- Email: [email protected]
Validation of parametric methods for [11C]PE2I positron emission tomography
NeuroImage 74 (2013) 172–178
Contents lists available at SciVerse ScienceDirect
NeuroImage journal homepage: www.elsevier.com/locate/ynimg
Validation of parametric methods for [ 11C]PE2I positron emission tomography My Jonasson a, d,⁎, Lieuwe Appel a, d, Jonas Engman b, Andreas Frick b, Dag Nyholm c, Håkan Askmark c, Torsten Danfors a, c, d, Jens Sörensen a, d, Tomas Furmark b, Mark Lubberink a, e a
Nuclear Medicine & PET, Uppsala University, 751 81 Uppsala, Sweden Psychology, Uppsala University, 751 81 Uppsala, Sweden Neuroscience, Neurology, Uppsala University, 751 81 Uppsala, Sweden d Molecular Imaging, Uppsala University Hospital, 751 81 Uppsala, Sweden e Medical Physics, Uppsala University Hospital, 751 81 Uppsala, Sweden b c
a r t i c l e
i n f o
Article history: Accepted 11 February 2013 Available online 19 February 2013 Keywords: PET [11C]PE2I Kinetic modeling Dopamine transporter Reference tissue model Parametric images
a b s t r a c t Objectives: The radioligand [11C]PE2I is highly selective for dopamine transporter (DAT) and can be used in vivo for investigation of changes in DAT concentration, progression of disease and validation of treatment using positron emission tomography (PET). DAT is an important protein for regulation of central dopamine concentration and DAT deficiency has been associated with several neurodegenerative and neuropsychiatric disorders. Accurate parametric images are a prerequisite for clinical application of [ 11C]PE2I. The purpose of this study was to evaluate different methods for producing [11C]PE2I parametric images, showing binding potential (BPND) and relative delivery (R1) at the voxel level, using clinical data as well as simulations. Methods: Investigations were made in twelve subjects either with social anxiety disorder (n=6) or parkinsonian syndrome (n=6), each receiving an 80 min dynamic PET scan. All subjects underwent a T1-weighted MRI scan which was co-registered to the PET images and used for definition of regions of interest using a probabilistic template (PVElab). Two basis function implementations (receptor parametric mapping: RPM, RPM2) of the simplified reference tissue model (SRTM) and three multilinear reference tissue models (MRTMo, MRTM and MRTM2) were used for computation of parametric BPND and R1 images. In addition, reference Logan and standard uptake value ratio (SUVr) were investigated. Evaluations of BPND and R1 images were performed using linear regression to compare the parametric methods to region-based analyses with SRTM and cerebellar gray matter as reference region. Accuracy and precision of each method were assessed by simulations. Results: Correlation and slope of linear regression between parametric and region-based BPND and R1 values in both striatum and extra-striatal regions were optimal for RPM (R2 =0.99 for both BPND and R1; slopes 0.99 and 0.98 for BPND and R1, respectively, in striatum). In addition, accuracy and precision were best for RPM and RPM2. Conclusion: The basis function methods provided more robust estimations of the parameters compared to the other models and performed best in simulations. RPM, a basis function implementation of SRTM, is the preferred method for voxel level analysis of [11C]PE2I PET studies. © 2013 Elsevier Inc. All rights reserved.
Introduction The dopamine transporter (DAT) is a transmembrane protein located on the presynaptic neurons responsible for re-uptake and removal of dopamine from the synaptic cleft. This process regulates the action of dopamine and has shown to be of interest in several physiological functions (Gulley and Zahniser, 2003). DAT is associated with different neurodegenerative and neuropsychiatric disorders such as Parkinson's disease (Antonini et al., 2001; Ribeiro et al., 2002), schizophrenia (Laakso et al., 2001), attention-deficit/hyperactive disorder (ADHD) (Jucaite et al., 2005; Spencer et al., 2005) and social anxiety disorders (SAD) (Warwick et al., 2012). ⁎ Corresponding author at: Nuclear Medicine & PET, Uppsala University, 751 81 Uppsala, Sweden. E-mail address: [email protected] (M. Jonasson). 1053-8119/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.neuroimage.2013.02.022
Positron emission tomography (PET) is a non-invasive molecular imaging method for diagnosis and monitoring treatment effects in-vivo. PET provides functional images of the body and has proven to be a powerful tool for neurotransmission studies including imaging of DAT. The cocaine analog N-(3-iodoprop-2E-enyl)-2βcarbomethoxy-3β-(4-methyl-phenyl)nortropane (PE2I) has demonstrated an excellent binding to DAT, and both in vitro autoradiography studies using [ 125I]PE2I (Hall et al., 1999), and in vivo PET studies with [ 11C]PE2I (DeLorenzo et al., 2009; Hirvonen et al., 2008; Jucaite et al., 2006; Seki et al., 2010) have shown marked accumulation in the striatum where concentration of DAT is high. An intermediate binding has been found in the midbrain and no, or very little, specific binding in the cerebellum. Previous studies have shown that PE2I has a high selectivity for DAT, and because of that, [ 11C]PE2I provides high contrast PET images and appears to be sufficiently sensitive for detection of DAT in small extrastriatal brain areas (Emond et al., 2008). [ 11C]PE2I has also
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
shown a good reproducibility and reliability in test–retest PET studies (Hirvonen et al., 2008). [ 11C]PE2I PET studies can thus be a valuable tool for early detection of neurological disorders where changes in DAT concentration are involved, for the assessment of disease progression, and for validation of treatments (Emond et al., 2008). Applying a simplified reference tissue model (SRTM) (Lammertsma and Hume, 1996) on regions of interest (ROIs), using cerebellum as reference region, has proven to be a robust method for parameter estimation (Hirvonen et al., 2008; Jucaite et al., 2006), showing a good correlation with compartmental modeling using a plasma input (DeLorenzo et al., 2009; Seki et al., 2010). To further facilitate clinical use of [ 11C]PE2I, the possibility of straightforward visual assessment of DAT availability is a prerequisite. To this end, parametric images have to be available, where the parameters of interest are estimated for each voxel. In addition to binding potential (BPND) images, showing DAT availability, images of relative delivery (R1), reflecting regional cerebral blood flow (rCBF) may be of interest in the differential diagnosis of neurodegenerative diseases (Meyer et al., 2011). For a voxel based analysis, using standard non-linear least square fitting for estimation of the parameters is sensitive to noise and very time consuming. Parametric methods, utilizing linearization of compartment models, provide faster analyses on a voxel level than non-linear regression analysis (Gunn et al., 1997). Voxel-based analysis of BPND has been performed previously for [ 11C]PE2I (Leroy et al., 2012; Odano et al., 2012; Ribeiro et al., 2009), but no extensive validation of parametric methods has been presented for this tracer. The purpose of this study was to evaluate different methods for producing [ 11C]PE2I parametric images, showing BPND and R1 at the voxel level, using clinical data as well as simulations. Methods Subjects and data acquisition Data from a total of twelve subjects were included in this study whereof six were SAD subjects (three women and three men, mean age 35 ±11 years) and six were subjects with parkinsonian syndrome (five women and one man, mean age 68± 6 years). Each subject signed a written informed consent and the study was approved by the local independent ethics and radiation safety committees. In addition, the scans of subjects with SAD were part of a clinical trial that was approved by the Swedish medical products agency. Together, these two groups are expected to display a large range of DAT availability in the striatum. Each subject underwent a dynamic PET scan on an ECAT Exact HR+ scanner (Siemens/CTI, Knoxville). After a 10 min transmission scan for attenuation correction, an 80 min emission scan in 3-dimensional acquisition mode was started simultaneously with the injection of about 350 MBq [ 11C]PE2I. Twenty-two frames of increasing durations (4× 1, 2 × 2, 4 × 3, 12 × 5 min) were acquired. Dynamic images were reconstructed using ordered subset expectation maximization (OSEM) with 6 iterations and 8 subsets and a 4 mm Hanning post-filter, applying all appropriate corrections. In addition, each subject underwent a T1-weighted magnetic resonance image (MRI) scan (3D-SENSE) on a 3 T Achieva scanner (Philips Healthcare, Best, The Netherlands). [ 11C]PE2I was synthesized based on previously described methods (Halldin et al., 2003). ROI analysis Dynamic PET images were realigned to correct for inter-frame patient motion using Voiager (GE Healthcare, Uppsala, Sweden). MRI images were co-registered to the sum of the first 3 min of the PET scan, mostly resembling a blood flow image, using statistical parametric mapping (SPM5; Wellcome Trust Center for Neuroimaging, University College London, UK). Gray matter ROIs were defined on
173
