Quantification of Positron Emission Tomography Data Using Simultaneous Estimation of the Input Function: Validation with Venous Blood and Replication of Clinical Studies

Abstract

Purpose:

To determine if one venous blood sample can substitute full arterial sampling in quantitative modeling for multiple positron emission tomography (PET) radiotracers using simultaneous estimation of the input function (SIME).

Procedures:

Participants underwent PET imaging with [¹¹C]ABP688, [¹¹C]CUMI-101 and [¹¹C]DASB. Full arterial sampling and additional venous blood draws were performed for quantification with the arterial input function (AIF) and SIME using one arterial or venous (vSIME) sample.

Results:

Venous and arterial metabolite-corrected plasma activities were within 6% of each other at varying time-points. vSIME- and AIF-derived outcome measures were in good agreement, with optimal sampling times of 12 min (¹¹C]ABP688), 90 min ([¹¹C]CUMI-101), and 100 min ([¹¹C]DASB). Simulation-based power analyses revealed that SIME required fewer subjects than the AIF method to achieve statistical power, with significant reduction for [¹¹C]CUMI-101 and [¹¹C]DASB with vSIME. Replication of previous findings and test-retest analyses bolstered the simulation analyses.

Conclusions:

We demonstrate the feasibility of AIF-recovery using SIME with one venous sample for [¹¹C]ABP688, [¹¹C]CUMI-101, and [¹¹C]DASB. This method simplifies PET acquisition while allowing for fully quantitative modeling, although some variability and bias are present with respect to AIF-based quantification, which may depend on the accuracy of the single venous blood measurement.

Introduction

Fully quantitative positron emission tomography (PET) relies on arterial blood sampling during the scan to measure the arterial input function (AIF), the unmetabolized radiotracer activity in blood needed to derive quantitative outcome measures (e.g., total volume of distribution (V_T) and binding potentials) [1]. Arterial sampling is invasive, costly, and often a deterrent to subject enrollment and retention [2]. Methods that obviate such sampling, generally split into reference region and AIF -recovery approaches, are invaluable. Reference region approaches solely quantify specific binding relative to nondisplaceable uptake (BP_ND), whereas alternate outcome measures can be needed, e.g. V_T and specific binding relative to radiotracer in plasma (BP_P) [1]. Further, the assumption that there exists a reference region completely devoid of specific binding is not valid for many radiotracers, such as [¹¹C]DASB (target: serotonin transporter) [3], [¹¹C]ABP688 (target: metabotropic glutamate receptor 5) [4], [¹¹C]PE2I (target: dopamine transporter) [5], and [¹⁸F]fallypride (target: dopamine D2/D3 receptor) [6]. Thus, less-invasive AIF-recovery methods are of interest.

AIF-recovery methods include population-based approaches, image-derived input functions [7–9] [10], and Simultaneous Estimation (SIME) [2], which recovers the AIF using simultaneous fitting of PET data in multiple brain regions and one arterial blood sample, with the assumption that the AIF is common to all brain regions [11–14]. SIME has been applied to several radiotracers, including [¹¹C]DASB [2, 15], [¹¹C]PIB (for amyloid-beta fibrils) [2], [¹¹C]CUMI-101 [2, 16–17] and [¹¹C]WAY-100635 [2, 17] (both for serotonin-1A receptor), and [¹⁸F]FDG (glucose analog) [2, 12, 18].

Although methods such as SIME with one arterial sample (aSIME) require less blood to be drawn, they still necessitate arterial catheterization or puncture. Thus, venous blood-based quantification strategies are advantageous. If equivalence between venous and arterial metabolite-corrected radiotracer activity is demonstrated, venous blood could substitute arterial for less-invasive quantification. This has been shown for [¹⁸F]FDG and [¹¹C]N-methylspiperone at late time-points [19–20]. As this equivalence likely varies by radiotracer, each radiotracer must be investigated independently.

