Introduction

Myocardial perfusion imaging (MPI) with positron emission tomography (PET) has gained increasing interest due to superior image quality over single-photon emission computed tomography (SPECT) and improved diagnostic accuracy owing to higher spatial resolution and correction for attenuation and scatter.1,2 Obese patients represent a growing amount of patients undergoing evaluation for coronary artery disease (CAD) and the most common myocardial perfusion PET tracer, rubidium-82 (82Rb), has demonstrated high sensitivity,3 specificity, 4 and improved diagnostic accuracy1,5 for the detection of CAD among these patients. A recent study of 82Rb PET comparing obese (body mass index (BMI) ≥ 30 kg/m2) with overweight (BMI 25-30 kg/m2) and normal (BMI < 25 kg/m2) patients showed that the prognostic value remained the same irrespective of BMI.6

Advanced PET image reconstruction methods include time-of-flight (TOF) and point spread function (PSF) modeling. The influence of TOF has been studied mainly for oncology, where phantom and patient data have demonstrated better image quality, improved accuracy and precision of regional quantification, and increased contrast and signal-to-noise ratio (SNR), particularly for obese patients.7,8,9,10,11,12 TOF images are less susceptible to artifacts due to inconsistent data used for detector normalization, scatter, and attenuation correction.13,14,15 PSF modeling in iterative reconstruction provides more accurate activity estimates,16,17,18 decreases image noise and reduces partial-volume effects, though it may cause edge artifacts and alter image noise.18,19,20 PSF modeling improves image quality of cardiac 18F fluorodeoxyglucose viability (average BMI 28.3 ± 5.9) and 82Rb perfusion studies (average BMI 25.3 ± 6.5).21

Phantom and patient studies with TOF and PSF modeling have demonstrated promising results for 82Rb cardiac PET,21,22,23,24 including obese patients;23 however, the magnitude of these effects for both normal and obese patients has not been well studied. Armstrong et al22 showed that TOF + PSF reconstruction resulted in greater flow values compared with OSEM, though myocardial flow reserve (MFR) was not affected. The separate influences of TOF and PSF were not studied, but these authors did show an example (their Figure 6) of increased flow in the septal wall of PSF + TOF compared with OSEM. In a study of 27 subjects, Oldan et al 25 found no significant overall or segmental differences between FDG cardiac images reconstructed with and without TOF. Examples of the differences between TOF and non-TOF 82Rb cardiac reconstructions (both with PSF) for two obese patients are shown by DiFilippo et al23 (their Figures 2, 3). In a study of SNR for small tumors, the SNR improvement with TOF observed in abdominal and head and neck regions as BMI increases does not hold true in the lungs.9 Cardiac SNR results reported recently by Armstrong et al26 are consistent with this observation and show that improvement in SNR in the myocardium is independent of BMI. As Wells and de Kemp note in an accompanying editorial,27 cardiac imaging with relatively intense uptake compared to low lung background is different in nature from a homogenous activity region from which early theoretical predictions of the effect of SNR improvement with TOF were derived,28 and further study is needed to address the potential benefit of TOF in cardiac imaging. Theoretical analyses of the potential SNR improvements with TOF are further complicated by the routine clinical use of iterative reconstruction algorithms with improved SNR characteristics vs. analytic filtered backprojection algorithms.

In a previous investigation using average and large anthropomorphic cardiac-torso phantoms, we found that TOF and PSF modeling improved tracer quantitation accuracy and uniformity in the myocardial wall, especially for the phantom representing an obese patient.29

This article presents results on the individual and combined effects of TOF and PSF modeling on myocardial perfusion as a function of patient BMI and cross-sectional area of transaxial slices containing the heart. Breast tissue in the field of view and the size of the heart may influence the effects of TOF and PSF modeling, motivating analysis by gender and left ventricular (LV) myocardial volume.

