Introduction

The role of myocardial perfusion imaging (MPI) in diagnosing patients with ischemic heart disease is well established.1 The clinical use of cardiac positron emission computed tomography (PET) especially 82Rb-PET/CT for semi-quantitative myocardial perfusion imaging has increased significantly over the past decades.2,3,4 Global functional parameters including left ventricular ejection fraction (LVEF), endsystolic volume (ESV), enddiastolic volume (EDV), and left ventricular (LV) mass are available from electrocardiographic gated 82Rb-PET/CT. However, the clinical validity of these data is less evident. A previous study has shown that LVEF, ESV, and EDV at rest and during stress can be measured reproducibly from gated 82Rb-PET/CT using different software packages compared with contrast-enhanced CT ventriculography, with some underestimation of LVEF and EDV by 82Rb-PET/CT.5 LVEF and LV volumes measured by PET using other tracers have also been shown to correlate well with both cardiac magnetic resonance imaging (cMRI) and single-photon emission computerized tomography (SPECT).6,7,8 Other studies have shown that LVEF reserve (change in LVEF from rest to stress) measured with 82Rb-PET/CT predicted adverse events and all-cause mortality9,10 and that a high LVEF reserve could be used to exclude left main/3-vessel coronary artery disease.11 Furthermore, LV mass has been shown to be constant with changes in LVEF and LV volumes from rest to stress.12

Different commercially available software packages for analyses of 82Rb-PET/CT scans have previously been compared with regard to LVEF and LV volumes.13 However, no former studies have addressed the reproducibility of global functional parameters between two different scans assessed by rest and adenosine stress 82Rb-PET/CT. Therefore, we investigated reproducibility of left ventricular ejection fraction (LVEF), endsystolic volume (ESV), enddiastolic volume (EDV), and left ventricular (LV) mass between two scans and between two commercially available software packages.

Methods

Study Population

In a time span from September 2016 to March 2017, we scanned 40 healthy volunteers using rest and adenosine stress myocardial perfusion 82Rb-PET/CT. All participants, with one exception, underwent two PET scans. Volunteers aged 20 to 40 years were recruited primarily from the surrounding universities using advertisements approved by the Scientific Ethics Committee [protocol number H-15009293]. The following inclusion criteria were used for recruitment: age >18 years, no known medical condition, no regular consumption of medicine, no use of tobacco or euphoric substances (except alcohol) within the last three months, and no current participation in drug testing studies. The following exclusion criteria were used: any prior medical history of asthma, allergy or intolerance to theophylline or adenosine, pregnancy, or inability to adhere to the protocol. Informed oral and written consents were obtained from all volunteers, and the Scientific Ethics Committee of the Capital Region of Denmark and the Danish Data Protection Agency approved the protocol [protocol number H-15009293]. One of the two scans investigated in this study was used as control scans in another study so that only one extra scan was made of each participant in order to conduct these reproducibility data. We performed the trial in accordance with the principles of the Declaration of Helsinki.

PET Imaging

Participants underwent two PET scans separated by at least 24 hours. Each scan consisted of a single session, where the volunteer underwent PET myocardial perfusion at rest and during adenosine stress. We instructed the participants not to consume substances and medications containing caffeine and theophylline for 24 hours before each scan. Volunteers received (mean±SD) 1058 ± 142 MBq and 1055 ± 145 MBq 82Rb for rest and stress imaging, supplied from a CardioGen-82Sr/82Rb generator manufactured for Bracco Diagnostics Inc., Princeton, NJ. We utilized a Siemens Biograph mCT/PET 128-slice scanner (Siemens Healthcare, Knoxville, TN, USA). Both rest and stress images were acquired ECG-gated in list mode for 7 minutes from the 82Rb infusion start point. For the stress sequence, we used adenosine infusion (0.14 mg/kg/min) for 6 minutes, and the stress 82Rb infusion was started 2.5 minutes after initiating the adenosine administration. Moreover, we performed a low-dose CT for attenuation correction before the rest study, and in case of participant motion we repeated it after the stress study. As per clinical routine, we acquired coronary artery calcium score (CACS) images from a non-contrast breath-hold CT in all volunteers at their first scan.14