Here, we determined the equivalence between venous and arterial plasma activity, parent fraction, and metabolite-corrected plasma activity for [¹¹C]ABP688, [¹¹C]CUMI-101, and [¹¹C]DASB, and examined the use of a venous sample instead of an arterial in SIME (vSIME). This is the first SIME implementation for [¹¹C]ABP688. We hypothesized that venous-to-arterial metabolite-corrected plasma equivalence will be found at late time-points and these points will provide optimal vSIME performance relative to AIF-based methods. An often overlooked consideration with less-invasive quantification methods is the impact on statistical power in clinical studies. Thus, we investigated the impact of SIME-based quantification on sample sizes required to detect group differences. We further sought to replicate previous AIF-based findings of diurnal variation with [¹¹C]ABP688 and test-retest repeatability with [¹¹C]CUMI-101 and [¹¹C]DASB.

Materials and Methods

Subjects

Table 1 displays information on participants and data. The Institutional Review Boards at Columbia, Stony Brook, and Yale Universities approved this study. Written, informed consent was received from each participant. Inclusion criteria were: ages 18–70 years and capacity to provide informed consent (details reported in electronic supplementary material (ESM)), yielding 10, 19, and 18 scans for [¹¹C]ABP688, [¹¹C]CUMI-101 and [¹¹C]DASB, respectively, in which both arterial and venous sampling was performed. PET scans used to estimate optimal sampling time-points were independent of those in prior SIME validations.

Table 1.

Subject Demographics, Data Acquisition, and Analysis Protocols

Radiotracer	# of Subjects	Age	Number (%) Female	Injected Dose (mCi)	Arterial plasma draw time (min)	Venous plasma draw time (min)	Scan Duration (min)	Modeling Method; Metabolite Fit	Scanner
[11C]ABP688	10	35.10 ± 9.86	5 (50%) Female	16.60 ± 2.68	4, 9, 12, 15, 20, 25, 30, 40, 50, 60	4, 12, 30, 60	60	2 Tissue Compartmen t Modeling; Hills	ECAT HRRT
[11C]CUMI-101	19	36.53±14.48	8 (42.1%) Female	13.65 ± 4.86	3, 12, 20, 30, 45, 60, 75, 90, 105, 120	3, 12, 30, 60, 90	120	LEGA; Hills	ECAT HR+
[11C]DASB	18	35.72±15.07	4 (40%) Female	12.98 ± 4.29	3, 12, 30, 35, 50, 65, 80, 90, 100	3, 12, 20, 50, 80, 100	100	LEGA; Damped biexponential	ECAT HR+

Abbreviations: LEGA: likelihood estimation in graphical analysis

PET Acquisition & Processing

Radiotracer preparation, image acquisition, pre-processing, and metabolite-corrected AIF (product of plasma and parent fraction) fitting were performed as previously described [3, 21–24] (for more details see ESM).

SIME was employed as described elsewhere [2]. In brief, SIME minimizes an objective function that incorporates both AIF parameters and compartmental model parameters for several regions of interest (ROIs), and includes one blood sample as an anchor for model identifiability [2] (the ROIs used within SIME are reported in ESM).

SIME was ran with the arterial and venous metabolite-corrected activities in closest correspondence and with the optimal anchor points previously identified for [¹¹C]CUMI-101 and [¹¹C]DASB [2, 25]. All time-points available were tested as potential anchors for this first SIME validation with [¹¹C]ABP688.

PET Quantification

Likelihood estimation in graphical analysis (LEGA) ([¹¹C]DASB and [¹¹C]CUMI-101) [26] or two-tissue compartmental (2TC) modeling ([¹¹C]ABP688) [27] quantified V_T. Corresponding BP_ND estimates were computed indirectly with V_T of the pseudo reference region previously suggested for these tracers (cerebellar grey matter) [3, 28–29]. The regions used to assess SIME performance are provided in ESM.

SIME Validation

The difference between the AIF- and SIME-derived estimates, normalized by the AIF-derived estimate, assessed the agreement between methods [2, 25]. The optimal venous and arterial sampling time-point for anchoring SIME was determined using V_T percentage difference mean and standard deviations, and bootstrapped linear regressions were fit between AIF- and SIME-generated estimates [30] (details on bootstrapping are reported in ESM). Analyses performed in Matlab 2017a (The MathWorks, Inc.; Natick, MA).

