Sage Journals: Discover world-class research

Abstract

Standardized Uptake Value Ratio (SUVR) is a widely reported semi-quantitative positron emission tomography (PET) outcome measure, partly because of its ease of measurement from short scan durations. However, in brain, SUVR is often a biased estimator of the gold-standard distribution volume ratio (DVR) due to non-equilibrium conditions, i.e., clearance of the radiotracer in relevant tissues. Factors that affect radiotracer metabolism and clearance such as medication or subject groups could lead to artificial differences in SUVR. This work developed a correction that reduces the bias in SUVR (estimated from a short 15–30 min PET imaging session) by accounting for the effects of tracer clearance observed during the late SUVR time window. The proposed correction takes the form of a one-step non-linear algebraic transform of SUVR that is a function of radiotracer dependent parameters such as clearance rates from the reference and target tissues, and population averaged reference region clearance rate ( $k_{2, ref}$ ). An important observation was the need for accurate estimation of radiotracer clearance rate in target tissue, which was addressed with a regression based model. Simulations and human data from two different radiotracers (healthy controls for [¹¹C]LSN3172176, healthy controls and Parkinson’s disease subjects for [¹⁸F]FE-PE2I) were used to validate the correction and evaluate its benefits and limitations. SUVR correction in human data significantly reduced mean SUVR bias across brain regions and subjects (from ∼25% for SUVR to <10% for corrected SUVR). This correction also significantly reduced the variability of this bias across brain regions for both tracers (approximately 50% for [¹¹C]LSN3172176, 20% for [¹⁸F]FE-PE2I). Future work should investigate the benefits of using corrected SUVR in other populations and with different tracers.

Keywords

PET quantification tissue clearance correction

Introduction

Distribution volume ratio (DVR) is an important PET outcome for radiotracers following reference region models.¹ By definition, it is the ratio of activity concentration in a target tissue to reference tissue under equilibrium conditions. Practically, true equilibrium is seldom established within a typical PET scan duration following a bolus injection, and therefore, compartment modeling of dynamic scan data is the gold standard for estimating DVR.² Compartment modeling, however, requires long dynamic PET scan data, which is often impractical to implement in a clinical setting. DVR is frequently approximated by late time standardized uptake value ratio (SUVR), defined as the ratio of activity concentration in a target tissue to reference tissue during a specified short time window. Due to radiotracer clearance in tissue during the SUVR time-window, SUVR can be biased with respect to the underlying DVR value.^3,4 Typically, the bias due to tracer clearance is smaller in the reference region compared to the target regions, resulting in a positive bias in SUVR at late times.

SUVR estimation is commonly established during a suitable time-window that provides minimal bias compared to DVR based on Bland-Altman and other difference plots.^4,5 Previous studies also relied on pseudo-equilibrium based concepts, focusing on the time where the derivative of specific binding in a target is close to zero, to define the SUVR time interval.^6
–8 However, for many radiotracers that exhibit substantial regional variation in kinetics, no single time interval sufficiently reduces the differences between SUVR and DVR across all regions of interest (ROI). Alternatively, external factors such as competing drugs⁹ or disease status can affect individual radiotracer metabolism and clearance.^10
–12 Finally, logistical challenges frequently complicate the ability to acquire imaging during a consistent late time window. These important factors can confound the interpretation of SUVR as a surrogate measure of the true DVR in human neuroimaging scans.

The aim of this study was to develop a theoretical approach that corrects SUVR bias in radiotracers with reversible-binding kinetics by accounting for non-equilibrium effects of tracer clearance. This approach extends a correction developed for effects of non-equilibrium tracer clearance on late time tissue-to-plasma activity in bolus-infusion protocols.¹³ The theory behind this correction was extended to adapt it for correcting SUVR after a bolus injection. Next, the SUVR correction was validated using simulations and human data for the PET radiotracers [¹¹C]LSN3172176, which has high affinity for the muscarinic (subtype M1) acetylcholine receptors,¹⁴ and evaluation in human datasets with [¹⁸F]FE-PE2I, which has high affinity for dopamine transporter (DAT).¹⁵ The study concludes with a general roadmap for adapting this correction to other radiotracers based on the generalizability of the most important findings.

Theory

Part 1: Correction of bias in apparent V_T under non-equilibrium radiotracer clearance

Volume of distribution (V_T) is defined as the ratio of radiotracer concentration in tissue (C_T) to plasma (C_P), at equilibrium (steady state). However, in the absence of equilibrium, a non-zero clearance rate of the radiotracer from the plasma ( $β$ , min⁻¹) and tissues ( $β_{T}$ , min⁻¹), leads to a bias between V_T and C_T/C_P (or, apparent V_T). Previous work has developed a correction for this bias that is applicable if the conditions for transient equilibrium are met.^3,13 Transient equilibrium is established when the rate of clearance of the radiotracer in all tissues and plasma is not only small, but also identical ( $β = β_{T}$ ). While this paradigm is useful in the context of an imperfect tracer infusion, the conditions for transient equilibrium are largely unmet after most bolus injections. In particular, differential rates of tracer clearance from the plasma and tissues are expected during practical scan time after a bolus injection ( $β$ ≠ $β_{T}$ ). Here, we developed a mathematical framework for correcting the bias between V_T and apparent V_T (C_T/C_P) at late time periods, under non-equilibrium conditions of radiotracer clearance ( $β$ ≠ $β_{T}$ ). This correction is a more general form of the previously derived result, and transforms to the previously reported correction under the special assumption of transient equilibrium.

The concentration of the radiotracer following a bolus injection approaches a mono-exponential clearance at late time.

C_{P} (t) \to B e^{- β t}

(1)

At the same time, the concentration of the radiotracer in a particular target tissue is equal to a convolution of the impulse response function of the tissue with C_P. For a tracer whose kinetics are well-described by a 1TCM (where generally $k_{2}$ > $β$ ),

C_{T} (t) \to B e^{- β t} \otimes K_{1} e^{- k_{2} t} = \frac{B K_{1}}{k_{2} - β} (e^{- β t} - e^{- k_{2} t})

(2)

The instantaneous value of the clearance rate of the tracer in the same tissue ( $β_{T}$ ) can be defined by the equation,

C_{T} (t) = C e^{- β_{T} t}

(3)

where

β_{T}

depends on time, such that

β_{T} = - \frac{1}{C_{T}} \frac{d C_{T}}{d t}

(4)

Substituting equation (2) in equation (4),

β_{T} = \frac{e^{- β t} β - e^{- k_{2} t} k_{2}}{e^{- β t} - e^{- k_{2} t}} = \frac{β - z k_{2}}{1 - z}

(5)

where z =

\frac{e^{- k_{2} t}}{e^{- β t}}

equation (5) can be rearranged to,

z = \frac{β_{T} - β}{β_{T} - k_{2}}

(6)