Quantitative and Semi-quantitative Analyses

To quantify global functional parameters, we used static ECG-gated perfusion image reconstruction performed between 2.5 and 7 minutes after start of Rb-tracer infusion for both stress and rest images using 3D Ordered Subset Expectation Maximization (OSEM) reconstruction (2 iterations, 21 subsets) with point spread function modeling and time of flight.15 A 128 × 128 matrix, 2 × zoom (voxel dimensions 3.18 × 3.18 × 2.03 mm), 6.5 mm Gaussian post filter, attenuation and scatter correction (including prompt gamma correction) and 8 bins for gating. Quantification of functional parameters was conducted with Corridor4DM v 2015.0.0.44 (INVIA, LLC, Ann Abor, USA) and Quantitative Gated SPECT (QGS) v 2015.5 (Cedars-Sinai Cardiac Suite, Los Angeles, USA) (Figure 1). All data quantification with 4DM was automatically processed with an accurate LV contour. Furthermore, manual valve plane adjustments were performed on all 4DM data for comparison with the automatic quantification. If nothing else is indicated, 4DM data are the automatically processed without manual correction. Using QGS in 217 cases of 480, the automatic data processing algorithm was unable to generate an accurate LV contour. In those cases, it was necessary to reprocess data or apply constraints. Indexed parameters were calculated as parameters divided by body surface area in m2 (BSA).

Figure 1
figure 1

Representative 82Rb-PET images from the two scans at rest during end-systole and end-diastole. Slices of horizontal long axis (first and fourth columns), vertical long axis (second and fifth column), and short axis (third and sixth columns), including contours as created by the two software packages

Full size image

Statistical Analyses

SAS version 9.4 (SAS Institute, Cary, NC, USA) was used for statistical analyses. Continuous variables were expressed as means with standard deviations. Differences between scans and between software packages were analyzed with paired t tests and the Bland-Altman method with 95% confidence intervals and limits of agreement. P values < .05 were considered statistically significant.

Results

Study Population and Hemodynamic Parameters

Characteristics of participants are summarized in Table 1. The 40 healthy volunteers had a mean (± SD) age of 24 ± 4 years and 50% were men. Mean body mass index was 23 ± 3 kg/m2. Time between the two scans was 17 ± 14 days. All participants had CACS of 0. There was a tendency of lower heart rate and blood pressures at the second scan compared with the first. However, we found no significant difference in hemodynamic parameters from rest to stress between the two scans.

Table 1 Characteristics of study population and adenosine-induced hemodynamic changes
Full size table

Reproducibility of LVEF, LV Volumes, and LV Mass Between Scans

Resting and stress mean values of LVEF, ESV, EDV, and LV mass for the two scans and by the two different software packages are summarized in Table 2. Furthermore, 4DM data are shown without and with manual valve plane adjustment. LV mass was only available from 4DM. LV mass was higher by stress scans compared with rest scans: mean difference (95% CI) 3.0 (1.5 to 4.5), P < .001, limits of agreement: − 9.8 to 15.8. Both at rest and during stress, we found high reproducibility of all parameters between scans and this was true for both software packages (Table 2). Additionally, reproducibility was similar for 4DM without and with manual valve plane adjustment. The reproducibility between scans of LVEF reserve was also high for both software packages. In Figure 2A and B are shown Bland-Altman plots comparing LVEF at rest from the two scans measured by 4DM as an example.

Table 2 Summary of left ventricular functional measures and reproducibility measures: Scan 2 vs Scan 1
Full size table
Figure 2
figure 2

Bland-Altman plots of LVEF at rest: (A) comparison of scan 1 and 2 using 4DM (automatically processed data); (B) comparison of scan 1 and 2 using QGS; (C) comparison of QGS and 4DM (automatically processed data)

Full size image

Reproducibility of LVEF and LV Volumes Between Software Packages

When comparing LVEF, ESV, and EDV at rest and during stress between the two different software packages, we found significant differences between software packages on all functional parameters both at rest LVEF: mean difference (95% CI) 8.7 (8.1 to 9.4), P < .0001, limits of agreement: 3.1; 14.4; ESV: mean difference (95% CI) − 18.2 (− 20.0 to − 16.5), P < .0001, limits of agreement: − 33.1; − 3.3; and EDV: mean difference (95% CI) − 17.2 (− 18.6 to − 15.9), P < .0001, limits of agreement: − 28.9; − 5.6; and during stress LVEF: mean difference (95% CI) 8.0 (7.4 to 8.5), P < .0001, limits of agreement: 3.2; 12.7; ESV: mean difference (95% CI) − 13.4 (− 14.7 to − 12.1), P < .0001, limits of agreement: − 24.6; − 2.1; and EDV: mean difference (95% CI) − 6.9 (− 8.3 to − 5.4), P < .0001, limits of agreement: − 19.8; 6.1. Results were similar after manually adjusting the valve plane for 4DM measures. The minimum and maximum LVEF for QGS were 48% and 77% during rest and 53% and 82% during stress. For 4DM minimum and maximum values were 34% and 69% at rest and 46% and 74% during stress. After manually adjusting the valve plane, 4DM minimum and maximum values were 48% and 74% at rest and 52% and 79% during stress. LVEF reserve also differed significant between software packages: mean difference (95% CI) − 0.8 (− 1.6 to − 0.04), P < .04, limits of agreement: − 7.3 to 5.7. This difference, however, became non-significant after adjustment of 4DM valve plane: mean difference (95% CI) 0.5 (− 0.6 to 1.6), P < .36, limits of agreement: − 9.1 to 10.1. Figure 2C illustrates the Bland-Altman plot comparing LVEF at rest from 4DM and QGS as an example.

