Abstract

MATERIALS AND METHODS

Data Analysis

The PET and MR image volumes were coregistered using an automated method (6). The 18 regions of interest (ROIs) used in this study are defined on the coregistered MR image in Figure 1. The cerebellum was used as the reference region V_ND (free and non-specific uptake). The ROIs were applied to the coregistered dynamic PET scans to generate regional time–activity curves.

Open in a separate window

FIGURE 1

ROIs used in this study are shown superimposed on sample MR image. AVS = anterior ventral striatum; DFC = dorsal frontal cortex; LTC = lateral temporal cortex; MTC = mesial temporal cortex; OCC = occipital cortex; OCP = occipital pole; PAC = pregenual anterior cingulate; PAR = parietal cortex; PCL = precuneus, lower; PCU = precuneus, upper; SAC = subgenual anterior cingulate; SMC = sensory-motor cortex; THL = thalamus; VFC = ventral frontal cortex.

Five 20-min increments (30–50, 40–60, 50–70, 60–80, and 70–90 min after injection) and four 30-min increments (30–60, 40–70, 50–80, and 60–90 min after injection) were evaluated. The SUVR in each target ROI was calculated for each 20-min and 30-min time window by taking the ratio of the integrated activities between the target ROI and the reference cerebellar ROI. The equation for the SUVR calculation can be written as follows:

$$\text{SUVR}=\frac{{\text{SUV}}_{\text{ROI}}}{{\text{SUV}}_{\text{CER}}}=\frac{{\int }_{\text{t}1}^{\text{t}2}{\text{C}}_{\text{ROI}}(\text{t})\text{dt}}{{\int }_{\text{t}1}^{\text{t}2}{\text{C}}_{\text{CER}}(\text{t})\text{dt}}.$$

Eq. 1

In this equation, t1 and t2 are the start and end times, respectively, of the experimental time window, and C_ROI(t) and C_CER(t) are the time–activity concentrations of the target and cerebellar ROIs, respectively. Because variables such as injected dose or physical parameters (e.g., body weight or estimated surface area) cancel out in the ratio calculation, these were not applied in the individual calculations. Thus, SUVR is the summed tissue activity ratio. The SUVR is related to the DVR because V_T is equivalent to the tissue-to-plasma ratio (TPR) at equilibrium, as achieved using a constant infusion paradigm (14). The equilibrium TPR_ROI-to-TPR_REF ratio would be equal to DVR.

Two types of DVRs were calculated for each ROI over the entire 90-min PiB scan interval. One DVR was based on the measured arterial plasma input function (DVR_ART), and the other was based on cerebellar input (DVR_CER) estimated from image-based measurements. DVR_ART was obtained by first calculating the V_T over each ROI using the Logan graphical method (15), then taking the ratio of V_T in the target ROI to the V_T (or V_ND) in the cerebellar ROI. The Logan analysis regression was applied to the 35- to 90-min integration intervals (8 points) (5).

RESULTS

The scatter plots for the pooled ROI evaluation of the PiB DVR and SUVR are shown in Figure 2 for 3 time windows (30–50, 50–70, and 70–90 min). The top row compares SUVR with DVR_ART, and the bottom row compares SUVR with DVR_CER. The figure demonstrates that SUVR and DVR were strongly correlated across all time windows within varying degrees. When the SUVR and DVR_ART were compared (Fig. 2, top row), the correlation coefficient tended to increase for later time windows. When the SUVR and DVR_CER were compared (Fig. 2, bottom row), the correlation was highest for the intermediate time window. Although the plots in Figure 2 show the pooled ROIs, scatter plots of individual ROIs showed these same general patterns.

Open in a separate window

FIGURE 2

Scatter plots of DVRs and SUVRs for all 18 ROIs of 40 subjects with regression analysis results. Example plots are shown for 3 time windows (30–50, 50–70, and 70–90 min) in each column. First and second rows compare SUVRs at different time windows with DVR_ART and DVR_CER, respectively. Solid and dotted lines are for regression and identity lines, respectively.