co-registered MRI images using an automated probabilistic ROI template as implemented in the PVElab software (Svarer et al., 2005). Seven ROIs were included: the putamen, caudate, thalamus, midbrain, hypothalamus, amygdala and hippocampus, averaged over the left and the right hemisphere. ROIs were transferred to the dynamic PET images, yielding regional time-activity curves (TACs). Data were analyzed using SRTM with the cerebellar gray matter as reference tissue, since the cerebellum is a region known for low DAT concentration and thus lacks specific binding of PE2I (Halldin et al., 2003). Regions with standard error (SE) >25% for BPND estimates were excluded from further analysis. Parametric images For evaluation of parametric images, voxel level analyses of clinical data were performed using two receptor parametric mapping methods RPM (Gunn et al., 1997) and RPM2 (Wu and Carson, 2002), three multilinear reference tissue models MRTMo, MRTM and MRTM2 (Ichise et al., 1996, 2003), reference Logan analysis (Logan et al., 1996) and standard uptake value ratio (SUVr). In all analyses the cerebellum was used as reference region. The three and two parameter versions of receptor parametric mapping (RPM and RPM2) are implementations of SRTM and SRTM2 (Wu and Carson, 2002) using a set of predefined basis functions to linearize the models. One hundred basis functions were predefined for each scan with a discrete set of values for the exponential variable ranging from 0.01 to 0.5 min −1. For RPM2, the estimated number of parameters was reduced to two by setting the efflux rate constant from reference tissue, k2′ (min −1), to a constant value. This was performed by taking a volume-weighted mean value for k2′ including the putamen, caudate, thalamus, midbrain, hypothalamus, amygdala and hippocampus from the previous RPM analysis. The multilinear reference tissue models (MRTMo, MRTM and MRTM2) estimate the parameters using multilinear regression after a certain equilibrium time that for the present analyses was set to 30 min. For MRTM2, the parameters were reduced the same way as described above for RPM2 but with the k2′ value estimated from previous MRTM analysis. The estimated parameters for these five methods were thus R1 and BPND for each voxel. The reference Logan model is a graphical method and the analysis was performed with a time interval of 30–80 min for which BPND was indirectly estimated as the distribution volume ratio DVR-1. Additionally, the standardized uptake value ratio (SUVr) was estimated for the last four frames, 60–80 min, and BPND was estimated as SUVr-1. No estimates of R1 can be obtained by reference Logan or SUVr. To take into account the different counts in each frame, weights were included in the data analysis. The corresponding weighting factor for each frame was calculated as
w¼
Δt 2 f 2T
where Δt is the frame duration, T is the total counts in the frame and f is the decay correction factor for the frame which is computed as follows,
f ¼
λΔt expð−λt s Þ− expð−λt e Þ
where ts and te are the start and the end time of the frame and λ is the decay constant for 11C.
174
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
Fig. 1. [11C]PE2I parametric images of (a) binding potential, BPND, and (b) relative tracer delivery, R1, from a subject with SAD.
Quantitative evaluation of parametric images A quantitative evaluation of each parametric method on clinical data was performed by retrieving regionally averaged voxel values from the generated BPND and R1 parametric images by projecting the set of ROIs as described previously. Square of the correlation coefficient (R 2) and slope of a linear regression of ROI values between the different parametric methods and SRTM were calculated for the subjects. In addition, R 2 and slope were calculated separately for two substantial ROIs, the striatum and thalamus, representing regions with high and low uptake of [ 11C]PE2I. Accuracy and precision For assessment of accuracy and precision of BPND and R1 estimates, simulations were performed for all seven parametric methods as well as SRTM. TACs were created using a two tissue compartment model (2TCM) with plasma input curve and parameters from literature (Jucaite et al., 2006; Seki et al., 2010). For each set of parameters, 100 noisy TACs were generated. The noise was normally distributed with a mean value of zero and a scale factor to determine the level of noise according to a previously described method (Ichise et al., 2003). The noise level was set to 2% or 10% to resemble noise in a typical striatum ROI-TAC or voxel-TAC, respectively. Two set of rate constants were used to reflect both the high and the low DAT availability; K1 = 0.3, k2 = 0.15, k3 = 0.35 and 0.18 and k4 = 0.025 and 0.06, together with a blood volume fraction (Vb) of 0.05. In addition, a reference TAC was created with K1′ = 0.3, k2′ = 0.15, k3′ = 0.02, k4′ = 0.06 and Vb = 0.05, with a noise level of 2%. Since BPND = k3/k4, the two levels of BPND were 14 and 3, and R1 = k2/k2′ = 1. Bias and
coefficient of variation (COV) for BPND and R1 were calculated for all parametric imaging methods. Previous studies have shown that a scanning time of 80 min for [ 11C]PE2I may be insufficient for obtaining reliable quantification of DAT binding in the striatum (DeLorenzo et al., 2009; Seki et al., 2010). A minimum total acquisition time ranging from at least 70 min (Hirvonen et al., 2008) to up to 120 min (DeLorenzo et al., 2009) has been suggested in order to reach equilibrium. Therefore, additional TACs with scan durations of 120 min and 150 min were generated and analyzed with parametric methods. Simulations of RPM2 and MRTM2 were also performed to explore the effect of an inaccurate estimation of the efflux rate constant k2′. In this case an erroneous k2′ value was used, departing ± 10% and ±25% from its true value. To investigate the true effect of k2′ deviation, bias in BPND and R1 using the given k2′ was subtracted from the bias of the inaccurate k2′. All parametric image calculations and simulations were performed using Matlab (Mathworks Inc., Natick, MA, USA). Results Parametric images Parametric images of BPND and R1 from a subject with SAD and a subject with parkinsonian syndrome are given in Figs. 1 and 2 respectively. RPM and RPM2 provided qualitatively the best results, especially with regard to R1 images, for both subject groups. MRTM produced relatively noisy BPND images and showed extreme values in both the striatum and the rest of the brain. MRTM2 also showed noisy spots, but mostly in the skull.
Fig. 2. [11C]PE2I parametric images of (a) binding potential, BPND, and (b) relative tracer delivery, R1, from a subject with parkinsonian syndrome.
M. Jonasson et al. / NeuroImage 74 (2013) 172–178 Table 1 R2 and slope within subjects between SRTM and parametric images for seven brain regions: the putamen, caudate, thalamus, midbrain, hypothalamus, amygdala and hippocampus. Mean values of all subjects. R2
RPM RPM2 MRTMo MRTM MRTM2 Logan DVR-1 SUVr-1
Slope
BPND
R1
BPND
R1
1.00 1.00 0.99 0.98 0.99 1.00 0.99
1.00 0.96 0.37 0.28 0.71 – –
1.03 1.02 0.75 1.05 1.13 0.90 1.52
1.00 1.03 5.85 0.69 0.96 – –
Quantitative evaluation of parametric images Validation of each parametric method for clinical data was performed by comparison with ROI-based SRTM. A total of 14 regions were excluded from further analysis due to BP values with a standard error exceeding 25%. Those regions were the amygdala in four scans, hypothalamus in four scans, midbrain in three scans, hippocampus in two scans and thalamus in one. Both BPND and R1 values for these regions were excluded. R2 and slope of a linear regression between SRTM-based values and corresponding values based on each parametric image method for seven ROIs were calculated for each subject. R2 and slopes (mean of all subjects) are given in Table 1. BPND values based on all parametric methods showed a high correlation with SRTM (R2 ≥0.98). For R1, only RPM and RPM2 showed good correlation (R2 ≥0.96) with SRTM. MRTM2 had the highest R2 with SRTM of the multilinear reference tissue methods (R2 =0.71) and a poor correlation was shown for the other methods (R2 ≤0.37). The slope values showed a high variation between methods for both BPND and R1 values. Also here, RPM and RPM2 were optimal with slope values slightly higher than one. The mean R2 and slope values were also calculated for the two subject groups separately. These
175
results showed the same trend, with RPM as the optimal method closely followed by RPM2 (data not shown). Reference Logan and MRTMo images, shown in Figs. 1 and 2, produced an underestimation of BPND while SUVr showed an overestimation of BPND in the striatum compared to RPM. MRTMo and MRTM were not able to give robust estimations of R1 at the voxel level. MRTM2 underestimated R1 compared to RPM. The relationship between BPND values obtained by SRTM and by the various parametric methods in the striatum and thalamus, across subjects, is shown in Fig. 3 with a high agreement for RPM and RPM2 in both regions. The corresponding R 2 and slopes for the striatum and thalamus are given in Table 2. Also here, RPM and RPM2 showed the highest R 2 with regard to BPND values (R 2 ≥ 0.78), but R 2 for R1 were higher in the striatum than in the thalamus. The multilinear reference tissue models showed relatively high correlations for BPND in both the striatum and the thalamus (R 2 ≥ 0.93), except in the thalamus for MRTM, while R 2 for R1 were close to zero in five out of six cases. R 2 for BPND regarding Logan and SUVr were in the same range (R 2 ≥ 0.95), except in the striatum for the Logan approach. Slope values were still close to one for RPM and RPM2 considering the striatum, but more variable in the thalamus. For the multilinear reference tissue models, very variable slope values were found regarding both the striatum and the thalamus.