Let the apparent V_T be denoted by the symbol V_T(A). Then,

V_{T (A)} = \frac{C_{T} (t)}{C_{P} (t)} = \frac{B K_{1}}{k_{2} - β} \frac{(e^{- β t} - e^{- k_{2} t})}{B e^{- β t}} = \frac{K_{1}}{k_{2} - β} (1 - z)

(7)

Substituting z from equation (6) to equation (7) and simplifying,

V_{T (A)} = \frac{K_{1}}{k_{2} - β_{T}}

(8)

Since, the true V_T is equal to $K_{1} / k_{2}$ , the previous equation can be rearranged as,

V_{T} = V_{T (A)} (1 - \frac{β_{T}}{k_{2}})

(9)

Note that this derivation did not assume transient equilibrium. However, if the additional assumption of $β = β_{T}$ were imposed (transient equilibrium), then $z = 0$ , and equation (9) would transform to,

V_{T} = V_{T (A)} (1 - \frac{β}{k_{2}})

(10)

which was derived previously.^3,13 The present work focused on correcting the bias in SUVR at late time points after a bolus injection. Under these conditions, non-equilibrium tracer clearance rates are expected and therefore, the more general form (equation (9)), was used to derive the correction for SUVR bias.

Part 2: Correction of bias in SUVR values under non-equilibrium radiotracer clearance

To adapt the framework developed above (equation (9)) for correction of non-equilibrium bias in SUVR values, consider two distinct tissues of interest – from target and reference regions, respectively (subscript tar-target, ref-reference). Each of these tissues can have distinct clearance rates of the radiotracer during the SUVR time window ( $β_{tar}$ and $β_{ref}$ , respectively). In that case, the relationships between $V_{T}$ and $V_{T (A)}$ for the target and reference tissues can be expressed as,

V_{T, tar} = V_{T (A), tar} (1 - \frac{β_{tar}}{k_{2, tar}})

(11)

V_{T, ref} = V_{T (A), ref} (1 - \frac{β_{ref}}{k_{2, ref}})

(12)

Dividing the two equations,

DVR = SUVR \frac{(1 - \frac{β_{tar}}{k_{2, tar}})}{(1 - \frac{β_{ref}}{k_{2, ref}})}

(13)

where

DVR = \frac{V_{T, tar}}{V_{T, ref}}

and

SUVR = \frac{V_{T (A), tar}}{V_{T (A), ref}}

. Further, for a tracer that is well described by 1TCM kinetics,

$k_{2, tar} = R_{1} \frac{k_{2, ref}}{DVR}$ _, where $R_{1} = \frac{K_{1, tar}}{K_{1, ref}}$ . Substituting $k_{2, tar}$ in the previous formula and rearranging,

DVR = \frac{SUVR}{1 - \frac{β_{ref}}{k_{2, ref}} + β_{tar} \frac{SUVR}{k_{2, ref} R_{1}}}

(14)

The denominator on the right side of this formula represents a correction of the non-equilibrium errors in $SUVR$ . Note that at equilibrium ( $β_{ref} = β_{tar} = 0$ ) DVR = SUVR as expected. Due to the assumptions made in this model (late time clearance rates expressed as a single exponential, 1TCM kinetics for the tracer), the left hand side is better represented as an estimator for the underlying DVR, henceforth referred as the corrected SUVR or SUVR_C.

SUV R_{C} = \frac{SUVR}{1 - \frac{β_{ref}}{k_{2, ref}} + β_{tar} \frac{SUVR}{k_{2, ref} R_{1}}}

(15)

$β_{tar}$ and $β_{ref}$ are the clearance rates of the tracer from the target and reference regions, respectively, that can be estimated from the regional time-activity curves, while $R_{1}$ and $k_{2, ref}$ can be approximated a priori based on previously published literature estimates. It should be noted that this correction cannot be performed on SUVR values measured from a static PET image since dynamic data are required to estimate the clearance rates of the radiotracer in the target and reference tissues. Instead, the correction could be performed on SUVR values obtained from a short dynamic scan (15–30 min) with three or more PET frames.

Materials and methods

Simulations

Simulations were used to evaluate the benefits and limitations of the proposed correction formula. Simulated data were based on published human data with [¹¹C]LSN3172176 ¹⁴. K₁ was set to 0.407 mL min⁻¹cm⁻³ for all simulated regions and k₂ was calculated to vary V_T (K₁/k₂) in the range of 5 to 50 mL cm⁻³, with V_T = 5 mL cm⁻³ used to simulate the cerebellum as the reference region.¹⁴

Correction performance across varying tracer clearance rates in tissue ( $β_{tar}$ ) was implemented as follows. First a baseline input function was generated from a population average of metabolite-corrected arterial plasma time activity curves for [¹¹C]LSN3172176 from human studies (n = 22 healthy controls). The later phase (t = 6 min to t = 120 min) of this averaged input function was described with a two-exponential fit. Simulated plasma input functions were then generated by varying the slower exponent’s rate constant ( $β$ ) across a range of terminal clearance rates [−0.003, 0.007] min⁻¹, bridged with the early phase (t = 0 min to t = 6 min) using cubic spline interpolations (see Figure S1). The differences in the terminal clearance rates ( $β$ ) of the input functions cause the tracer clearance rate in tissue at late times ( $β_{tar}$ ) to vary in the simulations.

Bias characterization

Simulations were performed for [¹¹C]LSN3172176 to characterize bias in SUVR and SUVR_C compared with true DVR. Noiseless simulated time activity curves were generated for tissue regions with varying DVR (range: 1 to 10) and over the range of simulated $β$ values (input functions). For each combination of DVR and $β$ values, SUVR was calculated as the mean activity ratio between the target and reference region (cerebellum) using 60–80 minutes of simulated data. This time period was chosen to ensure radiotracer clearance from all ROIs, as the activity in the region with the highest DVR peaked around 60 min. SUVR values were then corrected to SUVR_C using the correction formula (equation (15)) with k_2,ref based on population average from human studies, $β_{tar}$ and $β_{ref}$ estimated from single exponential decay fits to 60–80 min of simulated noiseless time activity curves for the target and reference regions, and R₁ = 1.

