94 (0.88 to 0.99; p < 0.0001).
This study has derived, for the first time, a CMR noninvasive composite parametric regression model for mPAP and demonstrated that it has good diagnostic accuracy when used prospectively in a heterogeneous cohort of patients with suspected PH. In addition, we have shown that PVR can be predicted using CMR estimates of mPAP, CO, and PCWP and that this prediction also has high diagnostic accuracy. These results demonstrate that invasively measured variables, traditionally made at RHC, that are crucial for clinical evaluation of patients with PH (pressure, vascular resistance, and flow) can be non-invasively estimated using CMR imaging in this patient cohort. Importantly, CMR is noninvasive, reproducible (23,24), sensitive to change (25–27), and does not require ionizing radiation, making it an ideal modality for ongoing assessment of patients with PH.
In current clinical practice echocardiography is used as the first line investigation for suspected PH, allowing assessment of RV morphology and function and estimation of pulmonary arterial pressure. Using the modified Bernoulli equation the pressure gradient between the right ventricle and right atrium based on the velocity of the tricuspid regurgitant jet can be calculated. Estimated right atrial pressure is added to this equation to predict pulmonary artery systolic pressure (28) and a number of studies have subsequently used this measure to estimate mPAP. Several studies have identified a strong relationship between echocardiographic measures of mPAP made at RHC (29,30). However, the diagnostic utility of echocardiography-derived mPAP has been questioned in specific underlying etiologies (4). Moreover, a recent prospective study of unselected patients with PH, has shown the under and over estimation of Doppler echocardiography-derived mPAP with limits of agreement ranging from –40 to 38.8 mm Hg (28). Concerns regarding the reproducibility of echocardiographic measures have been expressed and challenges exist in accurate delineation of RV morphological features. Consequently, echocardiography has not yet been adopted as a stand-alone tool capable of accurately measuring pulmonary hemodynamic metrics and RV morphology and function. RHC has been advocated as the gold standard to assess the response to treatment when results of noninvasive assessments such as clinical assessment, 6-min walking test and echocardiography have been equivocal. Concerns regarding falls in PVR measured at RHC in the setting of worsening right ventricular function have resulted in some physicians advocating an approach based on using both RHC and MR metrics.
Several studies have identified significant relationships, between CMR imaging metrics and mPAP (6,8–10,31–35) and PVR (11,36,37) in patients with PH. However, this is the first study to devise a CMR-based prediction model for estimation of mPAP and to validate the results prospectively. The numerical linear regression model, a composite index of VMI (the ratio of RV:LV mass) and interventricular septal angle correlated well with mPAP in the derivation and prospective validation cohorts and showed good diagnostic utility. This model includes 2 variables namely, VMI and the degree of septal curvature, that are physiologically linked to pulmonary artery pressure in addition to having a strong statistical relationship. VMI has been shown in previous work by other investigators to correlate strongly with mean pulmonary arterial pressure, likewise septal curvature has been shown to relate well to pulmonary arterial pressure (6,31). This study describes a novel, simple, and reproducible method of evaluating septal curvature based on the angle of the septum in relation to the 2 septal hinge points, a similar correlation between septal angle verses mPAP in comparison to septal curvature versus mPAP from previous work was found (7,38). We also sought to estimate PCWP in order to replicate the key measurements made at RHC, which would also allow a noninvasive estimation of PVR. LA size adjusted for BSA was used as a plausible physiological correlate and demonstrated a moderate correlation with invasively measured PCWP. Further study of CMR measures to try and improve estimation of LA pressure, estimated at RHC by PCWP, such as transmitral flow measurements with the addition of septal tissue velocity, may be of value for improved estimation of LA pressure.
A recent study (36) used a derivation and validation cohort to devise and test a noninvasive CMR model of PVR. The study identified a prediction rule based on RV ejection fraction and the natural log of average pulmonary arterial velocity, the model performed well in the validation cohort. The limits of agreement at Bland-Altman analysis were –6.0 to 4.9 WU, and area under the ROC curve analysis for the diagnosis of PH was 0.97. Our study derives a prediction rule for mPAP based on physiological measurements, an estimate of PCWP based on LA volume and phase contrast CMR CO (39,40) to derive PVR. In comparison, narrower limits of agreement were found in the present study at Bland-Altman analysis, –5.1 to 4.6 WU. Additionally, CMR-derived PVR was found to have a high degree of accuracy for predicting raised PVR as measured by catheterization (area under the ROC curve: 0.94), while diagnostic accuracy for PH was also strong (area under the ROC curve: 0.96).
