Methodology Poster Abstracts

P099. Imaging the sensitivity of [¹²³I]5-IA-85380 to increases in acetylcholine at the beta2-nicotinic acetylcholine receptors: physostigmine studies in human subjects

Irina Esterlis¹, Jonas Hannestad¹, Frederic Bois¹, John Seibyl², Marc Laruelle¹, Richard E. Carson³ and Kelly Cosgrove⁴

¹Yale University Department of Psychiatry, New Haven, Connecticut, USA; ²Institute for Neurodegenerative Disorders New Haven, Connecticut, USA; ³Yale University Department of Radiology, New Haven, Connecticut, USA; ⁴Yale University Departments of Psychiatry and Radiology, New Haven, Connecticut, USA

Background: Competition between neurotransmitter and radioligands has provided a very useful method to assess synaptic changes in dopamine, but this approach has been slow to extend to other neurotransmitter system. Previously, Fujita and colleagues showed that the high affinity beta2-nicotinic acetylcholine receptor (β2-nAChR) radioligand [¹²³I]5-IA-85380 (5-IA) may be sensitive to extracellular increases in ACh in baboons; 1 however, such an examination in humans has lagged. Given that acetylcholine is one of the major neurotransmitters in the brain and has been implicated in the psychiatric and medical illnesses, we developed a paradigm to interrogate the ACh system in vivo via use of 5-IA SPECT imaging and physostigmine, a centrally-acting acetylcholinesterase inhibitor.

Methods: Six healthy subjects (3 men, 3 women; 31±4.1 yrs) participated in one 5-IA SPECT study. 5-IA was administered as a bolus plus constant infusion (B/I 7.0 h); total injected dose was 390.2±13.2 MBq. After three 30-min baseline scans at 6-8 h post infusion, physostigmine (1-1.5 mg) was administered IV over 60 min, and nine additional 30-min scans were collected during the next 6 h. The outcome measure was BP_F (specific volume of distribution), calculated as VT/fp (estimated receptor availability) minus VND/fp (nondisplaceable binding; previously estimated in a smoking to satiety paradigm²).

Results: We observed a significant reduction in BP_F after physostigmine administration (25±15% reduction in cortical regions, 15±11% thalamus (Figure 1), 16±14% in striatum, and 35±34% in cerebellum; p<.05). This effect reflected a combination of a significant decrease in tissue concentration of 5-IA (7-16% region specific, p<.05) and a significant increase in plasma parent concentration (8%, p<.05).

Conclusions: These data suggest that physostigmine-induced increases in extracellular ACh might compete with 5-IA for binding to β2-nAChRs in humans, although other mechanisms, such as a direct effect of physostigmine on nicotinic acetylcholine receptors, should be ruled out. Additional validation of this paradigm is warranted, but we suggest that nicotinic imaging could be used to interrogate changes in synaptic ACh.

References

1. Fujita M, Al-Tikriti M, Tamagnan G et al. Influence of acetylcholine levels on the binding of a SPECT nicotinic acetylcholine receptor ligand [¹²³I]5-I-A-85380. Synapse 2003; 48:116-122.

2. Esterlis I, Cosgrove K, Batis J et al. Quantification of smoking induced occupancy of β2-nicotinic acetylcholine receptors: estimation of nondisplaceable binding. Journal of Nuclear Medicine 2010; 51:1226-1233.

P100. Correction for radiolabeled O-methyl metabolite in human brain during L-[β-¹¹C]DOPA PET

Keisuke Matsubara¹, Hiroshi Ito², Yoko Ikoma², Maki Okada², Masanobu Ibaraki¹, Kazuhiro Nakamura¹, Hiroshi Yamaguchi¹, Tetsuya Suhara² and Toshibumi Kinoshita¹

¹Department of Radiology and Nuclear Medicine, Research Institute for Brain and Blood Vessels Akita, Akita, Japan; ²Molecular Imaging Center, National Institute of Radiological Sciences, Chiba, Japan

Background: Dopamine synthesis rate, one of the presynaptic functions of the central dopaminergic system, can be measured by positron emission tomography (PET) measurement with L-[β-¹¹C]DOPA [1]. The radiolabeled O-methyl metabolite of L-[β-¹¹C]DOPA in peripheral organ, L-[β-¹¹C]O-methyl-DOPA (L-[β-¹¹C]OMD), suggested to penetrate the blood-brain barrier (BBB). Thus, L-[β-¹¹C]OMD may affect tissue radioactivity measured by PET and the endogenous dopamine synthesis rate estimated by kinetic analyses. However, the influence of L-[β-¹¹C]OMD in the living tissue for the kinetic analyses has not been investigated in detail. In the present study, to evaluate the influence of L-[β-¹¹C]OMD on the tissue time-activity curve (TAC) with L-[β-¹¹C]DOPA PET, the metabolite correction was applied to tissue TACs acquired from healthy volunteers.

Methods: The metabolite correction method proposed by Kumakura et al. in [¹⁸F]FDOPA PET study [2] was employed. In this method, component for O-methyl metabolite in the tissue TAC is estimated by compartmental analysis with two kinds of arterial input function for L-[β-¹¹C]DOPA and L-[β-¹¹C]OMD, and TAC in occipital cortex as a reference region with no irreversible binding. TAC in each brain region for L-[β-¹¹C]DOPA PET studies with ten healthy volunteers [1] was corrected by using the estimated L-[β-¹¹C]OMD TAC. This method assumes the distribution of O-methyl metabolite is uniform around the brain [3]. The endogenous dopamine synthesis rate (K_i) was estimated by Gjedde-Patlak plot analysis with the arterial input function and the metabolite-corrected TAC. The K_i was also estimated from the non-corrected TAC. For comparison to conventional analysis, relative influx constant (k_ref) was also estimated by Gjedde-Patlak plot analysis using TAC in occipital cortex, regarded as reference tissue input function. Data of 29–64 min and 29–89 min were used for linear regression in Gjedde-Patlak plot analysis.

Results: Calculated OMD component had only a marginal effect on tissue TAC (fraction of area under curve (AUC) of L-[β-¹¹C]OMD TAC in putaminal TAC: 9.4±2.1%). K_i with no metabolite corrected TAC correlated significantly to K_i with the metabolite correction (p<0.001, r=0.99 (29–64 min), p<0.001, r=0.99 (29–89 min)), and K_i were overestimated in all brain regions with no metabolite correction, see Figure A. k_ref also correlated to K_i with the metabolite correction, as shown in Figure B (p<0.001, r=0.99 (29–64 min), p<0.001, r=0.99 (29–89 min)).

Conclusions: The results suggest that the influence of the O-methyl metabolite L-[β-¹¹C]OMD to tissue TAC and kinetic parameters in L-[β-¹¹C]DOPA PET is marginal. This finding is accounted for by low fraction of L-[β-¹¹C]OMD in plasma in case of L-[β-¹¹C]DOPA, in contrast with high fraction of O-methyl metabolite for [¹⁸F]FDOPA. The results also suggest the net endogenous dopamine synthesis rate can be determined without the metabolite correction in case of L-[β-¹¹C]DOPA, not same as [¹⁸F]FDOPA.

References

[1]Ito H. et al., 2006, Nucl Med Commun 27:723-731.

[2]Kumakura Y. et al., 2005, J Cereb Blood Flow Metab 25:807-819.

[3]Doudet D.J. et al., 1991, J Cereb Blood Flow Metab 11:726-734.

P101. Does σ₁ ligand [¹¹C]SA4503 bind to the emopamil binding protein in the brain in vivo?

Jun Toyohara, Muneyuki Sakata and Kiichi Ishiwata

Tokyo Metropolitan Institute of Gerontology, Japan

Background: Carbon-11 labeled 1-[2-(3,4-dimethoxyphenyl)ethyl]-4-(3-phenylpropyl)piperazine ([¹¹C]SA4503) was shown to be a promising PET ligand for the mapping σ₁ receptors (σ₁Rs), and was applied to human subjects [1]. Using PET with [¹¹C]SA4503, we demonstrated that [¹¹C]SA4503 is useful for studying pathophysiology of neurological disorders such as Alzheimer's disease and Parkinson's disease and for evaluation of σ₁R occupancy by several therapeutic drugs [1]. However, [¹¹C]SA4503 has high affinity not only to σ₁Rs (K_i between 4 and 14 nM) but also to the emopamil binding protein (EBP) with K_i of 1.7 nM [2] and to the vesicular acetylcholine transporter (VAChT, K_i=50 nM) [3], which may affect the interpretation of neuroimaging findings. A rodent study confirmed that [¹¹C]SA4503 shows negligible binding to VAChT sites in vivo [4]. To our knowledge, no information is available about the possibility of [¹¹C]SA4503 binding to EBP in the brain in vivo.

Methods: To further confirm the selectivity of [¹¹C]SA4503, we carried out an in vivo blocking experiment using five blockers with high affinity for EBP and σ₁Rs: haloperidol (σ₁/EBP=0.006), trifluperidol (σ₁/EBP=0.002), ifenprodil (σ₁/EBP=2.3), tamoxifen (σ₁/EBP=12) and trifluoperazine (σ₁/EBP=52). Receptor blocking by each blocker was determined based on its ability to reduce the radioactivity in brain uptake of [¹¹C]SA4503. [¹¹C]SA4503 was co-injected with different amounts of each blocker into mice. The mice were killed by cervical dislocation at 30 min after injection. IC₅₀ values from binding displacement by blockers were determined using GraphPad Prism.

Results: The brain uptake of [¹¹C]SA4503 was dose-dependently decreased by SA4503 and high-affinity σ₁ blockers haloperidol (K_i, 1.6 nM), ifenprodil (K_d, 9 nM), and trifluperidol (K_i, 0.83 nM). The SA4503 and σ₁-selective haloperidol and trifluperidol showed the same in vivo IC₅₀ values (between 132 and 145 nM). The full block rates were more marked for SA4503 (73%) and ifenprodil (72%) than haloperidol (65%) and trifluperidol (60%). The rank order of in vivo IC₅₀ values of these four blockers tended to correlate with the in vitro σ₁R binding affinity. On the other hand, the blocking effects of the EBP blockers tamoxifen (K_i, 2.8 nM) and trifluoperazine (K_i, 3.9 nM) on the brain uptake of [¹¹C]SA4503 were negligible. The blood levels of [¹¹C]SA4503 were dose-dependently slightly increased by co-injection of σ₁ blockers. On the other hand, co-injection of EBP blockers did not affect the blood levels of [¹¹C]SA4503. As a result, dose-dependent blocking effects and in vivo IC₅₀ values calculated from the brain-to-blood ratio were not markedly different.

Conclusions: The contribution of in vivo EBP binding of [¹¹C]SA4503 was negligible in the mouse brain. Furthermore, in vivo IC₅₀ values of σ₁R-selective blockers tended to correlate with the in vitro K_i values toward σ₁Rs. From these results, we confirmed the σ₁R-selective binding of [¹¹C]SA4503 in the brain.

References

[1] Toyohara J et al. CNS Agents Med Chem 2009; 9:190–6.

[2] Berardi F et al. Bioorg Med Chem 2001; 9:1325–35.

[3] Shiba K et al. Bioorg Med Chem 2006; 14:2820–6.

[4] Ishiwata K et al. Nucl Med Biol 2006; 33:543–8.

P102. Reproducibility of PET studies with L-[β-11C]DOPA and [¹⁸F]FE-PE2I in healthy humans

Masayuki Suzuki¹, Hiroshi Ito¹, Harumasa Takano¹, Hironobu Fujiwara¹, Yasuyuki Kimura¹, Fumitoshi Kodaka¹, Andrea Varrene², Yoshiro Okubo³, and Tetsuya Suhara¹

¹Molecular Imaging Center, Molecular Neuroimaging Program, National Institute of Radiological Sciences, Chiba, Japan; ²Karolinska Institutet, Department of Clinical Neuroscience, Centre for Psychiatry Research, Karolinska Hospital, Stockholm, Sweden; ³Department of Neuropsychiatry, Nippon Medical School, Tokyo, Japan

Background: The dopaminergic neurotransmission system is of central interest in schizophrenia. While postsynaptic dopamine D2/3 receptors are assumed to be a major target for antipsychotic agents, recent positron emission tomography (PET) studies suggest that presynaptic dopaminergic function could also be modulated by antipsychotic agents [1]. Major presynaptic dopaminergic function could be measured by dopamine synthesis capacity and dopamine transporter availability. The L-[β-11C]DOPA uptake rate reflects dopamine synthesis capacity, and [¹⁸F]FE-PE2I binding reflects dopamine transporter availability [2]. Although quantification methods of both radioligands have been established, test-retest reproducibility, crucial for psychopharmacological trials, has not been confirmed. In the present study, we tested the reproducibility of L-[β-11C]DOPA and [¹⁸F]FE-PE2I in healthy humans.

Methods: Six healthy male volunteers (23.8±0.98 years) participated in the study. All subjects underwent two PET studies on separate days (7-day interval). Each PET study comprised two PET scans, one with L-[β-11C]DOPA and one with [¹⁸F]FE-PE2I. Volumes of interest (VOIs) were defined for the caudate and putamen. Reference regions were defined for the occipital cortex (L-[β-11C]DOPA) and the cerebellar cortex ([¹⁸F]FE-PE2I). VOIs were manually drawn on the co-registered individual magnetic resonance images corresponding to the PET images. Dopamine uptake rate constant (Ki) and dopamine transporter binding potential (BPND) were calculated using Gjedde-Patlak plot and simplified reference tissue model method, respectively. Then, reproducibility was assessed in terms of intra-subject variability (absolute variability) and reliability (intra-class correlation coefficient, ICC).

Results: Ki and BPND values were in agreement with previous studies. Ki and BPND of scans 1 and 2 for caudate were 0.0098±0.0009, 0.0103±0.0016, 3.07±0.60 and 2.99±0.41, respectively. Ki and BPND of PET scans 1 and 2 for putamen were 0.0107±0.0010, 0.0102±0.0008, 3.15±0.56 and 3.26±0.56, respectively. The absolute variability values of Ki and BPND were small for caudate (12.2%±8.8 and 9.9%±10.2, respectively) and putamen (4.5%±2.2 and 13.1%±11.0, respectively). The ICC values in the caudate and putamen showed fair-to-excellent reproducibility for L-[β-11C]DOPA (0.40 and 0.84) and [¹⁸F]FE-PE2I (0.80 and 0.68).

Conclusions: The present study demonstrated the rest-retest reproducibility of L-[β-11C]DOPA and [18F]FE-PE2I in healthy humans. Both radioligands showed minimum intra-subject variability and fair-to-excellent reproducibility. The results suggest that L-[β-11C]DOPA and [¹⁸F]FE-PE2I are both applicable for neuropharmacological evaluation of dopaminergic presynaptic function.

References

[1] Ito, H. et al. J Neurosci 2009; 29(43):13730-4.

[2] Varrone, A. et al. J Nucl Med 2011; 52(1):132-9.

P103. Correction of head movement by frame-to-frame image realignment on human brain PET images with [¹¹C]raclopride and [¹¹C]FLB457

Yoko Ikoma, Yasuyuki Kimura, Takahiro Shiraishi, Tetsuya Suhara and Hiroshi Ito

National Institute of Radiological Sciences, Chiba, Japan

Background: Positron emission tomography with [¹¹C]raclopride and [¹¹C]FLB457 has been utilized for imaging dopamine D₂ receptors in the striatum and extrastriatum, respectively. PET scans with these ligands require 60-90 min consecutive acquisition for a quantitative analysis. Therefore, head movement is often observed during scanning, and it hampers the reliability of quantitative outcomes, especially for small brain structures. Image-based motion correction with frame-to-frame realignment is a practical method, since it dose not require motion information obtained by an online tracking system. However, frame-to-frame realignment for emission PET images causes a mismatch between the emission scan and transmission scan used for the attenuation correction, and it may result in the error of quantitative analysis. In the present study, optimal image-based motion correction method including the transmission scan was evaluated for PET studies with [¹¹C]raclopride and [¹¹C]FLB457, and investigated the effect of this correction method on quantitative analysis outcomes.

Methods: In 60-min dynamic PET scan with [¹¹C]raclopride and 90-min dynamic PET scan with [¹¹C]FLB457, emission sinograms were reconstructed with or without attenuation correction. In the first method, 6 motion parameters (x-translation, y-translation, z-translation, roll, pitch, and yaw) were estimated frame-by-frame by realigning each frame of attenuation corrected image (AC) with a reference image using normalized mutual information. Four kinds of images were used as the reference image for realignment, those are, summed image of all frames, summed image of the first 2 min, one frame image with high count, and MR image coregistered to PET summation image. Meanwhile, in the second method, motion parameters were estimated by realigning each frame of non-attenuation corrected images (NAC) by a similar way. Then μ-map obtained from a transmission scan was realigned using estimated motion parameters to correct a mismatch between the transmission and emission scans [1, 2], and attenuation correction and reconstruction were performed sequentially. After motion correction, time-activity curves (TAC) for the striatum or frontal cortex were obtained. The binding potentials (BP) were estimated for these TACs by a simplified reference tissue model with the cerebellum as a reference region, and compared with BP estimates before the correction.

Results: The motion correction using NAC images could realign each frame image to every reference images for both [¹¹C]raclopride and [¹¹C]FLB457 studies, and the mismatch between the emission and transmission scan was also corrected using estimated motion parameters. However, the motion correction using AC images did not realign each frame image. After the correction with NAC, a discontinuity of TACs in the striatum and frontal cortex was improved. In the subject with the large movement during PET scanning, BP value of the striatum increased about 25% after the motion correction.

Conclusions: Head movement during the PET dynamic scan with [¹¹C]raclopride and [¹¹C]FLB457 could be accurately corrected by applying the image-based realignment to non-attenuation corrected images.

References

1. Mourik JEM et al. Eur J Nucl Med Mol Imaging 36:2002-2013, 2009.

2. Wardak M et al. J Nucl Med 51:210-218, 2010.

P104. Building and utilizing in silico tool to identify potential molecules for evaluation as preclinical tracers: facilitating rapid receptor occupancy assay development

Stuart Morton, Elizabeth Joshi, Dana Benesh, Vanessa Barth and Thomas Raub

Eli Lilly, Indianapolis, Indiana, USA

Background: To facilitate rapid iterations within a discovery SAR, it is critical to establish a direct translation between experimental in vivo pharmacology and target engagement. To this end, establishment of a target engagement assay is one way discovery teams are able to demonstrate a direct interaction, while coupling the outcome to measured pharmacokinetic/pharmacodynamic (PK/PD) data sets. A significant bottleneck in the development of such assays is the identification of appropriate tracers to evaluate and establish these in vivo receptor occupancy (RO) assays. In this study, we characterized a series of known PET ligands and non-radiolabeled cold tracers, covering a variety of targets including G protein-coupled receptors (GPCRs), intracellular enzymes, and transporters, in a mouse model of CNS exposure. Additionally, these compounds also were evaluated in a battery of in vitro assays to measure permeability, substrate recognition via P-glycoprotein (Pgp), and unbound fraction in plasma and brain homogenate.

Methods: Brain and plasma concentrations of known, non-radiolabeled PET tracers, CNS literature compounds, in-house CNS compounds and other non-radiolabeled cold tracers were measured by LC-MS/MS following an intravenous dose of approximately 1-mg/kg to mice in a previously described calibrated mouse brain uptake assay (MBUA). Additionally, cellular permeability and Pgp substrate recognition were evaluated in vitro using a Madin-Darby canine kidney (MDCK) cell line, while the unbound fraction in plasma and mouse brain homogenates were measured using equilibrium dialysis. These data were then compiled with in-house data from the above mentioned assays and used to develop a series of related in silico regression and classification models for each assay.

Results: A comparison of the data obtained for the known PET ligands to CNS literature compounds showed an ideal tracer ligand profile had marginal or no Pgp transport efficiency. Not too surprisingly, the considered tracer compounds also showed a higher effective permeability and free fraction giving rise to suitable free drug levels to engage the target. The data have resulted in a screening cutoff of f_u,brain >1% and lack of Pgp recognition in order to prioritize new compound selection. A comparison across the datasets also revealed that the PET ligands showed a higher percent of dose compared to the general CNS compounds.

Conclusions: The in vivo evaluation of preclinical and clinical tracers using the calibrated MBUA showed that the ligands had a different profile in regards to calculated and measured parameters as compared to general CNS targeted compounds. These data allowed us to establish general screening criteria for new tracer compound selection/prioritization that has been instrumental in driving our early tracer identification efforts in discovery.

P105. Quantification of human brain cannabinoid CB₁ receptors using a novel PET radioligand, [¹¹C]SD5024

Tetsuya Tsujikawa¹, Sami Zoghbi¹, Jinsoo Hong¹, Sean Donohue¹, Christer Halldin², Victor Pike¹, Robert Innis¹ and Masahiro Fujita¹

¹Molecular Imaging Branch, National Institute of Mental Health, Bethesda, Maryland, USA; ²Karolinska Institutet, Department of Clinical Neuroscience, Psychiatry Section, Stockholm, Sweden

Background: The cannabinoid sub-type-1 (CB₁) receptor is one of the most abundant G protein-coupled receptors in brain and is a promising target of therapeutic drug development. We have developed a novel CB₁ receptor radioligand from a 3,4-diarylpyrazoline structural class, namely [¹¹C]SD5024, [cyano-¹¹C](-)-3-(4-chlorophenyl)-N-[(4-cyanophenyl)sulfonyl]-4-phenyl-4,5-dihydro-¹H-pyrazole-1-carboxamidine (Donohue et al., J. Med. Chem. 2008, 51, 5608). This study aimed to evaluate the ability of [¹¹C]SD5024 to quantify CB₁ receptors in human brain.

Methods: Seven healthy subjects (30±6 y) had a brain PET scan with 11.3±4.8 mCi of [¹¹C]SD5024. Using a radiometabolite-corrected arterial input function and one and unconstrained two-tissue compartment models (1 and 2-TCMs), total distribution volume, V_T, was measured in ten regions. As methods with faster computation time, linear regression analysis of Logan and Ichise's bilinear analysis (MA1) were also used to calculate V_T.

Results: Brain activity peaked at ∼30 min at modest levels (1.5–3 SUV) and then decreased slowly to ∼80% of the peak at 120 min. 1-TCM showed good fitting in all datasets (Figure 1) whereas 2-TCM did not converge in ∼30% of the fittings. Regional V_T values from 1-TCM were consistent with known distribution of CB1 receptors, showing 3.2 in putamen, 2.6 in frontal cortex, and 1.7 in thalamus. Despite slow washout of the radioligand from brain, V_T by 1-TCM calculated from 60 and 120 min data differed by<10% (Figure 2) indicating that 60 min data provide adequate information on the pharmacokinetics and that influence from radiometabolites to V_T was low. Logan and MA1 analyses only slightly underestimated V_T across all brain regions compared with 1-TCM (∼6% and ∼5%, respectively), and regional V_T values by Logan and MA1 showed excellent correlations with those from 1-TCM (R²=0.97, P<0.0001 and R²=0.95, P<0.0001, respectively). Intersubject variability was small and similar among different methods by all of 1-TCM, Logan, and MA1 (∼26%). These intersubject variabilities were markedly smaller than that of the previously used [¹¹C]MePPEP (∼60%).

Conclusions: Binding of [¹¹C]SD5024 was measured accurately by all of 1-TCM, Logan, and MA1. [¹¹C]SD5024 is a promising radioligand for quantifying CB₁ receptors in human brain.

P106. Reproducibility measurements of [¹¹C]NOP-1A, a new PET radioligand to image nociceptin/orphanin FQ peptide (NOP) receptors in healthy subjects

Talakad Lohith¹, Sami Zoghbi¹, Cheryl Morse¹, Nancy Goebl², Johannes Tauscher², Victor Pike¹, Robert Innis¹ and Masahiro Fujita¹

¹Molecular Imaging Branch, National Institute of Mental Health, National Institutes of Health, Bethesda, Maryland, USA; ²Eli Lilly & Co., Indianapolis, Indiana, USA

Background: We have recently developed a new PET radioligand (S)-3-(2′-fluoro-6′,7′-dihydrospiro[piperidine-4,4′-thieno[3,2-c]pyran]-1-yl)-2-(2-fluorobenzyl)-N-methylpropanamide ([¹¹C]NOP-1A) to image and quantify nociceptin/orphanin FQ peptide (NOP) receptors in vivo in humans.^1,2,3 The aim of this study was to assess reproducibility of measuring [¹¹C]NOP-1A binding in healthy subjects based on both regional and voxel data.