Bias in fixed parameters: R₁ and k_2,ref

Simulations were also performed to evaluate the sensitivity of the correction formula to errors in the implemented values of R₁ and k_2,ref, which were assumed a priori to be 1 and 0.081 min⁻¹ (population average), respectively. To account for individual variability in these parameters, the correction formula was evaluated across a wider range of R₁ and k_2,ref parameters: [75%, 125%] of their initially simulated values. This range encompasses the standard deviation (SD) of R₁ and k_2,ref observed from [¹¹C]LSN3172176 in human healthy control data (SD: 18.2 and 15.8 percent of their respective mean values; R₁ values computed over nine brain ROIs). For sensitivity analysis of R₁, this parameter was varied from 0.75 to 1.25 by varying K₁ for all target ROIs (K_1,ref was held constant at 0.407 min⁻¹). However, the correction was always performed with R₁ = 1 to evaluate bias introduced by this a priori choice. The k₂ parameter of all target regions was simultaneously varied to keep V_T constant. For sensitivity analysis of k_2,ref, this parameter was varied in the range of 0.081 ± 0.020 min⁻¹ in the simulated time activity curves of the reference region. The k₂ parameter of all target regions was simultaneously adjusted to simulate regions with varying DVR (range: 1 to 10). Each region’s simulated SUVR was corrected to SUVR_C using the fixed value of k_2,ref = 0.081 min⁻¹ in the correction formula. Population averaged plasma arterial time activity curve was used as the input function for these simulations.

Variability in estimated parameters: $β_{ref}$ and $β_{tar}$

The tracer clearance rates in reference and target tissues, $β_{ref}$ and $β_{tar}$ , can theoretically be estimated with a single exponential decay fit to the respective time activity curves during the SUVR time window. However, noise in the time activity curves in human data and restriction to a short time window introduce substantial uncertainty in the estimation of $β_{ref}$ and $β_{tar}$ . Thus, simulations were performed to study the sensitivity of the proposed correction to errors in the estimation of tracer clearance rates in reference and target tissue. For $β_{ref}$ , noiseless time activity curves were generated for tissues with varying DVR (range: 1 to 10) for the population-averaged input function. The average coefficient of variation (CV) for $β_{ref}$ in the human data (60–80 min) was about 21%, therefore, a standard error (s.e.) equal to 25% of the noiseless estimate of $β_{ref}$ was used to characterize the performance of the correction formula in this simulation.

The mean value and s.e. in $β_{tar}$ strongly depends on the specific ROI. The s.e. in $β_{tar}$ estimation depends on ROI volume since time activity curves are noisier for smaller ROIs. Three different ROIs (DVR = 6.2, 7.2, 9.0) were simulated to mimic radiotracer dynamics of the frontal lobe, caudate and the putamen regions, respectively, after a bolus of [¹¹C]LSN3172176 using the baseline input function. The expected CV in $β_{tar}$ for all three regions were estimated using 60–80 minutes of PET data in the human cohort. The largest value of CV (123.8%, estimated for caudate) was used to define a s.e. equal to 125% of the noiseless $β_{tar}$ value in the simulations for all three regions, which simulates the worst-case variability in $β_{tar}$ estimation.

Human data

For [¹¹C]LSN3172176, a combination of previously reported human data,¹⁴ and data from ongoing studies on healthy subjects [overall n = 30, mean (SD) age: 39.8 (11.6) years, 14 F] who underwent a dynamic scan (High Resolution Research Tomograph (HRRT); Siemens CTI, Knoxville, TN, USA) were re-analyzed. [¹¹C]LSN3172176 was synthesized as previously described.¹⁴ Subjects were administered an i.v. bolus injection of [¹¹C]LSN3172176 (injected activity: 549.4 ± 153.2 MBq; molar activity: 244 ± 136 MBq/nmol), following which imaging data and arterial blood samples were acquired for 120 minutes. Metabolite analysis was performed to estimate the parent input functions using previously published methods.¹⁴ Kinetic modeling of dynamic data was performed using the 1TCM with metabolite-corrected plasma input function.¹⁴ Cerebellum was used as the reference region in the computation of DVR. A time-window of 60–80 min post injection was selected for SUVR correction, consistent with the simulations.

The correction formula was further evaluated with human data from an ongoing study with the [¹⁸F]FE-PE2I tracer, in separate cohorts of subjects diagnosed with Parkinson’s disease (PD) and healthy controls. Twenty PD subjects [mean age (SD): 63.4 (7.7) years, mean disease duration: 5.1 (3.3) years, Hoehn and Yahr Scale: 2 (all patients, mild bilateral disease), 10 F] and 12 healthy controls [mean age (SD): 58 (6.1) years, 6 F] underwent a 60 min scan on the HRRT after a bolus injection of [¹⁸F]FE-PE2I (injected activity: 151 ± 33 MBq, molar activity: 164 ± 110 MBq/nmol). Kinetic modeling was performed using the simplified reference tissue model (SRTM) with cerebellum as the reference region.^15,16 A late time-window of 40–60 min post injection was selected for correcting [¹⁸F]FE-PE2I SUVR since only 60 min of dynamic data were collected.

All studies operated under protocols approved by the Yale University Human Investigation Committee and the Yale University Radiation Safety Committee in accordane with the United States federal policy for the protection of human research subjets in Title 45 Part 46 of the Code of Federal Regulations (45 CFR 46). Written informed consent was obtained from all study participants after complete explanation of the procedures.

A linear-regression estimator for $β_{tar}$

Due to the high sensitivity of the SUVR_C formula to errors in $β_{T}$ (see Results: Errors in $β_{ref}$ and $β_{tar}$ ), direct estimation of $β_{tar}$ from noisy PET data acquired during a short time window was not a practical approach. A novel linear regression-based estimator of $β_{tar}$ based on the clearance rate from a composite region was proposed as a solution. The form of the estimator underscores the inverse relationship between the local clearance rate of the tracer from a tissue and its total binding in SUVR units, i.e., regions with higher SUVR than the composite region tend to have lower clearance rates than the composite region. The estimator takes the following form,

β_{tar} = β_{CR} + a (\frac{1}{{SUVR}_{tar}} - \frac{1}{{SUVR}_{CR}})

(16)

where subscript ‘tar’ and ‘CR’ refer to the target region and a composite region, respectively, and a is a positive constant (unit: min⁻¹) that depends on the radiotracer and the SUVR time-window. For a given tracer, this estimator leverages a composite region where the noise is sufficiently low to improve the accuracy of β_CR estimation from a short time window of PET data. β_CR can then be estimated with a single exponential decay fit to the composite region's time activity curve during the SUVR time window.

Determination of the regression slope (a)

A cross-validation approach was adopted to determine a for [¹¹C]LSN3172176. A training dataset (n = 15) was used to estimate the regression slope in the estimator of $β_{tar}$ for a different testing dataset (n = 15). Nine brain regions (thalamus, hippocampus, caudate, putamen, ventral striatum, occipital, parietal, frontal and temporal lobes) were selected for primary analysis. The temporal cortex was used as a composite region for [¹¹C]LSN3172176 due to its large volume (low noise in β_CR estimation) and sufficiently high uptake (DVR: 5.2 ± 0.7). For each ROI, an accurate estimate of β_tar (clearance rate during 60–80 minutes) was obtained from the low-noise 1TCM fit to the time activity curve for that ROI (0–120 min), using the subject’s metabolite-corrected plasma input function. β_CR, ${SUVR}_{CR}$ and ${SUVR}_{tar}$ were estimated directly from PET data between 60–80 minutes. These variables for the training set were then fit to the linear regression estimator of β_tar (equation (16)) and the least-squares estimate for the regression slope ( a ) was obtained. This value for a was then used to perform corrections in the testing dataset.