Power Analysis

To examine the detection of group differences with SIME, one ROI per radiotracer was selected to mimic hypothesis-testing scenarios ([¹¹C]ABP688: hippocampus, [¹¹C]CUMI-101: anterior cingulate cortex, [¹¹C]DASB: amygdala).20% and 5% group differences were generated from the AIF-estimated data, and the relationships between AIF and SIME estimates, determined during validation, were used to simulate SIME-based group differences (see ESM for details on simulated power analyses).

Results

Venous-to-Arterial Comparison

The ratio of venous-to-arterial plasma activity approached equivalence at late time-points for all radiotracers (Fig. 1): 1.01 ± 0.05 at 60min ([¹¹C]ABP688); 0.96 ± 0.07 at 90min ([¹¹C]CUMI-101); 0.96 ± 0.10 at 100min ([¹¹C]DASB). The [¹¹C]ABP688 parent fraction remained higher in venous than arterial plasma (ratios range: 1.34–1.41 from 4–60 min; Fig. 1), while parent fraction was approximately equivalent in venous and arterial blood for [¹¹C]CUMI-101 and [¹¹C]DASB (ratios range: 0.91 – 1.02; Fig. 1). The venous-to-arterial metabolite-corrected plasma activity approached equivalence at 12min (ratio = 1.03 ± 0.28) for [¹¹C]ABP688 and increased thereafter, while the same ratio converged at the latest time points for [¹¹C]CUMI-101 (ratio = 0.97 ± 0.11 at 90min) and [¹¹C]DASB (ratio = 0.97 ± 0.21).

Figure 1.

a Venous-to-arterial ratios of total plasma activity; b venous-to-arterial ratios of parent radiotracer fraction; c venous-to-arterial ratios of metabolite-corrected plasma activity shown for all time-points across the 3 tracers. Mean and standard deviations shown with horizontal and vertical solid lines. A ratio of 1 is shown with dashed lines in all plots.

SIME V_T

vSIME estimated V_T with a bias below 15%, on average, for all radiotracers, and the optimal blood time-points matched those with the closest metabolite-corrected venous-to-arterial equivalence (Table 2, Fig. 2).

Table 2.

Percentage Difference between Simultaneous Estimation (SIME) & Arterial Input Function-Generated V_T across Blood Sampling Times

Radiotracer	Sampling Time (min)	Arterial				Venous
		Mean ± SD	Median	MAD	[min, max]	Mean ± SD	Median	MAD	[min, max]
[11C]ABP688
	4	46.03 ± 9.13	49.62	4.49	[21.33 ,59.41	−131.99 ±1212.69	31.02	24.11	[−1078.17 ,71.40]
	12	−5.88 ± 14.41	−8.77	8.36	[−28.85 ,22.79]	−2.79 ± 18.69	−6.26	7.48	[−33.42 ,42.44]
	30	−72.80 ± 20.75	−66.65	12.12	[−116.91 ,−41.92]	−39.27 ± 43.80	−36.86	15.52	[−150.44 ,28.85]
	60	−123.56 ± 51.28	−122.92	40.77	[−214.31 ,−35.29]	−69.41 ± 56.52	−47.42	13.04	[−241.26 ,1.99]
[11C]CUMI-101
	30	2.84 ±11.74	3.79	7.05	[−25.41, 23.90]	−15.35 ± 31.68	−15.33	12.06	[−248.72, 25.31]
	60	−1.30 ± 21.70	−5.16	7.47	[−37.01, 80.26]	−20.37 ± 19.93	−18.92	13.42	[−77.65, 16.86]
	90	−9.86 ± 24.20	−13.54	6.38	[−86.13, 78.01]	−14.79 ± 24.09	−19.46	8.13	[−50.43, 67.85]
[11C]DASB
	50	−8.51 ± 15.15	−10.84	4.33	[−40.01, 54.41]	−42.48 ± 29.04	−38.16	19.20	[−119.34, 24.32]
	80	−8.00 ± 13.55	−7.03	9.61	[−39.22, 24.45]	−20.90 ± 24.07	−23.67	14.49	[−76.58, 47.46]
	100	−12.40 ± 16.52	−6.71	12.79	[−59.38, 17.55]	−11.66 ± 21.72	−9.72	6.96	[−86.18, 20.35]

Abbreviations: SD = standard deviation, MAD = median absolute deviation, V_T = volume of distribution

Figure 2.

