Reference Ranges for LVEF and LV Volumes from Electrocardiographically Gated ⁸²Rb Cardiac PET/CT Using Commercially Available Software

MATERIALS AND METHODS

Data Analysis

Attenuation-corrected PET images were reconstructed by an iterative algorithm (ordered-subset expectation maximization, 2 iterations, 21 subsets), with postprocessing filtering (Butterworth, order 10; cutoff, 0.25 cycles/bin). Four commercially available products—the CardIQ Physio package (a dedicated PET software; GE Healthcare) and the 3 major SPECT software packages (Emory Cardiac Toolbox [ECTb; Syntermed, Inc.], Quantitative Gated SPECT [QGS; Cedars-Sinai Medical Center], and 4DM-SPECT [4DM; INVIA, LLC])—were used for further analysis of electrocardiographically gated datasets. No dedicated PET version of the latter 3 software packages was available. This analysis included oblique reorientation of the datasets on the transversal planes, first parallel to the septum and then parallel to the inferior wall; definition of valve plane (automatically processed by all 4 algorithms), with manual adjustment in the case of inadequate anatomic delineation, except for QGS (which did not have a manual correction option); quality control of automated contour detection; and software-derived calculation of LVEF, ESV, and EDV from rest and stress datasets. Figure 1 shows specific examples of contour findings for all 4 software programs. To assess for sex differences, left ventricular (LV) volumes were normalized to body surface area by dividing ESV and EDV (ESV index and EDV index [in mL/m²], respectively) by body surface area.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Representative horizontal long-axis (first and fourth columns), vertical long-axis (second and fifth columns), and short-axis slices (third and sixth columns), including contours as created by all 4 software packages, during end-systole and end-diastole in representative case.

Interobserver reproducibility from 24 cases was determined using 2 independent observers unaware of prior clinical interpretation. One reader repeated the analysis to determine intraobserver agreement. In both instances, the analysis consisted of reconstruction and oblique reorientation of the datasets, manual definition of valve plane when needed, and quality control of automated contour detection.

RESULTS

Reference Ranges for LVEF, ESV, and EDV

The mean values of LVEF, ESV, and EDV for all 62 patients are summarized in Tables 2 and 3. In general, LV volumes and LVEF measurements showed significant differences among all 4 software packages (P < 0.001), with the following exceptions: LVEF was comparable between CardIQ Physio and 4DM, whereas ECTb yielded the highest and QGS the lowest mean LVEF values (Table 2). LV volumes were also significantly different among the software, with some exceptions, as detailed in Table 3.

View this table:

• View inline
• View popup

TABLE 2

Mean LVEF Using 4 Different Software Packages

View this table:

• View inline
• View popup

TABLE 3

Mean LV Volumes (n = 62) Using 4 Different Software Packages in the Study Group

The response of LVEF to vasodilator stress, a potential marker of disease severity, was relatively similar among the software (Fig. 2).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Differences in LVEF response to vasodilator stress among software products (P value was not significant between subgroups).

Comparison of Subgroups of Healthy Subjects

The mean values of normal LVEF in patients with a low pretest probability of CAD (subgroup 1) and normal coronary CTA findings (subgroup 2) are depicted in Figures 3 (rest) and 4 (vasodilator stress), respectively. Subgroup 1 has higher values in almost all settings. When the 20 patients within subgroup 2, which received β-blockade, were compared with the 8 patients of subgroup 2 without β-blockade, no significant differences in LVEF were observed with any of the 4 software packages. ESV and EDV, finally, were not statistically different between the 2 subgroups, although there was a trend for higher ESV in subgroup 2 (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

Comparison of mean LVEF at rest between subjects with low pretest probability of CAD (subgroup 1) and normal coronary CTA results (subgroup 2). NS = not significant.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4. 

Comparison of mean LVEF at stress between subjects with low pretest probability of CAD (subgroup 1) and normal coronary CTA results (subgroup 2). NS = not significant.