A leave-one-out cross-validation approach was adopted to calibrate a for [¹⁸F]FE-PE2I due to a smaller cohort of scans in healthy controls (n = 12). Four brain regions (putamen, caudate, ventral striatum and substantia nigra) were selected for primary analysis. The striatum (defined as combination of the caudate and putamen) was selected as the composite region for [¹⁸F]FE-PE2I. For every healthy control subject where regional SUVR values were corrected (S_i), regional data from all other subjects (S_j, j ≠ i) was used to define the training set. For each ROI, an accurate estimate of β_tar (clearance rate during 40–60 min) was obtained from the low-noise SRTM fit to the complete time activity curve (0–60 min), with cerebellum as the reference region. β_CR, ${SUVR}_{CR}$ and ${SUVR}_{tar}$ were estimated directly from PET data between 40–60 minutes. These variables were then fit to the linear regression estimator of β_tar (equation (16)) and the regression slope ( a ) was estimated. This estimated value of a is then used in the correction formula for the remaining healthy control. The mean of these estimated values for a (n = 12) was used for performing SUVR correction in the PD cohort.

Sensitivity of the errors in DVR prediction to the choice of regression slope (a)

To characterize the effect of errors in a on bias and SD of the errors in DVR prediction by SUVR_C, a sensitivity analysis was performed for [¹¹C]LSN3172176 (test dataset) and [¹⁸F]FE-PE2I (PD cohort) by varying a in the range of [0, 0.075] min⁻¹. This range represents about [−100,150]% difference from the estimated values of a for each radiotracer. For each value of a , β_tar was estimated using the linear regression model (equation (16)) and this value of β_tar was then used in the correction formula (equation (15)). The bias and SD of bias in the prediction of DVR by SUVR_C were then computed for the different values of a across multiple ROIs for each radiotracer.

Correction of regional SUVR

For [¹¹C]LSN3172176, the bias in SUVR and SUVR_C compared to DVR were computed for all nine brain regions across 15 subjects in the testing dataset. DVR was calculated using the kinetic parameters from 1TCM fit to dynamic PET data (0–120 minutes) with cerebellum as the reference region. SUVR was computed based on 60–80 minutes of PET data. To calculate SUVR_C, $k_{2, ref}$ = 0.081 min⁻¹, R₁ = 1 $,$ β_ref was estimated using a single exponential fit to 60–80 minutes of cerebellar time activity curve, and β_tar was computed using the regression-model with a = 0.03 min⁻¹ (see Results: Determination of the regression slope) with temporal cortex as the composite region.

For [¹⁸F]FE-PE2I, ground truth regional DVR was calculated based on SRTM fit to dynamic PET data (60 min). SUVR was computed based on 40–60 minutes of PET data. To calculate SUVR_C, $k_{2, ref}$ = 0.10 min⁻¹ (mean SRTM $k_{2}^{'}$ across brain regions and subjects), R₁ = 1 was used, β_ref was estimated using a single exponential fit to 40–60 minutes of cerebellar time activity curve, and β_tar was computed using the regression-model with a based on the leave-one-out analysis in the control subjects and a = 0.026 min⁻¹ in the Parkinson’s disease cohort (see, Results: Determination of the regression slope). Striatum was selected as the composite region for both cohorts.

Comparison of bias in corrected SUVR and early SUVR in [¹⁸F]FE-PE2I

To further demonstrate the benefit of using corrected SUVR in clinical studies with [¹⁸F]FE-PE2I, regional bias in SUVR_C (calculated 40–60 min post injection) was compared to the bias in early SUVR (15–45 min post injection) in healthy controls and Parkinson’s disease patients. This time window for early SUVR had best agreement compared to DVR in previous studies.¹⁷

Voxel-wise correction of SUVR

The correction formula, when used with the regression-based estimator for β_tar, can be directly applied at the voxel level. Exploratory analysis was performed to assess voxel-wise performance of the correction for [¹¹C]LSN3172176 and [¹⁸F]FE-PE2I with composite regions of temporal cortex and striatum, respectively. All images were transformed to a standard template (AAL MNI 2 mm) ¹⁸ to facilitate composite and reference region extraction and comparison across subjects. The transformation first performed a linear registration between subject’s PET and MR images, followed by a non-linear registration of subject’s structural MR image (3 T image acquired on Siemens Prisma) to a T1 in template space using previously published algorithms.¹⁹ For each tracer, we first identified two subjects for whom the correction formula resulted in the best and worst performance across primary ROIs, quantified by the root mean square error (RMSE) across all primary ROIs. Voxel SUVR values (60–80 min p.i. for [¹¹C]LSN3172176, 40–60 min p.i. for [¹⁸F]FE-PE2I) for these two subjects were corrected to SUVR_C ( a = 0.03 min⁻¹ for [¹¹C]LSN3172176, a = 0.026 min⁻¹ for [¹⁸F]FE-PE2I). For [¹⁸F]FE-PE2I, this analysis was performed in the control cohort due to the higher range of DVR in controls for this radiotracer. The SUVR and SUVR_C images were then compared to the corresponding DVR images for each subject. For [¹¹C]LSN3172176, DVR images were generated from voxel-level 1TCM fit to 120 min dynamic PET data, using metabolite-corrected arterial plasma activity as input function (ref = cerebellum). [¹⁸F]FE-PE2I DVR images were generated by voxel level SRTM fits to 60 min dynamic PET data (ref = cerebellum).

Diagnostic performance of corrected SUVR in Parkinson’s disease with [¹⁸F]FE-PE2I

The diagnostic performance of SUVR, SUVR_C and DVR in differentiating Parkinson’s disease subjects from healthy controls were compared by calculating the Cohen’s d measure for group differences in the primary brain regions for [¹⁸F]FE-PE2I. Recent studies have shown that early SUVR (15–45 min) provides better agreement with DVR for [¹⁸F]FE-PE2I,¹⁷ and better test-retest reliability and longitudinal assessment.²⁰ Therefore, early SUVR was also included alongwith late-time SUVR (40–60 min) in this analysis.

Results

Bias characterization

Simulations based on [¹¹C]LSN3172176 kinetics indicated that SUVR was positively biased for all simulated regions and plasma clearance rates (see Figure 1(a)). On average (SD), SUVR overestimated DVR by 51.6% (22.4%). Recall, that β in this figure represents the terminal clearance rate of the tracer in plasma (rate constant associated with the slowest exponent in the input functions). Increasing β, however, also increases the tracer clearance rate from the tissue (β_tar) during the SUVR time window. Therefore, SUVR bias increases as β_tar becomes more positive. SUVR_C reduced this bias to less than 1% [average bias (SD): 0.46% (0.13%); see Figure 1(b)] in all simulated regions and for all values of β (note the different display scale for SUVR_C).