Figure 3A further demonstrates the effect of time window on the correlation coefficients between the DVR and the SUVR from Figure 2. The comparison of SUVR with DVR_ART showed that the correlation coefficient improved as the study progressed and began to plateau toward the end of the study (Fig. 3A). For the SUVR-to-DVR_CER comparison, the differences in R² across different time windows were fairly small, with the intermediate time windows (40–60 and 40–70 min) showing the highest correlations (Fig. 3A).

Open in a separate window

FIGURE 3

Linear regression results (R² [A] and slope [B]) for comparison between SUVR and each DVR are shown here for nine 20- or 30-min time windows that range from 30–50 to 70–90 min (only 20-min time windows are labeled on abscissa). First 2 points show comparison of DVR_ART and DVR_CER and are plotted for reference purposes.

In addition to the correlation coefficients, the slopes in Figure 2 can be used to assess bias in the SUVR measurement. The slope was higher than the identity line in all plots except the top left, meaning that the computed SUVRs generally overestimated the DVRs in all time windows except the one in which an early-time-window SUVR calculation (30–50 min) was compared with DVR_ART. This overestimation was shown again in Figure 3B, in which the slopes of the regression lines for each time window are plotted. The overestimation was evident for all time windows when SUVR was compared with DVR_CER, whereas overestimation was evident only in time windows greater than or equal to 40–60 min when SUVR was compared with DVR_ART (Fig. 3B, solid line with square data points).

Figure 3B also shows that the slope of the regression line increased with later times, until the 60- to 80-min interval. In all windows, the degree of the overestimation in SUVR was consistently larger for DVR_CER than for DV_ART, indirectly suggesting that the DVR_CER underestimated DVR_ART in general, consistent with expectation (6). The DVR_CER underestimated DVR_ART in this study by approximately 13% (data not shown).

These results were generally true across individual ROIs as well. Specifically, when comparing DVR_ART with SUVR, correlation coefficients for most ROIs increased as a function of time window and began to plateau toward the end of the study. Nearly all of the 18 ROIs showed correlation coefficients in the range of 0.7–1.0, with the exception of the SWM, pons, mesial temporal cortex, and occipital cortex. Correlation coefficients between DVR_CER and SUVR did not vary much as a function of time window, with the intermediate time windows showing the highest correlations. As with DVR_ART, SWM and pons demonstrated lower correlations than did other ROIs.

For individual ROIs, the slopes of the regression lines increased in general with later time windows until the 60-to 80-min interval, as was noted for the pooled ROI comparison. The slope values were relatively tightly grouped across each time window for most ROIs, with the exception of the SWM, thalamus, pons, and anterior ventral striatum. DVR_ART was again overestimated by the SUVR in time windows greater than 40–60 min for all ROIs except the SWM, and DVR_CER overestimated SUVR in all time windows over 30–50 min for all ROIs.

The observations from the correlation analysis were reflected in the concordance analysis results that were expressed as 95% CIs. Strong agreement was evident between the earliest SUVRs (30–50 min) for both DVR_ART (95% CI, 0.84–0.95) and DVR_CER (95% CI, 0.87–0.96). Strong agreement was also evident at the latest times (70–90 min) (DVR_ART: 95% CI, 0.85–0.95), with somewhat lesser agreement between the SUVR and DVR_CER as reflected by the lower CI range (DVR_CER: 95% CI, 0.72–0.92). Across all 18 ROIs, strong agreements were also evident (DVR_ART: 95% CI, 0.82–0.95, and DVR_CER: 95% CI, 0.73–0.92) as were the concordance results for the 13 primary ROIs (DVR_ART: 95% CI, 0.86–0.96, and DVR_CER: 95% CI, 0.77–0.93), which did not include the pons, sensory-motor cortex, thalamus, or SWM.