Materials and Methods

Study Population

Thirty-six patients referred for assessment of CAD with 82Rb PET MPI at the University of Maryland Medical Center were included in the study on a rolling basis as they met enrollment criteria of weight > 113.6 kg (250 lb) or BMI > 30 and provided informed consent under a protocol approved by the University of Maryland, Baltimore Institutional Review Board. Patient demographics are given in Table 1.

Table 1 Patient demographics
Full size table

82Rb PET/CT Acquisition

Rest and pharmacological stress 82Rb PET MPI was performed on a Siemens Biograph mCT PET/CT scanner with four rings of block detectors and a 21.6 cm axial PET field of view.30 Patients were instructed to abstain from caffeine products for 12 h prior to the study. Imaging was performed with arms above the head unless this could not be tolerated. First, a CT scan for attenuation correction was performed under free breathing over a 22.4 cm axial field of view. Scan parameters were x-ray tube voltage 120 kVp, tube current 300 mA, exposure time 500 ms, total collimation width 19.2 mm, table speed 57.6 mm/seconds, table feed per rotation 28.8 mm, and pitch factor 1.5. The CTDIvol was 7.28 mGy and the dose-length product was 174 mGy·cm.

Next, two 8-minute 3D listmode PET studies were performed at rest and under pharmacological stress with adenosine or regadenoson. Activity was administered from a Cardiogen 82Sr-82Rb generator (Bracco Diagnostics). The activity administration parameters were a maximum administered activity of 2220 MBq (60 mCi), maximum flow rate of 50 mL/minute, and a maximum injected volume of 50 mL. Weight-dependent dosing was not performed. The actual injected activity (mean ± SD) for each scan was 1974 ± 312 MBq (53.3 ± 8.4 mCi) with a range of 1095-2231 MBq (29.5-60.3 mCi) and a median of 2203 MBq (59.6 mCi). The smallest amounts of activity (~ 1110-1480 MBq; ~ 30-40 mCi) were administered when the generator was the oldest (~ 40 days old; the half-life of 82Sr is 25.3 days) and the dose volume reached the 50 mL limit.

Detector saturation is a concern for large administered doses and fan sum plots showing the coincidence events for each detector block can show saturation and stunning effects.31 The fan sum plots were examined in different time frames for each study. Although there is evidence of dose saturation in some detector blocks for a few patients at early times, no saturation effects were observed during the 1.5-8.0 minute period that was used for generation of the perfusion images.

Registration of PET and CT images was checked by the technologist and adjusted via manual rigid-body registration if necessary. PET and CT data were anonymized and transferred to a workstation equipped with Siemens e7 software tools, enabling offline image reconstruction with clinical algorithms.

Image Reconstruction

Static perfusion images were reconstructed using data from 1.5-8 minute post-injection using four different algorithms. These were (1) Ordered Subsets Expectation Maximization (OSEM), (2) OSEM with TOF (OSEMTOF), (3) OSEM with PSF modeling (PSF), and (4) OSEM with PSF modeling and TOF (PSFTOF). The data were corrected for normalization, randoms, scatter, prompt gamma rays, dead time losses, and attenuation. The manner of implementation of corrections for scatter and prompt gamma rays may be found in Refs.32,33. PSF modeling used the HD-PET option (Siemens Medical Solutions).16 All iterative reconstructions were performed with 4 iterations, 21 subsets, and an 8-mm full width at half maximum (FWHM) 3D Gaussian post-reconstruction filter, which matches clinical practice for 82Rb TOF image reconstructions.

Image Processing and Analysis

Polar maps of perfusion were generated using PMOD (PMOD Technologies), which extracted peak values (Bq/mL) in the LV myocardium along rays perpendicular to the long axis of the LV. There were 36 samples at 10-degree angular increments in each of 21 equally spaced slices from apex to base plus a single sample at the apex.

Qualitative and quantitative analyses were performed to assess the effects of TOF and PSF modeling. The short-axis images and polar maps were scaled to their own maximum value only for qualitative display.