Figure 1.

Heatmaps characterizing the bias in SUVR (a) and SUVR_C (b) during 60–80 min time window post injection, compared to DVR, in simulations for [¹¹C]LSN3172176 across different specific binding levels and tracer clearance rates in plasma (β). β refers to the terminal clearance of the tracer from plasma at late time (rate constant for the slowest exponential term). Color represents the percent bias in SUVR and SUVR_C compared to DVR. Note the difference in color scales.

Errors in fixed parameters: R₁ and k_2,ref

Simulations examining the effects of errors in R₁ on SUVR_C bias revealed a modest effect (Figure 2(a)), with the bias exceeding ten percent (range for bias: [−10.4 12.3]%) only when R₁ was sufficiently (∼25%) different from 1. These findings did not change appreciably when the simulations were repeated with different input functions (data not shown). Similarly, errors in k_2,ref implementation (∼25%) resulted in only a small bias in SUVR_C (range for bias: [−8.3 6.5]%) for most k_2,ref values (see Figure 2(b)).

Figure 2.

Heatmaps characterizing the robustness of the correction formula to errors in R₁ and k_2,ref for [¹¹C]LSN3172176. Color represents the percent bias in SUVR_C, compared to DVR. a: Bias in SUVR_C when the implemented value of R₁ = 1 in the correction is different from the simulated R₁ (simulated by varying regional K₁ across all DVR values). b: Bias in SUVR_C across all DVR values (y-axis) when the implemented value of k_2,ref = 0.081 min⁻¹ in the correction is different from the simulated k_2,ref values. The bias in SUVR_C is generally small and exceeds 10% only when the error in the simulated values of R₁ and k_2,ref is large (∼ 25%).

Variability in estimated parameters: $β_{ref}$ and $β_{tar}$

Errors in $β_{ref}$ resulted in a negligible effect on the performance of the correction (see Figure 3(a)), with a maximum bias of only about 5% in SUVR_C. On the other hand, errors in estimation of $β_{tar}$ caused bias in SUVR_C that was unacceptably large in regions with high binding (caudate and putamen), even for 25 to 50% departures from the true $β_{tar}$ value (see Figure 3(b)). This observation motivated the development of a linear regression-based estimator of $β_{tar}$ for implementing the correction.

Figure 3.

Heatmap characterizing the robustness of the correction to errors in $β_{ref}$ and $β_{T}$ for [¹¹C]LSN3172176. a: Less than 5% bias in SUVR_C was observed for a wide range of implemented values of $β_{ref}$ in the correction formula (x-axis), across ROIs (y-axis). b: Heatmap demonstrating the high sensitivity of the correction formula to errors in $β_{T}$ estimation in the putamen, caudate and frontal lobe (DVR = 9.0, 7.2 and 6.2, respectively) for [¹¹C]LSN3172176. The coefficient of variation (CV) for each ROI was varied by [−125 125]%, which was decided based on the worst (highest) CV in $β_{T}$ estimates in these three ROIs from 60–80 min of human PET data. Color represents the percent bias in SUVR_C, compared to DVR.

Determination of the regression slope (a)

The regression slope of the $β_{tar}$ estimator ( a ) was estimated from training sets for each radiotracer. For [¹¹C]LSN3172176, the mean (s.e.) value of a was 0.03 ± 0.003 min⁻¹ (Figure S2). For [¹⁸F]FE-PE2I, a leave-one out approach yielded a values of 0.026 ± 0.001 min⁻¹ for healthy controls (See Figure S3). This mean value of a = 0.026 min⁻¹ was also used to estimate β_tar for SUVR correction in the PD cohort.

Sensitivity of the errors in DVR prediction to the choice of regression slope (a)

Least-squares estimation of a (0.03 min⁻¹ for [¹¹C]LSN3172176, 0.026 min⁻¹ for [¹⁸F]FE-PE2I) led to less than 10% bias in corrected SUVR, and less than 10% variance in the bias of corrected SUVR, consistently across different ROIs, for either tracer. The performance of the correction formula was similar (less sensitive) in the range of a = [0.02, 0.04] min⁻¹ for both tracers. Notably, the standard deviation in the bias across subjects increased considerably at higher values of a for all ROIs. Figure S4 shows the trajectories for mean bias and standard deviation of the bias in SUVR and SUVR_C, as a function of a , for both [¹¹C]LSN3172176 (test dataset) and the PD cohort of [¹⁸F]FE-PE2I.

Correction of regional SUVR

[¹¹C]LSN3172176 SUVR overestimated the true DVR across different brain regions in the testing dataset (mean bias: 24.6%, range: [9.6, 35.4]%; See Figure 4). With the exception of the ventral striatum (the region with highest binding), SUVR_C reduced the bias across all regions (mean bias: −1.7%, range: [−16.0, 8.0]%). Reduction in the bias magnitude was statistically significant for the caudate and putamen (p < 0.01), brain regions in the cerebral cortex (p < 10⁻⁴) and the hippocampus (p < 10⁻⁵) using paired t-tests (α = 0.05, two-tailed). SUVR_C also reduced the standard deviation of the bias from an average value of 15.3% (range: [10.5, 17.6]%) to 6.9% (range: [6.1,8.3] %) across ROIs (note error bars in Figure 4). Bland-Altman plot (Figure S5) confirms better agreement of SUVR_C with DVR (mean difference: −0.2, range:[−1.5 1.2]) compared to SUVR (mean difference: 1.3, range:[−0.8 3.4]) in the test cohort. The highest differences between SUVR_C and DVR were observed in the ventral striatum.

Figure 4.

Performance of the correction formula (SUVR_C) in reducing the bias and variance in DVR estimation in different brain regions for [¹¹C]LSN3172176 compared to raw SUVR values calculated between 60–80 min post injection. Results reported across n = 15 subjects (testing dataset). VS = Ventral Striatum.

For [¹⁸F]FE-PE2I evaluated in the control and PD cohorts, SUVR overestimated the true DVR across all four brain regions (mean bias [range], HC: 28.1% [26.4, 30.4]%; PD: 25.1% [19.2, 29.9]%; see Figure 5). SUVR_C reduced this bias in all brain regions in both samples (mean bias in HC: 3.0%, range: [−1.2, 8.8]%; mean bias in PD: 2.7%, range: [−0.3, 7.3]%). Reduction in the bias magnitude (absolute value of bias) was statistically significant for all brain regions and across disease states using paired t-tests (p < 10⁻⁵, α = 0.05, two-tailed). The standard deviation of the bias in SUVR was, on average, 9.5% of DVR (range: [7.3, 11.3]%) in the healthy controls, and 8.0% (range: [6.9, 9.3]%) in PD patients. SUVR_C, on average, reduced this variability to 8.4% (range: [8.2, 11.1]%) in HCs, and 6.3% (range: [5.4, 6.8]%) in PD patients. Bland-Altman plot (Figure S6) confirms better agreement of SUVR_C with DVR (mean difference: −0.04, range:[−0.2 0.3]) compared to SUVR (mean difference: 0.6, range:[−0.1 1.2]) in subjects with PD. The difference between SUVR_C and DVR was similar in all regions.