Accuracy and precision Accuracy and precision of BPND and R1 in cases with high and low BPND as well as high and low noise levels, are presented in Fig. 4. The receptor parametric methods performed best in the simulations, with a better accuracy for RPM. The multilinear reference tissue models showed large variation in the performance within each model. Reference Logan and SUVr had in general a reasonable precision but reference Logan showed a poor accuracy. SRTM showed accuracy and precision similar
Fig. 3. Across subject relationship between BPND obtained by SRTM and BPND, Logan DVR-1 and SUVr-1 obtained by the various parametric methods for the striatum (a, b and c) and the thalamus (d, e and f).
176
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
Table 2 R2 and slope across subjects between SRTM and the seven different methods for striatum and for thalamus. Striatum R2
RPM RPM2 MRTMo MRTM MRTM2 Logan DVR-1 SUVr-1
Thalamus R2
Slope
Slope
BPND
R1
BPND
R1
BPND
R1
BPND
R1
0.99 0.99 0.93 0.96 0.99 0.97 0.95
0.99 0.93 0.09 0.01 0.38 – –
0.99 0.97 0.59 1.17 1.16 0.81 1.18
0.98 0.94 −3.69 −0.15 0.55 – –
0.98 0.98 0.98 0.31 0.99 0.98 0.37
0.78 0.27 0.00 0.01 0.08 – –
1.08 1.14 0.98 0.74 1.04 1.01 0.53
1.10 0.94 0.56 0.47 −0.85 – –
to the receptor parametric mapping methods except in the case of high binding and high noise level where both bias and COV were higher. Effect of different scan durations is shown in Fig. 5 in the case of high BPND and high noise level. The differences in bias and COV between 80, 120 and 150 min scanning alternatives were in general small. Also in this case RPM performed best, followed by RPM2. Bias of the simulations of over- and underestimations of the efflux rate constant from the reference tissue, k2′, are given in Fig. 6. RPM2 showed consistently an underestimation of BPND and R1 for both over- and underestimations of k2′. MRTM2 showed a low sensitivity for an inaccurate k2′ for BPND estimation due to the small contribution of the term containing k2′ in the model equation. However, the good accuracy in BPND was accompanied by a higher bias in R1. Discussion In this study, seven different reference tissue methods were evaluated for the generation of [11C]PE2I parametric images, where the parameters of interest were BPND and R1. The reference tissue model approach has a clear advantage in that there is no need for highly invasive arterial blood sampling, which can mean discomfort for the patient and requires
plasma metabolite analysis. The performance of each method was investigated using clinical data and simulations, and compared with the results of region-based SRTM analyses. A previous study (DeLorenzo et al., 2009) has shown a slightly better correlation between DVR-1 as based on reference Logan analysis and plasma-input BPND, than between SRTM and plasma-input BPND. However, since our purpose is to evaluate parametric images for both BPND and R1, we opted to use SRTM as the gold standard in the present work. The receptor parametric methods, RPM and RPM2, showed robust and quantitatively accurate results, both for clinical and simulated data and RPM showed the highest correlation compared to SRTM. RPM2 showed an accuracy and precision close to that of RPM, but has a disadvantage when k2′ is fixed for the reference tissue efflux rate value of the 2TCM. In the analysis of clinical data this value is taken from previously estimated k2′ by RPM and therefore RPM2 shows a higher correlation to SRTM in clinical data than in simulations. MRTM2 performed best of the multilinear reference tissue models but a drawback in the two parameter version of the models is that they rely on accurate parameter estimation in the previous analysis for fixing the correct k2′ value. This is a challenge for MRTM2 where the fixed k2′ value is derived from previous analysis using MRTM, which is not able to accurately estimate all parameters. With an inaccurate estimate of k2′ carried on to the MRTM2 analysis, this resulted in considerable, BPND-value dependent, underestimation of R1. Simulations, using an erroneous k2′ value, also showed that especially R1 estimation with MRTM2 was very sensitive for an inaccurate reference efflux parameter as shown in Fig. 6. It has been shown previously that scanning time up to 120 min is needed in order to obtain reliable results of DAT binding in the striatum (DeLorenzo et al., 2009). However, according to the performed simulations in this study, only MRTM2 showed some improvement in the simulation with longer scan durations, but accuracy and precision were still not better than the 80 min simulated scans of the receptor parametric mapping methods. One issue that is not addressed in this study, but important for the outcome of the results, is the kinetics of [11C]PE2I in cerebellum. Previous
Fig. 4. Bias (in units of BPND or R1) of (a) BPND, (b) R1 and COV of (c) BPND and (d) R1 for simulations with a BPND = 3 and 14 and a noise level (nl) of 2% and 10% for 80 min simulated scan time.
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
177
Fig. 5. Bias (in units of BPND or R1) of (a) BPND, (b) R1 and COV of (c) BPND and (d) R1 for 80, 120 and 150 min simulated scan duration with a BPND = 14 and a noise level of 10%.
studies have shown that the kinetics of [11C]PE2I in the cerebellum needs more than one compartment in order to be properly described (DeLorenzo et al., 2009; Jucaite et al., 2006) which contradicts the assumptions made for the reference tissue models (Lammertsma and Hume, 1996). If the reference tissue has specific binding, there will be an underestimation in BPND when using reference tissue models for estimation of the parameter (Gunn et al., 1997; Lammertsma et al., 1996). Previous reports have shown a tendency of 30–50% underestimation of BPND for DAT-rich regions when using reference tissue models compared with plasma input models (Hirvonen et al., 2008; Jucaite et al., 2006). Another study has shown that blocking of DAT did not alter cerebellum kinetics (Halldin et al., 2003) indicating that the second compartment is not caused by specific binding. Possible cerebral uptake of metabolites, retaining the 11C label but not binding specifically to DAT, may explain this apparent second compartment, in which case reference tissue methods will show a bias compared to plasma-input methods directly estimating BPND (Yaqub et al., 2009) as has indeed been observed
for [11C]PE2I (Jucaite et al., 2006). Other explanations of a second compartment can be a slow non-specific binding or intravascular activity. Biased BPND has also been reported for linearized graphical methods, suffering from noise induced negative bias (Slifstein and Laruelle, 2000) and overestimations of BPND using SUVr have been shown before (Yaqub et al., 2008). However, the relative importance of an underestimation of BPND is small as long as there is a good correlation between BPND obtained from reference tissue models and BPND obtained using a plasma input function (Yaqub et al., 2009) which previous studies also have shown.
Conclusions Both receptor parametric mapping methods, RPM and RPM2, were able to generate robust and quantitatively accurate parametric images. RPM showed the highest correlations with region-based
Fig. 6. Bias (in units of BPND or R1) of (a) BPND and (b) R1 for simulated over- and underestimations of k2′ and with a BPND = 3 and 14.