Methods: Ten healthy subjects underwent test and retest scans, each of 2 hrs duration after bolus injection of 19±4 mCi of [¹¹C]NOP-1A along with arterial blood sampling for metabolite corrected input function. The test and retest scans were performed on same day separated by 1 h except for one subject where in scans were separated by 10 days. Using brain and arterial plasma data, the retest variability (RV, absolute difference between test and retest divided by mean of the two) of total distribution volume V_T was determined by both compartmental modeling on region-of-interest (ROI) data and bilinear MA1 analysis on voxel data.

Results: Similar to the previous study of single brain scan in each subject,³ in the current study, an unconstrained two-tissue compartment model provided good fitting and good identifiability of V_T based on ROI data. The rank order of the regional V_T values, ranging 5 to 13 mL cm⁻³, was similar to known distribution of the NOP receptors.⁴ The compartmental V_T showed moderate reproducibility with a fair RV averaging 14% for all brain regions. Also in line with the previous study,³ regional distribution of V_T by voxel-wise MA1 analysis correlated well with the V_T values by the compartmental analysis on ROI data. Voxel wise MA1 yielded reproducibility with RV ranging 13–0% (See Figure).

Conclusions: The reproducibility of V_T for [¹¹C]NOP-1A from both ROI and voxel data was moderately good. [¹¹C]NOP-1A is a potentially useful PET radioligand to image NOP receptors in clinical studies.

References

1.Pike VW et al. J Med Chem 2011, 54(8):2687-700.

2.Kimura Y et al. J Nucl Med 2011, 52(10):1638-45.

3.Lohith TG et al. J Nucl Med 2012 (E-pub ahead of print 6 February 2012).

4.Berthele A et al. Neuroscience 2003, 121(3):629-40.

P107. Feedback-controlled bolus plus infusion (FC-B/I) method for quantitative measurement of receptor binding with F-18-labeled PET ligand in living brain of conscious monkey

Hiroyuki Ohba, Shigeyuki Yamamoto, Norihiro Harada, Takeharu Kakiuchi and Hideo Tsukada

Central Research Laboratory, Hamamatsu Photonics K.K., Hamamatsu, Japan

Background: For the assessment of pharmacological effects in the living brain with PET, the steady-state method in combination of bolus injection and the following constant infusion of PET ligand (BI method) was proposed [1]. This method allows us to evaluate the drug effect on PET ligand binding in single PET scan, however the achievement of equilibrium state of radioactivity level in the target region of interest (ROI) is not so easy. We previously proposed the feedback-controlled bolus plus infusion (FC-B/I) method for quantitative assessment of drug-PET ligand interaction in the living brain of conscious monkey [2]. This system successfully detected the decrease in [¹¹C]raclopride binding in the conscious monkey striatum with cold-raclopride in a dose-dependent manner, however the PET ligand labeled with short half-life of C-11 (20 min) hampered the equilibrium state with large data fluctuation at late phase ca. 60 min and later. To solve the problem, N-2-[¹⁸F]fluoroethyl-4-piperidyl benzilate ([¹⁸F]4-FEPB; [3]), a PET ligand labeled with F-18 (H.L.=109 min) for muscarinic cholinergic receptor (mACh-R) imaging, was applied to thr FC-B/I method.

Methods: Young male monkeys (Macaca mulatta) were subjected to the PET study under conscious condition. The FC-B/I method used real-time reconstructed PET image data as an input to the feedback control for infusion rate of PET ligand to achieve equilibrium state in the ROI. The method using an animal PET scanner (SHR-7700, Hamamatsu Photonics, Japan) was carried out as follows. At the start of PET scan, one fifth volume of [¹⁸F]4-FEPB solution in 12 mL syringe was intravenously injected into monkey in bolus, followed by feedback controlled infusion. The radioactivity in ROI on the real-time reconstructed PET image (FBP Hanninig 8.0 mm) was monitored and the computer calculated the appropriate infusion rate of [¹⁸F]4-FEPB obtained by PID control rule every 1 min. When the equilibrium was obtained, the infusion rate was fixed. After the fixation of infusion rate of [¹⁸F]4-FEPB, scopolamine, an anti-mACh-R agent, was administered at the dose of 0.01 and 0.03 mg/kg.

Results: The equilibrium state of radioactivity levels were observed within 40 min. When scopolamine was intravenously administered, the radioactivity level of [¹⁸F]4-FEPB in the cortex was gradually lowered with time, and reached the second equilibrium state. The reduction degrees of radioactivity levels in the cortex showed significant dose-dependency (ca. 50 and 75% reduction at 0.01 and 0.03 mg/kg, respectively).

Conclusions: The conventional bolus injection method with C-11-labeled ligands can repeat twice and more in same subject in a day, however it cannot do with F-18-ligand because of its long half-life. In contrast, the FC-B/I method can provide the data of pre- and post-drug conditions in single scan. The stable equilibrium state as well as double data acquisition in single scan should be very beneficial for pharmacological PET research with F-18.

References

[1] Carson, RE et al., 1993, J Cereb Blood Flow Metab 13:24-42.

[2] Ohba, H et al., 2010, NeuroImage 52(Supplement 1):S28.

[3] Skaddan, MB et al., 2000, J Med Chem 43:4552-4562.

P108. Evaluation of two methods for automated PET region of interest analysis

Katarina Varnäs¹, Martin Schain^1,2, Zsolt Cselényi^1,3, Christer Halldin^1,2, Lars Farde^1,2,3 and Andrea Varrone^1,2

¹Karolinska Institutet, Department of Clinical Neuroscience, Stockholm, Sweden; ²Stockholm Brain Institute, Stockholm, Sweden; ³AstraZeneca R&D, Södertalje, 15185 Sweden

Background: Manual definition of regions of interest (ROIs) has been considered the reference standard method in PET data evaluation. Manual ROI definition may be labor-intensive, prone to rater bias and show low reproducibility. The introduction of automated template-based methods for ROI definition may overcome these limitations. The aim of the present study was to validate two automated methods for definition of regions of interest for the analysis of PET data obtained using the 5-HT_1B receptor radioligand [¹¹C]AZ10419369¹ in human subjects.

Methods: Ten human control subjects underwent PET examinations with the HR system and [¹¹C]AZ10419369.² ROIs were defined on T1-weighted MRIs by manual delineation using the software Human Brain Atlas,³ and by using two semi-automatic methods for ROI definition, i.e. the software package FreeSurfer⁴ and the AAL template.⁵ Regional binding potentials (BP_ND) were estimated from the extracted PET time activity curves using the simplified reference tissue model and the manually defined cerebellum as reference region. For the AAL method, BP_ND values were also determined using as a reference region, a version of the cerebellar ROI (CERCRU1 and CERCRU2) incorporated in the AAL template in which the boundaries of the ROI were limited to the lowest half of cerebellum, and were cropped to reduce spill-in from vascular structures. BP_ND values obtained with automatic methods were compared to those obtained using manually defined ROIs.

Results: The spatial overlap between ROIs defined with automatic and manual methods was 30% and 40% for the AAL and FreeSurfer ROIs, respectively. Absolute difference in BP_ND values obtained using FreeSurfer ROIs compared to manually defined ROIs were on average less than 5% for most cortical regions analyzed, including the anterior cingulate, dorsolateral prefrontal, orbitofrontal and occipital cortex. Mean absolute differences were between 5-10% in the dorsomedial prefrontal cortex, hippocampus and putamen. For most of the regions analyzed, results were comparable for the AAL and FreeSurfer methods (Figure 1), although AAL ROIs yielded larger differences in the hippocampus (25% vs. 5.1%) and the occipital cortex (-5.0% vs. -0.29%). For all regions analyzed, the AAL method using automatically defined reference region yielded lower BP_NDcompared to corresponding values obtained using the manually defined cerebellar ROI.

Conclusions: The good correspondence between automated and manual methods for ROI analysis in the majority of regions suggests that automated methods for ROI definition can be used for the analysis of [¹¹C]AZ10419369 PET data. FreeSurfer provides more anatomically precise regions and lower measurement error than AAL for some regions, such as the occipital cortex and hippocampus. BP_ND obtained using automated methods for target ROIs and manually defined reference region shows good agreement to that obtained using manually defined ROIs. Use of the template-cerebellum as reference region may, however, underestimate BP_ND, possibly due to inclusion of vascular radioactivity in the ROI.

References

1. Piersson et al. (2008). NeuroImage 41:1075-85.

2. Varnäs et al. (2011). J Cereb Blood Flow Metab 31:113-123.

3. Roland et al. (1994). Hum Brain Mapp 1:173-184.

4. Fischl et al. (2004). Cereb Cortex 14:11-22.

5. Tzourio-Mazoyer et al. (2002). NeuroImage 15:273-89.

P109. Combining image-derived and venous input functions for serotonin-1A receptor quantification

Andreas Hahn¹, Lukas Nics², Pia Baldinger¹, Richard Frey¹, Wolfgang Birkfellner³, Matthias Schütz², Wolfgang Wadsak², Markus Mitterhauser², Siegfried Kasper¹ and Rupert Lanzenberger¹

Medical University of Vienna, Austria: ¹Department of Psychiatry and Psychotherapy; ²Department of Nuclear Medicine; ³Center for Medical Physics and Biomedical Engineering

Background: Image-derived input functions (IDIF) provide a promising alternative to arterial blood sampling, but neither full independence of arterial samples nor routine application have been achieved yet [1]. We evaluate the combination of image-derived and venous input functions (IDIF+VIF) for quantification of the main inhibitory serotonin receptor subtype (5-HT1A) before and after hormone treatment as a proof-of-principle.

Methods: Fifteen PET measurements with arterial and venous blood sampling where obtained from 10 healthy women (mean age±SD=54.4±3.1 years) using the radioligand [carbonyl-11C]WAY-100635 (GE Advance, FWHM=4.36 mm). Eight of these scans were taken before and seven after hormone replacement therapy (8 weeks). Cerebral blood vessels were extracted from PET images by linear discriminant analysis and corrected for partial volume effects [2]. This blood vessel region-of-interest (ROI) was further refined that each voxel contains at least 66% of the maximum within the peak frame. The tail of the input function was given by manual venous blood samples from 10 min onwards and altogether corrected for delay, plasma to whole-blood ratio and radioactive metabolites [3] using only venous samples (IDIF+VIF). For validation, arterial input functions (AIF) were defined by automatic and manual arterial blood samples and similarly corrected (delay, plasma/whole-blood ratio, metabolites) using arterial samples. 5-HT1A binding potentials (BPP) were quantified separately using AIF and IDIF+VIF in PMOD3.3. A two-tissue compartment model was applied with K1/k2 fixed to cerebellar white matter [4]. ROIs were taken from an atlas comprising frontal, orbitofrontal, parietal, temporal, occipital and cingulate cortices, insula, amygdala-hippocampus complex, midbrain (incl.dorsal raphe nucleus) as well as cerebellar gray (excl.vermis) and white matter. Differences in arterial and venous manual samples were assessed by paired-samples t-tests. Direct comparison of 5-HT1A BPP between AIF and IDIF+VIF was carried out by linear regression analysis.

Results: For manual samples, no significant differences were found for whole-blood activity and plasma to whole-blood ratio from 10 min onwards (p>0.1). Variation of metabolite fractions were in line with reported values (maximum=24% at 5 min, [4]) with no significant differences after 20 min (p>0.12). Regression analysis showed strong agreement in 5-HT1A BPP between AIF and IDIF+VIF at baseline (R²=0.95, figure), after treatment (R²=0.93) and when pooling all scans (R²=0.93). Also, slopes were close to unity (0.97…1.07) and intercepts close to zero (-0.05…0.16).

Conclusions: This study validates the combined use of image-derived and venous input functions for a minimally invasive quantification of neuronal receptors. The consistent results between IDIF+VIF and AIF obtained for both baseline and treatment condition indicate great potential for routine use in clinical research protocols with complete substitution of arterial blood sampling [1].

References

[1] J Cereb Blood Flow Metab 31:1986-1998, 2011.

[2] J Nucl Med 39:904-911, 1998.

[3] Eur J Nucl Med Mol Imaging 37(S2):242, 2010.

[4] J Cereb Blood Flow Metab 27:185-197, 2007.

P110. Validation of population-based and image-derived input functions in the tracer kinetic modeling of [¹¹C](R)-rolipram in healthy volunteers and depressed subjects

Paolo Zanotti-Fregonara¹, Christina Hines¹, Sami Zoghbi¹, Jeih-San Liow¹, Yi Zhang¹, Victor Pike¹, Wayne Drevets², Carlos Zarate¹, Masahiro Fujita¹ and Robert Innis¹

¹NIMH, Bethesda, Maryland, USA; ²Oklahoma University School of Community Medicine, Tulsa, Oklahoma, USA

Background: Quantitative PET studies of neuroreceptor tracers typically require an arterial input function. To avoid arterial catheterization, population-based input function (PBIF) and image-derived input function (IDIF) are the two most common alternative approaches. In the literature, however, these approaches are often applied to small populations of homogenous subjects. The current study assess the accuracy and reproducibility of V_T values for [¹¹C](R)-rolipram derived using the PBIF and IDIF approaches in a large sample (n=51) of subjects for whom arterial input functions were available, and who either were psychiatrically healthy or had major depressive disorder (MDD).

Methods: The PBIF was generated using [¹¹C](R)-rolipram parent time-activity curves from 12 healthy volunteers (Group 1). The PBIF was scaled using arterial samples, which in a previous study by our laboratory were shown to provide more accurate Logan-V_T results than non-invasive scaling factors. Image data from the subjects in Group 1 also were used to determine the test-retest variability of V_T values obtained using the PBIF and IDIF approaches. The PBIF then was prospectively applied to an independent sample of healthy subjects (Group 2; n=25) and a sample of subjects with MDD (Group 3; n=26). The robustness of the PBIF approach was assessed further by studying subgroups of subjects drawn from Groups 2 and 3 who had distinctly shaped (i.e. “flatter” or “steeper”) input functions compared to the PBIF from Group 1. Logan-V_T values also were obtained with an IDIF from the carotid artery. In some subjects, we measured arteriovenous differences in [¹¹C](R)-rolipram concentration to see whether venous samples could be used instead of arterial samples.

Results: Retest variability of V_T data derived using the PBIF approach was similar to that obtained using the arterial input function and the IDIF approaches (14.5%, 15.2% and 14.1%, respectively). Excellent results were obtained when the PBIF was prospectively applied to the subjects in Group 2 (PBIF/reference V_T ratio 1.02±0.05; mean±SD) and Group 3 (V_T ratio 1.03±0.04). Only 4 subjects out of 51 had a V_T error >10%. Accurate results also were obtained for the subgroups whose input functions appeared “flatter” or “steeper” than the PBIF (V_T ratio 1.07±0.04 and 0.99±0.04, respectively). The IDIF approach yielded results of similar accuracy: IDIF/reference V_T ratio 0.99±0.05 and 1.00±0.04 for healthy subjects and depressed subjects, respectively. Only 1 of the 51 subjects had a V_T error >10%. Due to significant [¹¹C](R)-rolipram arteriovenous differences, venous samples could not be substituted for arterial samples.

Conclusions: Although an arterial line cannot be avoided, both PBIF and IDIF provide accurate and precise alternatives to full arterial input function for [¹¹C](R)-rolipram PET studies. This holds for both healthy subjects and MDD subjects, as well as for subjects with distinctly shaped input functions.

P111. Wavelet-based resolution recovery using an anatomical prior provides quantitative recovery for human population phantom PET [¹¹C]raclopride data

Miho Shidahara¹, Charalampos Tsoumpas², Colm McGinnity³, Takashi Kato⁴, Hajime Tamura¹, Alexander Hammers⁵, Hiroshi Watabe⁶ and Federico Turkheimer³

¹Tohoku University Graduate School of Medicine, Japan; ²King's College London, UK; ³Imperial College London, UK; ⁴National Institute for Longevity Sciences, Aichi, Japan; ⁵The Neurodis Foundation, Lyons, France; ⁶Osaka University Graduate School of Medicine, Japan

Background: The objective of this study was to evaluate a resolution recovery (RR) method using a variety of simulated human brain [¹¹C]raclopride PET images.

Methods: Simulated datasets of 15 numerical human phantoms [1] were processed by a wavelet-based resolution recovery method using an anatomical prior [2]. The anatomical information was in the form of a hybrid segmented-atlas, which combined an atlas for anatomical labelling [3] and a PET image for functional labelling of each anatomical structure. We applied RR to both 60-minute-static and dynamic PET images. For BP_ND images, the simplified reference tissue model (SRTM) with basis pursuit [4] was performed using PMOD software, with the cerebellum as a reference region. Recovery was quantified in 84 regions, comparing the typical ‘true' value for the simulation, as obtained in normal subjects; simulated and RR PET images. In order to evaluate the clinical applicability of SFS-RR, we further applied it to two human [¹¹C]raclopride data.

Results: The radioactivity concentration in the white matter, striatum, and other cortical regions was successfully recovered for the 60-minute static image of all 15 human-phantoms; the dependency of the solution on accurate anatomical information was demonstrated by the difficulty of the technique to retrieve the subthalamic nuclei due to mismatch between the two atlases used for data simulation and recovery. SFS-RR improved quantification in the caudate and putamen, the main regions of interest, from -30.1% and -26.2% to -17.6% and -15.1%, respectively, for the 60-min static image and from -51.4% and -38.3% to -27.6% and -20.3% for the BP_ND image, respectively. After the recovery, the mean BP_ND of two healthy controls were improved by 46.9 and 29.6% in caudate and 29.0 and 22.1% in putamen, respectively (Figure 1).

Conclusions: The proposed methodology proved effective in the resolution recovery of small structures from brain [¹¹C]raclopride PET images. The improvement is consistent across the anatomical variability of a simulated population as long as accurate anatomical segmentations are provided.

References

[1] Reilhac A, Batan G, Michel C et al. IEEE Nucl Sci 2005, 52:1321-8.

[2] Shidahara M, Tsoumpas C, Hammers A et al, NeuroImage, 2009, 44:340-8.

[3] Hammers A, Allom R, Koepp MJ. Hum Brain Mapp 2003, 19:224-47.

[4] Gunn RN, Gunn SR, Turkheimer FE et al. J Cereb Blood Flow Metab 2002, 22:1425-39.

P112. Rapid metabolite analysis of PET radioligands by direct plasma injection combining micellar cleanup with fast-micellar liquid chromatography

Ryuji Nakao¹, Magnus Schou^1,2 and Christer Halldin¹

¹Karolinska Institutet, Stockholm, Sweden; ²AstraZeneca R&D, Södertälje, Sweden

Background: Radiometabolite analysis in plasma during a PET study is one of the most important components for accurate pharmacokinetic modelling of PET radioligands. There are some limitations arising from the short half-lives of radionuclides which limit the number of samples that can be analyzed as well as resulting in low radioactivity of plasma samples collected towards the end of a PET study. Radiometabolite analysis involves a sample pretreatment step aimed at quantitative reduction of matrix components. This is generally achieved manually and makes the analysis procedure long, tedious and may introduce additional sources of error. Recently, we investigated the use of fast-LC [1] and micellar-LC (MLC) [2] techniques for PET metabolite analysis. The fast radio-LC method benefit from high speed and sensitivity, thus allowing great numbers of samples to be analysed.In the MLC procedure, plasma samples are directly injected onto a LC column, providing a more accurate estimation of the metabolite corrected input function. The aim of this work is to develop a column-switching system for the metabolite analysis of a wide variety of PET radioligands in plasma employing direct injection based on the combination of fast-LC column with micellar mobile phase.

Methods: Step A; An aliquot of diluted plasma sample taken from monkey or human was directly injected onto the extraction column for sample cleanup. Proteins and polar radiometabolites were flushed with micellar mobile phase (M1) contained 0.5–10% acetonitrile with 25–50 mM sodium dodecyl sulfate (SDS), while parent radioligand and its lipophilic radiometabolites were retained on the extraction column. Step B; At 1 min the switching valve was switched and mobile phase was changed to allow forward flush the trapped compounds from the extraction column onto the analysis column using a mixture of micellar and sub-micellar eluents (M1/M2) containing low (0.5–10%) and high (40–70%) concentration of acetonitrile with 25–50 mM SDS. Step C; After the subsequent elution of unchanged radioligand and its lipophilic radiometabolites the mobile phase was changed to M1 and re-equilibrated for 1 min. And then switching and manual injection valves were switched back to their original positions.

Results: Under the optimum conditions, complete separations of target PET radioligands from their radiometabolites were achieved within 4 min with good peak efficiency. This system allowed metabolite fractions to be determined for all samples (up to 22 per 90 min PET study) to obtain an arterial input function for ¹¹C-labeled radioligands. The high sensitivity provided by the present method (5-times better than standard-LC) enabled radiometabolite analysis with lower sample volume (∼0.5 mL plasma).

Conclusions: Column switching radio-MLC is a suitable technique for the determination of PET radioligands in plasma. This technique allows direct plasma injection and highly sensitive radiometric measurement for routine metabolite studies of PET radioligands. Furthermore, it provides rapid and reliable analysis of a large number of plasma samples for improved estimation of metabolite-corrected input functions.

References

[1] Nakao R. et al. J Label Compd Radiopharm: 54:S138, 2011.

[2] Nakao R. et al. J Label Compd Radiopharm: 54:S139, 2011.

P113. Initial evaluation of [¹¹C]-CURB in human brain using positron emission tomography

Pablo Rusjan, Junchao Tong, Romina Mizrahi, Isabelle Boileau, Steve Kish, Sylvain Houle and Alan Wilson

PET Centre, Centre for Addiction and Mental Health, Toronto, Canada

Background: Fatty acid amide hydrolase (FAAH) is the enzyme responsible for metabolizing the endogenous cannabinoid, anandamide. As such FAAH plays a key role as a gatekeeper within the endocannabinoid signaling system and thus represents an important target for molecular imaging. Ex-vivo animal studies (Wilson AA, Nucl. Med. Biol. 2011) demonstrate that [¹¹C]-CURB has good brain uptake, regional heterogeneity, specificity and irreversibly bound to FAAH. In this work we present the initial evaluation of [¹¹C]-CURB human brain using PET.

Methods: Five healthy subjects (age 19-53 years; gender 3F/2M) were scanned in a High Resolution Research Tomograph (HRRT) for 90 minutes after an i.v. bolus injection of [¹¹C]-CURB (350±29 MBq, 1.6±0.5 μg). The bolus was administered with a syringe pump over 1 minute. Manual and automatic arterial samples were collected for [¹¹C]-CURB metabolization studies. An unmetabolized parent compound in plasma function was created and used as input function for the kinetic modeling. Regions of interest were automatically delineated using ROMI software (Rusjan PM, Psychiatry Res. 2006).

Results: Following an intravenous bolus injection of [¹¹C]-CURB, time-activity curves (TACs) showed a peak [standard uptake value (SUV) of 3.6-4.5] at 3 min followed by washout then a plateau (SUV 2.3-2.8) after 30 minutes. Activity was distributed in all the gray matter regions. Based on the average SUV after 30 minutes the highest uptake of radioactivity was seen in the thalamus and putamen, intermediate in the cerebellum, temporal cortex and insula cortex and lowest in the caudate, occipital cortex, frontal cortex and anterior cingulate cortex. Analysis of plasma with HPLC showed that the radioactive metabolites are more polar than the parent compound, unlikely to cross the blood-brain barrier, and consistent with our previous animal studies. The metabolization was faster at the beginning (only 48% of parent compound after 20 min) but became slower at later time (At 90 min post-injection there was still 37% of parent compound). An irreversible 2-tissue compartment model (2TCMi) with arterial input function fitted the TACs (Figure) and provides an identifiable net influx constant (Ki=K1.k3/(k2+k3)) (COV<4%). Average values of Ki ranged from 0.085 to 0.10 mL cm⁻³ min⁻¹ and the rank order of Ki values were similar to the SUVs described above. The values of Ki at 60 min were within 5% of those at 90 min with similar identifiability (COV<5%).

Conclusions: PET scanning of human with [¹¹C]-CURB is feasible. Initial kinetic modeling of the TAC with arterial input function shows that TACs can be fitted with a 2TCMi with 60 minutes of data.

Acknowledgements: This project was supported by Canada Foundation for Innovation, and the Ontario Ministry of Research and Innovation.

P114. New insights into levodopa induced dopamine release in Parkinson's disease

Vesna Sossi, Marjorie Gonzalez, Katherine Dinelle, Nicole Heffernan, Jess McKenzie, Silke Appel-Cresswell, Martin McKeown and A. Jon Stoessl

University of British Columbia, Vancouver, Canada

Background: Dopamine (DA) release is associated with motor function, expectation and reward. Stimulus-induced DA release can be estimated by positron emission tomography (PET) using a double ¹¹C-raclopride (RAC – a D₂-type receptor antagonist) scan, with one scan performed at baseline and the second after or during stimulus administration. Changes in synaptic DA levels, loosely referred to as ‘DA release', are traditionally estimated from the difference in the tissue input binding potential (BP_ND) between the baseline and the stimulus-related scan due to RAC and DA competition for the same binding sites. We developed a novel method to evaluate response to stimulus, based on the change in shape of the tracer distribution within an anatomically defined region. This method was applied to the estimate of levodopa induced DA release.