Quantitative analysis was performed using the polar map values of absolute activity concentrations (Bq/mL). These were first averaged in the 17 segments of the American Heart Association (AHA) heart model.34 The effect of applying TOF, PSF or both was computed using segmental data normalized to the maximum (peak perfusion) segmental value from OSEM (algorithm (1)). The mean percent change in segmental perfusion relative to OSEM is given by

$$ (\% \;{\text{Change}})_{i} = \frac{1}{N}\sum\limits_{n = 1}^{N} {\left({\frac{{{\text{Segment}}_{M,n} (i) - {\text{Segment}}_{{{\text{OSEM}},n}} (i)}}{{{\text{Max}}\left\{{{\text{Segment}}_{{{\text{OSEM}},n}} (j): = 1, \ldots,17} \right\}}}} \right)} \times 100, $$
(1)

where n = patient number, N  = number of patients, i  = AHA segment number, M  = reconstruction algorithm with TOF and/or PSF.

Differences in segmental perfusion were analyzed according to patient BMI and body cross-sectional area in the field of view (FOV) of the scanner. The latter was motivated by the fact that a small, obese patient could have a high BMI yet not have a large body area in the PET FOV. Errors in scatter and attenuation correction, and the influence of TOF, will likely be greater when there is more body mass in the FOV. The patient cross-sectional area in a transaxial slice through the heart was determined by segmenting a reconstructed x-ray CT slice into air and non-air components using MATLAB (Mathworks). The voxels inside the outer body contour and inside the arm(s) (if in the field of view) were counted and used to compute the cross-sectional area. The perfusion data also were analyzed with respect to gender and LV myocardium size. The LV myocardial volume was determined by Corridor 4DM (INVIA) from automatically determined epicardial and endocardial contours.35 Due to the non-linear associations between continuous predictors (such as BMI) and the outcomes, patients were grouped based on tertiles, each with 12 patients, to evaluate whether perfusion changes were associated with the magnitude of BMI groups (< 41, 41-52.1, > 52.1 kg/m2), cross-sectional area groups (< 1070, 1070-1186, > 1186 cm2), and LV myocardial volume groups (< 54, 54-87,  > 87 mL), respectively.

Summed stress score (SSS), summed rest score (SRS), and summed difference score (SDS)36 were automatically computed as in Ref. 37,38 for each patient and each reconstruction method. Using the 17-segment model, uptake in each segment was normalized to the maximum segment value for that study and multiplied by 100. Each segment was scored as 0 = normal uptake (85-100%), 1 = mild decrease (70-85%), 2 = moderate decrease (50-70%), 3 = severe decrease (15-50%), and 4 = no uptake (0-15%). SSS, SRS, and SDS were computed and analyzed according to patient BMI, cross-sectional area, gender, and LV size.

Statistical Analysis

Summary statistics for demographic data were reported and compared by gender, BMI groups, cross-sectional area groups, and LV size groups, respectively, using Mann-Whitney U test or Kruskal-Wallis H test for continuous variables and Fisher’s exact test for categorical variables as appropriate. For each summed score (SSS, SRS, and SDS), linear mixed effects models (LMMs) were used to compare the four reconstruction algorithms (OSEM, OSEMTOF, PSF, PSFTOF) with and without adjusting for gender, BMI groups, cross-sectional area groups, and LV size groups, respectively, while taking into account the correlation among the measures from the same patient. The model-based pairwise comparisons among the four reconstruction algorithms were examined using Scheffé’s method.

Similarly, for the segmental data from rest and stress studies, LMMs were used to compare the four reconstruction algorithms while adjusting for segments and each demographic data of interest (i.e., gender, BMI groups, cross-sectional area groups, and LV size groups). The model-based pairwise comparisons among the four reconstruction algorithms within each segment were examined using Scheffé’s method. All statistical tests were performed with a two-sided significance level of 0.05. All analyses were performed using Stata/SE (version 15, Stata Corporation).

The percent change for each segment was displayed in an AHA 17-segment polar map. The values for non-significant changes are given in a white font while significant changes are given in a black font.