The second evaluation focused on the ability to differentiate AD patients and controls using SUVRs computed over each time window. Contrast capability was evaluated by investigating the simple group mean difference, pooled SD, and the effect size or effective contrast for each ROI, which are shown, respectively, as a function of time window in Figures 4A, 4B, and 4C for a group of representative ROIs. The figure shows that certain ROIs, such as the pregenual anterior cingulate, dorsal frontal cortex, and anterior ventral striatum, had larger differences in the group means; other ROIs, for example, the pons and mesial temporal cortex, had smaller or even negative (pons) differences. When compared across different SUVR time windows, the simple differences were fairly stable, with the largest values around the 50- to 70-min time window for most ROIs. In Figure 4B, the pooled SDs were clearly graded across different SUVR time windows, with most ROIs displaying the largest values at later time windows.

Open in a separate window

FIGURE 4

Simple group mean difference (A), pooled SD (B), and effective contrast (C) between AD patient (n = 11) and control (n = 16) groups are shown for SUVRs over 9 time windows (only 20-min windows labeled on abscissa) for representative group of ROI. (D) Averaged mean effective contrast and simple difference across all ROIs are shown. Difference data were scaled by arbitrary number (4.087) to show this curve on same scale as effect size curve. AVS = anterior ventral striatum; DFC = dorsal frontal cortex; MTC = mesial temporal cortex; OCC = occipital cortex; OCP = occipital pole; PAC = pregenual anterior cingulate; PAR = parietal cortex; SMC = sensory-motor cortex; THL = thalamus.

The effective group contrasts across different ROIs are shown in Figure 4C. The effective contrasts mainly decreased with increasing time window because of larger SDs in the later windows. These overall patterns can again be verified in Figure 4D, which shows the average of all ROIs as a function of time window. The simple group mean difference peaked near the 50- to 70-min time window, and the effective contrast continually decreased with later time windows.

Results from the final SUVR evaluation method are shown in Figure 5. As expected from previous studies (5), the simulated curves clearly showed higher uptake of PiB in PCM for the AD patient over the scan period. In Figure 5B, the DVR line for the average AD patient intersected the SUVR curve between 20 and 30 min, and the line for the average control intersected the SUVR curve between 30 and 40 min, suggesting better agreement between the SUVR and DVR around the 20- to 40-min time for PCM. Also, the SUVR curves exhibited a dynamically changing range until around 40–50 min and became more constant after the 50-min point. Thus, SUVR measurements at earlier time windows, although less biased relative to later measurements, might suffer from a lack of stability.

Open in a separate window

FIGURE 5

Generated SUV and SUVR curves using averaged rate constants (K₁, k₂, k₃, and k₄) determined for AD patient (n = 11) and control (n = 16) groups. DVRs were calculated also using rate constants (with a blood volume fraction of 0.05). Curves show SUV (A) and SUVR or DVR (B) for average control (gray) and average AD patient (black) for PCM and cerebellar (SUV only) ROIs.

DISCUSSION

In this study, we have evaluated the effect of choosing different time windows for the SUVR calculation in PiB PET. Different time windows were used to compute the SUVR, and the computed values over each time interval were quantitatively compared with estimated DVRs and their ability to contrast data from AD patients and control groups. A qualitative investigation of the temporal dynamics of SUVR measurements was also performed. We found the best compromise among physiologic validity, agreement of outcomes, and study feasibility to be achieved by the 50- to 70-min SUVR.

The arterial input model was assumed to be the gold standard because this method was verified relative to arterial input–based compartmental modeling analyses that allowed us to evaluate PiB plasma clearance, nonspecific uptake, and specific binding and to also refine data-acquisition and analysis parameters. The 2-tissue, 4-parameter model proved to be an appropriate choice for the PiB PET kinetics and described the 90-min PiB data well; however, this model did not perform as well when applied to only 60 min of data (difficulty determining tissue clearance, that is, k₄). The compartmental analyses helped to verify that specific binding differences were not likely the result of plasma clearance or nonspecific effects.