Methods: Five Parkinson's disease (PD) subjects (UPDRS 18±11, range 11-38) were scanned with ¹¹C-dyhydrotetrabenazine (DTBZ, a VMAT2-based marker of dopaminergic integrity ) RAC at baseline and 1 h after levodopa. PET data, summed over 30-60 min after tracer administration, were realigned to individuals' MRIs for accurate identification of the caudate, putamen and ventral striatum. A 3D moment invariant analysis was applied to the tracer distribution within each region providing region specific scaling, translation and rotation invariant spatial descriptors of the tracer distribution. The descriptors were a combination of variance, skewness and kurtosis. The change in their values between the two RAC scans was taken as a measure of the effect of DA release and was evaluated as a function of DTBZ BP_ND and DA release. All three PET measures were evaluated as a function of clinical motor scores.

Results: The change in spatial variance (Δvar) of the RAC distribution within each ROI provided the most statistically robust results. Significant correlations were only observed in the putamen, consistent with the expected action of levodopa in PD. A positive correlation was observed between Δvar and DTBZ BP_ND (r=0.7, p=0.03) and a negative correlation between Δvar and UPDRS (r=0.81, p=0.005). Likewise a strong negative correlation was observed between DTBZ BP_ND and UPDRS scores (r=0.78, p=0.008) and no correlation between Δvar and DA release. No significant correlation was observed between DA release and UPDRS or DTBZ BP_ND, except for the most affected side, where a trend toward a negative correlation between DA release and DTBZ BP_ND was found (r=0.83, p=0.08).

Conclusions: The change in the variance of the tracer distribution was correlated significantly with clinical and PET measures of disease severity with the largest change observed for a more intact putamen and better clinical performance likely reflecting localized DA release consistent with the presynaptic function being relatively intact. There correlations opposite to what has been found for traditionally estimated magnitude of DA release in a larger population sample. While the interpretation of the findings will be refined with a larger population sample, these data show that the magnitude and the spatial distribution of DA release provide complementary information.

P115. lp-ntPET finds significant fast dopamine responses to cigarette smoking in sub-regions of the striatum

Su Jin Kim¹, Jenna M. Sullivan², Kelly P. Cosgrove^2,3 and Evan D. Morris^2,3

¹Yale University, New Haven, Connecticut, USA; ²Yale PET Center, Yale University, New Haven, Connecticut, USA; ³Department of Psychiatry, Yale University, New Haven, Connecticut, USA

Background: To use PET imaging to detect and characterize transient dopamine elevations in the brains of smokers smoking cigarettes.

Methods: Three male smokers received a bolus plus constant infusion of [¹¹C]raclopride in two conditions: a baseline and a scan that included smoking two consecutive cigarettes inside the PET scanner. Subjects began smoking 45 minutes after the initial tracer bolus. Event-by-event motion correction was incorporated into reconstruction using data from the Polaris tracking system. Dynamic data were filtered with HYPR (Christian, 2010) to increase the SNR of PET TACs while preserving spatial resolution and relevant temporal behavior. Constrained estimation of dopamine responses by voxel was accomplished with our previously published and validated linearized ntPET (‘lp-ntPET') model (Normandin, 2012). Ninety minutes of dynamic data were fitted everywhere within an anatomically-based dorsal striatal mask. The operational equation was implemented with basis functions that encode a wide range of dopamine responses. The possible responses were constrained to take-off no earlier than 5 minutes prior to the start of smoking and no later than a latest-allowable time (e.g., 70 min). Only significant responses were retained. Significance was determined by F-test comparing the fit of each voxel-wise TAC with lp-ntPET to its corresponding fit with standard reference model, MRTM (no time-variation in dopamine).

Results: All three subjects displayed rapid responses in ventral striatum shortly after smoking. One subject showed significant responses in dorsal caudate as well. Data are not corrected for multiple comparisons however it is important to note that there were no statistically significant responses at any voxels found in the same area during a baseline condition. These transient elevations of dopamine are generally too short-lived to be quantified by the conventional change in binding potential. In fact, the mean binding potential change in right ventral striatum in the three subjects was 0+/-15%. Shape and prevalence of significant responses were not dependent on choice of “latest-allowable” time.

Conclusions: In a preliminary cohort of three smokers undergoing two [¹¹C]raclopride scans, our recently-introduced linearized ntPET model detects consistent, short-lived dopaminergic responses in ventral striatum during smoking. These responses are not reliably detectable with conventional analysis because they are localized in space (to activated sub-regions of the ventral striatum) and localized in time (they don't produce long-term depression in raclopride binding). Inconsistent results across previously published reports on the dopaminergic response to smoking may be attributable to the inadequacies of the conventional methods. Our results suggest that new modeling methods are necessary to characterize non-steady responses to drugs and behavior captured by dynamic PET.

P116. 5-HT_1A sex based differences in BP_ND are due to K_Dapp and not B_max

Dustin Wooten, Ansel Hillmer, Jeffrey Moirano, Elizabeth Ahlers, Maxim Slesarev, Dana Tudorascu, Todd Barnhart, Mary Schneider and Bradley Christian

University of Wisconsin, Madison, USA

Background: Previous PET studies have shown sex based differences in 5-HT_1A binding potential (Parsey et al. Jovanovic et al.). The nondisplaceable binding potential (BP_ND) is a binding index representing receptor density (B_max), in vivo ligand-receptor affinity (1/K_Dapp), and nonspecific binding (f_ND). In this work, we use a multiple-injection (MI) PET protocol and the 5-HT_1A receptor antagonist, [¹⁸F]mefway (MEF), to compare sex differences of K_Dapp and B_max in addition to BP_ND in a cohort of rhesus monkeys.

Methods: Three-injection MI PET studies were performed on 18 (7 m, 11f) rhesus monkeys which included partial saturation injections of MEF. The studies were optimized for precise measurement of K_Dapp (via k_off/k_on) and B_max. Arterial plasma sampling of MEF was acquired to provide input functions for the modeling of the mefway concentration. A group based input function was derived from 11 of the studies in which arterial sampling was acquired and used for each subject after scaling for weight and injected specific activity. Compartmental modeling was performed using a model to account for nontracer doses of MEF for the estimation of B_max and K_Dapp. BP_ND estimates were acquired using the MRTM and data from the first injection (high specific activity MEF, 90 minute duration). The cerebellum was used as a reference region for BP_ND estimation. Voxel-based analysis of the spatially normalized BP_ND images was used to determine regions of significant sex differences in binding and to guide regional selection for ROI analysis for the estimation of K_Dapp and B_max.

Results: The parameter estimates of B_max and K_Dapp were within 15% when comparing the individual and group based plasma input function and there were no sex based differences in the shape or clearance of the MEF input functions. Voxel-based analysis of BP_ND revealed the greatest difference (females>males) within the amygdala-hippocampal complex (am-hip), using a threshold of p<0.001 (660 mm³ bilateral cluster size). Stepwise reduction of the significance threshold revealed higher MEF binding in females in most regions of high 5-HT_1A expression, including mesial temporal, cingulate, frontal and parietal cortices and raphe nuclei. MI analysis of the am-hip region revealed no difference in B_max (m:f, 22.9±5.7; 19.2±4.3 pmol/mL, p<0.14), however, there was a significant difference in K_Dapp (m:f, 4.0±0.6; 2.9±0.6 pmol/mL, p<0.0012). Uncertainties in the individual parameter estimates, calculated using Monte Carlo techniques, were approximately 5% and 7% for K_Dapp and B_max, respectively.

Conclusions: Our results indicate that in the am-hip, the increased MEF BP_ND in females is due primarily to lower K_Dapp and not increased B_max compared to males. The reduced K_Dapp in females could be due to decreased endogenous serotonin levels compared to males and is consistent with reports measuring serotonin synthesis (Sakai et al.).

References

Parsey R et al. (2004) Brain Research.

Jovanovic H et al. (2008) NeuroImage.

Sakai Y et al. (2006) NeuroImage.

P117. Novel spatial analysis method for PET data using 3D moment invariants: applications to Parkinson's disease

Marjorie Gonzalez¹, Katherine Dinelle², Nicole Heffernan², Jessamyn McKenzie², Silke Cresswell², Martin McKeown², Jon Stoessl² and Vesna Sossi¹

¹University of British Columbia,Vancouver, Canada; ²Pacific Parkinson's Research Center, Vancouver, Canada

Background: We propose to use the spatial characteristics of positron emission tomography (PET) tracers' distributions within anatomically-defined Regions Of Interest (ROIs) to extract additional information about pathological states beyond that available from pharmacokinetic modeling. We hypothesize that these spatial characteristics will be sensitive to disease-related changes and apply them to PET images of Parkinson's disease (PD), which is characterized by progressive loss of dopaminergic terminals starting in the posterior putamen and spreading towards the caudate.

Methods: We used 3D moment invariants (3DMIs) to characterize the spatial distribution of PET tracers. 3DMIs are mathematical spatial descriptors designed to be invariant to scaling, translation and rotation; those used here are a combination of terms describing spatial variance, skewness and kurtosis. This method has been used in fMRI studies to assess the spatial characteristics of voxel-based statistics and has been shown to be a powerful and sensitive method to characterize brain activation. Crucially, this allows characterization of the spatial distribution of activation without the need to warp brain images to a common brain template. We use 3DMIs to characterize PD-related dopaminergic deficit within anatomically-defined ROIs for the presynaptic tracers ¹⁸F-fluorodopa (FD) and ¹¹C-dyhydrotetrabenazine (DTBZ) and the post-synaptic tracer ¹¹C-Raclopride (RAC). We studied five PD patients (UPDRS=18±11, range 13-38) and five healthy controls. All subjects underwent an MRI scan for coregistration of the PET images and accurate identification of the putamen, caudate and ventral striatum.

Results: We found clear differences in the 3DMIs of PD patients from those of healthy controls: the spatial variance between the two groups was significantly different for DTBZ (var_DTBZ) and FD (var_FD) in the putamen (p<10⁻⁵) and caudate (p<0.007). No significant differences were found in the spatial variance of PD patients and healthy controls for RAC (var_RAC) in the putamen (p=0.55) or caudate (p=0.15). In the ventral striatum, significant differences in the spatial variance of PD patients and healthy controls were seen for all tracers (p<0.02). In addition, we found a significant positive correlation between UPDRS and var_DTBZ in the putamen (r=0.85, p=0.001), caudate (r=0.63, p=0.03), and ventral striatum (r=0.60, p=0.04). In contrast, less significant or no correlations were found between UPDRS and BP_DTBZ in the putamen (r=0.75, p=0.008), caudate (r=0.56, p=0.05), and ventral striatum (r=0.36, p=0.16). Significant correlations were also found between UPDRS and var_FD in the putamen (r=0.68, p=0.04). These results show that the tracers' spatial distribution can be a more sensitive measure of PD clinical stage than traditional pharmacokinetic parameters.

Conclusions: We find that the spatial characteristics of the radiotracer distribution provide valuable information about pathological states. Here we show that this method can distinguish between healthy controls and PD patients, as well as provide an independent and more clinically-meaninful measure of disease progression. This novel analysis method shows great promise to extract additional information from PET data that has traditionally been omitted and can have a wide range of applications in diagnosis, staging, treatment assessment, etc. Promising initial results for its use in helping to characterize treatment-related responses in PD patients are shown separately.

P118. Parametric mapping of [¹⁸F]FP-CIT dynamic PET based on graphical analysis for the assessment of abnormal dopamine transporter binding in Parkinson's disease

Seongho Seo¹, Jae Sung Lee², Hye Bin Yoo¹, Jee-Young Lee³, Yu Kyeong Kim², Beom S. Jeon³ and Dong Soo Lee²

Seoul National University, South Korea: ¹Department of Brain and Cognitive Sciences; ²Department of Nuclear Medicine, ³Department of Neurology

Background: [¹⁸F]FP-CIT PET studies have drawn increasing interest for dopamine transporter (DAT) imaging to study Parkinson's disease (PD), due to the tracer's fast kinetics and high affinity to DAT. Since FP-CIT binding has mostly been quantified by analyzing region of interest (ROI) kinetics, it is hard to find studies using parametric image approach. Here, the applicability of parametric image method in quantification of DAT binding and detection of PD-related abnormality in the binding, was investigated by evaluating conventional and new graphical analysis methods based on reference tissue input.

Methods: For 9 controls and 18 PD patients, dynamic 90-min PET scans were acquired without blood sampling. For striatal ROI kinetics, the binding potential (BP) of DAT was estimated by various graphical methods using the cerebellum as a reference region, including Logan analysis, the recently developed RE plot and RE plot combined with Gjedde-Patlak analysis (RE-GP) [1]. Each method was evaluated by comparing the resulting BP estimates across the methods and to that of simplified reference tissue model (SRTM) considered the most reliable for ROI kinetics of FP-CIT [2]. Next, BP parametric images were generated by each method and compared with each other. The images were spatially normalized to FP-CIT template and smoothed with 12 mm FWHM Gaussian kernel for statistical analysis; PD group was compared to control group to test for significant difference in BP parametric image, for Logan and RE. Due to the necessity of further optimization, RE-GP method was excluded from the comparison.

Results: In all ROI kinetics, BP estimates by RE and RE-GP were highly correlated with those obtained by Logan: y=0.69x+0.11, r=0.99 for RE; y=1.03x+0.01, r=1.00 for RE-GP (Figure 1A). Although all of the graphical approaches underestimated BP compared to SRTM, they were well correlated and RE-GP BPs were best matched to SRTM BPs (Logan: y=0.60x+0.18, r=0.93; RE: y=0.42x+0.24, r=0.91; RE-GP: y=0.65x+0.15, r=0.96). For parametric images, Both RE and RE-GP showed smoother images than Logan avoiding spatial inconsistency by noise-induced bias (Figure 1B). For detection of abnormal binding in PD, both Logan and RE hardly found significant regions in the comparison between whole PD group and control group at significance level of uncorrected p=0.05. For the reduced binding in subgroup of PD patients, however, RE detected significant regions in striatum comparable to those by Logan, regardless of relative underestimation in RE parametric image (Figure 1C).

Conclusions: RE and RE-GP plots provided parametric mapping of striatal DAT binding that were more robust to noise than Logan analysis despite underestimating BP than SRTM. RE plot shows ability of detecting binding changes similar to Logan.

References

[1] Zhou et al., 2010, Multi-graphical analysis of dynamic PET. Neuroimage 49:2947-2957.

[2] Yaqub et al., 2007, Quantification of dopamine transporter binding using [¹⁸F]FP-β-CIT and positron emission tomography, JCBFM 27:1397-1406.

P119. Visual assessment of parametric [¹¹C]PIB images in a memory clinic population

Marissa D. Zwan^1,2, Rik Ossenkoppele^1,2, Ronald Boellaard², Albert D. Windhorst², Pieter G. Raijmakers², Human Adams², Adriaan A. Lammertsma², Philip Scheltens¹, Wiesje M. van der Flier¹ and Bart N.M. van Berckel²

¹Department of Neurology & Alzheimer Center, VU University Medical Center, Amsterdam, The Netherlands; ²Department of Nuclear Medicine & PET Research, VU University Medical Center, Amsterdam, The Netherlands

Background: Alzheimer's disease (AD) is characterized by amyloid plaques, which can be visualized by [¹¹C]PIB and PET. The purpose of this study was to evaluate the agreement between visual assessments of parametric [¹¹C]PIB images derived from three different analytical methods in a memory clinic population.

Methods: Dynamic [¹¹C]PIB PET scans of 90 minutes duration were obtained for 30 AD patients, 30 patients with non-AD dementia, 30 patients with mild cognitive impairment (MCI) and 30 healthy controls. Parametric images of non-displaceable binding potential (BP_ND) were generated using receptor parametric mapping with fixed efflux rate constant for the cerebellum, which was used as reference tissue (RPM2). In addition, averaged 60-90 min activity concentration (summed) images and standardised uptake value ratio images (SUVr) were generated, in the latter case again using cerebellum as reference tissue. All images were classified in a binary fashion as positive or negative by three independent nuclear medicine physicians, who were blinded to diagnosis. Inter-reader agreement was determined using the Fleiss kappa ((f)κ). In addition, the optimal [¹¹C]PIB BP_ND cut-off value was determined for correct classification of images using ROC analysis.

Results: Visual agreement between readers differed between analytical methods with excellent agreement for parametric BP_ND images ((f)κ=0.88), but only good agreement for SUVr ((f)κ=0.68) and moderate agreement for summed images ((f)κ=0.57). Disconcordance between readers was only seen in groups other than AD. The optimal global [¹¹C]PIB BP_ND cut-off value was 0.19, with a sensitivity of 98% and a specificity of 97%.

Conclusions: Visual assessment of parametric BP_ND images showed substantially better inter-reader agreement than that of SUVr and summed images. Parametric BP_ND images are clearly better for assessment of amyloid pathology in non-AD dementia patients, MCI patients, and healthy controls. Caution is needed when interpreting SUVr and summed images in these groups.

P120. Validation of an automated method for quantitative evaluation of ¹¹C-raclopride PET data based on a PET-template

Felix Pierre Kuhn¹, Cyrill Burger², Chantal Martin Sölch³, Katharina Ledermann³ and Alfred Buck¹

¹Medical Radiology Department, Nuclear Medicine, University Hospital Zurich, Switerland; ²PMOD LTD, Zurich, Switzerland; ³Department of Psychiatry, University Hospital Zurich, Switzerland

Background: ¹¹C-raclopride is used for the differentiation of Parkinson disease (PD) from atypical parkinsonian syndromes. An accurate quantification of the receptor binding and the comparison with reference values is mandatory to establish a reliable clinical diagnosis. The widely applied indirect normalization of PET data into stereotactic space with the help of a MRI is primarily based on the cortical outline of the MR images and therefore the subcortial regions are often not adequately transformed. Direct image normalization based on a ¹¹C-raclopride template might yield a more reliable normalization of the subcortical regions and enable an automated placement of volumes of interest (VOI) outlining the putamen and the caudate nucleus (striatum).

Methods: PET-data of 5 healthy subjects acquired 40-50 min. post injection of 250 MBq ¹¹C-raclopride was used to generate a raclopride-PET template (PET/CT Discovery STE, GE Healthcare). To validate this template 10 healthy subjects underwent a raclopride-PET and volumetric T1 weighted MR imaging (MR 3T Achieva, Philips Medical Systems). First, all PET images were normalized to the stereotaxic Montreal Neurological Institute (MNI) brain template using the corresponding MR images. VOIs determined by the previously validated Hammers-atlas were manually adjusted in 3 spatial dimensions according to the MRI to precisely outline the striatum. Second, all PET-images were directly normalized to the raclopride-template and the Hammers-VOIs were automatically positioned. Finally, mean tissue radioactivity concentrations were obtained from all VOIs on the direct and indirect spatially normalized PET-data. The striatal-to-cerebellar ratio (SCR) of raclopride uptake was calculated as this relationship is widely used as a quantitative parameter in PET imaging. All data processing was done using PMOD software (PMOD LTD, Zurich, Switzerland).

Results: The SCR values obtained based on the PET template had good agreement and high correlation with the manual analysis based on the MRI. Pearson's correlation coefficient yielded 0.95, which is significant at the 0.01 level (2-tailed). The Bland–Altman plot showed narrow limits of agreement with 39 of 40 measurements within the 1.96 standard deviation range.

Conclusions: The PET-template based method for automated quantification of ¹¹C-raclopride receptor binding is simple to implement and allows for fast, robust and reliable image analysis. There was no significant difference between the SCR values obtained by this method and by the more demanding MRI-based VOI placement. The presented method alleviates the clinical workflow and increases the reproducibility of the measurements.

P121. Spectral analysis iterative filter for voxel-wise quantification of PET tracers with irreversible uptake

Mattia Veronese¹, Kathleen Schmidt², Carolyn Smith², Gaia Rizzo¹, Federico Turkheimer³ and Alessandra Bertoldo¹

¹Department of Information Engineering, University of Padova, Padova, Italy; ²Section on Neuroadaptation & Protein Metabolism, National Institute of Mental Health, Bethesda, Maryland, USA; ³Division of Experimental Medicine, Imperial College London, London, UK

Background: Spectral Analysis Iterative Filter (SAIF) is a Spectral Analysis (SA) method developed for quantifying rates of cerebral protein synthesis (rCPS) at both ROI and voxel levels in L-[1-¹¹C]Leucine PET studies [1, 2]. The method implements a filtering procedure that optimizes estimation of irreversible uptake of tracer in tissue (K_i, ml/g/min), transport of tracer from plasma to tissue (K₁, ml/g/min), and apparent distribution volume of tracer (V_T,a, ml/cm³). As with other SA techniques, SAIF provides useful information about tracer kinetics without assuming a specific compartmental model. This study examines applicability of SAIF for voxel-wise quantification of several PET tracers with irreversible uptake.

Methods: Three datasets were analyzed: L-[1-¹¹C]Leucine [2] (6 subjects), [¹¹C]SCH442416 [3] (5 subjects) and [¹⁸F]FDG [4] (1 subject). Voxel-wise analysis was performed with SAIF and unfiltered SA, and results compared to other voxel-wise methods: a Basis-Function-Method (BFM) that assumes a homogenous tissue kinetic model for L-[1-¹¹C]Leucine, Weighted Non-Linear Least Squares (WNLLS) applied to a two tissue compartment model for [¹¹C]SCH442416, and a Patlak plot for [¹⁸F]FDG. Relative differences between methods of voxel estimates were evaluated for the principal parameters of interest: rCPS (L-[1-¹¹C]Leucine), V_T,a ([¹¹C]SCH442416), and K_i ([¹⁸F]FDG). Means of parameter estimates over all voxels within a ROI were also computed.

Results: Analysis of L-[1-¹¹C]Leucine data showed very good agreement between SAIF and BFM: relative differences in rCPS voxel estimates in a representative subject were 5%±24% (mean±SD). Means of all voxels within a ROI determined with SAIF and BFM were highly correlated (Figure, Panel A). Unfiltered SA estimates were not in the physiological range and excluded from further analysis. With [¹¹C]SCH442416, high correlation and limited relative differences in V_T,a across ROIs (+2%±4%) were detected between WNLLS and SAIF (Figure, Panel B). Unfiltered SA did not provide reliable estimates (V_T,a mean relative difference: +54%±12%). [¹⁸F]FDG data analysis with SAIF demonstrated excellent agreement with Patlak results (K_i mean±SD relative differences over all brain voxels: 6.7%±12.1%) (Figure, Panel C). For all tracers SAIF provided high quality parametric maps consistent with those of their respective reference methods.

Conclusions: SAIF results were in good agreement with reference voxel analytic methods and better than results based on unfiltered SA. SAIF was shown to be adaptable for analysis of different irreversibly trapped tracers without requiring a specific compartmental model. Unlike Patlak analysis, SAIF provides a complete description of both reversible and irreversible tracer kinetic components. SAIF, therefore, represents a robust and valid alternative for voxel analysis of PET tracers with irreversible uptake.

References

[1] Veronese et al., JCBFM 2010.

[2] Veronese et al., JCBFM 2012 (in press).

[3] Hinz et al., Molecular Imaging and Biology 2003.

[4] Bertoldo et al., IEEE Trans Biomed Engr 1998.

P122. Improved diagnostic performance of florbetaben β-amyloid PET via partial volume effect correction

Henryk Barthel¹, Michael Rullmann¹, Jürgen Dukart², Stefanie Herbert¹, Julia Luthardt¹, Hermann-Josef Gertz¹, Cornelia Reininger³ and Osama Sabri¹

¹University of Leipzig, Germany; ²University Lausanne, Switzerland; ³Bayer Healthcare, Berlin, Germany

Background: In recent years, β-amyloid (Aβ) brain PET imaging has emerged as a valuable tool to support the clinical diagnosis of Alzheimer's disease (AD). Florbetaben is a promising 18F-labeled Aβ-targeted PET tracer [Lancet Neurol 2011] currently in Phase 3 clinical development. In principle, in Aβ PET imaging, concomitant neocortical atrophy reduces the PET signal and as such the potential of this “hot spot” imaging technique to diagnose AD. The question of whether partial volume effect correction (PVEC) might improve the accuracy of Aβ PET has not been systematically addressed so far.