Using multivariate linear regression analysis, a parametric model, to derive an equation to fit a physiological process inherently has its limitations, as the models are not physically founded. However, the variables identified by the multivariate model VMI and interventricular septal position have been shown in previous work to be strong consistent markers of mPAP (6,7,34). Furthermore RHC, estimation of PVR is made with the variables PCWP, mPAP, and CO, using the equation: PVR = (mPAP – PCWP)/CO. PCWP measurements can be inaccurate in estimating LV end-diastolic pressure. Ideally a CMR surrogate of LV end-diastolic pressure measured at left heart catheterization would be superior to a surrogate of PCWP. Further study assessing the correlation between LA CMR measurements and LV pressures is required. Estimates of CO using CMR have been shown to have moderate correlation with invasive measurements (3), however, phase contrast measurements may be affected by turbulent flow, which may result in inaccuracies in this measurement. Finally it is also necessary to consider the logistic and financial issues associated with implementation of CMR in the assessment of patients with PH, such as cost, availability, and expertise.
This study demonstrates the feasibility of CMR to estimate key pulmonary hemodyamic measurements in addition to morphological and functional RV measurements, making this an ideal noninvasive modality for the assessment and follow-up of patients with PH. In clinical practice follow-up CMR may reduce the need for repeat RHC by providing the physician with an estimate of hemodynamic parameters previously only confidently measured at cardiac catheterization, in addition to CMR prognostic indexes (14).
CMR imaging, in addition to providing detailed functional and morphological information, can accurately estimate mean pulmonary artery pressure in patients with PH and calculate PVR by estimating all major pulmonary hemodynamic metrics measured at RHC.
Dr. Elliot has received support from Actelion Pharmaeuticals and Glaxo Smith Klein to attend conferences and has also received lecture fees. Dr. Kiely has served on advisory boards and given lectures for Actelion, Bayer, GlaxoSmithKline, Pfizer, Lily, and United Therapeutics; and the department in which he works has received unrestricted educational grants from Actelion, Bayer, and Pfizer. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- body surface area
- cardiac magnetic resonance
- cardiac output
- left atrial
- left ventricle/ventricular
- mean pulmonary arterial pressure
- pulmonary capillary wedge pressure
- pulmonary hypertension
- pulmonary vascular resistance
- right heart catheterization
- receiver-operating characteristic
- right ventricular
- short axis
- steady-state free precession
- ventricular mass index
- Received September 20, 2012.
- Revision received January 17, 2013.
- Accepted January 24, 2013.
- American College of Cardiology Foundation
- Galie N.,
- Hoeper M.M.,
- Humbert M.,
- et al.
- Fisher M.R.,
- Criner G.J.,
- Fishman A.P.,
- et al.
- Saba T.S.,
- Foster J.,
- Cockburn M.,
- Cowan M.,
- Peacock A.J.
- Hagger D.,
- Condliffe R.,
- Woodhouse N.,
- et al.
- Hurdman J.,
- Condliffe R.,
- Elliot C.A.,
- et al.
- van Wolferen S.A.,
- Marcus J.T.,
- Boonstra A.,
- et al.
- Appleton C.P.,
- Galloway J.M.,
- Gonzalez M.S.,
- Gaballa M.,
- Basnight M.A.
- van de Veerdonk M.C.,
- Kind T.,
- Marcus J.T.,
- et al.
- Berger M.,
- Haimowitz A.,
- Van Tosh A.,
- Berdoff R.L.,
- Goldberg E.
- Currie P.J.,
- Seward J.B.,
- Chan K.L.,
- et al.
- Laffon E.,
- Vallet C.,
- Bernard V.,
- et al.
- Swift A.J.,
- Rajaram S.,
- Condliffe R.,
- et al.
- Garcia-Alvarez A.,
- Fernandez-Friera L.,
- Mirelis J.G.,
- et al.
- Bell A.,
- Beerbaum P.,
- Greil G.,
- et al.
- Gatehouse P.D.,
- Rolf M.P.,
- Graves M.J.,
- et al.