The arterial Logan and 2-tissue, 4-parameter cortical DVRs strongly agreed, and the former proved more stable (5). As previously described, a difference of about 1% was observed between the average (n = 5) AD posterior cingulate DVR determined for 50–90 min (5 points) and that determined for 35–90 min (8 points). Our data also showed that average absolute percentage differences in the arterial blood as input over 60 min (ART60) and over 90 min (ART90) DVRs for controls (n = 8) ranged from 1% to 2% across cortical areas (n = 8) and from 2% to 5% in subcortical areas (n = 3, including white matter), although temporal variation was apparent in the regional V_T values used to compute the DVR. The ART60 V_T was consistently less than the ART90 V_T. This variation was most pronounced for white matter (14%), and differences in the cerebellar V_T values were 6% and 9% for controls (n = 8) and AD patients (n = 6), respectively. It is unclear what is driving the V_T differences, although potential sources include factors that affect tissue or plasma clearance (e.g., vascular metabolites or input function inaccuracies).

The primary radiometabolite of PiB (6-sulfato-PiB) is produced peripherally in mice, rats, baboons, and humans. This radiometabolite is produced in rat brains but not produced in mouse or human brain tissue and does not cross the blood–brain barrier to a significant extent (19). The plasma parent fraction was initially determined using reverse-phase high-performance liquid chromatography (HPLC) measurements, and the aqueous or organic extraction measurements that were subsequently used did not differ from the HPLC measurements by more than 2% at any time. Use of the arterial DVR in the verification of simpler measurements is supported by our recent study that showed a strong correlation between PiB DVR_CER values and region-matched postmortem correlates of amyloid deposition for a deceased AD patient who underwent PiB imaging before death (20).

Comparisons among SUVR, DVR_ART, and DVR_CER demonstrated that SUVRs in most time windows generally overestimate DVR (either DVR_ART or DVR_CER), whereas DVR_CER generally underestimated DVR_ART. The degree of SUVR overestimation for both estimates of DVR increases with later time windows but tends to stabilize around 50–70 min, with minimal changes throughout the remainder of the study. Therefore, if the proximity of SUVR to DVR is the key factor to consider, SUVR should be calculated in the earlier time windows such as 30–50 or 40–60 min to minimize bias in the SUVRs.

Correlations between SUVR and DVR were generally stronger in later time windows, especially for DVR_ART. Specifically, the correlation coefficient between SUVR and DVR_ART increases more or less asymptotically as the time windows gets later, and the coefficient between SUVR and DVR_CER increases initially but becomes relatively stable by the 40- to 60-min range. Therefore, in terms of correlation strength, using later time windows to compute SUVR will give better results. When considering the optimal time window for calculation, note that the correlation coefficient from the SUVR-to-DVR_ART comparison does not gain much improvement at windows later than 50–70 min, and the correlation coefficient from the SUVR-to-DVR_CER comparison stays relatively strong through a wide range of time windows before 60–80 min.

Assessment of effective contrasts between AD patients and controls using SUVRs at different timewindows demonstrated that, in general, contrast tends to decrease with later time windows. At later times, counting statistics of the PET scan are inferior to those at earlier times, leading to increases in the pooled SD. This, in turn, decreases the effective group contrast. When the effective contrasts for only a few interested regions in the frontal, parietal, anterior cingulate, and precuneus cortices, such as the dorsal frontal cortex, parietal cortex, pregenual anterior cingulate, and PCM, are plotted separately (not shown), this superiority in the earlier time windows was also clear, suggesting the earlier time windows would be better in terms of the AD–to–control group separation.

Effective group contrast is an important measurement when trying to make distinctions between 2 subject groups. The ability to distinguish between an AD patient and a control or to classify a subject into a certain group is a critical consideration when studying a large number of subjects longitudinally. Distinctions between subject groups will be necessary when longitudinally tracing the progression of a subject from a normal or an MCI state to AD or when tracing any reversal of plaque deposition due to clinical intervention. Although results from the 2 other evaluations of SUVR time windows cannot be ignored, results from this particular assessment of the SUVR calculation are important for the purposes of our studies.