Methods: To answer this question, we analyzed florbetaben PET and MRI data obtained in a recently published Phase 0 trial [Eur J Nucl Med Mol Imaging 2011] of 10 patients with probable AD and 10 age-matched healthy volunteers (HVs). The image data were co-registered, and the grey matter (GM) was segmented using SPM5. PVEC was performed employing the PVELab tool by using the voxel-based modified Müller-Gärtner method. Regional standardized uptake values (SUVs) were compared between GM-segmented and GM-segmented + PVEC PET data. Mesial temporal lobe atrophy was determined by the Sheltens scale, and regional GM volumes by voxel-based morphometry.

Results: PVEC had no influence on the reference region cerebellar cortex SUVs. In contrast, different neocortical SUVs in AD patients increased after PVEC to a larger extent than these in HVs. The resulting composite SUV ratios (SUVRs) increased after PVEC by 12.2±6.2 for ADs vs. 3.4±3.2% for HVs (p=0.001). Effect sizes (Cohen's d) for AD vs. HV separation increased by PVEC from 1.44 to 1.56 for the composite and from 0.08 to 0.69 for the mesial temporal cortex SUVRs. Of interest, the mesial temporal cortex SUVR increase by PVEC was correlated with the Sheltens scale (r=0.92, p<0.001), and that of the composite SUVRs with the composite GM volume (r=-0.71, p<0.001).

Conclusions: From these results we conclude that, in florbetaben Aβ PET imaging, PVEC is able to correct for brain atrophy. Consequently, PVEC increases the power of florbetaben PET to discriminate between AD patients and HVs. To substantiate these preliminary findings, larger sample size PET data from a multi-center trial are currently being analyzed.

Acknowledgements: This research was supported by Bayer Healthcare (Germany).

P123. Evaluation of PET quantification sensitivity to the arterial input function modeling

Gaia Rizzo¹, Mattia Veronese¹, Federico E. Turkheimer², Paolo Zanotti-Fregonara³, Masahiro Fujita³, Sami S. Zoghbi³, Robert B. Innis³, Kathleen C. Schmidt⁴, Carolyn B. Smith⁴ and Alessandra Bertoldo¹

¹Department of Information Engineering, University of Padova, Padova, Italy; ²Division of Experimental Medicine, Imperial College London, London, UK; ³Molecular Imaging Branch, National Institute of Mental Health, Bethesda, Maryland, USA; ⁴Section on Neuroadaptation & Protein Metabolism, National Institute of Mental Health, Bethesda, Maryland, USA

Background: Quantification of PET data commonly requires parent tracer measurement, which is obtained via manual or automatic blood sampling. Sometimes, a model-based description of the arterial samples is preferred to actual measured data because of the errors and noise inherent to blood measurements. We evaluated the impact of the most commonly used arterial models on the quantification results.

Methods: Five models were considered: segmented (straight line from time zero to the peak followed by the sum of three exponentials to describe washout [1]), model proposed by Feng et al [2], Gamma-variate [3], and the Feng and Gamma models constrained to the peak of arterial measurements. Four datasets with different blood sampling methods (manual: L[1-11C]leucine, [11C]-(R)-rolipram; automatic: [11C]WAY100635, [11C]FLB457) were used to measure the Residual Sum of Squares (RSS) between the arterial samples and each model-predicted values and then the impact on quantitative parameters, i.e. rate of cerebral protein synthesis (rCPS) for L[1-11C]Leucine, obtained region-wise with Spectral Analysis Iterative Filter (SAIF. [4]) and the distribution volume (V_T) obtained with a two-tissue compartmental model with the other tracers. Mean and standard deviation of the percent relative difference in parameters estimated with the arterial model predictions and those estimated with either arterial measurements (manually sampled data) or the segmented model (automatically sampled data) were computed.

Results: The Feng and segmented models best described the arterial data with a comparable RSS, although their impacts on PET quantification were different. We must stress that, due to measurement error, we do not know the true arterial values and we cannot quantify which methods best describes the true input function. In L[1-11C]leucine Feng and segmented results were in agreement with arterial measures for both rCPS and all the microparameters (Figure, Panel A). For [11C]-(R)-rolipram there was a good agreement for V_T (Figure, Panel B) but for the microparameters the discrepancy was up to 10% (with 50% of variability). For [11C]WAY100635 and [11C]FLB457 the methods substantially differed for both V_T (difference >10%) and microparameters (>30%) (Figure, Panel C/D). This was due to different model fitting to the data in both the peak and the tail of the curve. Notably, the higher noise of automatic sampling prevented the direct use of the measures without preliminary modeling. As compared to Feng, the other methods showed a relative RSS difference >300% and unreliable quantification results.

Conclusions: Quantification of PET kinetic data may be heavily influenced by the choice of the input function fitting model. Different models characterized by a good agreement with measured data may nevertheless lead to important differences on the final estimates of both macro and microparameters.

References

[1] Parsey et al (2000) JCBFM 20:1111-1133.

[2] Feng et al (1997) IEEE Trans Inform Technol Biomed 1:243-254.

[3] Lee et al (2010) Magn Reson Med 63:1305–1314.

[4] Veronese et al (2010) JCBFM 30:1460-1476.

P124. Image derived input function using a multivariate analysis method based on pair-wise correlation between PET-image voxels

Martin Schain^1,2, Simon Benjaminsson^2,3, Katarina Varnäs¹, Anton Forsberg^1,2, Christer Halldin^1,2, Anders Lansner^2,3,4, Lars Farde^1,2,5 and Andrea Varrone^1,2

¹Karolinska Institutet, Department of Clinical Neuroscience, Stockholm; ²Stockholm Brain Institute, Stockholm; ³Royal Institute of Technology, Stockholm; ⁴Stockholm University, Stockholm; ⁵AstraZeneca R&D, Södertalje, 15185, Sweden

Background: Measurement of arterial input function is a prerequisite for kinetic modeling of PET-data, particularly for radioligands without suitable reference regions. The measurement is laborious and requires cannulation of a peripheral artery, a procedure associated with patient discomfort and potential adverse events. A non-invasive procedure would thus be preferable. Several methods aiming at deriving the blood curve directly from PET-images have been suggested. These methods have provided successful results in a limited number of cases and seem to be applicable only to few radioligands.¹ The challenge has been to reliably extract PET-signals originating from blood vessels, due to factors related to radioligand behavior, blood vessel anatomy and PET-system resolution. A novel voxel clustering algorithm has recently been described for fMRI data. The method calculates the correlation between all voxel pairs in order to visualize connectivity patterns in brain.² The present study was aimed at evaluating whether a similar approach applied to PET-data acquired with the High Resolution Research Tomograph (HRRT) could cluster voxels suitable for extracting image derived input functions (IDIFs). Objectives: 1) To apply a pair-wise correlation method to PET-data of [¹¹C]raclopride (dopamine D₂-receptor), [¹¹C]flumazenil (GABA_A-receptor), and [¹¹C]AZ10419369 (5-HT_1B-receptor)³; 2) To compare the quantifications using IDIF and measured arterial input function (MIF).

Methods: For each radioligand, 6 subjects were examined using the HRRT. MIFs were measured in all cases. For the [¹¹C]AZ10419369-data, venous samples were withdrawn to verify the agreement between arterial and venous blood radioactivity at late time points. Using a supercomputer, the Pearson correlation coefficient between time-activity curves was calculated for all voxel pairs located in an image volume covering the neck. The carotid artery was segmented from co-registered MR-images by applying an intensity threshold. Subsequently, those voxels displaying a similar temporal pattern to the segmented MR-voxels were selected as an artery mask. To compensate for signal loss during the early phase, the first 3 minutes of the measurement were corrected for spill-out of radioactivity using the method described by Rousset et al. (reduced to one compartment).⁴ For all radioligands, late arterial samples were used to scale the IDIF. IDIFs obtained from [¹¹C]AZ10419369-data were also scaled using one venous sample obtained at the end of the PET-acquisition. Distribution volume (V_T) was calculated with Logan graphical analysis, using both MIFs and IDIFs.

Results: IDIFs were successfully extracted for all subjects. Average (SD) percent difference between V_T -values obtained with IDIF and MIF were 4.0(11.9) for [¹¹C]flumazenil, 0.8(11.2) for [¹¹C]raclopride, and -2.8(5.2) and -0.2(10.7) for [¹¹C]AZ10419369 when scaling with arterial and venous samples respectively (Figure 1).

Conclusions: The aforementioned method enables estimation of regional V_T values with good agreement to those obtained with MIFs. The [¹¹C]AZ10419369-data provides further support that radioactivity levels in late venous samples can approximate that in arterial blood for scaling of the IDIF. Similar applications to other radioligands are however needed for further verification.

References

1 Zanotti-Fregonara et al. (2011). J Cereb Blood Flow Metab 31:1986-1998.

2 Benjaminsson et al. (2010). Front Syst NeuroSci 4:1-7.

3 Piersson et al. (2008). NeuroImage 41:1075-85.

4 Rousset et al. (1998). J Nucl Med 39:904-911.

P125. Improved mapping and quantification of serotonin transporter availability in the human brainstem with the HRRT

Martin Schain^1,2, Miklos Tóth¹, Zsolt Cselényi^1,3, Ryosuke Arakawa⁴, Christer Halldin^1,2, Lars Farde^1,2,3 and Andrea Varrone^1,2

¹Karolinska Institutet, Department of Clinical Neuroscience, Stockholm, Sweden; ²Stockholm Brain Institute, Stockholm, Sweden; ³AstraZeneca R&D, Södertalje, 15185 Sweden; ⁴Molecular Neuroimaging Group, Molecular Imaging Center, National Institute of Radiological Sciences, Chiba, Japan

Background: The serotonin system is involved in many physiological functions such as sleep, mood and cognition, and clinical conditions such as depression, anxiety and Parkinson's disease [1]. The serotonergic neurons originate from the raphe nuclei in the brainstem, and reliable in vivo estimates of the receptor/transporter availability in the raphe are thus of interest in many research areas [2]. Previous studies have shown that binding potential (BP_ND) cannot be reliably estimated in such small regions using region-based kinetic modeling [3]. The unreliability is primarily due to high noise levels in PET data, which has previously been addressed with parametric imaging. Lastly, due to the very small size of the brainstem nuclei, the resolution of the system might be a profound factor for accurate quantification. The overall aim of this study was to examine the potential to evaluate radioligand binding in the small brain nuclei, and to assess the effect of improved resolution on the outcome measures.

Methods: For comparative purposes, radioligand binding was measured with both the ECAT EXACT HR (resolution ∼4.5 mm FWHM) and the High Resolution Research Tomograph (HRRT, ∼1.5 mm FWHM [4]) systems. Six subjects were examined with both systems on the same day under similar experimental settings, using the serotonin transporter radioligand [¹¹C]MADAM [1]. Parametric images of binding potential (BP_ND) were obtained using a wavelet aided approach (WAPI) which has previously been shown to successfully reduce noise levels in PET-images [5]. Regions of interest (ROIs) were delineated with a threshold-based semiautomatic delineation procedure developed in our group, and included the ventral midbrain, superior colliculi, dorsal raphe, median raphe, and a joint ROI covering raphe magnus and raphe obscurus. Regional BP_ND-values were estimated by applying the ROIs to the parametric images, and the percent difference in BP_ND between the systems was calculated.

Results: Signal for [¹¹C]MADAM binding was obtained for five brainstem structures (Figure 1). Overall, the HRRT provided 30-40% higher BP_ND values than the HR (p=0.0017). The difference between the systems was not dependent on the threshold used in the ROI selection procedure.

Conclusions: The aforementioned methodology enables estimation of [¹¹C]MADAM binding in the small brainstem nuclei. Using the recovery coefficient in a 10 mm sphere in the NEMA phantom as a measure of a system's resolution, the difference inBP_ND between the HR and the HRRT was mainly attributable to their disparate resolutions. BP_ND-values estimated with WAPI provided substantially reduced across-subject variability than those previously reported using kinetic analysis. This procedure could therefore be considered when studying receptor/transporter availability in the brainstem. Test-retest reproducibility studies are required to further understand the reliability of the present approach.

References

[1] Halldin, C. et al., Synapse 2005; 58(3):173-83.

[2] Hornung, J.P., J Chem Neuroanat 2003; 26(4):331-43.

[3] Schain, M. et al., Neuroimage 2011; 60(1):800-807.

[4] Varrone, A. et al., Eur J Nucl Med Mol Imaging 2009; 36(10):1639-50.

[5] Cselenyi, Z. et al., Neuroimage 2006; 32(4):1690-708.

P126. Voxel-wise quantification of [¹¹C](R)-rolipram PET data in human brain

Gaia Rizzo¹, Mattia Veronese¹, Paolo Zanotti-Fregonara², Masahiro Fujita², Robert B. Innis² and Alessandra Bertoldo¹

¹Department of Information Engineering, University of Padova, Padova, Italy; ²Molecular Imaging Branch, National Institute of Mental Health, Bethesda, Maryland, USA

Background: [¹¹C](R)-rolipram is a PET ligand for the in vivo quantification of phosphodiesterase 4, an enzyme that metabolizes cAMP, a prevalent second messager [1]. Our study aimed at selecting the best methodology for voxel-wise quantification of [¹¹C](R)-rolipram, comparing standard methods versus data-driven and hierarchical approaches.

Methods: Ten healthy subjects underwent a 90-min PET scan following injection of 421±144 MBq of [¹¹C](R)-rolipram, complete with metabolite-corrected plasma input functions. First, ROI analysis was performed to set and validate the application of data-driven quantification methods (Spectral Analysis, SA [2], and Rank Shaping, RS [3]) compared to the 2TCM. Then, SA, RS, Logan analysis [4] and 2TCM solved with Hierarchical Basis Function Method (2TCM-HBFM, [5]) were used to measure the volume of distribution (V_T) at the voxel level. The results were compared with the estimates obtained voxel-wise with 2TCM solved with weighted nonlinear estimator (2TCM-WNLS), corrected for failures and outliers before the comparison.

Results: At the ROI level all methods provided comparable V_T results with 2TCM, with high correlation (R²>0.95). Logan analysis tended to underestimate V_T, especially in high binding ROIs, while RS showed a constant bias (+9%) probably due to a bias introduced by the method filtering procedure. SA provided the best performance with a computational time x3 lower than 2TCM. 2TCM-WNLS applied voxel-wise required high computational time (up to x5 compared to the other methods), and image quality was generally poor. On average, 50% of the voxels were eliminated because their values were physiologically implausible or inaccurate. Although 6% of the voxels had a negative V_T, 2TCM-HBFM showed one of the highest correlations (R²>0.97), the best agreement with 2TCM-WNLS estimates (Figure, Panel A) and the lowest percentage relative difference (Figure, Panel B). Logan plot results presented good correlation but the higher underestimation among the methods. SA performed better than RS (SA: R²>0.98; RS: R²>0.96) but the image quality was poorer, and 5% of the voxels had to be eliminated. Similarly to ROI results, RS showed a constant bias (-5%) of V_T, probably due to the filtering procedure.

Conclusions: 2TCM-WNLS high computation time and poor image quality point to the necessity of an alternative method for quantification at the voxel level. Compared to the other methods, 2TCM-HBFM showed more accurate V_T quantification with good quality parametric maps and excellent agreement with 2TCM-WNLS results. Consequently 2TCM-HBFM was selected as the best alternative to the gold standard approach for voxel-wise quantification of [¹¹C](R)-rolipram data. Among the data-driven methods, SA represented the best approach both at the ROI and voxel level.

References

[1] Zanotti-Fregonara et al. (2011) Neuroimage 54: 1903-1909.

[2] Cunningham and Jones (1993) JCBFM 13:15-23.

[3] Turkheimer et al. (2003) Phys Med Biol 48:3819-3841.

[4] Logan et al. (1990) JCBFM 10:740-747.

[5] Rizzo et al. (2011) BrainPET Conference 2011, 707.

P127. Can non-invasive analysis detect group differences in 5HT_2A receptor binding measured using ¹¹C-MDL100907 PET?

Sophie Holmes, Jose Anton-Rodriguez, Philip Noonan, Rainer Hinz and Peter Talbot

Wolfson Molecular Imaging Centre, University of Manchester, UK

Background: ¹¹C-MDL100907 is a PET radioligand with high selectivity for the serotonin 2A receptor (5-HT_2AR). The two-tissue compartment model (2TCM) with arterial plasma input function has been validated as the optimum analytic method.^1,2 However, as the need for invasive methods generally reduces widespread clinical utility and patient acceptability, investigation of the sensitivity and accuracy of non-invasive methods to detect group differences is warranted. In comparison to the 2TCM, several reference tissue analytic methods underestimate the binding potential (BP) although it remains highly correlated with the 2TCM BP.^2,3 The aim of this study was therefore to investigate whether those simplified methods, despite underestimating BP, are still are able to provide a significant group discrimination.

Methods: ¹¹C-MDL100907 data were analysed from a previous clinical study which demonstrated significantly lower (-8.6%, p=0.034) cortical 5-HT_2AR availability (BP_ND) in a clinical cohort (high impulsive aggression [high-IA], n=13) in comparison to a control (low-IA, n=11) cohort using 2TCM with arterial plasma input function. Cortical BP_ND values were calculated using the following reference tissue models, each using the cerebellum time-activity curve as reference region input, and compared to the 2TCM: i) region of interest (ROI) analysis using the simplified reference tissue model (SRTM); ii) voxel-wise parametric mapping using Logan's graphical analysis (LGA); and iii) a target to cerebellum ratio during the final 30 minutes of the 110-minute scan to provide estimates of apparent BP_ND. BP_ND was compared between reference tissue methods and 2TCM using linear regression to obtain correlations (r²), and statistical bias measured. Group differences across all regions were measured for each analytic method using univariate analysis of variance (ANOVA).

Results: Compared to the 2TCM, estimates of BP_ND using SRTM showed substantial negative bias (-50±8%, see Table) across all 8 ROIs, and moderate correlation (r²=0.65). The magnitude of the between-group difference was slightly underestimated (-6.1% vs -8.6%) and reached trend statistical significance (F=2.78, p=0.098). LGA BP_ND estimates also showed substantial negative bias, although to a lesser extent than SRTM (-33%±11%) and was moderately correlated with 2TCM (r²=0.61). The group difference was highly underestimated (-1.6% vs -8.6%) and non-significant (F=0.17, p=0.68). The ratio method yielded apparent BP_NDvalues with considerable negative bias (-49±10%), displayed the least correlation with 2TCM (r² =0.56), and the group difference was highly underestimated (-3.5% vs -8.6%) and non-significant (F=2.05, p=0.15).

Conclusions: Consistent with previous reports, when compared to the 2TCM with arterial input function the non-invasive reference tissue methods give ¹¹C-MDL100907 BP_ND values with a substantial negative bias. For group discrimination, the non-invasive methods underestimated the magnitude of the group difference and failed to detect a significant group difference. Nevertheless, for SRTM the underestimation was small and the significance reached trend level. LGA and the ratio method performed poorly. Our data suggest that SRTM may potentially be suitable for discriminating group differences using ¹¹C-MDL100907 with larger cohorts or clinical groups with a greater between-group difference in 5-HT_2AR availability, and warrants further investigation.

References

1. Hinz R et al. (2007) Journal of Cerebral Blood Flow and Metabolism 27:161-172.

2. Talbot PS et al. (2012) Neuroimage 59:71-285.

3. Meyer PT et al. (2010) Neuroimage 50:984-993.

P128. In vivo characterization of binding potential, standardized brain uptake (SUV), and brain kinetics in rodents by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) of non-labeled positron emission tomography (PET) imaging tracers predicts translational imaging outcomes in nonhuman primates and humans

Vanessa Barth, Elizabeth Joshi, Dana Benesh, Charles Mitch, Zhaogen Chen, Thomas Raub, Lee Phebus, Anne Need, Karen Rash and John Schaus

Eli Lilly, Indianapolis, Indiana, USA

Background: Positron emission tomography (PET) imaging has become a useful noninvasive technique to explore molecular biology within living systems; however, the utility of this method is limited by the availability of tracers to probe specific targets and disease biology. Improvements in the ability to predict small molecule performance as tracers prior to radiolabeling would help alleviate this rate limiting factor. In this study, we characterized the brain penetration, or peak SUV (standardized uptake value), binding potential (BP), and brain kinetics of a series of known, non-radiolabeled PET ligands using in vivo LC-MS/MS and correlated these parameters with reported PET ligand performance in non-human primates and humans.

Methods: Brain concentrations of known, non-radiolabeled PET tracers were measured by LC-MS/MS following an intravenous micro-dose to rodents. The PET tracers studied included those reported to label G protein-coupled receptors (GPCRs), intracellular enzymes, and transporters. The aforementioned parameters were compared to published values in the literature, where available, for both nonhuman primate and human.

Results: The LC-MS/MS values for both BP and SUV were different from the published values for the PET tracers; however, the data set enabled the identification of cut off values for the in vivo LC-MS/MS data output. Binding potentials of greater than 1.5 and peak SUVs of greater than approximately 150% at 5 minutes post-dose in rodents positively predicted PET tracer performance in higher order species. The brain kinetics appeared similar between both techniques despite differences in tracer dose.

Conclusions: In vivo LC-MS/MS evaluation of nonlabeled tracers in rodents can predict PET ligand performance of small molecules by assessing three key parameters: BP, peak SUV, and kinetics. This approach reduces the need for rodent PET imaging studies while increasing the chemical space from which to identify tracers.

P129. Parametric methods for [¹¹C]PE2I positron emission tomography

My Jonasson¹, Lieuwe Appel¹, Jonas Engman², Andreas Frick², Dag Nyholm³, Håkan Askmark³, Torsten Danfors¹, Jens Sörensen¹, Tomas Furmark² and Mark Lubberink¹

Uppsala University, Sweden: ¹Nuclear Medicine & PET; ²Psychology; ³Neurology

Background: The dopamine transporter (DAT) is an important protein for regulation of central dopamine concentration and DAT deficiency has been associated with several neurodegenerative and neuropsychiatric disorders. [¹¹C]PE2I is a highly selective DAT ligand¹ that can be used in vivo for investigation of changes in DAT concentration, progression of disease and validation of treatment using Positron Emission Tomography (PET). Accurate parametric images, showing binding potential (BP_ND) and relative delivery (R₁) at the voxel level, are a prerequisite for clinical application of [¹¹C]PE2I. The purpose of the present study was to determine the optimal method for generation of these images.

Methods: Data was used from six social phobia subjects and six patients with parkinsonian syndrome, each receiving an 80 min dynamic PET scan after injection of 350 MBq [¹¹C]PE2I. All patients 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^2,3: RPM, RPM₂) of the simplified reference tissue model (SRTM) and three multilinear reference tissue models⁴ (MRTM_o, MRTM and MRTM₂) were used for computation of parametric BP_ND and R₁ images. In addition, the reference Logan method and standard uptake value ratio (SUV_r) were investigated. Cerebellar grey matter was used as reference region. Accuracy and precision of each method were assessed by simulations and clinical data were evaluated by comparing the parametric methods to region-based analyses with SRTM using linear regression.

Results: Correlation and slope of linear regression between parametric and region-based BP_ND and R₁ values in both striatum and extra-striatal regions were optimal for RPM (R²=0.99 for both BP_ND and R₁; slope 0.99 and 0.98 for BP_ND and R₁, respectively, in striatum). In addition, accuracy and precision as determined using simulations were best for RPM and RPM₂.

Conclusions: The basis function methods provided more stable 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.

References

1 Halldin et al. Eur J Nucl Med Mol Imaging, 2003.

2 Gunn et al. Neuroimage, 1997.

3 Wu & Carson, J Cereb Blood Flow Metab, 2002.

4 Ichise et al. J Cereb Blood Flow Metab, 2003.

P130. Common reference tissue kinetic model for comparison of three opioid receptor tracers in the rat brain

Frode Willoch, Trine Hjornevik, Ina Lindbom, Brian Jeffery Reed and Christopher Coello

Preclinical PET/CT Unit, Dept. of Anatomy, University of Oslo, Norway

Background: In order to study binding kinetics of endogenous agonistic peptides and exogenous antagonistic/agonistic drugs on opioid receptors (ORs), three OR tracers have been considered: the non-selective antagonist [¹⁸F]DPN (Wester, 2000), the μ-partial agonist [¹⁸F]BPN and the μ- and κ-agonist [¹⁸F]PEO (Schoultz, 2012). The present study aims to identify a common reference tissue kinetic model of in-vivo binding potential (BP) for the three previous OR tracers to measure intraindividual differences. The simplified reference tissue model (SRTM) has been successfully validated against invasive two-tissue compartmental model only for [¹⁸F]DPN in humans (Spilker, 2004). However, the multilinear reference tissue model (MRTM2) has been recently shown to be a more robust model than SRTM for reversible ligands (Ichise, 2003). In consequence of this, the non-displaceable BP (BPnd) values are compared between MRTM2 and SRTM for [¹⁸F]DPN to validate MRTM2. Also, BPnd values of the three tracers calculated with MRTM2 are compared.