Figure 5.

Performance of the correction formula (SUVR_C) in reducing the bias and variance of bias in DVR estimation in different brain regions for [¹⁸F]FE-PE2I compared to raw SUVR values calculated between 40–60 min post injection. Results reported separately for 12 healthy controls (a) and 20 patients diagnosed with Parkinson’s disease (b). SN = Substantia Nigra, VS = Ventral Striatum.

Comparison of bias in corrected SUVR and early SUVR in [¹⁸F]FE-PE2I

A lower bias was observed for SUVR_C across brain regions in each cohort (mean bias [range], HC: 3.1% [−1.1, 8.7]%; PD: 2.1% [−0.9, 6.4]%; see Figure S7), compared to early SUVR (mean bias [range], HC: −7.8% [−14.4, 6.6]%; PD: 4.7% [−1.0, 9.9]%; see Figure S7). Additionally, the standard deviation of the regional bias was also smaller for SUVR_C (mean [range], HC: 8.7% [5.9, 11.6]%; PD: 6.2% [5.5, 7.1]%; see error bars, Figure S7) compared to SUVR (mean [range], HC: 9.7% [7.6, 11.2]%; PD: 7.5% [5.9, 8.8]%; see error bars, Figure S7).

Voxel-Wise correction of SUVR

Figure 6 illustrates SUVR, SUVR_C, and true DVR images for [¹¹C]LSN3172176 (A-B) and [¹⁸F]FE-PE2I (C-D) in subjects where the regional performance of the correction formula was best (A, C), i.e., smallest error with respect to DVR, and worst (B, D), respectively. Qualitatively, SUVR_C images matched better with the DVR images. SUVR tends to overestimate DVR in most relevant brain voxels. Scatter plots comparing voxel SUVR and SUVR_C values with the true underlying DVR values confirm that on average, SUVR_C reduced the mean bias in SUVR, and standard deviation of the bias, across all voxels (Figure S8). Even in subjects with the worst performance (B, D), SUVR_C performed either better than ([¹¹C]LSN3172176), or similar to ([¹⁸F]FE-PE2I), using raw SUVR values.

Figure 6.

(a, b) Images comparing [¹¹C]LSN3172176 DVR (1TCM) with raw SUVR and SUVR_C images for the subjects with best (A) and worst (B) performance of the correction formula based on average root mean square errors in the estimation of regional DVR by SUVR_C and (c, d) Images comparing [¹⁸F]FE-PE2I DVR (SRTM) with raw SUVR and SUVR_C images for the healthy controls with best (C) and worst (D) performance of the correction formula based on average root mean square errors in the estimation of regional DVR by SUVR_C. SUVR and SUVR_C are calculated from only 20 min of PET data, 60–80 min post-injection for [¹¹C]LSN3172176 and 40–60 min post injection for [¹⁸F]FE-PE2I.

Diagnostic performance of corrected SUVR in Parkinson’s disease with [¹⁸F]FE-PE2I

Table 1 compares the effect sizes (Cohen’s d) for group differences between a cross-sectional cohort of Parkinson’s disease subjects (n = 20) and healthy controls (n = 12) using early SUVR, SUVR, SUVR_C and DVR. The performance of SUVR_C was similar to DVR in the ventral striatum and caudate, and similar to SUVR in the putamen. In substantia nigra, the Cohen’s d value for SUVR_C was inflated compared to DVR, likely due to the relatively higher bias of SUVR_C in controls compared to PD subjects (Figure 5). Early SUVR had the lowest effect sizes in all regions. Further analyses performed in test-retest data and longitudinal follow-up of PD subjects would strengthen this initial evaluation of SUVR_C compared to SUVR and DVR.

Table 1.

Effect size comparison [¹⁸F]FE-PE2I PET outcomes.

Region	Cohen’s dSUVR_15–45	Cohen’s dSUVR_40–60	Cohen’s dSUVR_c,40–60	Cohen’s dDVR
Substantia nigra	1.03	1.78	1.93	1.58
Ventral Striatum	1.11	1.74	1.97	2.14
Caudate	1.51	2.11	2.35	2.25
Putamen	2.69	3.76	3.72	4.11

Comparing the Cohen’s d measure of effect size between 12 healthy controls and 20 Parkinson’s disease patients for early SUVR (15–45 min), SUVR (40–60 min), SUVR_C (40–60 min) and DVR (0–60 min).

Discussion

This study presents an improved estimation of SUVR from a short time window of PET data based on a one-step non-linear algebraic transform. This approach corrects for non-equilibrium effects of tracer clearance that can introduce substantial bias in SUVR compared to ground-truth DVR. While the correction was strictly derived for a radiotracer following 1TCM kinetics ([¹¹C]LSN3172176), this work also demonstrates application to a radiotracer following 2TCM kinetics ([¹⁸F]FE-PE2I), correcting SUVR values to the DVR value calculated using SRTM. For both radiotracers this correction not only significantly reduced bias in SUVR estimates, but it also reduced population-level variability in the bias for most ROIs in our study population. In addition to regional analyses, initial results suggest that correcting SUVR can also provide superior quantification than SUVR in voxel-based pipelines.

There are important applications for more accurate and precise SUVR estimates that are enabled by this correction. Lower variability in the bias, in particular, improves accuracy of PET scans for early diagnosis,^21,22 and measuring longitudinal changes within a subject.²³ Additional scenarios where SUVR correction is important is when either the use of drugs or differential disease states affects radiotracer availability and metabolism. Changes in radiotracer clearance rate across subjects due to these factors could introduce differential biases in SUVR that can be minimized if SUVR for each subject is corrected for such effects. Additionally, this correction could be highly useful for radiotracers with large variation in kinetics across brain regions, such as [¹¹C]LSN3172176 (10-fold dynamic range in V_T). For such radiotracers, it is often implausible to find a time window during which regional SUVR values are close to the equilibrium DVR for all ROIs. Future studies could compare how raw and corrected SUVR values compare with DVR in clinical paradigms and with different tracers.

The validity of the correction was demonstrated in two radiotracers with established reference regions, i.e., reference regions with minimal displaceable binding. However, some radiotracers ([¹¹C]UCB-J, [¹⁸F]FPEB) make use of pseudo-reference region approaches, where the reference region may have a small amount of specific binding. SUVR correction could be applied to correct SUVR values calculated from a pseudo-reference region to their corresponding DVR values obtained via kinetic modeling using the same reference. It should be noted, however, that typical issues encountered when using DVR from an imperfect reference can also be expected for corrected SUVR in these cases.