Temporal dynamics of the theoretic SUVR curve were considered qualitatively. Stable ranges of SUVR measurement began around the 40- to 50-min time, as verified by the steep slope of the curve before 40 min. This issue was more evident for the AD group, for whom PiB uptake was more significant (and prolonged), than for the controls. A larger slope (ΔSUVR/Δtime) could indicate sensitivity of the measure to errors in timing that a later acquisition would be less sensitive to because measures are more stable at later times. This means that variations in the PET technique that could affect the temporal dynamics of the tracer delivery, such as the differences in the tracer injection duration, could also change the SUVR measure unexpectedly. Although not a serious concern in a well-controlled study, variations in the PET technique could be an important issue when data are collected across multiple clinical sites. Temporal stability of the outcome measure would help to ensure uniformity in this aspect of the data collection across sites, leading us to seriously consider later time intervals for SUV data collection for such studies.

On the basis of these evaluations, the optimal time window for SUVR estimation may vary depending on the experimental parameters and criteria used. In the correlation-based assessment, the nearness of SUVR to DVR was observed at early times (e.g., 30–50 min), and the correlations between SUVR and DVR were generally higher at later times. However, as noted earlier, the correlations between SUVR and DVR_ART were not remarkably improved after the 50- to 70-min window, and correlations between SUVR and DVR_CER were relatively stable over the wide range of time windows. For effective contrast, earlier time windows were generally better; however, SUVRs before 40 min are changing rapidly and can vary significantly with differences in the imaging procedure. Forty to sixty minutes is a natural time interval to consider for the simplified SUVR measurement, given that the PiB tissue-to-plasma and tissue-to-cerebellum ratios in primary cortical areas begin to plateau at about 40 min, although this plateau can occur a bit later (50 min) for subjects with significant PiB uptake. In consideration of the agreement of outcomes, physiologic validity, measurement stability, and study feasibility, the 50- to 70-min time window was chosen as the overall best compromise for the SUVR calculations in this study.

Our site, in addition to several other sites, has used the 40-to 60-min SUVR because its advantages (relative to the 50- to 70-min SUVR) include lower bias, better counting statistics for lower PiB dose administrations (e.g., 185–370 MBq [5–10 mCi]), and shorter imaging sessions that can improve study feasibility. The 40- to 60-min window also has the same benefits as the 50–70 min window in that the SUVRs are most similar to DVRs at earlier times (Figs. 2 and ​and3),3), and the effective group contrast between AD patients and controls is more distinct for earlier time windows, compared with later time windows (Fig. 4). On the basis of the data in this study, the 40-min simulated SUVRs are close to the steep slope region of the temporal SUVR curve (Fig. 5). We chose to begin our SUVR calculation at 50 min (rather than at 40 min) to avoid SUV computations over times when the SUVs may be changing rapidly. The choice of a more stable SUVis at the expense of greater bias, relative to earlier (though potentially more unstable) measurements. The choice of the SUVR imaging interval should consider experimental parameters and overall study aims, such as injected dose and the subject groups of interest. Overall, we expect both the 40- to 60-min and the 50- to 70-min windows to provide good distinction of PiB retention in controls and AD patients.

CONCLUSION

Multiple time intervals were considered for the PiB PET SUVR calculation across controls, MCI patients, and AD patients. The examined time windows were 20 or 30 min and were evaluated between 30 and 90 min after injection. Although results varied according to the type of evaluation performed, a 20-min window proved adequate. A 50- to 70-min time window provided a good compromise between agreement of outcomes, physiologic validity, measurement stability, and study feasibility. The 40- to 60-min PiB SUVR is a natural time interval to consider for the simplified SUV measurement, particularly if studies are limited by low injected doses. The data in this study indicated that the 50- to 70-min interval was an optimal window for PiB PET studies, particularly those that might involve a large number of longitudinal scans collected at multiple, technically varying, clinical sites in controls, MCI patients, and AD patients.

Acknowledgments

References