Results

Qualitative Analysis

Short-axis cardiac images and polar maps for two representative subjects are shown in Figure 1 and polar maps for the four iterative reconstruction methods for all subjects are shown in Figure 2. Low perfusion regions are less prominent and polar maps are more uniform with TOF (Figure 1). The latter observation applies to polar maps from all patients (Figure 2). In this study, 11 out of 36 patients had one (patients 15, 22) or both (patients 5, 11, 25, 26, 27, 29, 33, 34, 35) arms in the FOV. With TOF reconstruction, the size and intensity of apparent perfusion defects are often reduced and in general perfusion appears higher in the septal wall.

Figure 1
figure 1

Transaxial CT slice, short-axis slices, and polar maps of rest-stress 82Rb PET MPI for two patients and the four reconstruction algorithms. (A) Subject 5, female, with BMI 38.5 kg/m2 and cross-sectional area 1342 cm2, and (B) subject 4, male, with BMI 36.0 kg/m2 and cross-sectional area 905 cm2

Full size image
Figure 2
figure 2

(A) Rest and (B) stress polar maps for OSEM, OSEMTOF, PSF, and PSFTOF reconstructions for the 36 study subjects

Full size image

Quantitative Analysis, Segmental Uptake

The percentage differences in myocardial uptake between (a) TOF and non-TOF (OSEMTOF-OSEM, and OSEMTOF-PSF), (b) PSF and non-PSF (PSF-OSEM and PSFTOF-OSEMTOF), (c) PSFTOF and OSEM (PSFTOF-OSEM), and (d) PSF only and TOF only (PSF-OSEMTOF) algorithms categorized by BMI, patient cross-sectional area, gender, and LV size are summarized in 17-segment polar maps (Figures 3, 4, 5, and 6).

Figure 3
figure 3

17-segment polar maps of segmental differences in cardiac perfusion between the different reconstructions for patients grouped by BMI. The values for non-significant changes are given in white font while significant changes are given in black font

Full size image
Figure 4
figure 4

17-segment polar maps of segmental differences in cardiac perfusion between the different reconstructions for patients grouped by cross-sectional area in the PET FOV. The values for non-significant changes are given in white font while significant changes are given in black font

Full size image
Figure 5
figure 5

17-segment polar maps of segmental differences in cardiac perfusion between the different reconstructions for males and females. The values for non-significant changes are given in white font while significant changes are given in black font

Full size image
Figure 6
figure 6

17-segment polar maps of segmental differences in cardiac perfusion between the different reconstructions for patients grouped by LV volume. The values for non-significant changes are given in white font while significant changes are given in black font

Full size image

General trends are summarized before a detailed analysis by classification scheme. Whole heart perfusion values averaged over the 17 segments were compared with OSEM values. For the rest case, perfusion was 1.9% greater for OSEMTOF, 1.6% greater for PSF, and 3.9% greater for PSFTOF. The differences between each method were significant (P < .05) except for PSF vs. OSEMTOF. For the stress case , perfusion was 2.0% greater for OSEMTOF, 1.8% greater for PSF, and 4.2% greater for PSFTOF. The differences between each method were significant (P < .05) except for PSF vs. OSEMTOF. The magnitude of the differences for stress was not significantly different than the magnitude of the differences for rest.

Iterative reconstructions with TOF showed marked increased uptake in the septal wall and a slight decrease in the inferior and anterolateral walls. PSF modeling resulted in a small uptake increase in almost all 17 segments. This is consistent with higher values expected with resolution recovery. The change in uptake with PSF was comparable for rest and stress studies. The combined effect of TOF and PSF modeling was approximately additive. Due to the greater magnitude of the increase with TOF compared to that with PSF in some septal segments, the difference polar maps of PSF-OSEMTOF showed negative values in some of these segments that reached statistical significance. These general trends for TOF and PSF modeling were the same across BMI, cross-sectional area, gender, and LV size groups.