178
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
SRTM analysis and is thus the method of choice for [ 11C]PE2I parametric analysis. Acknowledgment The authors would like to thank the staff at the PET center at Uppsala University Hospital for the production of [ 11C]PE2I and for the acquisition of the PET data. Scans performed in this work were financed by grants from Uppsala University Hospital, the Swedish Parkinson Foundation and the Swedish Research Council. References Antonini, A., Moresco, R.M., Gobbo, C., De Notaris, R., Panzacchi, A., Barone, P., Calzetti, S., Negrotti, A., Pezzoli, G., Fazio, F., 2001. The status of dopamine nerve terminals in Parkinson's disease and essential tremor: a PET study with the tracer [11-C] FE-CIT. Neurol. Sci. 22, 47–48. DeLorenzo, C., Kumar, J.S., Zanderigo, F., Mann, J.J., Parsey, R.V., 2009. Modeling considerations for in vivo quantification of the dopamine transporter using [(11)C]PE2I and positron emission tomography. J. Cereb. Blood Flow Metab. 29, 1332–1345. Emond, P., Guilloteau, D., Chalon, S., 2008. PE2I: a radiopharmaceutical for in vivo exploration of the dopamine transporter. CNS Neurosci. Ther. 14, 47–64. Gulley, J.M., Zahniser, N.R., 2003. Rapid regulation of dopamine transporter function by substrates, blockers and presynaptic receptor ligands. Eur. J. Pharmacol. 479, 139–152. Gunn, R.N., Lammertsma, A.A., Hume, S.P., Cunningham, V.J., 1997. Parametric imaging of ligand-receptor binding in PET using a simplified reference region model. NeuroImage 6, 279–287. Hall, H., Halldin, C., Guilloteau, D., Chalon, S., Emond, P., Besnard, J., Farde, L., Sedvall, G., 1999. Visualization of the dopamine transporter in the human brain postmortem with the new selective ligand [125I]PE2I. NeuroImage 9, 108–116. Halldin, C., Erixon-Lindroth, N., Pauli, S., Chou, Y.H., Okubo, Y., Karlsson, P., Lundkvist, C., Olsson, H., Guilloteau, D., Emond, P., Farde, L., 2003. [(11)C]PE2I: a highly selective radioligand for PET examination of the dopamine transporter in monkey and human brain. Eur. J. Nucl. Med. Mol. Imaging 30, 1220–1230. Hirvonen, J., Johansson, J., Teräs, M., Oikonen, V., Lumme, V., Virsu, P., Roivainen, A., Någren, K., Halldin, C., Farde, L., Hietala, J., 2008. Measurement of striatal and extrastriatal dopamine transporter binding with high-resolution PET and [11C] PE2I: quantitative modeling and test–retest reproducibility. J. Cereb. Blood Flow Metab. 28, 1059–1069. Ichise, M., Ballinger, J.R., Golan, H., Vines, D., Luong, A., Tsai, S., Kung, H.F., 1996. Noninvasive quantification of dopamine D2 receptors with iodine-123-IBF SPECT. J. Nucl. Med. 37, 513–520. Ichise, M., Liow, J.S., Lu, J.Q., Takano, A., Model, K., Toyama, H., Suhara, T., Suzuki, K., Innis, R.B., Carson, R.E., 2003. Linearized reference tissue parametric imaging methods: application to [11C]DASB positron emission tomography studies of the serotonin transporter in human brain. J. Cereb. Blood Flow Metab. 23, 1096–1112. Jucaite, A., Fernell, E., Halldin, C., Forssberg, H., Farde, L., 2005. Reduced midbrain dopamine transporter binding in male adolescents with attention-deficit/hyperactivity disorder: association between striatal dopamine markers and motor hyperactivity. Biol. Psychiatry 57, 229–238. Jucaite, A., Odano, I., Olsson, H., Pauli, S., Halldin, C., Farde, L., 2006. Quantitative analyses of regional [11C]PE2I binding to the dopamine transporter in the human brain: a PET study. Eur. J. Nucl. Med. Mol. Imaging 33, 657–668.
Laakso, A., Bergman, J., Haaparanta, M., Vilkman, H., Solin, O., Syvälahti, E., Hietala, J., 2001. Decreased striatal dopamine transporter binding in vivo in chronic schizophrenia. Schizophr. Res. 52, 115–120. Lammertsma, A.A., Hume, S.P., 1996. Simplified reference tissue model for PET receptor studies. NeuroImage 4, 153–158. Lammertsma, A.A., Bench, C.J., Hume, S.P., Osman, S., Gunn, K., Brooks, D.J., Frackowiak, R.S., 1996. Comparison of methods for analysis of clinical [11C]raclopride studies. J. Cereb. Blood Flow Metab. 16, 42–52. Leroy, C., Karila, L., Martinot, J.L., Lukasiewicz, M., Duchesnay, E., Comtat, C., Dollé, F., Benyamina, A., Artiges, E., Ribeiro, M.J., Reynaud, M., Trichard, C., 2012. Striatal and extrastriatal dopamine transporter in cannabis and tobacco addiction: a high-resolution PET study. Addict. Biol. 17, 981–990. Logan, J., Fowler, J.S., Volkow, N.D., Wang, G.J., Ding, Y.S., Alexoff, D.L., 1996. Distribution volume ratios without blood sampling from graphical analysis of PET data. J. Cereb. Blood Flow Metab. 16, 834–840. Meyer, P.T., Hellwig, S., Amtage, F., Rottenburger, C., Sahm, U., Reuland, P., Weber, W.A., Hüll, M., 2011. Dual-biomarker imaging of regional cerebral amyloid load and neuronal activity in dementia with PET and 11C-labeled Pittsburgh compound B. J. Nucl. Med. 52, 393–400. Odano, I., Varrone, A., Savic, I., Ciumas, C., Karlsson, P., Jucaite, A., Halldin, C., Farde, L., 2012. Quantitative PET analyses of regional [(11)C]PE2I binding to the dopamine transporter — application to juvenile myoclonic epilepsy. NeuroImage 59, 3582–3593. Ribeiro, M.J., Vidailhet, M., Loc'h, C., Dupel, C., Nguyen, J.P., Ponchant, M., Dollé, F., Peschanski, M., Hantraye, P., Cesaro, P., Samson, Y., Remy, P., 2002. Dopaminergic function and dopamine transporter binding assessed with positron emission tomography in Parkinson disease. Arch. Neurol. 59, 580–586. Ribeiro, M.J., Thobois, S., Lohmann, E., du Montcel, S.T., Lesage, S., Pelissolo, A., Dubois, B., Mallet, L., Pollak, P., Agid, Y., Broussolle, E., Brice, A., Remy, P., Group F.P.s.D.G.S., 2009. A multitracer dopaminergic PET study of young-onset parkinsonian patients with and without parkin gene mutations. J. Nucl. Med. 50, 1244–1250. Seki, C., Ito, H., Ichimiya, T., Arakawa, R., Ikoma, Y., Shidahara, M., Maeda, J., Takano, A., Takahashi, H., Kimura, Y., Suzuki, K., Kanno, I., Suhara, T., 2010. Quantitative analysis of dopamine transporters in human brain using [11C]PE2I and positron emission tomography: evaluation of reference tissue models. Ann. Nucl. Med. 24, 249–260. Slifstein, M., Laruelle, M., 2000. Effects of statistical noise on graphic analysis of PET neuroreceptor studies. J. Nucl. Med. 41, 2083–2088. Spencer, T.J., Biederman, J., Madras, B.K., Faraone, S.V., Dougherty, D.D., Bonab, A.A., Fischman, A.J., 2005. In vivo neuroreceptor imaging in attention-deficit/ hyperactivity disorder: a focus on the dopamine transporter. Biol. Psychiatry 57, 1293–1300. Svarer, C., Madsen, K., Hasselbalch, S.G., Pinborg, L.H., Haugbøl, S., Frøkjaer, V.G., Holm, S., Paulson, O.B., Knudsen, G.M., 2005. MR-based automatic delineation of volumes of interest in human brain PET images using probability maps. NeuroImage 24, 969–979. Warwick, J.M., Carey, P.D., Cassimjee, N., Lochner, C., Hemmings, S., Moolman-Smook, H., Beetge, E., Dupont, P., Stein, D.J., 2012. Dopamine transporter binding in social anxiety disorder: the effect of treatment with escitalopram. Metab. Brain Dis. 27, 151–158. Wu, Y., Carson, R.E., 2002. Noise reduction in the simplified reference tissue model for neuroreceptor functional imaging. J. Cereb. Blood Flow Metab. 22, 1440–1452. Yaqub, M., Tolboom, N., Boellaard, R., van Berckel, B.N., van Tilburg, E.W., Luurtsema, G., Scheltens, P., Lammertsma, A.A., 2008. Simplified parametric methods for [11C]PIB studies. NeuroImage 42, 76–86. Yaqub, M., Boellaard, R., van Berckel, B.N., Tolboom, N., Luurtsema, G., Dijkstra, A.A., Lubberink, M., Windhorst, A.D., Scheltens, P., Lammertsma, A.A., 2009. Evaluation of tracer kinetic models for analysis of [18F]FDDNP studies. Mol. Imaging Biol. 11, 322–333.