Percentage differences between Simultaneous Estimation (SIME) and arterial input function derived V_T values shown for a [¹¹C]ABP688, b [¹¹C]CUMI-101, and c [¹¹C]DASB. The blood sampling times are shown on the x-axis. Arterial data is plotted left of the sampling time in red-to-yellow and venous is right in blue-to-green. Color gradients represent stratification by subject across the regions of interest examined. Medians are shown with horizontal solid line and median absolute deviations (MADs) are shown with vertical solid lines. Two outliers are excluded from the 4-min venous sample plot for [¹¹C]ABP688, where % differences are −5561.5% and −10781.7%. These points are included in the calculation of the medians and MADs.

The optimal vSIME sampling times varied from the optimal aSIME times for [¹¹C]CUMI-101 and [¹¹C]DASB, where earlier arterial draws at 30 and 80min, respectively, yielded the most accurate aSIME outcomes (Table 2, Fig. 2). Generally, aSIME and vSIME overestimated V_T relative to the AIF method (slopes=0.79 to 1.08, Fig. 3).

Figure 3.

V_T estimates from Simultaneous Estimation (SIME) and the arterial input function plotted for a the optimal arterial and b venous sampling times across the three tracers. Colors in the gradients from red-to-yellow and blue-to-green show individual subjects. Identity is shown with a black line. Mean best-fit approximation from bootstrapped regression is shown with a thick solid line. The 95% bootstrapped regression confidence interval (CI) for the best-fit slope is shown in grey.

Power Analysis

Across radiotracers and a- and vSIME, when accounting for the bias as described in ESM, SIME required fewer subjects to detect 20% and 5% group differences relative to the AIF method (Table 3). This reduction in sample size was significant for [¹¹C]CUMI-101 (p = 0.004 and 0.005 for 20 and 5%, respectively) and [¹¹C]DASB (p = 0.024 and 0.045 for 20 and 5%, respectively) with vSIME. For aSIME, only [¹¹C]DASB (p = 0.010 and 0.012 for 20 and 5%, respectively) required significantly fewer subjects.

Table 3:

Simulated Group Difference Results

				Group Difference = 20%			Group Difference = 5%
	Blood Sample	Estimated slope of SIME vs AIF for the single region (mean ± SD)	Ratio of VT Variance (AIF/SIME)	Estimated sample size: SIME (mean ± SD)	Estimated sample size: AIF	p-value	Estimated sample size: SIME (mean ± SD)	Estimated sample size: AIF	p-value
[11C]AB P688	arterial	1.031 ± 0.367	1.795	35 ± 23	51	0.255	458±324	807	0.17
	venous	0.692 ± 0.225	0.740	37 ± 25	51	0.310	485±339	807	0.2
[11C]CU MI-101	arterial	0.705 ± 0.103	0.624	12 ± 4	16	0.108	169 ± 62	244	0.119
	venous	0.244 ± 0.303	0.397	6 ± 4	18	0.004*	93 ± 76	274	0.005*
[11C]DA SB	arterial	0.750 ± 0.299	2.030	6 ± 3	14	0.010*	82 ± 49	210	0.012*
	venous	0.354 ± 0.225	0.721	5 ± 2	9	0.024*	60 ± 37	130	0.045*

^*p < 0.05; Abbreviations: SIME – Simultaneous Estimation; AIF – Arterial Input Function Method; SD – standard deviation.

SIME BP_ND

At the optimal V_T-determined sampling times, aSIME and vSIME estimated BP_ND within 13% and 10%, respectively, of full sampling ([¹¹C]ABP688: −9.67 ± 20.7% (aSIME) and −9.08 ± 19.46% (vSIME); [¹¹C]CUMI-101: 12.11 ± 7.59% (aSIME) and 4.91 ± 21.08% (vSIME); [¹¹C]DASB: 9.05 ± 9.98% (aSIME) and 9.37 ± 9.62% (vSIME)). Analogous to V_T estimation, BP_ND was overestimated (slopes = 0.79 to 0.93), although the 95% confidence intervals (CIs) of the regression slopes were narrower and intercepts were lower than that of V_T, indicating higher accuracy in estimating BP_ND (Suppl. Fig. A in ESM) with this indirect approach.