Indexed Functional Parameters in Men and Women

Table 3 shows LVEF and indexed ESV, EDV, and LV mass for men and women at rest and during stress for the two scans using both software packages for LVEF and volume indexes and using 4DM for LV mass index. 4DM LVEF and volume data are valve plane-adjusted measures. In all scan parameters in both scans at all times and using both software packages, women differed significantly from men. LVEF was significantly higher in women than in men and ESV index and EDV index were significantly smaller in women as was LV mass index.

Table 3 Mean LVEF, LV volume index, and LV mass index for men and women
Full size table

Discussion

In this study, we investigated the reproducibility of different global functional parameters assessed by 82Rb-PET/CT between two scans and between two commercially available software packages in 40 healthy young volunteers. We found high reproducibility between scans for all parameters but poor concordance between software packages. In addition, we found that women differed significantly from men in all parameters and that LV mass was systematically higher measured during stress than at rest.

The reproducibility that we showed between scans has not previously been assessed and previous correlation studies with other scan modalities6,7,–8 combined with our results substantiate that global functional parameters measured by PET can be suitable for clinical use. In accordance with a study by Bravo et al, we also found poor reproducibility between software packages.13 Explanations for these differences between software packages may to some extend be explained by nuances of edge detection or valve plane determination; however, we did not have access to the algorithms of the software.

When comparing our results analyzed by 4DM both with and without manual valve plane adjustment with the results by Bravo et al also using 4DM, our results for LVEF were comparable. Contrary, we found higher ESV and EDV with 4DM than Bravo et al. This could in part be explained by the fact that 91% of patients in the study by Bravo et al were women contrary to 50% in our study. In our study and in the literature, investigated by SPECT and cMRI, women has been shown to have lower ESV and EDV and higher LVEF than men.16,17,–18 That LVEF was not correspondingly higher in the study by Bravo et al compared with ours may be explained by the fact that these patients were older. Another explanation of the differences could be that we used a newer version of 4DM than Bravo et al and that the two versions may differ. When comparing our results on LVEF, ESV, and EDV assessed by QGS with results from QGS by Bravo et al, LVEF was markedly higher in our study, whereas ESV was equivalent and EDV was higher. This may also been explained by the fact that we used a newer software version.

Previous studies have suggested associations between LVEF reserve and clinical outcomes.9,10,–11 We found high reproducibility of this parameter between scans but a significant, although not pronounced, difference between software packages. However, this significant difference disappeared, when using 4DM data with manual valve plane adjustment. Our findings question whether LVEF reserve can be used as a prognostic marker in general or only within the software packages previously investigated.

Our finding that LV mass analyzed with 4DM was systematically higher during stress than at rest was in contrast to a previous study, suggesting that LV mass remains unchanged between the two conditions and thereby can be used as a quality control check on rest and stress LVEF computed from 82Rb-PET data.12 Our results indicate that this may not be true for all software packages. An explanation could be that mass is calculated on the basis of the endo- and epicardial borders, which will naturally vary slightly between rest and stress, as they are probably not taken in the exact same frame due to change in heart rate from rest to stress.

The use of 3D acquisition has been shown to be feasible for quantifying myocardial blood flow.19 However, studies comparing functional parameters between 2D and 3D acquisition has not been performed. Therefore, our findings may not be applicable for systems using 3D acquisition.

Study Limitations

This study has some limitations that should be considered. This study was conducted in healthy young volunteers with LVEF in the normal range and we cannot rule out that the results would have differed in older individuals and in patients with comorbidities and in patients with different degrees of heart failure; therefore, results cannot necessarily be used as reference values for patients in general. However, we find establishment of data in normal subjects important to test reproducibility and for later comparison with older individuals and patients with cardiovascular disease.

New Knowledge Gained

This study is, to our knowledge, the first to compare functional parameters from 82Rb-PET/CT between two scans and between different software packages in healthy young volunteers of both sexes equally represented. Our results indicate that the global functional parameter measures are reliable, but that each software package should have its own normal reference range and that the same software package should be used when interpreting functional parameters. Thus, it is possible to use functional parameters from 82Rb-PET/CT in a clinical setting, but sex-specific normal range in older individuals needs to be better established.

Conclusion

In healthy volunteers, we found high reproducibility between scans when measuring the global functional parameters LVEF, LVEF reserve, ESV, EDV, and LV mass from 82Rb-PET/CT using the same software package. In contrast, we found poor concordance between parameter measures when comparing the two commercially available software packages.