Results by patient BMI

For TOF iterative reconstructions, the change in uptake was greater for the two tertiles with greater BMI, with peak values slightly larger for the BMI-II group (Figure 3). The trend was for increased perfusion throughout the septal wall, though increases reached statistical significance (P < .05) only for AHA segments 8 (mid anteroseptal) and 14 (apical septal) walls. When only PSF was modeled in image reconstruction, the change in uptake was comparable among the groups and did not reach statistical significance in any segment. For PSFTOF, the magnitude of the changes was greater than for TOF or PSF alone, reflecting their combined influence, and only reached statistical significance in segments 8 and 14 in BMI-II and BMI-III groups.

Patient characteristics for the BMI tertiles are summarized in Table 2. Not unexpectedly the weight (P < .001) and area of the transaxial slice containing the heart (P < .001) were significantly different among the BMI tertiles. The distribution of male and female patients was not associated with the BMI group (P = .32).

Table 2 Patient groups by BMI
Full size table

Results by patient cross-sectional area

The observation that the magnitude of uptake changes with TOF was not greater for BMI tertile III than BMI tertile II was initially puzzling. This led us to consider the physical factors leading to changes with TOF. Corrections for attenuation and scatter and the effect of misregistration between emission and transmission scans due to respiration, for example, will be greater when more of the body is in the FOV of the PET scanner. Although body cross-sectional area is related to BMI (Figure 7), short patients could have a high BMI yet have less body area in the scanner than a tall patient with a small BMI. This provided the rationale for evaluating changes due to TOF and PSF as a function of body cross-sectional area in a transaxial slice through the heart.

Figure 7
figure 7

Patient cross-sectional area in the PET scanner field of view vs. BMI

Full size image

Patients were grouped into cross-sectional area tertiles and results are shown in Figure 4. For TOF iterative reconstructions, the change in uptake generally increased in patient groups with greater cross-sectional area (Patient Area-III > Patient Area-II > Patient Area-I). As for BMI, the trend was for increased perfusion in the septal wall with statistically significant differences (P < .05) only for segments 8 and 14. For reconstructions with PSF modeling, the change in uptake was comparable among patient groups. For PSFTOF, a similar trend as with TOF was observed.

Patient characteristics for the cross-sectional area tertiles are summarized in Table 3. As expected, there were statistically significant differences in weight and BMI among the groups. The differences in LV myocardial volume were not statistically significant. The distribution of male and female patients was not associated with the LV size group (P = .32).

Table 3 Patient groups by cross-sectional area
Full size table

Results by gender

For TOF iterative reconstructions, the change in uptake in the septal wall was generally greater for females than males and reached statistical significance for AHA segments 8 (mid anteroseptal) and 14 (apical septal) walls for females, but not for males (see top two rows of polar maps in Figure 5). The effect of PSF modeling was a slight increase in uptake that was comparable between males and females and did not reach statistical significance for any segment. For PSFTOF, the magnitude of change in uptake was greater than for TOF or PSF alone and reached statistical significance in females for heart segments in the anteroseptal and apical anterior walls (see row 5 of the polar maps, segments 2, 8, 13, and 14).

The differences between genders could be due to breasts in the field of view or other factors that correlate with gender. For example, LV myocardial volumes were generally larger for males than for females, as expected (Figure 8). This makes it difficult to decouple the effect of breasts from those of spillover and spatial blurring, which have a greater relative effect for small hearts. Patient characteristics for the two genders are summarized in Table 1. Differences in weight and LV myocardial volume were statistically significant (both P < .05), while differences in BMI, cross-sectional area and age were not.

Figure 8
figure 8

LV myocardial volume distribution for males and females. Dashed lines represent the grouping into tertiles based on LV size

Full size image

Results by LV myocardial volume

For TOF iterative reconstructions, the change in uptake was greater in patient groups with a smaller LV size (Figure 6, rows 1 and 2). Once again there were increases in the septal wall area that reached statistical significance in AHA segments 8 and 14. With PSF modeling, the slight increase in uptake was comparable among patient groups but was not statistically significant (Figure 6, rows 3 and 4). For PSFTOF, a similar trend as with TOF was observed, with slightly larger increases in perfusion (Figure 6, row 5). Patient characteristics for the LV size tertiles are summarized in Table 4. Weight was smaller in the LV size-I group (P = .05) and the distribution of male and female patients was different in the LV size-III group, with more males and fewer females (P = .004).