This work characterized the correction method for two specific radiotracers, but the general roadmap presented here can be adapted for different radiotracers in the future. Simulations characterizing correction performance indicate low sensitivity to errors in the choice of R₁, k_2,ref and $β_{ref}$ . Thus, these choices can be based on previously published literature or on a small time-window of PET data. Assuming a population averaged value for k_2,ref is already a common practice in DVR measurement, particularly for reference region based methods.⁵ However, since the formula was sensitive to errors in $β_{tar}$ , a linear-regression based estimator was developed to improve $β_{tar}$ estimates based on a composite region from a short duration of PET data. Future translation of this method to each new tracer might involve a similar estimation of the radiotracer-specific regression slope ( a ). For this, complete dynamic PET data (and arterial blood samples if needed for the kinetic model) would ideally be acquired in a small cohort of subjects. A linear regression estimation of ( a ) (see equation (16)) can then be performed by pooling regional data from all subjects, and $β_{sim}$ can be computed during the SUVR time window using the low-noise kinetic model fit to regional time activity curves. Once estimated, we expect a to be constant as long as the SUVR time window and population demographics are consistent. Another important consideration is the identification of factors which govern the choice of a suitable composite region. Based on data shown here, it is likely that, (i) the composite region should have large enough volume (lower noise) to facilitate reliable estimation of the tracer’s clearance rate from a small window of PET data, and (ii) it should have moderate to high radiotracer uptake. Several comparable choices for the composite region may be possible. For example, replacing the temporal cortex with a different cortical lobe still led to a similar overall performance for the correction for [¹¹C]LSN3172176 (data not shown). The estimates of a using standard least-squares minimization resulted in values (0.03 min⁻¹ for [¹¹C]LSN3172176 SUVR [60–80 min], 0.026 min⁻¹ [¹⁸F]FE-PE2I SUVR [40–60 min]) that provided a reasonable tradeoff between reduction in the bias and variability of bias in SUVR_C (Figure S4)_. For each tracer and time window, a sensitivity analysis for performance across a range of values for a is recommended to fine-tune the results of the least-squares optimization. Finally, the choice of composite region and a could potentially vary between sub-populations (healthy controls vs diseased states). While this was not observed for [¹⁸F]FE-PE2I in a cohort of Parkinson’s disease patients, this possibility should be considered in future studies.

This approach was designed to facilitate analysis of PET data acquired during a pre-determined late time window. To assess the effects of timing for this window on SUVR correction, exploratory analysis was performed with [¹¹C]LSN3172176 human data for SUVR correction at different times. The window starting points were varied between 60–100 min, and the duration was maintained at 20 min. The value of the regression slope ( a ), estimated from the training dataset, decreased for later time windows (Figure S9). The estimates of a were reliable, indicated by a constant coefficient of variation of about 9% across time windows. These estimated a values were then used in the $β_{tar}$ estimator for performing SUVR correction in the testing cohort across the different time windows, respectively. For most ROIs (except ventral striatum), the average bias in corrected SUVR was within 15% of DVR, and the correction reduced the magnitude of the bias in SUVR values across these time windows (Figure S10). Thus, the correction demonstrated a good performance at several late time windows which are far from transient equilibrium. For future application of this method to a new tracer, the SUVR time window could be defined at a reasonable time post injection that ensures active clearance of the radiotracer from all ROIs, and a can be calibrated for that time-interval. A time dependent value of a could also be used for correcting SUVR bias in PET data with inter-subject differences in acquisition time, which if left uncorrected can lead to differential bias. We further speculate that the constant a should depend only on the radiotracer and SUVR time window, without additional variation across centers with different PET scanners. Contingent on validation from future multi-center studies, this method could be an important step in harmonizing the differential biases in SUVR data from different centers. Future work to this end will examine SUVR bias reduction in multi-center data across multiple time windows.

The SUVR correction introduced in this work could be improved in future studies. For example, instead of using a least-squares approach to estimate a , a loss function that better captures the trade-off between bias and variance in DVR estimation needed for a particular application could be minimized. Errors in the accurate estimation of $β_{tar}$ from the proposed linear regression-based estimator can lead to worse performance for corrected SUVR. For example, the higher magnitude of SUVR_C bias in the ventral striatum in [¹¹C]LSN3172176 human data was likely due to biased estimates of $β_{tar}$ in this region. Alternative approaches, such as a neural network based estimation of $β_{tar}$ , or a network to denoise PET data, could lead to more accurate estimation of $β_{tar}$ in future studies. Similarly, direct generation of DVR images from SUVR images using 3 D-UNet and other comparable networks could be investigated. Thus, this work is an important step in analytically correcting effects of non-equilibrium tracer clearance on SUVR estimates, providing a benchmark for comparison with future methods.

Conclusion

This study developed and validated a correction for SUVR that accounts for the non-equilibrium effects of tracer clearance and reduces both bias and variability in SUVR quantification. The parameters required for correction can be estimated from a combination of previous literature estimates and a short duration (∼20 min), dynamic, late time-window PET scan, and can be easily implemented for correcting regional or voxel-level SUVR estimates. This approach to correct SUVR is a promising improvement for semi-quantification of future brain PET imaging studies.

Supplemental Material

sj-pdf-1-jcb-10.1177_0271678X231196804 - Supplemental material for Improving SUVR quantification by correcting for radiotracer clearance in tissue

Supplemental material, sj-pdf-1-jcb-10.1177_0271678X231196804 for Improving SUVR quantification by correcting for radiotracer clearance in tissue by Praveen Honhar, David Matuskey, Richard E Carson, Ansel T Hillmer in Journal of Cerebral Blood Flow & Metabolism

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: These studies were supported by National Institutes of Health (NIH) grants (nos. K01AA024788 and 1R01NS124819) and AbbVie.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Authors’ contributions

PH, REC, ATH conceived and developed TCCr. PH performed simulations and data analysis. PH and ATH drafted the initial manuscript. PH, DM, REC, ATH reviewed and approved the final version of this manuscript.

ORCID iDs

Praveen Honhar

Ansel T Hillmer

Supplementary material

Supplemental material for this article is available online.

References

Innis

Cunningham

Delforge

et al. Consensus nomenclature for in vivo imaging of reversibly binding radioligands. J Cereb Blood Flow Metab 2007; 27: 1533–1539.

Gunn

and Cunningham

VJ.

Positron emission tomography compartmental models. J Cereb Blood Flow Metab 2001; 21: 635–652.

Carson

Channing

Blasberg

et al. Comparison of bolus and infusion methods for receptor quantitation: application to [18F] cyclofoxy and positron emission tomography. J Cereb Blood Flow Metab 1993; 13: 24–42.