Table 4 shows results according to sex. There was a trend for mildly higher LVEFs and smaller LV volumes in women than in the smaller group of men among the different software packages, reaching statistical significance in a few instances.

View this table:

• View inline
• View popup

TABLE 4

Mean LVEF and LV Volume Index Between Men and Women

DISCUSSION

Our study defines reference values and ranges for global functional parameters from gated ⁸²Rb PET/CT, using various currently available commercial software products in a typical referral population for PET (high body mass index, mostly women). Our study also provides further insights: it suggests that because values differ among software packages, these values should not be used interchangeably. And it suggests that the criteria used to define subjects for inclusion into a reference database may influence results.

With the exception of CardIQ Physio, which is a dedicated cardiac PET software application, the software products used in this study (ECTb, QGS, and 4DM) are all software that was originally designed for gated SPECT analysis. Our study clearly shows that significant differences may be encountered among these 4 software packages when applied to gated PET data. This finding is consistent with similar observations made in the analysis of gated myocardial SPECT data using those software packages (14). Although CardIQ Physio and 4DM seemed to be most comparable, ECTb consistently yielded the highest and QGS the lowest LVEF values, with a mean LVEF difference of 15% (regardless of stress or rest state) between the 2 software products. Differences in the algorithms for definition of endocardial and epicardial borders and definition of the base and valve planes are the most likely explanation for this observation. For example, QGS uses a 3-dimensional model of the heart without specific geometric assumptions of horizontal or transversal long axes, ECTb uses a 2-coordinate system (cylindric coordinate for basal and mid-myocardial segments but spheric coordinate for the apex), and 4DM requires the heart base to be perpendicular to the chosen long axes (21). Moreover, ECTb and QGS valve-plane definition models assume that the septal wall is shorter than the lateral wall; consequently, the basal limits are independently estimated on each side of the left ventricle, whereas 4DM assumes that the basal limits are the same in the septal and lateral walls (22,23).

Because of the physical properties of ⁸²Rb, which has an ultrashort half-life of 75 s and relatively high positron energy, the statistical quality of gated images may be limited. This limitation makes quantification of functional parameters a bigger challenge than when using SPECT or gated PET with other tracers. Nevertheless, reproducibility of the software algorithms was good in our analysis, and the validity of gated ⁸²Rb PET results was recently demonstrated by us in a comparison with contrast-enhanced CT ventriculography (11). This prior study agreed well with the reference technique but also showed a systematic underestimation of LVEF by PET. The need for defining method-specific reference ranges was emphasized by this prior study, but because of the inclusion of a range of health and disease such reference ranges could not be defined. Our current study provides this missing information; however, most subjects in our study were obese and female. Our results may, thus, be somewhat limited when applied to the general population, but they are representative of the typical population referred for PET because SPECT is limited by artifacts specifically in those individuals.

Other authors have described functional parameters using 2 of the software applications included in our study and reported values in similar ranges. Dorbala et al. used the ECTb software in a control group of 44 individuals with a low pretest likelihood of CAD evaluated with ⁸²Rb PET/CT (12). In this study the LVEF at rest and stress and the LVEF difference were 62% ± 9%, 69% ± 8%, and 7% ± 7%, respectively. Brown et al. used the QGS package in a mixed population of 87 individuals (15% had history of CAD) and found that the LVEF at rest and stress and the LVEF difference were 47% ± 13%, 54% ± 13%, and 7% ± 7%, respectively, (13). Our study, however, is the first—to our knowledge—to compare 4 common software packages directly, in the same patient group. It provides different reference ranges for each package and suggests that no common reference range exists. A resting LVEF of 50%, for example, would have to be considered low or depressed if using ECTb, whereas a resting LVEF of 45% would still be within the reference range for the other types of software. The tendency toward low LVEF values from ⁸²Rb PET has been recognized by some manufacturers, and novel dedicated algorithms are being developed. Those have not been included in our analysis, and new software may affect clinical practice in the future.