Replicating Findings with [¹¹C]ABP688

DeLorenzo et al. [22] found significantly higher [¹¹C]ABP688 V_T in the afternoon relative to the morning when using AIF-based quantification. We sought to replicate these findings with SIME. Using 7 of the 8 published subjects (1 subject did not have venous sampling and was excluded here) [22], a significant effect of time-of-day was found with AIF-based quantification for V_T (p = 0.001, β = 0.78), but not for BP_ND (p = 0.17), with afternoon uptake higher than morning uptake, agreeing with published results [22].

Using aSIME and vSIME, these findings were replicated for V_T (aSIME: p = 0.001, β = 1.60; vSIME: p < 0.001; β = 1.17) and BP_ND (aSIME: p = 0.24; vSIME: p = 0.08).

Replicating Test-Retest Findings

Scans from independent test-retest studies were used ([¹¹C]CUMI-101: 8 subjects and [¹¹C]DASB: 12 subjects with test and same-day retest scans), which overlap with the 7 [¹¹C]CUMI-101 scans and the 11 [¹¹C]DASB scans reported in previous studies [21, 31]. Only aSIME was assessed becase venous samples were not collected in the previous studies.

For [¹¹C]DASB, absolute test-retest percent differences using an AIF were 8.92 ± 8.53% (V_T) and 13.79 ± 17.00% (BP_ND). With aSIME, test-retest variability increased to 18.93 ± 14.69% (V_T) and 13.53 ± 17.94% (BP_ND). A significant difference was found in test-retest variability in V_T between SIME and AIF, but not in BP_ND (V_T: p = 0.03, 10.01% higher variability with aSIME; BP_ND: p = 0.90, 0.26% lower variability with aSIME).

For [¹¹C]CUMI-101, absolute test-retest percent differences using an AIF were 6.90 ± 4.56% (V_T) and 14.14 ± 9.63% (BP_ND). With aSIME, variability increased to 9.66 ± 7.89% (V_T) and 15.56 ± 11.65% (BP_ND)_. There were no significant differences in test-retest variabilities between AIF and aSIME (V_T: p = 0.26, 2.76% higher variability with aSIME; BP_ND: p = 0.53, 1.43% higher variability with aSIME).

Discussion

The purpose of this investigation across the radiotracers [¹¹C]ABP688, [¹¹C]CUMI-101, [¹¹C]DASB was to: 1) report the radiotracer dynamics in venous plasma relative to arterial, 2) validate the use of one venous blood sample for full quantification with SIME, and 3) illustrate the applicability of SIME in future studies. Our results indicate that venous and arterial metabolite-corrected plasma activities approached equivalence for all three radiotracers, albeit at different times post-injection. One venous sample generated outcome measures with good agreement to AIF-based quantification. Next, we found that SIME requires comparable, or even smaller, sample sizes, relative to AIF-based quantification, to accurately detect group differences. These effects were bolstered with analyses seeking to replicate prior findings.

Venous-to-Arterial Plasma Comparison

This blood data was examined in the context of choosing optimal post-injection blood sampling times for SIME, although it can be used to further: (1) other less-invasive quantification techniques and (2) our understanding of the mechanisms underlying radiotracer metabolism.

Across radiotracers, we observed higher total activity in arterial relative to venous plasma early in the scan, and equilibration by the end of the scan. We found that [¹¹C]ABP688 parent fraction was higher in venous plasma than arterial throughout the scan; because this ratio was stably higher than 1, it was at 12 min post-injection, when the venous-to-arterial total plasma activity was lower than one, that the corresponding venous and arterial metabolite-corrected plasma activities approximated each other. The relative deficit of radiometabolites observed in venous plasma appears to be biological. Examination of the high-pressure liquid chromatography (HPLC) data for venous and arterial plasma revealed clear separation of parent and metabolite peaks, despite a potential for pH-related differences in HPLC retention [32]. It is however difficult to identify potential biological explanations for this parent fraction imbalance. One possibility is that constant metabolite sequestration by the liver for excretion (80% of [¹¹C]ABP688 excretion is hepato-billiary [27]) may contribute to there being less metabolites in venous than arterial plasma. The fact that the arterial-to-venous equivalence of metabolite-corrected plasma activity is reached via the cancelling of an overestimation and underestimation is not ideal. One potential solution requiring validation is to normalize the venous parent fraction by the constant bias observed throughout the scan.