Table 4 Patient groups by LV myocardial volume
Full size table

Quantitative Analysis, Summed Scores

The differences in SRS, SSS, and SDS between the reconstruction algorithms were compared on a pairwise basis for all subjects (Table 5) and for BMI, patient cross-sectional area, gender, and LV myocardial volume subgroups (Tables 6, 7, 8, and 9) using LMMs.

Table 5 Comparison of summed score statistics over all patients
Full size table
Table 6 Summed score statistics by BMI
Full size table

Over all subjects (Table 5), the mean SRS and SSS were significantly smaller for reconstructions with TOF compared with reconstructions without TOF (OSEMTOF vs. OSEM, PSFTOF vs. PSF, PSFTOF vs. OSEM). The magnitude of these decreases ranged between 2.4 and 3.0 and was about the same for SRS and SSS; as a result, the mean SDS differences were near zero and not significantly different. The mean SRS, SSS, and SDS were not significantly different between PSF and non-PSF reconstructions (PSF vs. OSEM, PSFTOF vs. OSEMTOF). These findings are consistent with qualitative polar map and quantitative segmental analyses.

For BMI groups (Table 6), all significant changes in SRS and SSS were the result of including TOF in image reconstruction. Significant changes were observed only for the BMI-I and BMI-II tertiles. The magnitude of significant decreases ranged between 2.9 and 4.3. There were no significant changes in SDS in any BMI tertile for any reconstruction method.

For cross-sectional area groups (Table 7), again all significant differences in SRS and SSS were the result of including TOF in image reconstruction, though TOF did not always result in a significant difference. The number of comparisons for which the differences achieved statistical significance was one for tertile I, 3 for tertile II, and 5 for tertile III. The significant decreases with TOF ranged between 2.8 and 3.4. There were no significant changes in SDS for any cross-sectional area tertile or any reconstruction method.

Table 7 Summed score statistics by patient cross-sectional area in the FOV
Full size table

For gender groups (Table 8), there were only significant changes for females. These occurred only for SRS and SSS when TOF was additionally included in image reconstruction. The magnitude of the significant decreases ranged between 3.1 and 4.0. There were no significant changes in SDS.

Table 8 Summed score statistics by gender
Full size table

For LV myocardial volume groups (Table 9), there were changes in SRS for tertiles I and II and in SSS for tertiles II and III. All were associated with the application of TOF reconstruction, in some cases with PSF as well. The magnitude of the significant decreases in SRS and SSS ranged between 2.7 and 4.9. There were no significant changes in SDS.

Table 9 Summed score statistics by left ventricular myocardial volume
Full size table

Discussion

This study evaluated the effects of TOF and PSF modeling on 82Rb myocardial perfusion PET in obese patients. The most striking qualitative effects with TOF (Figures 1, 2) are (1) improved visual uniformity of the polar maps, (2) increased amplitudes in the septal wall, and (3) reduction in visual appearance of perfusion defects. For observation (1), improved uniformity is likely due to a reduction in artifact strength when there is mismatch between the emission and attenuation maps. The septal wall may be more sensitive to mismatch effects, which could account for observation (2), and many of the perfusion defects in the study population are in the septal wall. A relative increase in septal wall perfusion with TOF is consistent with previous observations.22,23 The effect of TOF is particularly evident in the quantitative 17-segment analyses when patients are grouped by size (Figures 3, 4). Septal wall perfusion increases by up to about 10% with TOF and is greater for larger patients. The increase with body size is more evident when patients are grouped by cross-sectional area of transaxial slices containing the heart. This is physically reasonable because PET-CT mismatch affecting attenuation and scatter correction is expected to be larger with more body mass in the scanner FOV.