One of the unique features of gated ⁸²Rb PET is that stress imaging, in contrast to SPECT, is conducted at a time close to peak vasodilator action. The dipyridamole-induced change of the LVEF is considered a marker of the severity of ischemia and extent of flow-limiting disease (12,13). The change was explained by an increase in EDV and mild decrease in ESV during vasodilator stress in our healthy subjects. Interestingly, the magnitude of the response of LVEF to vasodilator stress in our population was rather comparable among software packages, ranging from 5.9% ± 5.3% (ECTb) to 8.1% ± 5.4% (4DM). In contrast to absolute LVEF values, the LVEF response to vasodilation may thus be used independently of the software package, although the observed range is large and the value as an independent parameter requires further validation.

Finally, another interesting aspect of our study was that the definition of healthy subjects by a low pretest probability of CAD (subgroup 1) versus normal coronary CTA findings (subgroup 2) yielded slightly different functional results. Twenty of 28 individuals in subgroup 2, compared with subgroup 1, were prepared with β-blockers before CTA, and this difference may potentially explain borderline differences in the baseline and peak heart rate during vasodilator stress. Nevertheless, no significant differences were observed in LVEF within subgroup 2 between subjects who received β-blockers versus those who did not (data not shown). And other prior studies used CTA to define the validity of LVEF measurements, which did not differ from reference methods in which β-blockade was not used (24–27).

Because both subgroups had comparable baseline characteristics except for age, the reasons for the LVEF difference are not obvious. The definition of an individual having a low pretest probability of CAD is an entirely clinical assumption, which is based on the patient's age, sex, and chest pain characteristics as first described by Diamond and Forrester (18). Therefore, individuals with a low pretest probability of CAD are likely a good standard for the definition of healthy subjects, although no imaging test is available to confirm the absence of disease. In individuals with normal angiography results, on the other hand, the absence of disease is confirmed but the patients usually have symptoms or other factors that prompt the angiogram. Although not clearly identified by our analysis, the reason for the differences between both subgroups may lie in unknown factors that prompt the angiogram. It has been discussed controversially whether one group should be preferred over the other for the definition of a reference database (17). Because we were not able to identify a clear reason to disqualify either group, we decided to combine both for definition of the reference ranges reported in our study.

Some limitations of this study should be considered. First, it has a retrospective design and suffers from all the limitations of similar studies. Likewise, the findings may be limited by the relatively few patients and not be representative of the general population at any other site. Our healthy group consisted of a large fraction of female and obese individuals, which is an inherent bias in our referral population for PET myocardial perfusion evaluation. This bias is partly because PET, with its higher spatial resolution and attenuation correction, is considered superior to SPECT for the evaluation of obese individuals (9,28). Obese individuals may have higher LV volumes than healthy subjects; however, the LVEF is not adversely affected, even with severe degrees of obesity as previously reported by Dorbala et al. (29). Similar findings have been described using echocardiography (30).

Finally, multiple studies have shown that women have smaller EDV and ESV values and higher LVEFs than men (15,29,31,32), as agrees with our findings despite the fact we included a limited number of men. Some advocate an overestimation of LVEF in women, because, in general, women have a smaller ventricular chamber than men (15,32). Others have hypothesized that women have a higher stroke volume for any given EDV, translating into a higher LVEF (31). In either case, it is clear that sex-specific differences exist and additional studies that include a higher number of men are necessary. The results of our study nevertheless support the notion that individual reference values should be established for gated ⁸²Rb PET/CT.

CONCLUSION

There are significant differences in the reference range of the functional parameters LVEF, ESD, and EDV from gated ⁸²Rb PET/CT among most of the currently commercially available software applications. These differences have implications for the interpretation of, definition of abnormality on, and interchangeable use of quantitative results from gated ⁸²Rb PET/CT.