McNamee

Yee

S-H

Price

et al. Consideration of optimal time window for Pittsburgh compound B PET summed uptake measurements. J Nucl Med 2009; 50: 348–355.

Naganawa

Gallezot

J-D

Finnema

et al. Simplified quantification of ¹¹C-UCB-J PET evaluated in a large human cohort. J Nucl Med 2021; 62: 418–421.

Ito

Hietala

Blomqvist

et al. Comparison of the transient equilibrium and continuous infusion method for quantitative PET analysis of [11C] raclopride binding. J Cereb Blood Flow Metab 1998; 18: 941–950.

Nyberg

Farde

and Halldin

Test-retest reliability of Central [11C] raclopride binding at high D2 receptor occupancy. A PET study in haloperidol-treated patients. Psychiatry Res 1996; 67: 163–171.

Pappata

Samson

Chavoix

et al. Regional specific binding of [11C] RO 15 1788 to Central type benzodiazepine receptors in human brain: quantitative evaluation by PET. J Cereb Blood Flow Metab 1988; 8: 304–313.

Finnema

Halldin

Bang-Andersen

et al. Serotonin transporter occupancy by escitalopram and citalopram in the non-human primate brain: a [(11)C]MADAM PET study. Psychopharmacology (Berl) 2015; 232: 4159–4167.

10.

Heeman

Yaqub

Hendriks

et al. Impact of cerebral blood flow and amyloid load on SUVR bias. EJNMMI Res 2022; 12: 29–10.

11.

Haass-Koffler

Goodyear

Loche

et al. Administration of the metabotropic glutamate receptor subtype 5 allosteric modulator GET 73 with alcohol: a translational study in rats and humans. J Psychopharmacol 2018; 32: 163–173.

12.

Hillmer

Angarita

Esterlis

et al. Longitudinal imaging of metabotropic glutamate 5 receptors during early and extended alcohol abstinence. Neuropsychopharmacology 2021; 46: 380–385.

13.

Hillmer

and Carson

RE.

Quantification of PET infusion studies without true equilibrium: a tissue clearance correction. J Cereb Blood Flow Metab 2020; 40: 860–874.

14.

Naganawa

Nabulsi

Henry

et al. First-in-human assessment of (11)C-LSN3172176, an M1 muscarinic acetylcholine receptor PET radiotracer. J Nucl Med 2021; 62: 553–560.

15.

Sasaki

Ito

Kimura

et al. Quantification of dopamine transporter in human brain using PET with 18F-FE-PE2I. J Nucl Med 2012; 53: 1065–1073.

16.

Fazio

Svenningsson

Forsberg

et al. Quantitative analysis of (1)(8)F-(E)-N-(3-iodoprop-2-enyl)-2beta-carbofluoroethoxy-3beta-(4'-methyl-phenyl) nortropane binding to the dopamine transporter in Parkinson disease. J Nucl Med 2015; 56: 714–720.

17.

Sonni

Fazio

Schain

et al. Optimal acquisition time window and simplified quantification of dopamine transporter availability using 18F-FE-PE2I in healthy controls and Parkinson disease patients. J Nucl Med 2016; 57: 1529–1534.

18.

Tzourio-Mazoyer

Landeau

Papathanassiou

et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 2002; 15: 273–289.

19.

Papademetris

Jackowski

Schultz

et al. Integrated intensity and point-feature nonrigid registration. Med Image Comput Comput Assist Interv 2001; 3216: 763–770.

20.

Brumberg

Kerstens

Cselenyi

et al. Simplified quantification of [(18)F]FE-PE2I PET in Parkinson's disease: discriminative power, test-retest reliability and longitudinal validity during early peak and late pseudo-equilibrium. J Cereb Blood Flow Metab 2021; 41: 1291–1300.

21.

Lopes Alves

Heeman

Collij

et al. Strategies to reduce sample sizes in Alzheimer's disease primary and secondary prevention trials using longitudinal amyloid PET imaging. Alzheimers Res Ther 2021; 13: 82–20210419.

22.

Lammertsma

AA.

Forward to the past: the case for quantitative PET imaging. J Nucl Med 2017; 58: 1019–1024.

23.

van Berckel

Ossenkoppele

Tolboom

et al. Longitudinal amyloid imaging using 11C-PiB: methodologic considerations. J Nucl Med 2013; 54: 1570–1576.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.09 MB

Improving SUVR quantification by correcting for radiotracer clearance in tissue

Abstract

Abstract

Keywords

Introduction

Theory

Part 1: Correction of bias in apparent VT under non-equilibrium radiotracer clearance

Part 2: Correction of bias in SUVR values under non-equilibrium radiotracer clearance

Materials and methods

Simulations

Bias characterization

Bias in fixed parameters: R1 and k2,ref

Variability in estimated parameters: β ref and β tar

Human data

A linear-regression estimator for β tar

Determination of the regression slope (a)

Sensitivity of the errors in DVR prediction to the choice of regression slope (a)

Correction of regional SUVR

Comparison of bias in corrected SUVR and early SUVR in [18F]FE-PE2I

Voxel-wise correction of SUVR

Diagnostic performance of corrected SUVR in Parkinson’s disease with [18F]FE-PE2I

Results

Bias characterization

Errors in fixed parameters: R1 and k2,ref

Variability in estimated parameters: β ref and β tar

Determination of the regression slope (a)

Sensitivity of the errors in DVR prediction to the choice of regression slope (a)

Correction of regional SUVR

Comparison of bias in corrected SUVR and early SUVR in [18F]FE-PE2I

Voxel-Wise correction of SUVR

Diagnostic performance of corrected SUVR in Parkinson’s disease with [18F]FE-PE2I

Discussion

Conclusion

Supplemental Material

sj-pdf-1-jcb-10.1177_0271678X231196804 - Supplemental material for Improving SUVR quantification by correcting for radiotracer clearance in tissue

Footnotes

Funding

Declaration of conflicting interests

Authors’ contributions

ORCID iDs

Supplementary material

References

Supplementary Material

Part 1: Correction of bias in apparent V_T under non-equilibrium radiotracer clearance

Bias in fixed parameters: R₁ and k_2,ref

Variability in estimated parameters: $β_{ref}$ and $β_{tar}$

A linear-regression estimator for $β_{tar}$

Comparison of bias in corrected SUVR and early SUVR in [¹⁸F]FE-PE2I

Diagnostic performance of corrected SUVR in Parkinson’s disease with [¹⁸F]FE-PE2I

Errors in fixed parameters: R₁ and k_2,ref

Variability in estimated parameters: $β_{ref}$ and $β_{tar}$

Comparison of bias in corrected SUVR and early SUVR in [¹⁸F]FE-PE2I

Diagnostic performance of corrected SUVR in Parkinson’s disease with [¹⁸F]FE-PE2I