Contents lists available at SciVerse ScienceDirect
NeuroImage journal homepage: www.elsevier.com/locate/ynimg
Validation of parametric methods for [ 11C]PE2I positron emission tomography My Jonasson a, d,⁎, Lieuwe Appel a, d, Jonas Engman b, Andreas Frick b, Dag Nyholm c, Håkan Askmark c, Torsten Danfors a, c, d, Jens Sörensen a, d, Tomas Furmark b, Mark Lubberink a, e a
Nuclear Medicine & PET, Uppsala University, 751 81 Uppsala, Sweden Psychology, Uppsala University, 751 81 Uppsala, Sweden Neuroscience, Neurology, Uppsala University, 751 81 Uppsala, Sweden d Molecular Imaging, Uppsala University Hospital, 751 81 Uppsala, Sweden e Medical Physics, Uppsala University Hospital, 751 81 Uppsala, Sweden b c
a r t i c l e
i n f o
Article history: Accepted 11 February 2013 Available online 19 February 2013 Keywords: PET [11C]PE2I Kinetic modeling Dopamine transporter Reference tissue model Parametric images
a b s t r a c t Objectives: The radioligand [11C]PE2I is highly selective for dopamine transporter (DAT) and can be used in vivo for investigation of changes in DAT concentration, progression of disease and validation of treatment using positron emission tomography (PET). DAT is an important protein for regulation of central dopamine concentration and DAT deficiency has been associated with several neurodegenerative and neuropsychiatric disorders. Accurate parametric images are a prerequisite for clinical application of [ 11C]PE2I. The purpose of this study was to evaluate different methods for producing [11C]PE2I parametric images, showing binding potential (BPND) and relative delivery (R1) at the voxel level, using clinical data as well as simulations. Methods: Investigations were made in twelve subjects either with social anxiety disorder (n=6) or parkinsonian syndrome (n=6), each receiving an 80 min dynamic PET scan. All subjects underwent a T1-weighted MRI scan which was co-registered to the PET images and used for definition of regions of interest using a probabilistic template (PVElab). Two basis function implementations (receptor parametric mapping: RPM, RPM2) of the simplified reference tissue model (SRTM) and three multilinear reference tissue models (MRTMo, MRTM and MRTM2) were used for computation of parametric BPND and R1 images. In addition, reference Logan and standard uptake value ratio (SUVr) were investigated. Evaluations of BPND and R1 images were performed using linear regression to compare the parametric methods to region-based analyses with SRTM and cerebellar gray matter as reference region. Accuracy and precision of each method were assessed by simulations. Results: Correlation and slope of linear regression between parametric and region-based BPND and R1 values in both striatum and extra-striatal regions were optimal for RPM (R2 =0.99 for both BPND and R1; slopes 0.99 and 0.98 for BPND and R1, respectively, in striatum). In addition, accuracy and precision were best for RPM and RPM2. Conclusion: The basis function methods provided more robust estimations of the parameters compared to the other models and performed best in simulations. RPM, a basis function implementation of SRTM, is the preferred method for voxel level analysis of [11C]PE2I PET studies. © 2013 Elsevier Inc. All rights reserved.
Introduction The dopamine transporter (DAT) is a transmembrane protein located on the presynaptic neurons responsible for re-uptake and removal of dopamine from the synaptic cleft. This process regulates the action of dopamine and has shown to be of interest in several physiological functions (Gulley and Zahniser, 2003). DAT is associated with different neurodegenerative and neuropsychiatric disorders such as Parkinson's disease (Antonini et al., 2001; Ribeiro et al., 2002), schizophrenia (Laakso et al., 2001), attention-deficit/hyperactive disorder (ADHD) (Jucaite et al., 2005; Spencer et al., 2005) and social anxiety disorders (SAD) (Warwick et al., 2012). ⁎ Corresponding author at: Nuclear Medicine & PET, Uppsala University, 751 81 Uppsala, Sweden. E-mail address: [email protected] (M. Jonasson). 1053-8119/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.neuroimage.2013.02.022
Positron emission tomography (PET) is a non-invasive molecular imaging method for diagnosis and monitoring treatment effects in-vivo. PET provides functional images of the body and has proven to be a powerful tool for neurotransmission studies including imaging of DAT. The cocaine analog N-(3-iodoprop-2E-enyl)-2βcarbomethoxy-3β-(4-methyl-phenyl)nortropane (PE2I) has demonstrated an excellent binding to DAT, and both in vitro autoradiography studies using [ 125I]PE2I (Hall et al., 1999), and in vivo PET studies with [ 11C]PE2I (DeLorenzo et al., 2009; Hirvonen et al., 2008; Jucaite et al., 2006; Seki et al., 2010) have shown marked accumulation in the striatum where concentration of DAT is high. An intermediate binding has been found in the midbrain and no, or very little, specific binding in the cerebellum. Previous studies have shown that PE2I has a high selectivity for DAT, and because of that, [ 11C]PE2I provides high contrast PET images and appears to be sufficiently sensitive for detection of DAT in small extrastriatal brain areas (Emond et al., 2008). [ 11C]PE2I has also
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
shown a good reproducibility and reliability in test–retest PET studies (Hirvonen et al., 2008). [ 11C]PE2I PET studies can thus be a valuable tool for early detection of neurological disorders where changes in DAT concentration are involved, for the assessment of disease progression, and for validation of treatments (Emond et al., 2008). Applying a simplified reference tissue model (SRTM) (Lammertsma and Hume, 1996) on regions of interest (ROIs), using cerebellum as reference region, has proven to be a robust method for parameter estimation (Hirvonen et al., 2008; Jucaite et al., 2006), showing a good correlation with compartmental modeling using a plasma input (DeLorenzo et al., 2009; Seki et al., 2010). To further facilitate clinical use of [ 11C]PE2I, the possibility of straightforward visual assessment of DAT availability is a prerequisite. To this end, parametric images have to be available, where the parameters of interest are estimated for each voxel. In addition to binding potential (BPND) images, showing DAT availability, images of relative delivery (R1), reflecting regional cerebral blood flow (rCBF) may be of interest in the differential diagnosis of neurodegenerative diseases (Meyer et al., 2011). For a voxel based analysis, using standard non-linear least square fitting for estimation of the parameters is sensitive to noise and very time consuming. Parametric methods, utilizing linearization of compartment models, provide faster analyses on a voxel level than non-linear regression analysis (Gunn et al., 1997). Voxel-based analysis of BPND has been performed previously for [ 11C]PE2I (Leroy et al., 2012; Odano et al., 2012; Ribeiro et al., 2009), but no extensive validation of parametric methods has been presented for this tracer. The purpose of this study was to evaluate different methods for producing [ 11C]PE2I parametric images, showing BPND and R1 at the voxel level, using clinical data as well as simulations. Methods Subjects and data acquisition Data from a total of twelve subjects were included in this study whereof six were SAD subjects (three women and three men, mean age 35 ±11 years) and six were subjects with parkinsonian syndrome (five women and one man, mean age 68± 6 years). Each subject signed a written informed consent and the study was approved by the local independent ethics and radiation safety committees. In addition, the scans of subjects with SAD were part of a clinical trial that was approved by the Swedish medical products agency. Together, these two groups are expected to display a large range of DAT availability in the striatum. Each subject underwent a dynamic PET scan on an ECAT Exact HR+ scanner (Siemens/CTI, Knoxville). After a 10 min transmission scan for attenuation correction, an 80 min emission scan in 3-dimensional acquisition mode was started simultaneously with the injection of about 350 MBq [ 11C]PE2I. Twenty-two frames of increasing durations (4× 1, 2 × 2, 4 × 3, 12 × 5 min) were acquired. Dynamic images were reconstructed using ordered subset expectation maximization (OSEM) with 6 iterations and 8 subsets and a 4 mm Hanning post-filter, applying all appropriate corrections. In addition, each subject underwent a T1-weighted magnetic resonance image (MRI) scan (3D-SENSE) on a 3 T Achieva scanner (Philips Healthcare, Best, The Netherlands). [ 11C]PE2I was synthesized based on previously described methods (Halldin et al., 2003). ROI analysis Dynamic PET images were realigned to correct for inter-frame patient motion using Voiager (GE Healthcare, Uppsala, Sweden). MRI images were co-registered to the sum of the first 3 min of the PET scan, mostly resembling a blood flow image, using statistical parametric mapping (SPM5; Wellcome Trust Center for Neuroimaging, University College London, UK). Gray matter ROIs were defined on
173