There are differences in the effect of TOF between males and females. The increase in septal wall perfusion with TOF is larger and only statistically significant for females (Figure 5). With females, the breasts contribute to soft tissue mass in the cardiac FOV and move with respiration. Thus, the mismatch between PET and CT images may be greater on average for females, possibly accounting for the larger effect of TOF in minimizing the inconsistencies. Since scatter is greater as body mass increases and scatter correction uses the attenuation map, TOF may also partially compensate for errors in scatter correction due to emission-attenuation map mismatch. With our limited dataset it is not possible to decouple the influence of gender-related effects from heart size (Figure 8).

There are technical factors that make scatter correction more challenging for large patients, in particular the fitting and scaling of a simulated scatter tails to the observed scatter tails. For the Biograph mCT the x-ray CT scan for attenuation correction is reconstructed out to a diameter of 78 cm (the FOV for clinical diagnosis is 50 cm), which is the patient bore diameter of the scanner. The PET activity is reconstructed out to a 70 cm diameter field of view. The extension of single scatter simulation to TOF scatter correction implemented by Siemens is described in Ref.32 For our study, the largest lateral dimension of a patient in a transaxial slice containing the heart had a mean of 55.2 cm, a median of 53.5 cm, and a range of 38.4-69.0 cm. The corresponding largest anterior-posterior dimension had a mean of 32.0 cm, a median of 32.0 cm, and a range of 22.4-39.6 cm. On average, the patients filled 30% (range 14-43%) of the PET FOV (using data from Table 1 and the area inside the 70 cm diameter PET FOV). There were five patients with a small part of the body outside of the 70 cm PET FOV, and this was always part of an arm. The mean distance outside the 70 cm diameter circle was 2.2 cm with a range of 2.0-2.7 cm. For some patients, this would likely have a small effect on scatter correction. Previous work has shown that TOF reconstructions are less sensitive to data inconsistencies than non-TOF reconstructions.13,14 Perhaps due to its more central location in the body, the septal wall is more sensitive to inconsistencies in scatter, prompt gamma, and attenuation corrections than other parts of the heart, and any resulting bias is partially compensated with TOF image reconstruction. Although the heart was always near the center of the field of view, the large body habitus with part of the body sometimes near the edge and outside of the reconstructed PET FOV makes scatter and prompt gamma correction more difficult and would be a suitable topic for future investigation.

The effect of PSF modeling was a small increase in perfusion averaging 1.5% in almost all heart segments, though this did not reach statistical significance for any segment (Figures 3, 4, 5, and 6). Images were filtered with the clinically used 8-mm FWHM Gaussian filter, which may have diminished the effect of PSF modeling. PSF modeling was approximate because it was based on the system response for 18F,16 which has a smaller positron range than 82Rb.

TOF plus PSF modeling combined the individual effects of TOF and PSF, with the largest amplitude increases in the septal wall. The magnitude of the effects appears roughly additive and the response is dominated by the TOF effects.

TOF generally decreased SRS and SSS, with significant mean decreases between 2.4 and 3.0 (Table 5; OSEMTOF vs. OSEM, PSFTOF vs. PSF). This could affect patient risk stratification and move some patients into a less severe category. Changes in SRS and SSS tended to be in the same direction, with the result that SDS values tended to remain unchanged. The addition of PSF in image reconstruction (PSF vs OSEM, PSFTOF vs OSEMTOF) did not significantly change SRS, SSS, or SDS. The effect of TOF on SRS and SSS was greater for females (Table 8).

Reconstructed images used coincidence data from 1.5 to 8 minutes after Rb-82 injection as is standard of care at our institution. Use of an earlier starting time would result in an increased number of counts at the expense of decreased blood pool clearance, while a later starting time would result in fewer counts with increased blood pool clearance. Adjustment of the starting time would thus affect the image noise. It is likely that the study results would be similar if the starting time were adjusted, both because of the spatial smoothing employed as part of image reconstruction and because the quantitative analysis used a 17-segment model rather than a finer cardiac subdivision.