For [¹¹C]CUMI-101 and [¹¹C]DASB, given a relatively stable ratio of venous-to-arterial parent fraction (~1) throughout the scan, the metabolite-corrected plasma activities equilibrated at late time points, when the raw plasma activity, initially higher in arterial plasma, approached venous.

SIME Validation

We provide the first validation of SIME for [¹¹C]ABP688, while we build upon previous work validating aSIME for [¹¹C]CUMI-101 and [¹¹C]DASB. We found that the optimal venous sampling times were those with the closest metabolite-corrected plasma activity correspondence between venous and arterial blood (12, 90, and 100 min for [¹¹C]ABP688, [¹¹C]CUMI-101, and [¹¹C]DASB, respectively). SIME with blood samples acquired at these optimal times estimated V_T within 15% of the AIF method.

For [¹¹C]ABP688, vSIME was slightly more accurate in estimating V_T than aSIME (percentage difference from the AIF method: −2.8% for vSIME and −5.9% for aSIME), with the optimal blood sample time at 12 min.

For [¹¹C]CUMI-101 and [¹¹C]DASB, aSIME was previously found to optimally estimate V_T with a blood sample acquired at 60 and 50 min, respectively [2, 16, 25], while here we found that the optimal time-points were 30 and 80 min, respectively. This discrepancy may be attributable to different considered datasets and selection criteria (signed percent difference normalized by AIF-derived values used here, while unsigned percent difference normalized by the average of the two measurements used previously [2]). The fact that all late-scan time-points provided accurate quantification (within 13%), however, provides flexibility in applying aSIME to previously acquired datasets.

Although it is important to additionally examine binding potential estimation to test if SIME is robust to quantifying specific binding, when binding potentials are computed via subtraction, like here, with a reference region, biases introduced with SIME may be removed or reduced. As expected, BP_ND (as compared to V_T) was more stably estimated with SIME relative to full sampling, as evidenced by narrower ranges of regression slopes, higher R-squared values, and narrower CIs. Cerebellar grey matter has been shown to exhibit some specific binding for these tracers and is thus not an ideal reference region [3, 28, 33–34], so BP_ND estimates with full sampling and SIME will both be biased [35]. To remediate this bias, approaches have been proposed to estimate BP_ND noninvasively without a reference region [17, 36], but further binding potential estimation validation with SIME is necessary.

Power Analysis

Less-invasive PET quantification methods have the potential to reduced accuracy and precision in estimating outcomes [37]. Importantly, introduced bias and variance can negatively impact statistical power to detect group differences, an issue which is often unaddressed.

We found that when SIME group differences are simulated based on the established bias between SIME and AIF-based estimates, SIME required smaller sample sizes than the AIF method to achieve statistical power, with significant reductions observed for vSIME with [¹¹C]CUMI-101 and [¹¹C]DASB and for aSIME for [¹¹C]DASB. This effect can be understood by examining bias and variance introduced with SIME. For example, for [¹¹C]CUMI-101, in Table 3, the slope relating vSIME to AIF-based estimates was 0.24, indicating a large V_T overestimation with vSIME. The variance in V_T was also greater with vSIME than full sampling (ratio of variance = 0.40; Table 3). However, the variance in V_T does not appear to scale linearly with its overestimation, yielding a group difference that is visually, and statistically, more discernable (see Suppl. Fig. B.a and Table A in ESM).

Conversely, vSIME with [¹¹C]ABP688 (Suppl. Fig. B.b and Table A in ESM) slight overestimations were observed in V_T and V_T variance; thus, statistical power did not significantly differ between vSIME and full sampling. Importantly, the relationship between SIME and full sampling must be known and considered when weighing SIME’s utility within varying study designs. If alternate blood sampling times are used or this relationship were to change otherwise, then sample sizes would need to be reassessed.