Methods: Nine Sprague-Dawley rats (∼250 g) have been scanned dynamically for 60 minutes under gas anesthesia (isoflurane 2%) using the three OR tracers. Volumes were iteratively reconstructed (OSEM 3D) and extracted brains were spatial normalized. Three volumes-of-interest (VOI) were drawn in regions with high-density (thalamus), low-density (frontal) and free (cerebellum) of ORs. Firstly, the reference tissue (cerebellum) clearance rate (k_2′) was estimated for each subject using MRTM. Secondly, the BPnd values were calculated with SRTM and MRTM2. The same procedure was reproduced for all three tracers. The beginning time-point (t^*) of MRTM and MRTM2 methods has been varied (t^*=[0 s, 75 s and 150 s] ) in order to quantify its impact on k_2′, BP_nd estimation and Akaike criterion (AIC). For example, the k_2′ estimated with MRTM (t^*=0) was used to calculate BPnd and AIC with MRTM2 (t^*=0).

Results: Estimated mean k_2′ values with SRTM and MRTM (t^*=0 s, 75 s and 150 s) models are not significantly different (t(16)=0.21, p>0.5). High correlation (r²=0.88, t(15)=10.02, p<0.0001) has been found between SRTM and MRTM2 models for the [¹⁸F]DPN tracer (Figure 1A). For the three tracers, AIC criteria decreased between t^*=0 and t^*=75, and then was stable/increased with t^*=150. As expected, uncertainty of the mean BP_nd values (Figure 1B) decreased significantly between SRTM and MRTM2. When using MRTM2 (t^*=75), a significant difference (paired t-test, BPN: t(6)=6.56, p<0.001, PEO: t(6)=4.19, p<0.01) between DPN and the two other tracers was found only in the region with a high-density of ORs (Figure 1C).

Conclusions: A common reference tissue model (MRTM2 (t^*=75 s)) might be used to measure intraindividual differences between tracers but needs further validation with arterial sampling for [¹⁸F]BPN and [¹⁸F]PEO.

P131. Imaging dopamine transmission in the prefrontal cortex: a combined microdialysis and [¹¹C]FLB 457 PET study

Rajesh Narendran, Hank Jedema, Brian Lopresti, Neale Mason, Kate Gurnsey, James Ruskiewicz, Chi-Min Chen, Chester Mathis, William Frankle and Charles Bradberry

University of Pittsburgh, Pennsylvania, USA

Background: In a recent human PET study we demonstrated the ability to detect amphetamine-induced dopamine release in the prefrontal cortex as a reduction in the binding of the dopamine D_2/3 radioligand [¹¹C]FLB 457. A key requirement for validating this paradigm for use in clinical studies is demonstrating that the changes in [¹¹C]FLB 457 binding observed with PET following various doses of amphetamine are related to changes in extracellular dopamine concentration as measured with microdialysis.

Methods: Combined microdialysis and PET experiments were performed to compare, in five rhesus monkeys, amphetamine-induced dopamine release and [¹¹C]FLB 457 displacement in the prefrontal cortex after three doses of amphetamine (0.3 mg/kg, 0.5 mg/kg and 1 mg/kg). Amphetamine-induced change in dopamine concentration in dialysate was expressed as percent increase in dopamine release relative to the mean baseline dopamine concentration (Δ ECF DA). Amphetamine-induced change in [¹¹C]FLB 457 binding potential (Δ BP_ND) was calculated as the difference between the baseline and post-amphetamine BP_ND, and expressed as a percentage of the baseline BP_ND.

Results: Data was available from n=20 PET and n=18 microdialysis experiments (14 of which were acquired in simultaneous sessions). Amphetamine led to a dose-dependent increase in ECF DA (RMANOVA, p<0.0001) and decrease in [¹¹C]FLB 457 BP_ND (RMANOVA, p<0.0001). The relationship between amphetamine-induced peak Δ ECF DA and Δ BP_ND in the prefrontal cortex was linear (y=575.8+57 x, R²=0.32, p=0.035, n=14 simultaneous microdialysis and PET sessions).

Conclusions: The results of this study suggest that the magnitude of extracellular dopamine release is correlated with the magnitude of the reduction in [¹¹C]FLB 457 BP_ND in the prefrontal cortex. This is consistent with that previously reported for other D_2/3 radiotracers such as [¹²³I]IBZM and [¹¹C]raclopride that are widely used to measure dopamine release in the striatum. These data support the use of [¹¹C]FLB 457 to measure amphetamine-induced dopamine release in the prefrontal cortex.

P132. Measurement of absolute cerebral blood flow during cardiopulmonary bypass and selective cerebral perfusion using [¹⁵O]water and PET

Mark Lubberink¹, Thomas Tovedal², Arvid Morell³, Sandeep Golla¹, Sergio Estrada⁴, Veronika Asplund⁴, Gunnar Myrdal², Stefan Thelin², Gunnar Antoni⁴ and Fredrik Lennmyr²

Uppsala University, Sweden: ¹Nuclear Medicine & PET; ²Surgical Sciences; ³Radiology; ⁴Pre-clinical PET Platform

Background: Cerebral perfusion represents a crucial aspect of cardiopulmonary bypass (CPB) since adverse cerebral events are major outcome determinants after cardiothoracic surgery. Deep hypothermia combined with selective antegrade cerebral perfusion (SACP) is thought to provide partial protection through decreased metabolism. Knowledge of absolute cerebral blood flow (CBF) can aid in optimizing SACP. However, a standard dynamic PET scan after intravenous bolus injection of [¹⁵O]water, using continuous blood sampling from the radial artery for measurement of blood activity, can not be used during SACP. The aim of the present work was to validate absolute CBF measurements using intra-arterial bolus injection of [¹⁵O]water and arterial sampling from the arterial tubing during CPB and SACP.

Methods: Four anesthetized pigs underwent a standard 10 min dynamic [¹⁵O]water PET scan on a Hamamatsu SHR-7700 animal PET scanner, starting simultaneously with intravenous bolus injection of 1000 MBq [¹⁵O]water. Blood radioactivity concentrations were measured on-line using continuous sampling from the femoral artery. Subsequently, pigs underwent CPB and a second PET scan was made after injection of 500 MBq [¹⁵O]water in the arterial tubing. Then, pigs were cooled to 20°C and the measurement was repeated. During these first three scans, blood radioactivity concentrations were measured on-line using continuous sampling from the femoral artery. After this, the measurement was repeated again with 250 MBq [¹⁵O]water during SACP at a pump flow rate of 6 mL/kg/min. During this last scan, arterial blood radioactivity concentrations were measured on-line by continuous sampling directly from the arterial tubing, immediately proximal to the aortic cannula. CBF for left and right hemisphere cortex volumes of interest was calculated using the standard single-tissue compartment model with correction for blood volume and delay and dispersion of the input curve. Agreement of baseline and CPB CBF values, as well as CBP without and with SACP at hypothermia, was assessed using a paired t-test and Bland-Altman analysis.

Results: No significant differences between CBF values at baseline and CPB at 37°C, and between CPB at 20°C and CPB with SACP at 20°C, were found. Mean CBF was 0.31 (SD 0.08) mL cm⁻³ min⁻¹ at baseline and 0.35 (SD 0.05) mL cm⁻³ min⁻¹ during CPB at 37°C (p=0.22, bias 0.04±0.08 mL cm⁻³ min⁻¹). During hypothermia, mean CBF was 0.10 (SD 0.03) mL cm⁻³ min⁻¹ during both CPB only and CPB with SACP (p=0.85, bias 0.00±0.05 mL cm⁻³ min⁻¹).

Conclusions: Absolute measurement of CBF during CPB and SACP is possible by injection of [¹⁵O]water directly into the arterial tubing and continuous measurement of arterial blood concentrations directly sampled from this tubing. Absolute CBF was reduced by approximately 70% during hypothermia at 20°C.

P133. Establishing the reliability of an adapted [¹⁸F]-Fallypride imaging protocol in healthy older people

Joel T. Dunn¹, Chloe Clark-Papasavas², Marcel Cleij¹, Stacey Baker¹, Michael J. O'Doherty¹, Paul K. Marsden¹, Shitij Kapur³, R.M. Kessler⁴, R. Howard² and Suzanne J. Reeves²

¹PET Imaging Centre, Division of Imaging Sciences and Biomedical Engineering, King's College London, UK; ²Department of Old Age Psychiatry, Institute of Psychiatry, King's College London, UK; ³Institute of Psychiatry, King's College London, UK; ⁴Department of Radiology and Radiological Sciences, Vanderbilt University Medical Center, Nashville, Tennessee, USA

Background: PET imaging of striatal dopamine D2/3 receptors has played a crucial role in optimising antipsychotic drug treatment in young adults with schizophrenia (Kapur, 2000), and the development of high affinity D2/3 receptor PET radiotracers such as[¹⁸F]-Fallypride has allowed the clinical relevance of extrastriatal D2/3 receptor occupancy to be explored (Kessler, 2006) (Stone, 2009). In contrast to the abundance of data on young adults, this area has been largely neglected within the older population, who have a heightened sensitivity to antipsychotic drugs including extrapyramidal side effects (Howard, 2000) and could potentially benefit most from the clinical application of such imaging techniques. Currently, [¹⁸F]-Fallypride imaging involves multiple lengthy scanning sessions (60-80 min) over a total of 3-4 hours (Mukherjee, 2002) (Kessler, 2006), protocols which are not feasible for use in older adults, particularly those with cognitive impairment. The current study aims to establish the test-retest reliability of regional [¹⁸F]-Fallypride binding in healthy older adults, using a protocol adapted to minimise the duration of individual scans.

Methods: Eight healthy older participants (5 male, 76±5 years) were scanned twice, 45±11 days apart, on a GE VCT Discovery PET-CT scanner. Each session consisted of a single bolus intravenous injection of 250 MBq [¹⁸F]-Fallypride and 3 scans each of 30 min duration (t=0-30; 60-90; and 210-240 min). Optimal scan timings were determined using data on young adults (Dunn, 2010) (Kessler, 2006). Dynamic PET images were motion corrected frame-by-frame. ROIs were defined using an atlas in standard space (Tziortzi, 2011) spatially warped to each subjects' scans. BPND was calculated using the simplified reference tissue model (Lammerstma, 1996). Test-retest reliability was assessed using the Intraclass Correlation Coefficient (ICC) (Hammers, 2007) and absolute variability as a percentage, calculated as standard deviation of the absolute fractional error: 200^*∣scan1-scan2∣/(scan1+scan2). All preprocessing was performed using SPM8 (www.fil.ion.ucl.ac.uk/spm/software/spm8/) and all other analysis using Matlab (www.mathworks.com) and SPSS 19 (www.spss.com).

Results: Primary ROIs showed ICC >0.85 and showed an absolute variability of less than 10% (see Table 1).

Conclusions: The test-retest reliability of our adapted protocol is comparable with previous data in young adults (Mukherjee, 2002) and could be utilised to investigate D2/3 receptor parameters in a variety of clinical and research settings.

References

Dunn JT, J Psychopharmacology 2010; 24(S3):A33.

Hammers A, Neuroimage 2007; 38(1):82-94.

Howard R, Am J Psychiatry 2000; 157(2):172-8.

Kapur S, Am J Psychiatry 2000; 157(4):514-20.

Kessler RM, Neuropsychopharmacology 2006; 31(9):1991-2001.

Lammerstma AA, Neuroimage 1996; 4(3):153-8.

Mukherjee J, Synapse 2002; 46:170-188.

Stone JM, Schizophr Bull 2009; 35:789-97.

Tziortzi AC, Neuroimage 2011; 54(1):264-77.

P134. In vivo evaluation of α7 nicotinic acetylcholine receptors in rhesus monkey brain with [¹¹C]NS14492

Mika Naganawa, David Labaree, Daniel Holden, Nabeel Nabulsi, Yiyun Huang and Richard E. Carson

Yale PET Center, Yale University, New Haven, Connecticut, USA

Background: The α7 nicotinic acetylcholine receptor (α7 nAChR) modulates neuronal plasticity through Ca²⁺-dependent mechanisms [1] and α7 agonists are being investigated as therapeutic drugs for neuropsychiatric diseases, such as schizophrenia and Alzheimer's disease. [¹¹C]NS14492 is a selective α7 nAChR partial agonist, which has shown specific binding in pig brains [2]. The purpose of this study was to evaluate [¹¹C]NS14492 in rhesus monkeys.

Methods: Three anesthetized rhesus monkeys were scanned twice on the same day with [¹¹C]NS14492 at baseline and after pre-treatment with either the α7 nAChR partial agonist W-56203 [3] (3 mg/kg, n=1) or unlabeled NS14492 (1 mg/kg, n=1 and 3 mg/kg, n=1), given intravenously over 5 min starting at 15 min prior to scan 2. Dynamic PET images were acquired on the Focus 220 scanner over 120 min after tracer administration (139±57 MBq, 0.04±0.03 μg/kg, n=6). Arterial blood samples were collected for measuring the input function and time-activity curves (TACs) were measured in cerebellum, cortex, thalamus, hippocampus, and striatum. TACs were analyzed using one- and two- tissue compartment models (1T and 2T) and multilinear analysis (MA1) to calculate distribution volumes (V_T). Receptor occupancy was determined from occupancy plots [4].

Results: The highest uptake of [¹¹C]NS14492 was found in thalamus, moderate uptake was seen in the hippocampus, striatum, and cortex, and lowest uptake was observed in cerebellum. However, the regional differences were small. Blocking was not readily observable in SUV images. Plasma SUV was higher during the blocking scans. The free fraction of [¹¹C]NS14492 was not different between baseline and pretreatment scans (78±1% and 81±4%, n=3). The 2T model produced unstable parameter estimates, but with visually better fits than 1T. MA1 provided good fits. After administration of W-56203 or NS14492, a decrease in heart rate was observed, similar to that seen in rat heart [5]. Pretreatment with W-56203 did not show a reduction in V_T, whereas unlabelled NS14492 produced a reduction (∼20%) in V_T in the thalamus and hippocampus (Figure 1). Assuming that V_T in the cerebellum during the blocking experiments can be used as an estimate of V_ND, BP_ND was 0.1–0.6 at baseline and 0.1–0.3 after pretreatment across ROIs. Occupancy plots were applied to 4 ROIs (excluding striatum which deviated from the regression line). Pretreatment with 1 mg/kg and 3 mg/kg of NS14492 led to 29% (R²=0.69) and 46% (R²=0.86) receptor occupancy, respectively.

Conclusions: There exists some evidence of specific binding for [¹¹C]NS14492 in the monkey brain since pretreatment with NS14492 led to reduction in regional V_T values. However, [¹¹C]NS14492 does not appear to be a promising ligand for imaging α7 nAChR in humans because of its small binding potential (<0.6).

Acknowledgements: Research support provided by Bristol-Myers Squibb.

References

[1] Broide RS et al. Mol Neurobiol 1999; 20:1-16.

[2] Ettrup A et al. J Nucl Med 2011; 52:1449-1456.

[3] Tatsumi R et al. J Med Chem 2006; 49:4374-4383.

[4]Cunningham VJ et al. J Cereb Blood Flow Metab 2010; 30:46-50.

[5]Ji S et al. J Pharmcol Exp Ther 2002; 301:893-899.

P135. Evaluation of biases induced by noise reduction algorithms based on principal component analysis for PET tracers with small, high-contrast regions of interest on the HRRT scanner

Thomas Schafbauer, Richard E. Carson and Jean-Dominique Gallezot

Yale University, Department of Diagnostic Radiology, New Haven, Connecticut, USA

Background: The purpose of this study was to compare algorithms based on Principal Component Analysis (PCA) to reduce noise in Positron Emission Tomography (PET) images. A particular goal was to compare algorithm performance for tracers with small regions-of-interest (ROI) with different kinetics than most other brain voxels. We hypothesized that some PCA methods might not capture these unique kinetics, leading to large bias. The algorithms selected were A1) unnormalized, unweighted PCA; A2) the PCA method proposed by Joshi et al., 2008 [1] and A3) normalized, weighted PCA, based on the statistical model proposed by Benali et al., 1993 [2].

Methods: This study used real data from [¹¹C]P943 (5-HT_1B), [¹¹C]AFM (5-HT transporter) and [¹¹C]-(+)-PHNO (D2/D3) human brain scans performed on the High Resolution Research Tomograph (HRRT), and noisy simulated data based on [¹¹C]-(+)-PHNO kinetics. The primary outcome was parametric maps of the distribution volume V_T. For simulations, the mean (m) and standard deviation (sd) of V_T bias was computed, using noiseless simulation data as reference. For real data, V_T obtained by fitting ROI time-activity curves (TACs) was used as reference to compute biases. The optimal number Q of eigenvector for the PCA was selected by minimizing m²+sd². This criterion was averaged over selected ROIs to select the optimal Q for the whole image (Q_W). The optimal Q was also estimated for targeted small ROIs only (Q_S). The small ROIs were pallidum for [¹¹C]P943 and [¹¹C]-(+)-PHNO, and raphe for [¹¹C]AFM. ΔQ=Q_S−Q_W is the number of extra components needed to describe these small ROI TACs.

Results: ΔQ was equal to 3, 2, 10 and 4 for A1 (and 0, 24, 7 and 3 for A2) with [¹¹C]P943, [¹¹C]AFM, [¹¹C]-(+)-PHNO and the [¹¹C]-(+)-PHNO simulation, respectively. Conversely, ΔQ was always equal to 0 for A3. When Q_W was applied to all brain voxels, the bias in the targeted small regions had greater magnitude with A1 (-7±6%, -52±17%, -24±7% and -20±6%, for [¹¹C]P943, [¹¹C]AFM, [¹¹C]-(+)-PHNO and the [¹¹C]-(+)-PHNO simulation, respectively) and A2 (-5±19%, -46±11%, -26±9% and -32±4%, for [¹¹C]P943, [¹¹C]AFM, [¹¹C]-(+)-PHNO and the [¹¹C]-(+)-PHNO simulation, respectively) than for A3 (-2±5%, -8±13%, -6±12% and -11±8%, for [¹¹C]P943, [¹¹C]AFM, [¹¹C]-(+)-PHNO and the [¹¹C]-(+)-PHNO simulation, respectively). Of note, the lower bias observed with A1 and A2 for [¹¹C]P943 may be due to the narrower [¹¹C]P943 V_T range in the brain compared to [¹¹C]AFM and [¹¹C]-(+)-PHNO.

Conclusions: Small, high-contrast regions have very different kinetics than most other voxels. PCA methods that do not take into account data noise properties tend to include too few components in brain-wide analyses and do not capture the kinetics of small high-contrast regions, so A1 and A2 produce biased estimates. A3, which uses normalized, weighted PCA, is better able to preserve TACs from small atypical ROIs while still reducing noise.

References

[1] J Cereb Blood Flow Metab 28:852–65.

[2] Phys Med Biol 38:1065–80.

P136. A molecular imaging analysis pipeline aimed at maximising quality and throughput

Graham Searle, Cristian Salinas and Roger Gunn

Imanova Limited, London, UK

Background: The aim of the work presented here was to develop advanced software tools that enable a complete analysis workflow for quantitative neuroreceptor imaging studies. The analysis workflow encompasses image conversion, registration of associated structural images, motion correction, image segmentation, input function generation, kinetic modelling, parametric imaging, study level analyses and reporting. Many analysis software approaches already exist, but tend to be limited in their analysis functionality, the robustness and reproducibility of their application and by a need for significant manual intervention by the user. To address these issues, an analysis system with the following attributes was developed for application to academic and commercial neuroreceptor imaging studies:

• Flexibility and agility – to support a range of existing analysis methods and pipeline workflows, and enable rapid implementation of novel techniques.

• Audit trail – inbuilt, to record details of each analysis step applied to the data.

• Reproducibility – the ability to regenerate analysis results from primary data.

• Efficiency – to process large volumes of data with minimized user interaction.

• Quality – managed QC of the analysis workflow.

• Integration – with existing data management systems in order to support automated project management and reporting tools.

Methods: The analysis pipeline framework was developed using Matlab R2008b. It consists of low level functions which perform specific tasks (e.g. image registration), along with higher level functions to manage the analysis workflow. Subject-level data structures record the progress of an analysis, along with any QC steps that have been performed. The software is maintained using the Apache Subversion software version and control system, enabling multiple developers to work effectively in parallel. Crucially it also provides a mechanism by which the functions are ‘self-aware' in the sense that on execution, each function knows its own revision and records that within its output, as an audit trail. Thus, at any point in the future, the exact code used to generate analysis results can be easily obtained (and reapplied to primary data to reproduce results if required).

Results: The analysis pipeline has been used effectively for over 20 neuroimaging studies. For a routine analysis (e.g. a known tracer, with an a priori choice of kinetic model and atlas-based ROI definition), the complete analysis of a PET scan with associated structural MRI image requires less than an hour of computer time, plus around 10-30 min of human QC time. A GUI tool to facilitate and record the QC of the analysis workflow has been implemented, reducing administrative burden significantly. Data and project management tools were also developed, that can manage analysis workloads for users, for example by sending a daily email with a summary of the current analysis status of each study, highlighting outstanding tasks. Similar functions harvest study-level results for reporting.

Conclusions: The software systems described here enables robust, efficient and reproducible analyses without compromising the ability to easily incorporate novel analysis algorithms. The analysis pipeline enables users to devote their time to those activities that computers cannot do (e.g. critical QC steps), leaving other tasks to the software. This increases both throughput and quality.

P137. Test-retest reproducibility of kappa opioid receptor binding assessed with [¹¹C]GR103545 in humans

Mika Naganawa¹, Ming-Qiang Zheng¹, Jim Ropchan¹, Shannan Henry¹, Giampaolo Tomasi², Wendol Williams¹, Richard E. Carson¹ and Yiyun Huang¹

¹Yale PET Center, Yale University, New Haven, Connecticut, USA; ²Comprehensive Cancer Imaging Center, Imperial College, London, UK

Background: The kappa opioid receptor (KOR) has been implicated in several psychiatric and neurological disorders such as epilepsy, Alzheimer's disease, and drug abuse. [¹¹C]GR103545 has been developed as a novel, selective agonist tracer for KOR, previously evaluated in baboons [1] and human healthy controls [2]. The purpose of this study was to assess the test–retest reproducibility of [¹¹C]GR103545 binding measurements in humans.

Methods: Nine healthy volunteers (8M, 1F) each underwent two PET scans on the same day with [¹¹C]GR103545. Data were collected for 150 min on the HRRT scanner and arterial input functions were measured. Regional time-activity curves (TACs) were analyzed using a two-tissue (2T) compartment model and multilinear analysis (MA1, t^*=10, 20, 30, 40, 50, 60 min) to estimate distribution volume (V_T) and binding potential (BP_ND). Cerebellum and thalamus were evaluated as reference region. Test-retest variability for V_T and BP_ND was calculated as the difference between test and retest scans divided by the average (e.g., 2 × (V_T(test)−V_T(retest)) / (V_T(test)+V_T(retest)) × 100%.

Results: Activity doses and injected mass were 563±84 MBq and 3.0±0.78 μg for test scans, and 549±155 MBq and 3.0±0.77 μg for retest scans. Regional TACs of [¹¹C]GR103545 were well fitted with the 2T model and MA1. [¹¹C]GR103545 displayed slower kinetics in humans than baboons, and the micro-parameter estimates using 2T model were unreliable or had implausibly small values. The highest uptake of [¹¹C]GR103545 occurred in the amygdala, anterior cingulate cortex, and insula (mean MA1 V_T values: 30.7, 25.5, and 24.0, respectively), and the lowest in the posterior cingulate cortex, cerebellum, and thalamus (mean MA1 V_T: 10.7, 11.0, 7.2, respectively). The estimated V_T values from MA1 were ∼10% smaller than those from 2T (V_T(MA1, t^*=40 min)=0.89V_T(2T)+1.03, R²=0.95). V_T determinations (MA1, t^*=40 min) were reproducible with acceptable variability: across subjects, mean variability was 15% with SD 21% in all regions except for amygdala. V_T of cerebellum (MA1, t^*=40 min) was 11.3±3.7 (test) and 10.8±3.3 (retest). V_T of thalamus showed smaller intersubject SD: 7.4±1.4 (test) and 6.9±1.1 (retest). Between test and retest scans, V_T values for each subject changed in a consistent direction across regions, but variable direction across subjects, suggesting measurement error such as variability in the input function. In the high binding regions, BP_ND measured with cerebellum as a reference region was 1.57±0.73 (test) and 1.31±0.71 (retest) in anterior cingulate cortex and 1.35±0.61 (test) and 1.25±0.63 (retest) in insula. Reproducibility of BP_ND was better when thalamus was used as a reference region: 2.66±0.50 (test) and 2.37±0.59 (retest) in anterior cingulate cortex and 2.35±0.34 (test) and 2.30±0.39 (retest) in insula.