This was a clinical study and the true perfusion is not known. In general, however, better modeling of the physics of photon transport and detection within the body and PET instrumentation in image reconstruction will result in more accurate estimation of in vivo radiopharmaceutical uptake. In a recent study of anthropomorphic cardiac-torso phantoms of large patients, we showed that incorporating TOF and PSF modeling in image reconstruction improves quantitation accuracy.29 TOF reduced artifacts associated with deliberate spatial mismatch between the emission scan and the x-ray CT scan used for attenuation correction. It is therefore likely that clinical image reconstructions using TOF and PSF modeling provide the best estimates of myocardial perfusion. Nonetheless, additional phantom studies with perfusion defect inserts may shed additional light on the change in defect sizes observed in the polar map displays of these clinical studies for different image reconstruction methods, particularly for TOF vs. non-TOF reconstructions.

The results of this study show that both TOF and PSF modeling in image reconstruction have measurable effects on qualitative and quantitative observations of perfusion. Clinicians should consider these effects and gender differences when interpreting 82Rb perfusion studies. Further work is needed to investigate the observed dependence on gender. Work is ongoing to study the separate effects of TOF and PSF on absolute myocardial blood flow and flow reserve.

Study Limitations

This is a single-center study with a limited number of patients and the demographics are representative of obese patients presenting for PET MPI at the University of Maryland Medical Center. The patients were enrolled over a period of 12 months and in this study 9 of the patients were male and 27 were female. The gender distribution of obese patients presenting for MPI will differ at other institutions and further work is necessary to assess whether the conclusions of this paper apply to other patient populations.

This study employed clinical settings of iteration number, number of subsets, and post-reconstruction filter and did not systematically evaluate different settings. Convergence to a given noise level, image resolution, or other metric would depend on factors such as administered activity, patient weight, BMI, and other aspects of body habitus or physiology and is likely patient-dependent, though certain fixed values of iterations, subsets, and post-reconstruction filter are commonly employed in clinical practice, with some sites using fewer iterations with TOF reconstruction. Certain image reconstruction settings that were used in this study, such as 21 subsets for non-TOF image reconstructions, were possible with the Siemens e7 software toolkit and may not be available as part of the clinical software release on the Biograph mCT. A thorough investigation of image reconstruction parameters is non-trivial and would merit a separate investigation. The PET scans and implementation of TOF and PSF modeling used hardware and software of a single vendor (Siemens Medical Solutions) and applicability of the findings for PET systems of other manufacturers would need to be validated.

New Knowledge Gained

The individual and combined effects of TOF and PSF modeling in image reconstruction were investigated for 82Rb MPI in obese patients. The major effect of TOF was improved visual uniformity of polar maps and greater perfusion in the septal wall as the body cross-sectional area in the PET scanner FOV increased. This latter effect was more prominent for females than males. PSF modeling generally resulted in a small increase in perfusion in all cardiac segments. The general effect of TOF modeling was to decrease SRS and SSS, which could affect patient risk stratification. These changes were about the same magnitude and so SDS remained about the same. With PSF modeling SRS, SSS, and SDS were largely unchanged. This paper adds to the knowledge of how advanced image reconstruction algorithms affect 82Rb myocardial perfusion imaging of obese patients.

Conclusion

TOF and PSF modeling in image reconstruction both had significant effects on 82Rb PET MPI of obese patients. TOF improved visual uniformity of polar maps, increased amplitudes in the septal wall, and reduced the visual appearance of perfusion defects. Septal wall perfusion increased by up to about 10% with TOF and this increase was greater for larger patients. The increase was more evident for patients grouped by body cross-sectional area in transaxial slices containing the heart than by BMI. The increase in septal wall perfusion with TOF was greater for females than males. The effect of PSF modeling was a small increase in perfusion in almost all heart segments, averaging about 1.5%. TOF modeling generally decreased SRS and SSS, which could affect risk stratification. These changes were about the same magnitude and so SDS remained about the same. With PSF modeling, SRS, SSS, and SDS were largely unchanged. Clinicians should consider effects of TOF and PSF modeling and gender differences when interpreting 82Rb-82 perfusion studies.