co-registered MRI images using an automated probabilistic ROI template as implemented in the PVElab software (Svarer et al., 2005). Seven ROIs were included: the putamen, caudate, thalamus, midbrain, hypothalamus, amygdala and hippocampus, averaged over the left and the right hemisphere. ROIs were transferred to the dynamic PET images, yielding regional time-activity curves (TACs). Data were analyzed using SRTM with the cerebellar gray matter as reference tissue, since the cerebellum is a region known for low DAT concentration and thus lacks specific binding of PE2I (Halldin et al., 2003). Regions with standard error (SE) >25% for BPND estimates were excluded from further analysis. Parametric images For evaluation of parametric images, voxel level analyses of clinical data were performed using two receptor parametric mapping methods RPM (Gunn et al., 1997) and RPM2 (Wu and Carson, 2002), three multilinear reference tissue models MRTMo, MRTM and MRTM2 (Ichise et al., 1996, 2003), reference Logan analysis (Logan et al., 1996) and standard uptake value ratio (SUVr). In all analyses the cerebellum was used as reference region. The three and two parameter versions of receptor parametric mapping (RPM and RPM2) are implementations of SRTM and SRTM2 (Wu and Carson, 2002) using a set of predefined basis functions to linearize the models. One hundred basis functions were predefined for each scan with a discrete set of values for the exponential variable ranging from 0.01 to 0.5 min −1. For RPM2, the estimated number of parameters was reduced to two by setting the efflux rate constant from reference tissue, k2′ (min −1), to a constant value. This was performed by taking a volume-weighted mean value for k2′ including the putamen, caudate, thalamus, midbrain, hypothalamus, amygdala and hippocampus from the previous RPM analysis. The multilinear reference tissue models (MRTMo, MRTM and MRTM2) estimate the parameters using multilinear regression after a certain equilibrium time that for the present analyses was set to 30 min. For MRTM2, the parameters were reduced the same way as described above for RPM2 but with the k2′ value estimated from previous MRTM analysis. The estimated parameters for these five methods were thus R1 and BPND for each voxel. The reference Logan model is a graphical method and the analysis was performed with a time interval of 30–80 min for which BPND was indirectly estimated as the distribution volume ratio DVR-1. Additionally, the standardized uptake value ratio (SUVr) was estimated for the last four frames, 60–80 min, and BPND was estimated as SUVr-1. No estimates of R1 can be obtained by reference Logan or SUVr. To take into account the different counts in each frame, weights were included in the data analysis. The corresponding weighting factor for each frame was calculated as
w¼
Δt 2 f 2T
where Δt is the frame duration, T is the total counts in the frame and f is the decay correction factor for the frame which is computed as follows,
f ¼
λΔt expð−λt s Þ− expð−λt e Þ
where ts and te are the start and the end time of the frame and λ is the decay constant for 11C.
174
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
Fig. 1. [11C]PE2I parametric images of (a) binding potential, BPND, and (b) relative tracer delivery, R1, from a subject with SAD.
Quantitative evaluation of parametric images A quantitative evaluation of each parametric method on clinical data was performed by retrieving regionally averaged voxel values from the generated BPND and R1 parametric images by projecting the set of ROIs as described previously. Square of the correlation coefficient (R 2) and slope of a linear regression of ROI values between the different parametric methods and SRTM were calculated for the subjects. In addition, R 2 and slope were calculated separately for two substantial ROIs, the striatum and thalamus, representing regions with high and low uptake of [ 11C]PE2I. Accuracy and precision For assessment of accuracy and precision of BPND and R1 estimates, simulations were performed for all seven parametric methods as well as SRTM. TACs were created using a two tissue compartment model (2TCM) with plasma input curve and parameters from literature (Jucaite et al., 2006; Seki et al., 2010). For each set of parameters, 100 noisy TACs were generated. The noise was normally distributed with a mean value of zero and a scale factor to determine the level of noise according to a previously described method (Ichise et al., 2003). The noise level was set to 2% or 10% to resemble noise in a typical striatum ROI-TAC or voxel-TAC, respectively. Two set of rate constants were used to reflect both the high and the low DAT availability; K1 = 0.3, k2 = 0.15, k3 = 0.35 and 0.18 and k4 = 0.025 and 0.06, together with a blood volume fraction (Vb) of 0.05. In addition, a reference TAC was created with K1′ = 0.3, k2′ = 0.15, k3′ = 0.02, k4′ = 0.06 and Vb = 0.05, with a noise level of 2%. Since BPND = k3/k4, the two levels of BPND were 14 and 3, and R1 = k2/k2′ = 1. Bias and
coefficient of variation (COV) for BPND and R1 were calculated for all parametric imaging methods. Previous studies have shown that a scanning time of 80 min for [ 11C]PE2I may be insufficient for obtaining reliable quantification of DAT binding in the striatum (DeLorenzo et al., 2009; Seki et al., 2010). A minimum total acquisition time ranging from at least 70 min (Hirvonen et al., 2008) to up to 120 min (DeLorenzo et al., 2009) has been suggested in order to reach equilibrium. Therefore, additional TACs with scan durations of 120 min and 150 min were generated and analyzed with parametric methods. Simulations of RPM2 and MRTM2 were also performed to explore the effect of an inaccurate estimation of the efflux rate constant k2′. In this case an erroneous k2′ value was used, departing ± 10% and ±25% from its true value. To investigate the true effect of k2′ deviation, bias in BPND and R1 using the given k2′ was subtracted from the bias of the inaccurate k2′. All parametric image calculations and simulations were performed using Matlab (Mathworks Inc., Natick, MA, USA). Results Parametric images Parametric images of BPND and R1 from a subject with SAD and a subject with parkinsonian syndrome are given in Figs. 1 and 2 respectively. RPM and RPM2 provided qualitatively the best results, especially with regard to R1 images, for both subject groups. MRTM produced relatively noisy BPND images and showed extreme values in both the striatum and the rest of the brain. MRTM2 also showed noisy spots, but mostly in the skull.
Fig. 2. [11C]PE2I parametric images of (a) binding potential, BPND, and (b) relative tracer delivery, R1, from a subject with parkinsonian syndrome.
M. Jonasson et al. / NeuroImage 74 (2013) 172–178 Table 1 R2 and slope within subjects between SRTM and parametric images for seven brain regions: the putamen, caudate, thalamus, midbrain, hypothalamus, amygdala and hippocampus. Mean values of all subjects. R2
RPM RPM2 MRTMo MRTM MRTM2 Logan DVR-1 SUVr-1
Slope
BPND
R1
BPND
R1
1.00 1.00 0.99 0.98 0.99 1.00 0.99
1.00 0.96 0.37 0.28 0.71 – –
1.03 1.02 0.75 1.05 1.13 0.90 1.52
1.00 1.03 5.85 0.69 0.96 – –
Quantitative evaluation of parametric images Validation of each parametric method for clinical data was performed by comparison with ROI-based SRTM. A total of 14 regions were excluded from further analysis due to BP values with a standard error exceeding 25%. Those regions were the amygdala in four scans, hypothalamus in four scans, midbrain in three scans, hippocampus in two scans and thalamus in one. Both BPND and R1 values for these regions were excluded. R2 and slope of a linear regression between SRTM-based values and corresponding values based on each parametric image method for seven ROIs were calculated for each subject. R2 and slopes (mean of all subjects) are given in Table 1. BPND values based on all parametric methods showed a high correlation with SRTM (R2 ≥0.98). For R1, only RPM and RPM2 showed good correlation (R2 ≥0.96) with SRTM. MRTM2 had the highest R2 with SRTM of the multilinear reference tissue methods (R2 =0.71) and a poor correlation was shown for the other methods (R2 ≤0.37). The slope values showed a high variation between methods for both BPND and R1 values. Also here, RPM and RPM2 were optimal with slope values slightly higher than one. The mean R2 and slope values were also calculated for the two subject groups separately. These
175
results showed the same trend, with RPM as the optimal method closely followed by RPM2 (data not shown). Reference Logan and MRTMo images, shown in Figs. 1 and 2, produced an underestimation of BPND while SUVr showed an overestimation of BPND in the striatum compared to RPM. MRTMo and MRTM were not able to give robust estimations of R1 at the voxel level. MRTM2 underestimated R1 compared to RPM. The relationship between BPND values obtained by SRTM and by the various parametric methods in the striatum and thalamus, across subjects, is shown in Fig. 3 with a high agreement for RPM and RPM2 in both regions. The corresponding R 2 and slopes for the striatum and thalamus are given in Table 2. Also here, RPM and RPM2 showed the highest R 2 with regard to BPND values (R 2 ≥ 0.78), but R 2 for R1 were higher in the striatum than in the thalamus. The multilinear reference tissue models showed relatively high correlations for BPND in both the striatum and the thalamus (R 2 ≥ 0.93), except in the thalamus for MRTM, while R 2 for R1 were close to zero in five out of six cases. R 2 for BPND regarding Logan and SUVr were in the same range (R 2 ≥ 0.95), except in the striatum for the Logan approach. Slope values were still close to one for RPM and RPM2 considering the striatum, but more variable in the thalamus. For the multilinear reference tissue models, very variable slope values were found regarding both the striatum and the thalamus.