Conclusions: Test-retest reproducibility of V_T and BP_ND was similar in high-binding regions. The results suggest that thalamus may be more suitable as a reference region than cerebellum, although further blocking studies are needed to confirm the existence of a true reference region.

References

[1] Talbot PS et al., J Nucl Med 2005; 46:484-494.

[2] Tomasi G et al., NeuroImage 2010; 52:S172.

P138. Evaluation of linear and nonlinear spatial transformation models for high resolution PET studies of small brain structures

Dianne E. Lee¹, Jean-Dominique Gallezot¹, Beata Planeta-Wilson¹, Chiang-Shan R. Li² and Richard E. Carson¹

¹Department of Diagnostic Radiology; ²Department of Psychiatry, Yale University, New Haven, Connecticut, USA

Background: Appropriate inter- and intrasubject alignment is critical for localization and quantitative assessment of small brain regions, such as the locus coeruleus (LC) and substantia nigra (SN). However, inaccurate inter- and/or intra-subject registration will increase variability that can lead to false positive or negative results in cross-sectional comparisons. In this study, we evaluated the effect of linear vs. nonlinear MR registration approaches on inter- and intrasubject variability and sensitivity to detect group differences using norepinephrine (NET, [¹¹C]MRB) and D₂/D₃ ([¹¹C]PHNO) datasets. We tested the hypothesis that nonlinear transformations would produce higher BP_ND values with lower coefficients of variation (COV) and increase the significance of group differences in cross-sectional analyses.

Methods: Obese individuals (n=6), patients with cocaine dependence (PCD, n=12), and healthy controls (HC, n=5) were studied with [¹¹C]MRB using the High Resolution Research Tomograph (HRRT). In another protocol, six HC were studied twice (test/retest) with [¹¹C]PHNO. All subjects received a 3T MR scan that was used for co-registration to the MNI template. We applied the linear registration (FLIRT, FSL 3.2; Oxford, UK) and subsequently the nonlinear image transformation (Bioimage Suite 3.0). A template [¹¹C]MRB BP_ND image was obtained by averaging 12 subject's BP_ND images from a previous dataset^[1]; the high binding of [¹¹C]MRB to the LC allowed delineation of the LC ROI (0.38 cm³). The delineation for the SN (2.3 cm³) was made on the MR template. For [¹¹C]MRB, BP_ND images were calculated by MRTM2 (occipital cortex as reference). For [¹¹C]PHNO, BP_ND values were computed with SRTM (cerebellum as reference). The [¹¹C]MRB dynamic images were smoothed using a 5 voxel FWHM Gaussians kernel before computing the parametric images. The effect of registration methods on the mean BP_ND values, test/retest variability (TRV): 2[BP_test – BP_retest]/BP_test + BP_retest, and group differences were evaluated.

Results: Qualitative evaluation revealed nonlinear models provided visually better registration. BP_ND values were increased in all cases, and the % COV decreased in the HC groups with the nonlinear model (Table 1). More specifically, the nonlinear transformation produced statistically higher [¹¹C]MRB BP_ND in HC (42±22% p=0.01) and obese individuals (38±27% p<0.05) but not in the PCD (19±18% p=0.39). The cross-sectional comparisons between groups yielded consistent differences for either registration method; obese individuals compared to HC (p<0.01) for both linear and nonlinear cases, and no difference between PCD and HC with either transformations. For [¹¹C]PHNO, the COV was also reduced with the nonlinear method. Further, the BP_ND values were also markedly increased (26±25% p=0.06) between methods. TRV decreased with the nonlinear model (-8±28%) compared to the linear model (-9±32%).

Conclusions: Consistent with our hypothesis, nonlinear transformations produced higher BP_ND values with lower variability compared to the linear application. The test retest reliability improved moderately. However, the statistical significance of group differences remained consistent between methods.

Reference

[1] Ding YS. Synapse 2010; 64:30-38.

P139. Estimation of total volume of distribution with limited blood sampling: application to mGluR5 selective PET tracer [¹⁸F]FPEB

Aniket Joshi, Sandra Sanabria-Bohórquez, Jacquelynn Cook and Terence Hamill

Merck Research Laboratories, West Point, Pennsylvania, USA

Background: The quantification of reversible PET tracers without a reference region requires arterial input function measurement from time of tracer injection until end-of-scan. We propose a modified Logan plot (MLP) approach for estimating total volume of distribution (V_T) in brain PET studies using only limited late blood sampling. Furthermore, if the arterial and venous tracer concentrations coincide from the time Logan plot attains linearity, venous samples alone would suffice for V_T estimation. The proposed method was applied to quantify metabotropic glutamate receptor subtype 5 (mGluR5) receptor availability using [¹⁸F]FPEB (Hamill et al. 2005) in rhesus monkey. Theory - The operational Logan plot equation is shown in Eq. 1 (Logan et al. 1996) where, C_j(t) is the time-activity curve for the j^th region-of-interest (ROI), C_p(t) is the arterial input function, V_Tj and INT_j correspond to the slope and intercept of the Logan plot for the j^th ROI, and T_i is the midpoint time for the i^th frame in the dynamic PET scan. V_T, the total volume of distribution, is the parameter of interest. The Logan plot equation becomes asymptotically linear from time conventionally denoted as T* (45 min for [¹⁸F]FPEB). We modified the operational Logan plot equation such that the integral of the arterial input function on the right hand side in Eq. 1 is expressed as a sum of two integrals (Eq. 2). Of the two integrals, the first one is a constant for all the ROIs (represented as γ in Eq. 3). Eq. 3 is the operational equation for MLP. Estimation of γ, V_Tj and INT_j was performed by simultaneous fitting of Eq. 3 for five ROIs. For MLP-based analysis, arterial samples (C_p(t)) from T* onwards would suffice for parameter estimation. For tracers where the metabolite-corrected arterial and venous tracer concentrations coincide from T*, venous samples alone would suffice for quantification.

Methods: Nine 90 min [¹⁸F]FPEB scans were performed with arterial sampling in rhesus monkeys, of which seven were baseline and two were occupancy studies (with MPEP). V_T was estimated for all studies using Logan plot as well as MLP. In two baseline and two receptor occupancy studies, venous samples were also obtained. The V_T estimates from the proposed method and Logan plots were compared by Bland–Altman analysis and concordance correlation coefficients (CCC).

Results: V_T estimates using the Logan plot approach showed good agreement with estimates from arterial sampling-based MLP approach. This agreement was confirmed by the Bland Altman plots. The CCC values were between 0.83 and 0.95 for the five ROIs analyzed. In four scans where venous sampling was obtained, metabolite-corrected arterial and venous concentrations were similar from or before T* (45 min). Venous sampling-based analysis was also found to be in agreement with Logan plot analysis using Bland-Altman plots and CCC.

Conclusions: The authors caution that utility of MLP for a given tracer must be confirmed using the above approach before routine use.

References

Hamill TG, Krause S, Ryan C et al. (2005) Synapse 56:205-216.

Logan J, Fowler JS, Volkow ND et al. (1996) J Cereb Blood Flow Metab 16:834-840.

P140. Relationship between kinetic parameters of [¹¹C]PIB and amyloid β deposition studied in amyloid precursor protein transgenic mouse brains

Chie Seki, Masaki Tokunaga, Satoko Hattori, Masahiro Maruyama, Maiko Ono, Bin Ji, Jun Maeda, Tetsuya Suhara, Makoto Higuchi and Hiroshi Ito

National Institute of Radiological Sciences, Chiba, Japan

Background: [¹¹C]PIB is a widely-used PET ligand to evaluate amyloid β (Aβ) deposition in vivo. Although the nature of its binding properties is yet to be clarified, detailed quantitative PET assays of living animal models may provide critical insights into the molecular basis of the radioligand kinetics. We performed dynamic PET scans simultaneously with measurement of metabolite-corrected arterial input function (pTAC) to obtain kinetic parameters in two different types of amyloid precursor protein (APP) transgenic (Tg) and wild-type (WT) mice. APP23 Tg mice mainly develop dense-cored cortical and hippocampal Aβ plaques, while APP-SL Tg mice exhibit cerebrovascular and leptomenigeal Aβ deposition (cerebral amyloid angiopathy, CAA) with only a small number of parenchymal plaques. Regional kinetic parameters for [¹¹C]PIB were compared with postmortem plaque staining in the same individuals to investigate association of these parameters with amounts and types of Aβ deposits.

Methods: APP23 Tg (4 females at 24-26 mo), APP-SL Tg (1 male and 1 female at 23-24 mo), and WT (1 male and 3 females at 24-26 mo) mice were used. Immediately after intravenous injection of [¹¹C]PIB (38±13 MBq, 106±60 GBq/μmol), 60-min PET data acquisition with MicroPET focus 220 and serial arterial blood sampling were conducted. Regional brain time-activity curves (tTACs) were obtained from dynamic PET data with the aid of structural MR images. pTAC was obtained from radioactivity concentration and parent radioligand fraction in the plasma. Two-tissue compartment model analysis was then performed to estimate kinetic parameters (K₁-k₄) and total volume of distribution (V_T) for the radioligand. Distribution volume ratio (DVR) was also calculated using the pons as reference region, because abundant CAA in the cerebellum of APP-SL mice hampered the use of cerebellar data as reference, and because estimatedV_T values in the pons were similar among Tg and WT mice. Mouse brains were removed after PET scans, fixed with 4% paraformaldehyde, cryoprotected in 30% sucrose, and sectioned coronally, followed by staining with 0.01% (E,E)1-fluoro-2,5-bis(3-carboxy-4-hydroxystyryl)benzene (FSB), an amyloid- binding fluorescent dye.

Results: k₃/k₄, V_T and DVR in the neocortex and hippocampus of APP23 Tg mice were higher than those of WT mice, and the increase of these parameters was correlated with the density of FSB-stained parenchymal Aβ deposits. DVR showed small intragroup variability compared to k₃/k₄ and V_T, and is accordingly useful as a robust measure of plaque formation. k₄ values of these regions in APP23 Tg mice were significantly lower than other regions lacking notable plaque deposition, while no plaque-related changes of k₃ values were observed. These data suggest that plaques in APP23 Tg mice contain high-affinity but low-capacity binding components relative to nonspecific binding sites. V_T and DVR values in the cerebellum of APP-SL Tg mice were higher than those of APP23 Tg and WT mice, presumably reflecting accumulation of CAA in this area.

Conclusions: In summary, the present study supports the utility of Tg mice and small-animal PET system for validation of quantitative analytical methods by assessing correlations between kinetic parameter estimates and pathological changes.

P141. Preliminary evaluation of in vivo NET binding in the rat brain with [¹¹C]Dalene and [¹¹C]TAZA using PET imaging

Cristian C. Constantinescu, Min-Liang Pan, Bhavin Patel, Meenakshi Mukherjee, Himika Patel, Christopher Liang, M. Reza Mirbolooki and Jogeshwar Mukherjee

Radiological Sciences, University of California Irvine, USA

Background: [¹¹C]Dalene (4-aminomethyl-4′-(N-¹¹C-methyl-N-methylamino)stilbene) and [¹¹C]TAZA (4-amino-¹¹C-methyl-4′-dimethylaminoazobenzene) are two novel norepinephrine transporter (NET) radioligands. We preliminarily evaluated the in vivo NET binding characteristics of [¹¹C]Dalene and [¹¹C]TAZA in the rat brain and examined the effect of atomoxetine (ATX), a selective norepinephrine reuptake inhibitor, on the kinetics of both tracers.

Methods: Sprague-Dawley rats (308-468 g) received two 90 min PET scans each on two separate days (1-4 weeks apart) with an Inveon scanner. On the first day each rat received a baseline scan, one with [¹¹C]Dalene and the other with [¹¹C]TAZA (96±8 MBq). The second day each animal was pre-injected iv with a 50 μl bolus of ATX (1-2 mg/Kg), 17 min before [¹¹C]Dalene and 2 min before [¹¹C]TAZA (33±7 MBq). The images were reconstructed dynamically in 25 frames (4 × 0.5 min, 8 × 1 min, 5 × 2 min, 2 × 5 min, 6 × 10 min). Each subject received a CT scan that was used for attenuation and scatter correction. Images were normalized to Paxinos & Watson space via co-registration with an MR rat template (Schweinhardt et al. 2003). Common regions of interest were drawn on the MR template and placed on brainstem, thalamus (Tha), and midbrain (Mid), brain regions with highest tracer uptake as well as anterior cingulate cortex area 1 (Cg1), the region with low NET density and which was chosen as a reference based on previous studies (Logan et al. 2007;Gallezot et al. 2011). Time activity curves (TACs) from all regions were analyzed in PMOD using Logan non-invasive method (Logan et al. 1996) with Cg1's k₂, parameter estimated by fitting of brainstem data to MRTM (Ichise et al. 2003).

Results: Figures A and B show normalized [¹¹C]Dalene TACs from baseline and ATX condition, respectively. Similarly, [¹¹C]TAZA TACs are presented in Figures C and D. In the baseline condition both kinetics were slow and the ratio target/reference reached a constant level at 20 min post tracer injection. The kinetics in ATX condition were much faster and with a peak uptake 14-fold higher than in the control for [¹¹C]Dalene in the brainstem and 5-fold for [¹¹C]TAZA. [¹¹C]Dalene baseline BP_ND values were 1.42 (brainstem), 0.86 (Tha), 0.91 (Mid) with changes in BP_ND of -18% (brainstem), -12% (Tha), and 5.4% (Mid) in ATX condition. [¹¹C]TAZA baseline BP_ND values were 0.7 (brainstem), 0.41 (Tha), 0.50 (Mid) with increases in BP_ND of 11% (brainstem), 6% (Tha), and 18% (Mid) in ATX condition.

Conclusions: [¹¹C]Dalene and [¹¹C]TAZA binding pattern was consistent with the expected NET distribution in the rat brain. [¹¹C]Dalene binding values were higher than those of [¹¹C]TAZA. ATX pre-injection promoted a large and fast increase in [¹¹C]Dalene and [¹¹C]TAZA brain uptake. This effect could be attributed to systemic effects of excess norepinephrine caused by the large ATX dose, which generated an increase in blood flow and possibly changes in tracer free fraction. Further occupancy and reversibility of binding studies are needed for a complete evaluation of the two radioligands.

Acknowledgements: Research support provided by NIH/NIA R01AG029479 (JM), NIH/NIA R33AG030524 (JM).

P142. Quantification of dopamine release within the connectivity-derived functional subdivision of striatum

Roger N. Gunn¹, Mark Jenkinson², Eugenii A. Rabinner¹, Timothy Behrens², Paul Shotbolt¹, Christopher Long¹, Charalampos Tsoumpas³, Graham Searle¹, Suzanne N. Haber⁴ and Andri C. Tziortzi²

¹GSK Clinical Imaging Centre, Hammersmith Hospital, London, UK; ²FMRIB Centre, Department of Clinical Neurology, University of Oxford, UK; ³Division of Imaging Sciences & Biomedical Engineering, King's College London, UK; ⁴University of Rochester, School of Medicine and Dentistry, Rochester, New York, USA

Background: Over the last two decades the dopamine system has been extensively studied with Positron Emission Tomography (PET) with striatum being the focus of this research due to its abundance of dopamine receptors. The striatum acts in concert with the cortex to control and execute behaviors that in disorders such as Parkinson's and schizophrenia are impaired by abnormal dopamine function. To date, quantification of dopamine neurotransmission has been limited to anatomical striatal subdivisions. Here, tractography information derived from associated MRI data is used to functionally parcellate the striatum and explore dopamine neurotransmission in these functional subdivisions.

Methods: Diffusion-weighted (DWI) data in conjunction with pre and post-Amphetamine [C¹¹]PHNO PET data were acquired for 12 healthy humans. Striatum is subdivided into functional regions on the basis of anatomical connectivity using DWI-based probabilistic tractography. Cortex was anatomically subdivided into five regions of interest (ROI), where each is known to be associated with a particular function i.e. limbic, executive, rostral-motor, caudal-motor and sensory. The projections of fibers, from each cortical target to the striatum, were estimated and striatum was functionally subdivided according to the highest probability of connection. The simplified reference tissue model was applied to the [C¹¹]PHNO data to estimate the binding potential (BP_ND) within the anatomical (caudate, putamen and ventral striatum (vst)) and functional subdivisions and subsequently estimate dopamine release. To assess the homogeneity of dopamine release the %COV value within regions was estimated.

Results: The results show that the human functional organisation of the striatum is spatially coherent across individuals and largely congruent with non-human primates with distinctive and overlapping networks [1]. The limbic cortex, projects to the vst and ventral post-commissural putamen, contributing to 20±7% of the total striatal volume. Projections from the executive target were found from the most rostral pole of the striatum and extended post-commissural. The executive striatum occupies 49% of the total striatum, a finding that contradicts the concept that striatum is primarily a motor region. Rostral-motor occupies the dorsal post-commissural striatum and extends to dorsal post-commissural putamen. Caudal-motor and sensory projections reside in the post-commissural striatum. The highest dopamine release in the functional subdivisions was in the limbic ROI (20.75±5.94%) followed by the sensory (18.11±3.68%), caudal-motor (17.47±2.34%), rostral-motor (13.87±3.71%), and executive (13.87±3.71%). In the anatomical ROIs the highest release was in vst (21.17±6.93%) followed by putamen (16.36±4.42%) and caudate (14.29±6.84%). The average %COV in the functional areas was 20.69 while in the anatomical ROIs the average %COV was 35.86. A non-parametric statistical test (Kruskal-wallis) was performed to compare the homogeneity of dopamine release in the functional ROIs versus the anatomical ROIs. The results (p=0.05) show that homogeneity of dopamine release was significantly higher in the functional subdivisions.

Conclusions: These data provide evidence that there is an association between local dopamine release and the cortical connectivity profiles of these regions. The ability to delineate functional ROIs using DWI-tractography and apply them to neurotransmission experiments allows for a refined investigation of dopamine release that may be important in studies of pathophysiology in diseased populations.

Reference

[1] Haber 2006, JfN.

P143. Test-retest reliability of [¹¹C]flumazenil data acquired using the Delforge partial saturation method

S. Bouvard^1,2, N. Costes², F. Bonnefoi², F. Lavenne², F. Mauguière³, P. Ryvlin¹ and A. Hammers³

¹Translational and Integrative Group in Epilepsy Research, Lyon Neuroscience Research Center, CNRS 5292, INSERM U1028, Lyon, France; ²CERMEP-Imagerie du vivant, Lyon, France; ³Neurodis Foundation, Lyon, France; ⁴Department of Functional Neurology and Epileptology and Institute for Children and Adolescents with Epilepsy (IDEE), Hospices Civils de Lyon, Lyon, France

Background: [¹¹C]flumazenil (FMZ) PET images GABA_A receptors. Various methods of data analysis are used in the literature, including simple summed radioactivity images (SRI) and parametric images, obtained via compartmental modelling with or without an arterial input function, or via spectral analysis. The Delforge partial saturation method (1997) has been used extensively but no test-retest data on the reliability of the various possible output images has been available.

Methods: Ten healthy controls (22-47 years) were studied twice at one-week intervals. All had high resolution 1.5T MRI. After injection of 2.6 MBq/kg of [¹¹C]FMZ and 0.01 mg/kg of unlabeled FMZ, 3D data were acquired on a Siemens/CTI ECAT HR+ over 55 m, corrected for attenuation and scatter, and rebinned into 12 time frames. SRIs were created over two published time intervals (10-20 m, SRI-10-20, and SRI-20-45). Both were also expressed as standardized uptake values (SUV; SUV-10-20; SUV-20-45). The partial saturation model, based on a Scatchard plot, was used for the calculation of parametric Bmax images, with pons as a reference and with Kd either variable per voxel (BmaxKdvar) or using the same Kd throughout (BmaxKdfix). 83 regions were sampled with a frequency-based brain atlas (Hammers et al., 2003) warped onto each individual's MRI scan using Statistical Parametric Mapping software (SPM5), thresholded at 50% grey matter probability, and then coregistered onto each individual PET. Image quality was assessed visually. Average percentage test-retest differences were calculated as the standard deviation of (test-retest)/mean (test+retest) for all grey matter containing regions except pallidum and those under ten times scanner resolution (∼4.1³ mm³). Reliability was assessed per region via the intraclass correlation coefficient (ICC).

Results: Image quality was good for all types (Figure). The rank order of % test-retest differences (absolute values) was SUV-20-45>SRI-20-45>BmaxKdvar>SRI-10-20>SUV-10-20>BmaxKdfix. The rank order of ICCs was similar.

Conclusions: Considering left and right regions separately and including some regions with low binding (e.g. basal ganglia) may explain somewhat worse values than in other test-retest studies. Bmax is in theory an attractive parameter to quantify, being directly related to receptor concentration. However, in our hands Bmax parametric maps had relatively high test-retest variability and low reliability, with hardly any regions achieving ICCs considered as good, i.e. >0.70. We did, however, not test the reliability of measures of Bmax derived directly from less noisy time-activity curves at the ROI level. SRI and SUV images obtained between 20 and 45 m post injection performed best, with good ICCs despite relatively high average test-retest variability (∼10%). SRI-10-20 and SUV-10-20 images had intermediate values, unlikely to be sufficient for most studies (compare e.g. Hammers et al. 2008).

Reference

Hammers, A. et al., J Cereb Blood Flow Metab 2008; 28(1):207-16.

P144. PETmodel: an SPM toolkit for parametric imaging of dynamic PET data

Christopher J. Endres¹, Yun Zhou², Hassan Mohy-Ud-Din² and Arman Rahmim²

¹Dept of Radiology, Johns Hopkins Medical Institutions, Baltimore, Maryland, USA; ²Dept of Nuclear Medicine, Johns Hopkins Medical Institutions, Baltimore, Maryland, USA

Background: A crucial step towards the application of statistical parametric mapping (SPM) techniques to PET is the generation of parametric images from dynamic PET data. Unfortunately, there currently is no widely distributed toolkit incorporated into SPM that performs parametric modeling. Instead, investigators typically write custom modeling software that may be used exclusively at a single institution or even in an individual lab. In addition to being time-consuming to write the code, the existence of numerous software tools makes it likely that there are subtle differences and inconsistencies in the application of parametric models across research institutions. In order to promote the consistent application of such models, it would be far more expedient to have a toolkit that can be operated directly from the SPM menu to take advantage of the built in user interface and batch utilities. The freely available open source code distribution of a parametric PET modeling SPM toolkit, along with the expected frequent usage, would allow for rapid testing, debugging, and expansion of models/capabilities. In addition to promoting consistency in the application of specific models, incorporation into SPM would also allow the modeling to be streamlined in a batch job. Such implementation would allow investigators to test and apply models that they may not be familiar with and thus would otherwise not attempt. The numerous parametric modeling methods that are in common use should ideally be available freely to all PET investigators, and the PETmodel SPM toolkit aims to fulfill that need.

Methods: The basic design of the PETmodel SPM toolkit is as follows: Inputs - As per the SPM convention, dynamic PET data may be either a single 4D Nifti format image volume OR a sequence of 3D Nifti format image volumes. In addition, most models will typically require two text files containing the input function and the PET frame times, respectively. Both text files are in a two-column format. Outputs - Depending on the model, 2-3 parametric images will be written separately in Nifti format. For example, slope and Intercept images would be generated when applying conventional graphical methods.

Results: Models - The initial implementation supports several common models including Logan, Patlak, MRTM, and MRTM2. For formal release of the toolkit it is also planned to include LRRSC, SRTM2, the relative equilibrium as well as bi-graphical methods. The basis function approach will be added in the near future. Operation - PETmodel will be provided as an SPM8 toolkit and will appear as an option under the PET/SPECT toolbox menu. The entry dialogue will follow the SPM convention.

Conclusions: PETmodel is an SPM compatible open source Matlab toolkit that will be made freely available and will help to standardize application of PET parametric imaging methods.

P145. Optimal design in PET occupancy studies: a sensitivity study

Mattia Veronese¹, Stefano Zamuner², Roger N. Gunn³ and Alessandra Bertoldo¹

¹Department of Information Engineering, University of Padova, Padova, Italy; ²Clinical Pharmacology Modelling and Simulation, GlaxoSmithKline, Stockley Park, UK; ³Imanova Limited, London, UK; Imperial College, London, UK

Background: The application of optimal design algorithms to pharmacokinetic/pharmacodynamic experiments allows parameters to be estimated with minimum bias and variance [1]. In PET receptor occupancy (PET-RO) [2] studies it has been demonstrated that adaptive optimal design (AOD) algorithms allow a reliable selection of experimental design variables, such as dose levels or scan times post dose [3]. However the value of applying adaptive or non-adaptive optimal design methodologies to PET-RO studies depends on several factors including drug affinity to the target as well as feasibility constraints, such as sample size, number of scan per subjects and logistical constrains.