Replication of Findings

As fewer subjects were required in the simulation analyses to achieve statistical power, we hypothesized that SIME would be powered to estimate larger group differences in diurnal variation of [¹¹C]ABP688 V_T than full sampling. vSIME and aSIME estimated significant diurnal variations in [¹¹C]ABP688 V_T, replicating previous AIF-based findings [22], but the magnitude of the diurnal variation was overestimated with SIME. While Delorenzo et al. observed weak, but significant, effects with BP_ND, SIME and AIF using 7 of the 8 original subjects yielded non-significant diurnal variations with BP_ND. This may be a result of reduced statistical power using less subjects in this study, as the direction of the diurnal variation – afternoon uptake higher than morning – was consistent across all current and prior analyses [22]. These data support the simulation-based analyses, but while group differences are detectable with SIME, absolute estimates of V_T, as well as the magnitude of group-wise effects, are not necessarily estimated precisely.

SIME did not significantly alter test-retest variability in V_T or BP_ND relative to AIF-based quantification for [¹¹C]CUMI-101, supporting the non-significant difference in sample sizes detected in the simulation-based analyses. With [¹¹C]DASB, aSIME significantly increased the test-retest variability of V_T by 10% over the AIF method, but provided BP_ND test-retest repeatability within 0.3% of full sampling. Notably, for [¹¹C]DASB, vSIME V_T values are in closer agreement to AIF values than aSIME (vSIME slope=0.98; aSIME slope=0.84). Future analyses on [¹¹C]DASB test-retest datasets with venous blood may reveal that vSIME can robustly generate V_T estimates with lower test-retest variability.

Limitations

When weighing the reduced cost and burden relative to AIF-based quantification, without the need to increase sample sizes, SIME is a promising less-invasive approach. However, given the limited sample sizes considered, future validation is needed. Future work might also consider using more physiological models for the AIF, such as pharmacokinetic models [15]. Further, given that the optimal venous blood sampling times are 90 and 100 min for [¹¹C]CUMI-101 and [¹¹C]DASB, respectively, and that these tracers are carbon-11 labeled, it is possible that metabolite measurements may be less reliable at these late time-points, perhaps contributing to the relatively low correlation observed between vSIME and AIF-based V_T for these radiotracers. A potential solution to be investigated is to obtain and average duplicate or triplicate plasma and parent fraction measurements at the defined time-point, which may enhance the suboptimal V_T correspondence , and be especially advantageous if lower activity doses are injected.

[¹¹C]CUMI-101 and [¹¹C]DASB samples were composed of healthy control and depressed patients and SIME performance was not found to significantly differ across the populations (see details in ESM). Despite these preliminary results, we recommend validation of venous-to-arterial blood equivalence for specific populations/treatments before use in a large study, as this equivalence could vary across conditions.

Conclusions

Less-invasive PET quantification methods, like SIME, are essential to mitigate costs and discomfort associated with AIF-based quantification. We found that at the optimal blood sampling time, venous blood can substitute the previously validated arterial sample in SIME to provide outcomes in agreement with AIF-based quantification for [¹¹C]ABP688, [¹¹C]CUMI-101, and [¹¹C]DASB. One limitation is the observed variability in SIME performance across subjects, which may be related to the accuracy of venous blood measurement used for SIME anchoring. Future work could investigate acquisition of repeated blood draws at the anchor time to stabilize potential uncertainty introduced with a single plasma activity and parent fraction measurement. Further, work to develop noninvasive derivations of the SIME anchor could obviate the need for any blood sampling [15–16].

Pivotal to assessing the utility of a less-invasive method, SIME-introduced bias did not decrease statistical power. However, the fact that a bias is observed in the current SIME implementation necessitates careful consideration before application to studies in need of absolute quantification of outcomes. Despite this bias, however, SIME replicated previous findings of diurnal variation in [¹¹C]ABP688 binding and test-retest reliability of [¹¹C]CUMI-101 and [¹¹C]DASB binding, and thus, may be a promising alternative for minimally invasive PET quantification.