Accuracy and precision Accuracy and precision of BPND and R1 in cases with high and low BPND as well as high and low noise levels, are presented in Fig. 4. The receptor parametric methods performed best in the simulations, with a better accuracy for RPM. The multilinear reference tissue models showed large variation in the performance within each model. Reference Logan and SUVr had in general a reasonable precision but reference Logan showed a poor accuracy. SRTM showed accuracy and precision similar
Fig. 3. Across subject relationship between BPND obtained by SRTM and BPND, Logan DVR-1 and SUVr-1 obtained by the various parametric methods for the striatum (a, b and c) and the thalamus (d, e and f).
176
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
Table 2 R2 and slope across subjects between SRTM and the seven different methods for striatum and for thalamus. Striatum R2
RPM RPM2 MRTMo MRTM MRTM2 Logan DVR-1 SUVr-1
Thalamus R2
Slope
Slope
BPND
R1
BPND
R1
BPND
R1
BPND
R1
0.99 0.99 0.93 0.96 0.99 0.97 0.95
0.99 0.93 0.09 0.01 0.38 – –
0.99 0.97 0.59 1.17 1.16 0.81 1.18
0.98 0.94 −3.69 −0.15 0.55 – –
0.98 0.98 0.98 0.31 0.99 0.98 0.37
0.78 0.27 0.00 0.01 0.08 – –
1.08 1.14 0.98 0.74 1.04 1.01 0.53
1.10 0.94 0.56 0.47 −0.85 – –
to the receptor parametric mapping methods except in the case of high binding and high noise level where both bias and COV were higher. Effect of different scan durations is shown in Fig. 5 in the case of high BPND and high noise level. The differences in bias and COV between 80, 120 and 150 min scanning alternatives were in general small. Also in this case RPM performed best, followed by RPM2. Bias of the simulations of over- and underestimations of the efflux rate constant from the reference tissue, k2′, are given in Fig. 6. RPM2 showed consistently an underestimation of BPND and R1 for both over- and underestimations of k2′. MRTM2 showed a low sensitivity for an inaccurate k2′ for BPND estimation due to the small contribution of the term containing k2′ in the model equation. However, the good accuracy in BPND was accompanied by a higher bias in R1. Discussion In this study, seven different reference tissue methods were evaluated for the generation of [11C]PE2I parametric images, where the parameters of interest were BPND and R1. The reference tissue model approach has a clear advantage in that there is no need for highly invasive arterial blood sampling, which can mean discomfort for the patient and requires
plasma metabolite analysis. The performance of each method was investigated using clinical data and simulations, and compared with the results of region-based SRTM analyses. A previous study (DeLorenzo et al., 2009) has shown a slightly better correlation between DVR-1 as based on reference Logan analysis and plasma-input BPND, than between SRTM and plasma-input BPND. However, since our purpose is to evaluate parametric images for both BPND and R1, we opted to use SRTM as the gold standard in the present work. The receptor parametric methods, RPM and RPM2, showed robust and quantitatively accurate results, both for clinical and simulated data and RPM showed the highest correlation compared to SRTM. RPM2 showed an accuracy and precision close to that of RPM, but has a disadvantage when k2′ is fixed for the reference tissue efflux rate value of the 2TCM. In the analysis of clinical data this value is taken from previously estimated k2′ by RPM and therefore RPM2 shows a higher correlation to SRTM in clinical data than in simulations. MRTM2 performed best of the multilinear reference tissue models but a drawback in the two parameter version of the models is that they rely on accurate parameter estimation in the previous analysis for fixing the correct k2′ value. This is a challenge for MRTM2 where the fixed k2′ value is derived from previous analysis using MRTM, which is not able to accurately estimate all parameters. With an inaccurate estimate of k2′ carried on to the MRTM2 analysis, this resulted in considerable, BPND-value dependent, underestimation of R1. Simulations, using an erroneous k2′ value, also showed that especially R1 estimation with MRTM2 was very sensitive for an inaccurate reference efflux parameter as shown in Fig. 6. It has been shown previously that scanning time up to 120 min is needed in order to obtain reliable results of DAT binding in the striatum (DeLorenzo et al., 2009). However, according to the performed simulations in this study, only MRTM2 showed some improvement in the simulation with longer scan durations, but accuracy and precision were still not better than the 80 min simulated scans of the receptor parametric mapping methods. One issue that is not addressed in this study, but important for the outcome of the results, is the kinetics of [11C]PE2I in cerebellum. Previous
Fig. 4. Bias (in units of BPND or R1) of (a) BPND, (b) R1 and COV of (c) BPND and (d) R1 for simulations with a BPND = 3 and 14 and a noise level (nl) of 2% and 10% for 80 min simulated scan time.
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
177
Fig. 5. Bias (in units of BPND or R1) of (a) BPND, (b) R1 and COV of (c) BPND and (d) R1 for 80, 120 and 150 min simulated scan duration with a BPND = 14 and a noise level of 10%.
studies have shown that the kinetics of [11C]PE2I in the cerebellum needs more than one compartment in order to be properly described (DeLorenzo et al., 2009; Jucaite et al., 2006) which contradicts the assumptions made for the reference tissue models (Lammertsma and Hume, 1996). If the reference tissue has specific binding, there will be an underestimation in BPND when using reference tissue models for estimation of the parameter (Gunn et al., 1997; Lammertsma et al., 1996). Previous reports have shown a tendency of 30–50% underestimation of BPND for DAT-rich regions when using reference tissue models compared with plasma input models (Hirvonen et al., 2008; Jucaite et al., 2006). Another study has shown that blocking of DAT did not alter cerebellum kinetics (Halldin et al., 2003) indicating that the second compartment is not caused by specific binding. Possible cerebral uptake of metabolites, retaining the 11C label but not binding specifically to DAT, may explain this apparent second compartment, in which case reference tissue methods will show a bias compared to plasma-input methods directly estimating BPND (Yaqub et al., 2009) as has indeed been observed
for [11C]PE2I (Jucaite et al., 2006). Other explanations of a second compartment can be a slow non-specific binding or intravascular activity. Biased BPND has also been reported for linearized graphical methods, suffering from noise induced negative bias (Slifstein and Laruelle, 2000) and overestimations of BPND using SUVr have been shown before (Yaqub et al., 2008). However, the relative importance of an underestimation of BPND is small as long as there is a good correlation between BPND obtained from reference tissue models and BPND obtained using a plasma input function (Yaqub et al., 2009) which previous studies also have shown.
Conclusions Both receptor parametric mapping methods, RPM and RPM2, were able to generate robust and quantitatively accurate parametric images. RPM showed the highest correlations with region-based
Fig. 6. Bias (in units of BPND or R1) of (a) BPND and (b) R1 for simulated over- and underestimations of k2′ and with a BPND = 3 and 14.