In this work we presented a simulation study to evaluate the sensitivity of PET-RO studies to experimental scanning times. We also investigated the potentialities of optimal design algorithms when applied to PET-RO in presence of mis-specified drug kinetic assumptions.

Methods: A population k_on-k_off model relating the plasma concentration of the drug and the PET binding potential (BP) was applied to generate simulated data. Inter-subject variability was defined by an exponential distribution model (coefficient of variation, CV 30%), while noisy BP measures were simulated by assuming a proportional error model for the residual variability (CV 10%). Simulated experimental designs were chosen according to different levels of parameter mis-specifications with respect to the true simulated values (range: [-300% +300%]). For each design, 100 populations each with 12 subjects were considered. Only two PET scans after baseline were assumed per subject, chosen in a time window of 0-36 hours (minimum distance 4 hours). Analysis of the results included a comparison of the performance of adaptive, non-adaptive optimal designs and non-optimized designs. Design optimization was identified using the D-optimality criterion [4]. Three simulated compounds with different brain affinities (low, medium and high) were tested, with K_d(=k_off/k_on) equal to 15, 2.5 and 0.25 respectively . The dose level was held constant for all the simulations.

Results: For all the drugs and experiments considered, the best performances were achieved using optimal approaches (adaptive and non-adaptive) applied without parameter misspecification (Figure). The worst performances were reported by the non-adaptive method when initial parameter assumptions significantly underestimated the true kinetic of the tracer. However, when AOD was applied to the misspecified cases, precision and accuracy of parameters were recovered. K_d was the most robust parameter (bias range [1% 30%]), while k_on and k_off were much more sensitive to experimental choices (maximum bias 64% and 50% respectively). High-affinity compounds were more robust to experimental setting changes than medium or low affinity drugs.

Conclusions: Our results confirmed that an optimal choice of PET scanning times can improve the quality of parameter estimates in PET-RO and that the introduction of adaptive approaches can circumvent mis-specification of the estimated drug kinetics.

References

[1] Dodds MG et al. J Pharmacokinet Phar, 2005.

[2] Abanades S et al. JCBFM, 2011.

[3] Zamuner S et al. Clin Pharmacol Ther, 2010.

[4] Foracchia M et al. Comput Meth Prog Bio, 2004.

P146. Robust kinetic model for quantification of P-glycoprotein (P-gp) function under altered metabolism and after P-gp inhibition using (R)-[C-11]verapamil and PET

Shaonan Wang¹, Maria Feldmann², Rainer Hinz¹, Adam McMahon¹, Matthias Koepp², Alan Jackson¹ and Marie-Claude Asselin¹

¹Wolfson Molecular Imaging Centre, School of Cancer and Enabling Sciences, The University of Manchester, UK; ²Institute of Neurology, University College London, UK

Background: P-gp is acting as an efflux pump at the blood brain barrier (BBB) and has been hypothesized to play a role in pharmaco-resistance for various diseases, including epilepsy. P-gp function can be imaged in vivo using PET with the radiolabelled substrate (R)-[C-11]verapamil (VPM). Radiolabelled metabolites of VPM cross the BBB and VPM is extensively metabolized in epilepsy patients. P-gp inhibitors can be used to increase the low radioactivity in the brain. Lubberink (JCBFM 2007 27:424-33) proposed the volume of distribution (V_T) estimated from a single-compartment model with a non-polar input function to quantify VPM in healthy controls (HC). Shortening the scan duration and using the forward transport rate constant (K₁) estimated using the same model but with a parent input function was later advocated by Muzi (JNM 2009 50:1267-75). The aim of this study is to extend the kinetic model of VPM to epilepsy patients at baseline and after P-gp inhibition.

Methods: Six HC, six drug-refractory (DR) and six drug-sensitive (DS) epilepsy patients underwent PET scanning for 60 min at baseline after VPM injection. For HC and DR, a second VPM scan was performed after administration of the P-gp inhibitor tariquidar (2-3 mg/kg). For DS, two baseline scans were performed on the same day to assess test-retest reproducibility. Arterial blood was sampled to generate six plasma input functions: parent (V), polar metabolites (P), lipophilic metabolites (H), non-polar (V+H), all metabolites (H+P) or total (V+P+H). VPM-PET images were reconstructed using OP-OSEM with resolution modeling and frame-by-frame motion correction. The whole brain gray matter region was fitted to nine kinetic models comprising one-(2K) and two-(4K) tissue compartments combined with single or dual input functions, all with variable fractional blood volume (VB). The models were tested for the three groups under both conditions and two scan durations (10, 40 min). The Akaike information criteria (AIC) was used for model comparison.

Results: The single-compartment model with total plasma input function provided the highest AIC weights for all groups under both conditions and scan durations except when fitting 40 min data under P-gp inhibition for the DR group (Figure). In this case, the dual input models performed best. The single-compartment model fitted 40 min data only with total plasma input function whereas it failed when using either parent or non-polar input functions. In the DS group, the test-retest variability of K₁ and V_T were less than 8% and 15%, respectively, for single-compartment models. After P-gp inhibition, K₁ increased differentially by 60% for HC and 45% for DR patients, consistently across most models. In contrast, changes in V_T varied considerably between models.

Conclusions: K₁ was found to be a more robust measure of P-gp function than V_T, with less dependency on choice of kinetic model, input function and scan duration. Scan duration can be shortened to 10 min when estimating K₁ and the need for metabolite analysis can further be obviated when using the single-compartment model with a total plasma input function, which can be applied to all groups and under both conditions (baseline, blocking).

P147. Evaluation of PiB relative delivery value (R₁) as a proxy of relative cerebral blood flow

Yin Chen¹, Bedda Rosario², Charles Laymon², Wenzhu Bi³, Cristy Matan², Brian Lopresti², William Klunk⁴, Chester Mathis² and Julie Price²

University of Pittsburgh, Pennsylvania, USA: ¹School of Medicine; ²Department of Radiology; ³Department of Biostatistics; ⁴Department of Psychiatry

Background: PiB is a widely used PET amyloid imaging agent, and the 2-step simplified reference tissue model (SRTM2) provides robust parametric images of binding potential (BP_nd) and relative radioligand delivery (R₁) [1]. Meyer, et al., by assuming R₁ as a robust indicator of relative cerebral blood flow (CBF), observed strong correlation between PiB SRTM2 R₁ and normalized FDG uptake and suggested that PiB R₁ could be a biomarker of neuronal activity and thus neurodegeneration [2]. In the present work, quantitative K₁ of [¹⁵O]water and PiB were directly related to each other and to PiB SRTM2 R₁ in a cohort of elderly control (CON) subjects and patients with mild cognitive impairment (MCI) and Alzheimer's disease (AD).

Methods: Water (12 mCi, 3 min) and PiB (15 mCi, 90 min) PET (ECAT HR⁺) were performed in the same session for 19 subjects: 6 CON (63±12 yrs), 8 MCI (69±11 yrs), 5 AD (67±9 yrs). Water and PiB (metabolite-corrected) arterial input functions were generated. Regions of interest (ROIs) (n=14) were defined on each subject's MRI and included both cortical and subcortical areas and cerebellum (CER) as reference region. CTX5 was defined as the average of 5 primary cortical ROIs. One- and two-tissue compartment models were applied to water and PiB data, respectively. Regional K₁ (mL/cm³/min) values were normalized to CER K₁ to generate indices of relative CBF (water) and relative radioligand delivery (for PiB). PiB data were also analyzed by SRTM2. Data were corrected for atrophy-related CSF dilution [3]. Spearman's correlations (rho) were performed on a region-by-region basis across all subjects (no multiple comparison correction).

Results: For CTX5, water K₁=0.47±0.08, relative water K₁=0.89±0.1, PiB K₁=0.28±0.05, relative PiB K₁=0.91±0.08, and PiB SRTM2 R₁=0.89±0.07. 1) Water K₁ vs. PiB K₁. Correlation was low to moderate (rho≈0.2-0.7) for all ROIs and not significant for CTX5; however, 2 of the 5 ROIs within CTX5 showed significant correlation (frontal cortex, rho=0.47, p=0.043; parietal cortex, rho=0.51, p=0.025). 2) Relative water K₁ vs. relative PiB K₁. Except for pons, correlation was moderate (rho≈0.5-0.8) and significant (p≈0.001-0.019) for all other ROIs. 3) Relative water K₁ vs. PiB SRTM2 R₁. Correlation was moderate (rho≈0.5-0.8) and significant for all ROIs (p≈0.001-0.05). 4) Relative PiB K₁ vs. PiB SRTM2 R₁. Correlation was high (rho≈0.8-0.9) and significant for all ROIs (p<0.001). Similar results were observed without atrophy-related CSF dilution correction.

Conclusions: PiB SRTM2 R₁ can serve as a strong proxy of relative CBF and relative PiB K₁, although the one-tissue compartment SRTM2 assumptions are not met for PiB. This provides further evidence that PiB can be used to indicate neurodegeneration in addition to amyloid load.

References

[1] Yaqub M et al. J Cereb Blood Flow Metab. 2007; 27:1397-1406.

[2] Meyer PT et al. J Nucl Med 2011; 52:393-400.

[3] Meltzer CC et al. J Nucl Med 1999; 40:2053-2065.

P148. A multi-atlas based method for automated anatomical rat brain MRI segmentation and extraction of PET kinetics

Sophie Lancelot^1,2, Alexander Hammers⁴, Affifa Slimen¹, Anne-Charlotte Sahakian¹, Elise Levigoureux^1,2, Jean-Baptiste Langlois³, Caroline Bouillot³, Luc Zimmer^1,2,3 and Nicolas Costes³

¹Université de Lyon 1, INSERM, CNRS, Centre de Recherche en Neuroscience de Lyon, France; ²Hospices Civils de Lyon, France; ³CERMEP-Imagerie du Vivant, Lyon, France, ⁴NEURODIS Foundation, Lyon, France

Background: Data processing of rat brain PET images for pre-clinical imaging requires precise and reproducible delineation of emitting and non-emitting cortical and sub-cortical structures. This segmentation is necessary for extraction of regional time-activity curves and correction of spillover and partial volume effects. A template in reference space is required for the spatial normalization of individual brains for voxel-based analysis. We evaluated a set of manually created rat brain atlases, an anatomical template, and two methods for segmenting brain structures.

Methods: Seven individual rat brain high-resolution 7T 3D T2 MRIs were acquired (volumes of 192 × 256 × 45 voxels with voxel sizes 0.1 × 0.1 × 0.5 mm³). Individual atlases were created by manually drawing 14 anatomical structures (cortical and sub-cortical structures, white matter and ventricles leading to 29 labelled ROIs), using a detailed delineation protocol, and in reference to the Paxinos atlas.¹ The MRI template was created by choosing one MRI to define a reference space, and performing a 2-steps iterative procedure: affine registration of individual MRIs and averaging of the seven resampled MRI in the reference space. With the direct and inverse transformation matrices from individual spaces to reference space, each atlas was resampled in native and in reference space. Automated segmentation in native space was obtained in two ways: 1) Maximum probability atlases were created by decision fusion in the reference (template) space² and backtransformed into individual native space (MAXPROB)^. 2) Individual atlases were registered directly to individual native spaces, followed by decision fusion³ (MULTI). To estimate maximal accuracy, 126 combinations of two, three, four, five, six or all seven individual atlases were created. Accuracy was evaluated by computing 1) the Dice index (Dice; overlap of the automated and the manual structure delineation measured as intersection/average), 2) the volume bias (relative difference of region volumes measured automatically and manually). Robustness and reproducibility of PET regional measurement using automated segmentation were evaluated on an independent set of coregistered MRI/PET data.

Results: Dice indices in cortical and subcortical regions were always over 0.7 and reached maximal values of 0.9 using MULTI with the maximum possible of six individual atlases (Figure, left). There was no significant mean volume bias. Standard deviations of interindividual regional volumes decreased significantly when increasing the number of individual atlases (Figure, right). MAXPROB performances increased with increasing numbers of atlases used. They reached performances of the MULTI method with six atlases, but the performance gains did not reach significance beyond the use of three individual atlases. There were no significant differences in regional PET measures extracted with either automatic method, or manually.

Conclusions: Availability of multi-atlases simplifies and improves the methodology for rat image spatial normalization, automated extraction of regional values, and modelling of functional PET data. Introducing non-linear registrations, the MULTI capability may also enable automatic labelling of atrophied rat brains, in analogy to results in humans.⁴

References

1. Paxinos G, Watson C. The Rat Brain in Stereotaxic Coordinates. Academic Press (1986).

2. Hammers et al., Human Brain Mapping (2003).

3. Heckemann et al.. Neuroimage (2006).

4. Heckemann et al., Neuroimage (2010).

P149. PET quantification of (-)-[¹⁸F]-flubatine binding to nicotinic alpha4beta2 acetylcholine receptors in human brains

Georg Becker¹, Stephan Wilke¹, Peter Schönknecht², Marianne Patt¹, Julia Luthardt¹, Swen Hesse¹, Gudrun Wagenknecht³, Alexander Höpping⁴, Peter Brust⁵ and Osama Sabri¹

¹Departments of Nuclear Medicine, University of Leipzig, Germany; ²Departments of Psychiatry, University of Leipzig, Germany; ³Central Institute for Electronics (ZEL), Research Center Juelich, Germany; ⁴Advanced Biochemical Compounds (ABX), Radeberg, Germany; ⁵Helmholtz-Zentrum Dresden-Rossendorf, Research Site Leipzig, Germany

Background: Nicotinic alpha4beta2 acetylcholine receptors (nAChR) are an important target for diagnostic neuroimaging because of their involvement in Alzheimer's disease (AD) and Parkinson's disease. Using 2-[¹⁸F]F-A85380 PET a significant decline in alpha4beta2-nAChRs in early AD-patients which correlated to loss of cognitive function was shown (1, 2). However, this tracer is not suited for use as biomarker for early AD-diagnosis in a routine clinical set-up because of its unfavourable slow kinetics. Here we used the new radiotracer (-)-[¹⁸F]-Flubatine (formerly (-)-[¹⁸F]-NCFHEB) with significantly improved brain uptake, receptor affinity and selectivity (3). nAChR-parameters were determined by full kinetic modeling and the validity of the practically useful tissue ratio and tissue-to-plasma ratio as receptor parameters was evaluated.

Methods: After intravenous administration of ∼370 MBq (-)-[¹⁸F]-Flubatine, the PET brain imaging was performed in 20 healthy controls (age 70.6±4.6) using an ECAT EXACT HR+ system in 3D-acquisition mode. 23 frames were acquired from 0-90 min post injection and motion corrected with SPM2. Kinetic modeling using a 1-tissue compartment model (1TCM) with arterial input-function was applied to the volume of interest (VOI) based tissue-activity curves (TACs) generated for 29 brain regions (anatomically defined via MRI co-registration). Model-based receptor parameters used were the total distribution volume (V_T) and the distribution volume ratio (DVR) (reference: posterior corpus callosum). In addition the standardized uptake value ratio (SUVR) (50-70 min) as approximation of the DVR and the tissue-to-plasma concentration ratio (TTPR) (70-90 min) as approximation of V_T were used as non model-based receptor parameters.

Results: TACs of all 29 regions could be described adequately with the 1TCM and all kinetic parameters could be reliably estimated from 90 min PET data. V_T increased as expected with receptor density: corpus callosum (V_T: 5.68±1.01), frontal cortex (9.18±0.59), parietal cortex (9.10±0.61), pons (11.10±0.86), thalamus (25.03±3.33). Mean TTPR values in frontal and parietal cortices were 2% higher than the corresponding V_T values but 7% lower in the thalamus. There was a strong linear correlation between the two sets of TTPR and V_T values (r²=0.98, p<10⁻⁴) (Figure 1A). As V_T, DVR increased with receptor density. Frontal cortex (DVR: 1.66±0.27), parietal cortex (1.64±0.27), pons (2.01±0.35), thalamus (4.52±0.87). Mean SUVR values in frontal and parietal cortices were almost identical to mean DVR values (difference <0.1%) but ∼15 % lower in the thalamus. Accordingly there was a strong linear correlation between the SUVR and DVR values (r²=0.97, p<10⁻⁴) (Figure 1B).

Conclusions: For (-)-[¹⁸F]-Flubatine the receptor parameters TTPR and SUVR in cortical regions are in excellent agreement with corresponding parameters computed by full kinetic modeling. For unbiased estimates of TTPR and SUVR in the thalamus the use of a bolus/infusion scheme for tracer application should be considered.

References

1. O Sabri, P Brust. Acetylcholine receptors in dementia and mild cognitive impairment. Eur J Nucl Med Mol Imaging 2008; 35(Suppl. 1):30-45.

2. K Kendziorra, O Sabri. Decreased cerebral α4ß2 nicotinic acetylcholine receptors in living patients with mild cognitive impairment and Alzheimer's disease assessed with positron emission tomography. Eur J Nucl Med Mol Imaging 2010; 38: 515-525.

3. P Brust, O Sabri. In-vivo measurement of nicotinic acetylcholine receptors with [¹⁸F]norchloro-fluoro-homoepibatidine (NCFHEB). Synapse 2008; 62:205-218.

P150. Development and validation of an integrative software for automatic MRI and [¹¹C]PiB dynamic PET image processing and parametric Imaging

Amith Harsha¹, Yun Zhou², Jitka Sojkova³, Joshua Goh³, Arman Rahmim⁴, Dean F. Wong⁴, Susan M. Resnick³ and Jerry L. Prince⁵

¹Department of Radiology and Radiological Sciences, Division of Neuroradiology, Johns Hopkins University, Baltimore, Maryland, USA; ²Department of Radiology and Radiological Sciences, Division of Nuclear Medicine, Johns Hopkins University, Baltimore, Maryland, USA; ³National Institute on Aging, National Institutes of Health, Baltimore, Maryland, USA; ⁴Department of Radiology and Radiological Sciences, Division of Nuclear Medicine, Section of High Resolution Brain PET imaging, Johns Hopkins University, Baltimore, Maryland, USA; ⁵Image Analysis and Communication Lab, Department of Electrical Engineering, Johns Hopkins University, Baltimore, Maryland, USA

Background: Functional Positron Emission Tomography (PET) imaging is usually combined with high resolution Magnetic Resonance Imaging (MRI) in order to relate functional measures to anatomical structures. In light of emergent PET-MR imaging technology, it is important to use a combination of structural and functional images to obtain meaningful insight. Current image processing and analysis software for MRI and PET are separate and some steps are manual or semi-automatic. This practice places limitations on the size of studies and raises concerns over errors caused by rater variability. This paper describes the development of an automated pipeline of [¹¹C]PiB image processing integrating fully-automatic MR image processing tools with in-house PET image processing and kinetic modeling procedures.

Methods: The software was developed using Java Image Science Toolkit (JIST), a modular and cross-platform framework [1] developed for the Medical Image Processing, Analysis and Visualization (MIPAV) program [2]. The image processing involves the use of [¹¹C]PiB dynamic PET images and concurrent MR images collected for the same subject. The MR images underwent a skull stripping step using a brain extraction algorithm called SPECTRE [3]. These were then segmented using a topology preserving anatomy driven algorithm TOADS [4] to obtain the grey matter cerebellum (GM-CB) membership which was thresholded at 85% to avoid partial volumes. The GM-CB region masks were binarized and co-registered to the dynamic mean of the PET data and used as the reference region for normalizing the [¹¹C]PiB retention in the whole brain. The time activity curves (TAC) obtained for the reference region were used as input to a parametric model [5] and distribution volume ratio (DVR) and R1 images were produced.

Results: The resulting parametric images and histograms from both automatic and manual methods are comparable to one another. The manual delineation produces a very limited number of slices of the reference region when compared to the automated process, which encompasses the entire reference region. The TAC for both the manual and automated process are noted to be similar. An additional reference region for the whole cerebellum (WCB) is also included, highlighting the ease of adding a new ROI.

Conclusions: The automated process allows the possibility to use different reference tissue regions and also for switching out parametric models. The dynamic range of the [¹¹C]PiB images can also be adjusted based on the timing of the study. The pipeline allows us to readily test different modifications in a repeatable and reliable manner. The modular nature of this process pipeline expands the opportunities to explore different multi-modal imaging techniques.

References

[1] Lucas BC et al. Neuroinformatics 8:5-17 (2010).

[2] McAuliffe MJ et al. IEEE-CBMS 381-386 (2001).

[3] Carass A et al. 4th IEEE ISBI 656–659 (2007).

[4] Bazin PL, Pham DL. J Cogn Neurosci 19:1498–1507 (2007).

[5] Zhou Y et al. Neuroimage 18:975-989 (2003).

P151. Kinetic modeling without a reference region

R. Todd Ogden, Francesca Zanderigo and Ramin V. Parsey

Columbia University/NYSPI, New York, New York, USA

Background: The two-tissue compartmental model is commonly used to describe the kinetic behavior of radiotracers in neuroreceptor mapping applications. Estimation of BP_F for each region of interest (ROI) or voxel generally requires estimating the four rate parameters using standard nonlinear regression techniques applied to each region's observed time-activity curve. However, these four rate parameters are not always identifiable and “direct” estimation of BP_F (BP_F=K₁k₃/(k₂ k₄f_P)) can be unstable. This can be overcome if it is possible to identify a “reference region”, i.e., a region devoid of the receptor of interest. This can be done by estimating K₁ and k₂ for the reference region using a one-tissue model, then fitting each ROI/voxel separately using a two-tissue model while constraining the ratio K₁/k₂ to match the estimated K₁/k₂ ratio of the reference region. This reduces from four to three the number of free parameters that must be estimated for each ROI/voxel, and can greatly increase the stability of the estimation of BP_F. However for some radiotracers there is no true reference region, i.e., every region has at least some level of specific binding. In such a situation, applying the constrained estimation procedure outlined above will result in biased estimation of BP_F for every region.

Methods: Even without a true reference region, if nondisplaceable binding is constant throughout the brain [3], it is possible to estimate BP_F for all regions by fitting multiple time-activity curves simultaneously. This can be accomplished by constraining K₁/k₂ to be the same for all regions and estimating the three other kinetic parameters specific to each ROI. In particular, the objective function is the (weighted) sum of squared residuals, summing across time and across regions, giving 3R+1 free parameters when applied to R regions. This greatly increases the dimensionality of the parameter space and thus the computational complexity required for optimization. Since standard nonlinear regression techniques will not be adequate, we instead must consider algorithms for estimation in high-dimensional parameter spaces such as simulated annealing. We take an approach similar to [1, 2], who applied simulating annealing to estimate an arterial input function, but simpler, since when the input function is observed, there is only one shared parameter. We note that the same approach may also be taken when neither plasma data nor a reference region are available, by adding only a single shared parameter to those involved in the input function.

Results: We applied this methodology to data from a [¹¹C]WAY-100635 study of the 5HT_1A receptor and showed good agreement between this approach and the usual approach based on the reference region.

Conclusions: Even when a reference region is available, this approach can actually improve estimation of the nondisplaceable binding component: rather than relying only on information in a single region (the reference region) for that estimate, it can “borrow strength” across several regions, thereby increasing the precision of estimation.

References

[1] Hall et al. (1997). Brain Res 745:96-108.

[2] Wong et al. (2002). IEEE Trans Nucl Sci 49:707-13.

[3] Ogden et al. (2009). J Cereb Blood Flow Metab 30:816-826.

P152. Evaluation of the performance of a novel mobile brain PET-CT

Jinsong Ouyang¹, Matthew L. Keeler², Ali A. Bonab¹, Paul Domigan², Xuping Zhu¹, Marc Normandin¹, Thomas Brady¹ and Georges El Fakhri¹

¹Massachusetts General Hospital, Boston, Massachusetts, USA; ²PhotoDiagnostics Systems Inc., Boxborough, Massachusetts, USA

Background: The newly developed mobile NeuroPET-CT scanner is a full-ring brain PET-CT imaging system with a FOV of 25-cm in diameter and 21-cm in length. It uses 155316 2.3 × 2.3 × 10 mm dual-layer LYSO crystals and 12096 SiPMTs. The PET scanner was designed to achieve both high sensitivity and good spatial resolution. The CT has 3264 detector channels with 8 axial channels at spacing of about 1.25 mm. The X-ray source, capable of 140 keV at 7.0 mA, can rotate at 60 rpm for 1440 views/sec.