178
M. Jonasson et al. / NeuroImage 74 (2013) 172–178
SRTM analysis and is thus the method of choice for [ 11C]PE2I parametric analysis. Acknowledgment The authors would like to thank the staff at the PET center at Uppsala University Hospital for the production of [ 11C]PE2I and for the acquisition of the PET data. Scans performed in this work were financed by grants from Uppsala University Hospital, the Swedish Parkinson Foundation and the Swedish Research Council. References Antonini, A., Moresco, R.M., Gobbo, C., De Notaris, R., Panzacchi, A., Barone, P., Calzetti, S., Negrotti, A., Pezzoli, G., Fazio, F., 2001. The status of dopamine nerve terminals in Parkinson's disease and essential tremor: a PET study with the tracer [11-C] FE-CIT. Neurol. Sci. 22, 47–48. DeLorenzo, C., Kumar, J.S., Zanderigo, F., Mann, J.J., Parsey, R.V., 2009. Modeling considerations for in vivo quantification of the dopamine transporter using [(11)C]PE2I and positron emission tomography. J. Cereb. Blood Flow Metab. 29, 1332–1345. Emond, P., Guilloteau, D., Chalon, S., 2008. PE2I: a radiopharmaceutical for in vivo exploration of the dopamine transporter. CNS Neurosci. Ther. 14, 47–64. Gulley, J.M., Zahniser, N.R., 2003. Rapid regulation of dopamine transporter function by substrates, blockers and presynaptic receptor ligands. Eur. J. Pharmacol. 479, 139–152. Gunn, R.N., Lammertsma, A.A., Hume, S.P., Cunningham, V.J., 1997. Parametric imaging of ligand-receptor binding in PET using a simplified reference region model. NeuroImage 6, 279–287. Hall, H., Halldin, C., Guilloteau, D., Chalon, S., Emond, P., Besnard, J., Farde, L., Sedvall, G., 1999. Visualization of the dopamine transporter in the human brain postmortem with the new selective ligand [125I]PE2I. NeuroImage 9, 108–116. Halldin, C., Erixon-Lindroth, N., Pauli, S., Chou, Y.H., Okubo, Y., Karlsson, P., Lundkvist, C., Olsson, H., Guilloteau, D., Emond, P., Farde, L., 2003. [(11)C]PE2I: a highly selective radioligand for PET examination of the dopamine transporter in monkey and human brain. Eur. J. Nucl. Med. Mol. Imaging 30, 1220–1230. Hirvonen, J., Johansson, J., Teräs, M., Oikonen, V., Lumme, V., Virsu, P., Roivainen, A., Någren, K., Halldin, C., Farde, L., Hietala, J., 2008. Measurement of striatal and extrastriatal dopamine transporter binding with high-resolution PET and [11C] PE2I: quantitative modeling and test–retest reproducibility. J. Cereb. Blood Flow Metab. 28, 1059–1069. Ichise, M., Ballinger, J.R., Golan, H., Vines, D., Luong, A., Tsai, S., Kung, H.F., 1996. Noninvasive quantification of dopamine D2 receptors with iodine-123-IBF SPECT. J. Nucl. Med. 37, 513–520. Ichise, M., Liow, J.S., Lu, J.Q., Takano, A., Model, K., Toyama, H., Suhara, T., Suzuki, K., Innis, R.B., Carson, R.E., 2003. Linearized reference tissue parametric imaging methods: application to [11C]DASB positron emission tomography studies of the serotonin transporter in human brain. J. Cereb. Blood Flow Metab. 23, 1096–1112. Jucaite, A., Fernell, E., Halldin, C., Forssberg, H., Farde, L., 2005. Reduced midbrain dopamine transporter binding in male adolescents with attention-deficit/hyperactivity disorder: association between striatal dopamine markers and motor hyperactivity. Biol. Psychiatry 57, 229–238. Jucaite, A., Odano, I., Olsson, H., Pauli, S., Halldin, C., Farde, L., 2006. Quantitative analyses of regional [11C]PE2I binding to the dopamine transporter in the human brain: a PET study. Eur. J. Nucl. Med. Mol. Imaging 33, 657–668.
Laakso, A., Bergman, J., Haaparanta, M., Vilkman, H., Solin, O., Syvälahti, E., Hietala, J., 2001. Decreased striatal dopamine transporter binding in vivo in chronic schizophrenia. Schizophr. Res. 52, 115–120. Lammertsma, A.A., Hume, S.P., 1996. Simplified reference tissue model for PET receptor studies. NeuroImage 4, 153–158. Lammertsma, A.A., Bench, C.J., Hume, S.P., Osman, S., Gunn, K., Brooks, D.J., Frackowiak, R.S., 1996. Comparison of methods for analysis of clinical [11C]raclopride studies. J. Cereb. Blood Flow Metab. 16, 42–52. Leroy, C., Karila, L., Martinot, J.L., Lukasiewicz, M., Duchesnay, E., Comtat, C., Dollé, F., Benyamina, A., Artiges, E., Ribeiro, M.J., Reynaud, M., Trichard, C., 2012. Striatal and extrastriatal dopamine transporter in cannabis and tobacco addiction: a high-resolution PET study. Addict. Biol. 17, 981–990. Logan, J., Fowler, J.S., Volkow, N.D., Wang, G.J., Ding, Y.S., Alexoff, D.L., 1996. Distribution volume ratios without blood sampling from graphical analysis of PET data. J. Cereb. Blood Flow Metab. 16, 834–840. Meyer, P.T., Hellwig, S., Amtage, F., Rottenburger, C., Sahm, U., Reuland, P., Weber, W.A., Hüll, M., 2011. Dual-biomarker imaging of regional cerebral amyloid load and neuronal activity in dementia with PET and 11C-labeled Pittsburgh compound B. J. Nucl. Med. 52, 393–400. Odano, I., Varrone, A., Savic, I., Ciumas, C., Karlsson, P., Jucaite, A., Halldin, C., Farde, L., 2012. Quantitative PET analyses of regional [(11)C]PE2I binding to the dopamine transporter — application to juvenile myoclonic epilepsy. NeuroImage 59, 3582–3593. Ribeiro, M.J., Vidailhet, M., Loc'h, C., Dupel, C., Nguyen, J.P., Ponchant, M., Dollé, F., Peschanski, M., Hantraye, P., Cesaro, P., Samson, Y., Remy, P., 2002. Dopaminergic function and dopamine transporter binding assessed with positron emission tomography in Parkinson disease. Arch. Neurol. 59, 580–586. Ribeiro, M.J., Thobois, S., Lohmann, E., du Montcel, S.T., Lesage, S., Pelissolo, A., Dubois, B., Mallet, L., Pollak, P., Agid, Y., Broussolle, E., Brice, A., Remy, P., Group F.P.s.D.G.S., 2009. A multitracer dopaminergic PET study of young-onset parkinsonian patients with and without parkin gene mutations. J. Nucl. Med. 50, 1244–1250. Seki, C., Ito, H., Ichimiya, T., Arakawa, R., Ikoma, Y., Shidahara, M., Maeda, J., Takano, A., Takahashi, H., Kimura, Y., Suzuki, K., Kanno, I., Suhara, T., 2010. Quantitative analysis of dopamine transporters in human brain using [11C]PE2I and positron emission tomography: evaluation of reference tissue models. Ann. Nucl. Med. 24, 249–260. Slifstein, M., Laruelle, M., 2000. Effects of statistical noise on graphic analysis of PET neuroreceptor studies. J. Nucl. Med. 41, 2083–2088. Spencer, T.J., Biederman, J., Madras, B.K., Faraone, S.V., Dougherty, D.D., Bonab, A.A., Fischman, A.J., 2005. In vivo neuroreceptor imaging in attention-deficit/ hyperactivity disorder: a focus on the dopamine transporter. Biol. Psychiatry 57, 1293–1300. Svarer, C., Madsen, K., Hasselbalch, S.G., Pinborg, L.H., Haugbøl, S., Frøkjaer, V.G., Holm, S., Paulson, O.B., Knudsen, G.M., 2005. MR-based automatic delineation of volumes of interest in human brain PET images using probability maps. NeuroImage 24, 969–979. Warwick, J.M., Carey, P.D., Cassimjee, N., Lochner, C., Hemmings, S., Moolman-Smook, H., Beetge, E., Dupont, P., Stein, D.J., 2012. Dopamine transporter binding in social anxiety disorder: the effect of treatment with escitalopram. Metab. Brain Dis. 27, 151–158. Wu, Y., Carson, R.E., 2002. Noise reduction in the simplified reference tissue model for neuroreceptor functional imaging. J. Cereb. Blood Flow Metab. 22, 1440–1452. Yaqub, M., Tolboom, N., Boellaard, R., van Berckel, B.N., van Tilburg, E.W., Luurtsema, G., Scheltens, P., Lammertsma, A.A., 2008. Simplified parametric methods for [11C]PIB studies. NeuroImage 42, 76–86. Yaqub, M., Boellaard, R., van Berckel, B.N., Tolboom, N., Luurtsema, G., Dijkstra, A.A., Lubberink, M., Windhorst, A.D., Scheltens, P., Lammertsma, A.A., 2009. Evaluation of tracer kinetic models for analysis of [18F]FDDNP studies. Mol. Imaging Biol. 11, 322–333.