Methods: The performance of this new system has been investigated using procedures based on the NEMA PET standard. Extensions of the procedures are described. Spatial resolution was measured using a ²²Na point source. The transverse resolution is 2.5 mm near the center and becomes 3.3 mm at a distance of r=5 cm from the center. The axial resolution is 3.0 mm at the center and 3.3 mm at r=5 cm. The overall system sensitivity for true events, which is calibrated to the activity within the FOV, is 42 cps/kBq in the center of FOV using an energy threshold of 420 keV. The sensitivity has more than doubled compared to the previous NeuroPET scanner, which used CsI crystals. A 20-cm-diameter solid polyethylene cylinder with an overall length of 70 cm was used in the measurement of counting rate performance.

Results: For the NeuroPET-CT scanner, the maximum NEC rate of about 38 kcps was achieved for an activity concentration of 3.7 kBq/ml as shown in Figure A. HRRT, which is Siemens brain PET scanner, yields higher maximum NEC rate but at much higher activity concentration level. The variation of NEC rate between 3 and 6 kBq/ml, which is the clinical activity concentration range calculated for a 100-kg patient using SUV=1.5 and 200-400 MBq injection dose, is smaller than that for the HRRT. HR+, which is Siemens whole-body PET scanner, yields less maximum NEC rate at higher activity concentration level. Figure B shows four different slices of a Hoffman phantom. The Hoffman phantom, which was filled with 730 μCi ¹⁸F, was scanned for 85 min. The reconstruction was performed using 50 iterations of MLEM with 1.5 mm Gaussian inter-filter.

Conclusions: In summary, NeuroPET-CT is a mobile dedicated PET scanner that achieved high sensitivity, good spatial resolution, and high NEC rate at the activity concentration level used in the clinical setting. It is promising for monitoring response to new neuro-protective therapies.

P153. A kinetic model for mapping dopamine function with simultaneous PET/MR

Marc Normandin¹, Christin Sander², Ciprian Catana¹, Jacob Hooker¹, Bruce Jenkins¹, Wim Vanduffel¹, Georges El Fakhri¹, Bruce Rosen¹, Nathaniel Alpert¹ and Joseph Mandeville¹

¹Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts, USA; ²Massachusetts Institute of Technology, Cambridge, Massachusetts, USA

Background: The advent of simultaneous PET/MR permits concurrent and complementary views of neurobiological processes underlying the working brain. However, disparate physical measurements made by PET and fMRI that provide potentially synergistic information also raise challenges for interpreting data from both modalities within a common mechanistic framework. Notably, fMRI responses to stimuli are sensitive but exhibit complex dynamics and are thought to lack neurochemical specificity, whereas PET radioligand binding has molecular selectivity but suffers limited temporal and spatial resolution. We present initial support for a kinetic model of fMRI data that is readily coupled to PET analysis methods through a shared neurotransmitter activation signal.

Methods: From basic pharmacological principles we modeled fMRI cerebral blood volume (CBV) responses as excitatory effects of dopamine D₁-like receptors opposing inhibition exerted by D₂-like receptors. Simulations were performed using dopamine inputs having shapes and scales resembling different doses of dopaminergic drugs. Model binding parameters reflected the differential affinity of dopamine at D₁ vs. D₂. Saturable receptor densities were based on striatal D₁ and D₂ expression in rodent or primate.

A rhesus macaque underwent two simultaneous PET/MR scans using the Siemens BrainPET insert and 3T Trio. In each scan ∼5 mCi [¹¹C]raclopride was injected by i.v. bolus; in one study, 40 μg/kg unlabeled raclopride was co-administered to suppress the D₂ component of the CBV signal. MR acquisition with iron oxide contrast agent commenced several minutes before PET with whole brain covered by two-dimensional echo-planar imaging. PET data were acquired for 90 min and change in striatal binding potential (BP_ND) between baseline and blocking scans provided estimates of D₂ occupancy. In a separate fMRI study, contrast-enhanced CBV changes were determined in another monkey that received 0.6 mg/kg i.v. amphetamine.

Results: The scale and temporal dynamics of fMRI model simulations accurately reproduced CBV responses previously measured in rodents¹ and primates² following amphetamine or cocaine (Figure 1A, B). In PET/MR studies, unlabeled raclopride produced robust CBV increases in D₂-rich regions (Figure 1C), consistent with raclopride blockade attenuating baseline D₂-mediated neuronal inhibition. Concurrent [¹¹C]raclopride PET data indicated ∼90% D₂ occupancy. In the amphetamine study, striatal CBV decreased with magnitude and time course consistent with fMRI model simulations (Figure 1D).

Conclusions: A pharmacologically-inspired model of fMRI kinetics was developed with preliminary support from simulations, concurrent PET/MR with D₂ blockade, and fMRI with amphetamine challenge. Future PET/MR experiments will verify the model's generality using D₁ antagonists, D₁ or D₂ agonists, and dopaminergic drugs including amphetamine or cocaine. The validated fMRI model will be incorporated with PET analysis techniques into a unified mathematical structure for application in PET/fMRI studies of dopamine responses elicited by pharmacological or cognitive challenges.

Acknowledgements: Research support provided by NIH R21NS072148, R21EB012326.

References

1 Marota et al. (2000) NeuroImage 11(1):13-23.

2 Mandeville et al. (2011) Neuropsychopharmacology 36(6):1187-8.

P154. Coupling of neurovascular response and receptor occupancy with simultaneous PET/fMRI

Christin Y. Sander^1,2, Jacob M. Hooker¹, Ciprian Catana¹, Marc Normandin³, Wim Vanduffel¹, Bruce R. Rosen^1,4 and Joseph B. Mandeville¹

¹A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Massachusetts, USA; ²Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, Massachusetts; ³Division of Nuclear Medicine, Massachusetts General Hospital, Boston, Massachusetts, USA; ⁴Health Sciences and Technology, Harvard-MIT, Cambridge, Massachusetts, USA

Background: Simultaneous PET and functional MRI of the brain allow the analysis of complementary information about neurochemistry and function. We propose an imaging paradigm that assesses dynamic PET/fMRI responses to increasing mass doses of a PET ligand at variable specific activities. We determine correlations between receptor occupancy and neurovascular responses in the domains of space, time and dose using a D2 antagonist.

Methods: Dynamic PET/fMRI images of an anesthetized non-human primate (NHP, rhesus macaque) were acquired continuously for 90 min on a Siemens 3 Tesla with a BrainPET camera inside the scanner. Before injection of radiotracer, iron oxide was administered to improve fMRI detection power. [¹¹C]-raclopride (4.6-6.1 mCi injected activity) was administered intravenously at five different specific activities, with injected masses of 0.4 μg/kg (tracer dose), 1.4, 4.5, 16 and 41 μg/kg. PET/fMRI data were registered to a standardized NHP space. The simplified reference tissue model [1] was used for PET data analysis and to quantify binding potentials (BP_ND) with time activity curves derived from anatomical regions. Dopamine occupancies were simulated from occupancy data with inclusion of a dopamine compartment [2]. fMRI data were analyzed with a GLM using temporal information from the specific binding regions of PET data and relative change in cerebral blood volume (%CBV) was derived. The %CBV peak was determined from a regressor fit across the time profile.

Results: Specific binding of [¹¹C]-raclopride and increases in %CBV mainly occurred in the same D2-receptor rich striatal region of basal ganglia. This response is attributed to antagonism at D2 receptors, which invokes activation by inhibiting dopamine binding. Changes in CBV temporally aligned with specific binding of [¹¹C]-raclopride (Figure A). BP_ND values in whole putamen for the five doses of raclopride were 3.0 (tracer dose), 2.7, 1.8, 1.1 and 0.33 respectively. D2 occupancy values of raclopride were 10%, 40%, 63% and 90% for the four doses, relative to the tracer dose. Simulated dopamine peak occupancy levels were 18%, 12%, 7% and 2%, assuming basal dopamine occupancy of 20% (Figure B). Both raclopride and dopamine occupancies were linearly related to %CBV peak measures within the measured doses (Figure C).

Conclusions: We demonstrated an imaging methodology in which a PET antagonist or agonist can be administered as a pharmacological dose to measure neurovascular responses simultaneously with receptor occupancy, relative to a tracer dose. An endogenous neural response can be evoked without the need of a displaceable tracer and measured with fMRI in relation to PET across the full time response. A modified SRTM GLM that incorporates priors from fMRI data can enable an advanced dynamic analysis of PET binding. This may be particularly valuable towards the detection limits of PET (blocking dose) and fMRI (tracer dose). These preliminary data demonstrate a linear relationship between neurovascular responses and dopamine D2 occupancy.

References

[1] Gunn et al. (1997), Neuroimage 6:279-287.

[2] Endres et al. (1997), JCBFM 17:932-942.

P155. Is the cerebellum the best reference region for a PET study with [F-18]-fallypride?

Kenji Ishibashi^1,2, Chelsea Robertson³, Edythe London³ and Mark Mandelkern^2,4

¹Department of Psychiatry and Biobehavioral Sciences and Semel Institute at University of California at Los Angeles, USA; ²Veterans Administration Greater Los Angeles Healthcare System Los Angeles, California; ³Department of Molecular and Medical Pharmacology at University of California at Los Angeles, Los Angeles, California, USA; ⁴Department of Physics at University of California at Irvine, Irvine, California, USA

Background: The reference region method is widely used in PET imaging with radioligands for determining neuroreceptor binding potential. Images obtained using the selective dopamine D2-D3 receptor ligand [F-18]-fallypride (K_D∼30pM) are commonly analyzed using this method and the cerebellum as the reference region. Cerebellar activity is low, especially late in fallypride studies, and statistical uncertainty in measured binding potential values is dominated by the cerebellar measurements. There is substantial variation in published binding potentials for fallypride, suggesting the presence of systematic errors, possibly related to scanner-dependent inaccuracy in scatter correction for the cerebellum. We examined the potential of using reference regions other than the cerebellum for human fallypride studies. [I-125]-Epidepride autoradiography of postmortem human brain tissue (1) has indicated that several regions have comparably low D2 receptor density (<1% of putamen). These include the cerebellum, occipital cortex, calcarine cortex, and white matter.

Methods: A total of 79 scans were acquired on a Siemens ECAT EXACT HR+ PET Scanner (Siemens Healthcare, Erlangen, Germany) using [F-18]-fallypride to measure D2-D3 dopamine receptor availability. Scans were obtained from 43 healthy control participants and 36 methamphetamine-dependent participants. Anatomically defined volumes of interest for the caudate, putamen and nucleus accumbens (FSL FIRST) (2) were combined to create a volume of interest for the whole striatum. Volumes of interest for each reference region (cerebellum, occipital pole, intracalcarine cortex, and white matter of the superior longitudinal fasciculus) were created in standard space (MNI) and transformed into native space. We applied the simplified reference region method (SRTM) (3), implemented in the PMOD software package (PMOD Technologies, Zurich, Switzerland) to calculate the binding potential value in the whole striatum using each of 4 different reference regions (Table 1).

Results: Table 1 shows the mean binding potential values and the ratio of standard deviation to mean for each group. The striatal binding potential values calculated using white matter as a reference region are somewhat larger than for those calculated using cerebellum as a reference region, which are in turn substantially larger than the binding potential values obtained using occipital pole or intracalcarine cortex as a reference region. The ratio of standard deviation to mean binding potential is notably smaller for the white matter reference region than the other reference regions.

Conclusions: There are a number of potential reference regions available for [F-18]-fallypride modeling, all of which have comparable receptor density less than 1% of that of striatum. Which of these best matches the principal regions of interest, the striatal structures, in volume of distribution for free ligand, is uncertain. From the point of view of measurement uncertainty, we find that using white matter as a reference region appears better than using cerebellum.

References

[1]Hall H et al. Synapse 23:115–123 (1996).

[2]Patenaude B et al. Neuroimage 56:907-922 (2011).

[3]Lammertsma AA et al. Neuroimage 4:153-158 (1996).

P156. PK11195-PET modeling using a non-brain reference region

Anat Maoz¹, Lilja Solnes¹, Susan A. Gauthier², Paresh J. Kothari¹, Masanori Ichise³, Joseph Osborne¹, Michael Synan¹, Shankar Vallabhajosula¹ and Tracy Butler⁴

¹Radiology and Citigroup Biomedical Imaging Center, Weill Cornell Medical College of Cornell University, New York, New York, USA; ²Neurology and Neuroscience, Weill Cornell Medical College of Cornell University, New York, New York, USA; ³Nuclear Medicine, Columbia University, New York, New York, USA; ⁴Comprehensive Epilepsy Center, New York University, New York, New York, USA

Background: PK11195 (PK) is a radiotracer that binds to the translocator protein expressed by activated microglia in the brain. It has been used to image neuroinflammation in several diseases including Alzheimer's disease (AD), multiple sclerosis (MS) and epilepsy. Much research has been devoted to optimizing analysis methods for PK PET due to low specific to non-specific binding ratio. When arterial blood sampling is unavailable, reference tissue models are useful. The problem with reference tissue models is that it is difficult to identify a reference region devoid of PK specific binding, especially when studying brain diseases such as MS and epilepsy that often affect typically-used reference regions like the cerebellum. To address this issue, unbiased methods such as cluster analysis have been used to identify non-contiguous gray matter voxels with minimal PK binding that are appropriate to use as a reference region in reference tissue modeling. While this model was successful in many cases, it was found to underperform in many other scans. It may be that using a brain reference region can never be optimal when studying patients with disease processes potentially affecting the entire brain. Autopsy studies of patients with epilepsy and MS have demonstrated widespread histological abnormalities that are not currently detectable using in vivo methods. In such cases, using normal-appearing but actually abnormal brain as a reference region for PET analysis may be inappropriate. To address this issue, and because brain diseases such as MS and epilepsy typically are not associated with extra-cerebral systemic disease, we decided it would be reasonable to explore the use of a non-brain reference region included in the field of view (FOV) of PET brain scanning. Because there are few extracerebral regions of low PK specific binding, we tried using an extracerebral region with visibly high binding – the clivus bone of the skull, with the understanding that this would give rise to negative calculated binding potential images (BP_ND^*=(V_T−V_T(Clivus))/V_T(Clivus)).

Methods: PK PET scans of patients with focal epilepsy, MS and normal controls were used for the analysis using Pmod software (Pmod 3.1, PMOD Technologies Ltd.) High PET signal in clivus was traced using fused PET/CT images for each patient. Parametric images of BP_ND^* were generated using MRTM0 modeling.

Results: MRTM0 images generated using a clivus reference region were neurobiologically plausible, comparable to images generated using a cerebellar region and corresponding in some cases to individual patients' clinical scenarios.

Conclusions: We present preliminary evidence that use of a non-brain reference region (clivus) may be appropriate in reference tissue modeling of PK brain PET, especially in patients with neurological diseases potentially affecting most or all of the brain. The clivus is a skull bone included in the field of view of brain PET studies. Negative correlation of clivus SUV with age suggests that PK signal in clivus reflects bone marrow, which is known to atrophy with increasing age. Further studies examining non-brain reference regions including clivus in brain PK PET are warranted.

P157. Effect of variations in cerebellum kinetics on robustness of [¹¹C]DASB parametric images

My Jonasson¹, Lieuwe Appel¹, Jonas Engman², Andreas Frick², Jens Sörensen¹, Tomas Furmark² and Mark Lubberink¹

¹Nuclear Medicine & PET, Uppsala University, Sweden; ²Psychology, Uppsala University, Sweden

Background: [¹¹C]DASB is a selective serotonin transporter (SERT) radioligand for in vivo investigations of SERT using positron emission tomography (PET). Parametric images of DASB binding potential (BP_ND) and relative delivery (R₁) can be generated using linearizations of the simplified reference tissue model (SRTM), with cerebellum as reference tissue.¹ Previous studies on the biodistribution of [¹¹C]DASB have shown a substantial initial uptake of 30-70% of the injected amount of [¹¹C]DASB in lungs, followed by a slow release.² Drugs such as selective serotonin reuptake inhibitors may have substantial uptake in the lungs, thereby decreasing lung uptake and changing plasma kinetics and hence brain uptake of [¹¹C]DASB. The aim of this simulation study was to investigate the influence of different cerebellum kinetics on the robustness of parameter estimations.

Methods: Two different implementations of SRTM were considered: multilinear reference tissue model³ (MRTM₂) and receptor parametric mapping⁴ (RPM₂), applying a fixed reference tissue efflux rate constant (k_2′).Simulations were performed to assess accuracy and precision of BP_ND and R₁ estimates. Three different measured cerebellum time-activity curves (TACs) peaking at approximately 10, 20 and 30 min p.i. were used reflecting fast, medium and slow kinetics (Figure A). For each reference TAC, 100 noisy striatal TACs were generated using SRTM with scan durations of 60, 90 and 120 min. Simulations were also performed using an erroneous k_2′ value, departing ±10% and ±25% from its true value. Bias and coefficient of variation (COV) were calculated for BP_ND and R₁ for each set of simulations.

Results: The robustness of R₁ was consistently high regardless of kinetics, analysis method or scan duration, with COV of 2-5% and bias of -0.5 to 0.5%. For BP_ND, both accuracy and precision were reduced with the slow cerebellum TAC. RPM₂ provided the most robust estimations of BP_NDfor slow kinetics (Figure B). An underestimation of k_2′ of 25% had most effect on the parameter estimation and gave a positive bias of 60% for RPM₂ and >100% for MRTM₂, for a 60 min scan.

Conclusions: Robustness of parameter estimations for [¹¹C]DASB is dependent on the kinetics of the tracer in the cerebellum. For slow kinetics, RPM₂ resulted in the most robust parameter estimations.

References

1 Ichise et al. J Cereb Blood Flow Metab 2003.

2 Lu et al. J Nucl Med 2004.

3 Wu & Carson. J Cereb Blood Flow Metab 2002.

4 Gunn et al. Neuroimage 1997.

P158. Kinetic analysis of [¹⁸F]FEOBV, a PET radiotracer for imaging the vesicular acetylcholine transporter

Robert Koeppe, Myria Petrou, Peter Scott, Nicolaas Bohnen, Michael Kilbourn and Kirk Frey

Department of Radiology, University of Michigan, Ann Arbor, Michigan, USA

Background: [¹⁸F]FEOBV is a vesamicol analog that binds selectively to the vesicular acetylcholine transporter (VAChT) and has been used in preclinical studies to quantify presynaptic cholinergic nerve terminals. FEOBV scans were performed in three healthy human volunteers to determine the safety, whole body biodistribution, and verify dosimetry. Scans were then performed in seven healthy volunteers for assessment of FEOBV kinetics in humans and to investigate kinetic analysis alternatives with the goal of identifying a simplified protocol for human research, and eventually clinical studies.

Methods: Seven healthy controls underwent dynamic brain imaging from 0-120, 150-180, and 210-240 min following bolus injection of 240 mBq of [¹⁸F]FEOBV. Two subjects were also scanned from 330-360 min post-injection. Arterial blood sampling was performed in all subjects with chromatographic identification of authentic FEOBV and radiolabeled metabolites to determine the arterial plasma input. Venous samples were drawn at two late times (∼2.0 and 3.5 hrs) and analyzed for metabolites. Four general kinetic approaches were evaluated. I. Arterial input-based kinetic modeling was performed using a two-tissue compartment configuration, determining the standard rate parameters, K₁-k₄. Simultaneous fitting of multiple time-activity curves was used with constraints that K₁/k₂ and/or k₄ be the same for all regions for a given subject. Various VAChT binding indices were tested including k₃, the total volume of distribution, V_T, and the binding potential, BP_ND. II. The venous time points were used in conjunction with a routine that fits the shape of the input function (anchored to the scale of the metabolite-corrected venous samples) in addition to the kinetic rate constants. III. Reference tissue methods were tested, using cerebellum as the reference. IV. A late summed image normalized to cerebellum, aimed at testing whether a simple “FDG-like” approach may be feasible for FEOBV, was compared to the reference tissue approach.

Results: Kinetic analysis showed K₁ values of 0.2-0.3 ml ml⁻¹min⁻¹ in cortical regions. V_ND ranged from 5-9. The dissociation rate, k₄, ranged from 0.003-0.008 min⁻¹. The average k₃ estimates ranged from 0.113 min⁻¹ in basal ganglia to 0.0036 min⁻¹ in occipital cortex, with corresponding BP_ND values of 24.8 and 0.82, respectively (a dynamic range of ∼30).

Conclusions: COVs in k₃ estimates varied from 15-30% (higher COVs in regions with higher VAChT). V_T measures yielded a dynamic range of ∼15, with no reduction in variability. Reference tissue approaches yielded more stable estimates of DVR (1+BP_ND), with COVs ranging from 16% for basal ganglia to 11% for motor cortex. Corresponding COVs for BP_ND were 19 and 42%, respectively. The late static distribution of FEOBV correlated very highly with the DVR estimates from reference tissue models.

P159. Optimization of HYPR filter for detection of smoking-induced dopamine release in [¹¹C]raclopride PET studies

Shuo Wang¹, Su Jin Kim¹, Jenna M. Sullivan¹, Julia Gillard², Jean-Dominique Gallezot¹, Kelly P. Cosgrove³ and Evan D. Morris³

¹Yale PET Center, Yale University, New Haven, Connecticut, USA; ²Department of Psychology, University of Glasgow, Scotland, UK; ³Department of Psychiatry, Yale University, New Haven, Connecticut, USA

Background: HighIY constrained backPRojection (HYPR) is a promising image processing method for maximizing signal-to-noise ratio (SNR) of PET data without sacrificing spatial and temporal resolution (Christian et al., 2010). The goal of this study is to evaluate the impact of different HYPR filters on PET data for use in the detection of smoking-induced dopamine (DA) release in conjunction with our previously published ‘lp-ntPET' model (Normandin et al., 2012) and [¹¹C]raclopride imaging.

Methods: 1. Simulation – Striatal time activity curves (TACs) were simulated with an enhanced tracer kinetic model that included time-varying DA. Four-dimensional image data were simulated by applying noisy TACs to a dorsal-striatal template. Phantom data were then processed with 4 different HYPR filters (3 × 3 × 3, 5 × 5 × 5, 7 × 7 × 7, or 9 × 9 × 9 voxel boxcar kernels). 2. Human Data – HYPR filtering was also tested on experimental data from a healthy human subject who received two B/I [¹¹C]raclopride scans on the HRRT: a baseline scan and a scan that included smoking two consecutive cigarettes while inside the PET scanner. HYPR processing was applied to the dynamic data using 4 different HYPR filters. The HYPR processed dynamic data were fitted with lp-ntPET within a dorsal-striatal mask. ‘F-maps' were create by the F-test comparing the fit of each voxel-wise TAC with lp-ntPET to its corresponding fit with the standard reference model MRTM, which implicitly considers DA to be time-invariant.

Results: Results from simulations showed that 7 × 7 × 7 and 9 × 9 × 9 voxel kernel filters might negatively bias the TACs by over-smoothing the image. Results from the human data showed that the SNR of voxel-wise TACs was improved. The differences between F-maps from the smoking and control scans were accentuated by HYPR processing (larger contiguous activation clusters in the smoking scan, fewer random artifact voxels in the control scan) with the 5 × 5 × 5 voxel kernel filter producing the largest difference between smoking and control F-maps.

Conclusions: HYPR processing may be an effective image pre-processing method to aid in the detection of short-lived, smoking-induced DA release imaged during dynamic [¹¹C]raclopride PET studies. An optimal (5 × 5 × 5 voxel kernel) HYPR filter is useful for detecting significant deflections in striatal TACs due to DA release at the voxel level.

P160. Lactate receptor locations link neurotransmission, neurovascular coupling, and brain energy metabolism

Linda Hildegard Bergersen^1,2,3, Knut Husø Lauritzen¹, Fredrik Lauritzen¹, Maja Puchades⁴ and Albert Gjedde^2,3

¹The Brain and Muscle Energy Group, CMBN, University of Oslo, Oslo, Norway; ²Department of Neuroscience and Pharmacology, University of Copenhagen, Copenhagen, Denmark; ³Center for Healthy Aging, Faculty of Health Sciences, University of Copenhagen, Copenhagen, Denmark; ⁴The Glio and Neurotransmitter Group, CMBN, University of Oslo, Oslo, Norway,

Background: A major issue in neuroenergetics is the mechanism that links neurotransmission to brain energy metabolism. The modern understanding of the regulation of brain energy metabolism as it supports brain functional integrity suggests that a near-equilibrium feed-forward process establishes the necessary metabolic changes in advance of any changes of brain metabolite concentration. The outcome of such an adjustment is a homeostatically maintained environment where major metabolites such as ATP and ADP do not actually change under normal circumstances, because changing fluxes of metabolites are maintained by regulation of enzyme, transporter and receptor activities, rather than by changing concentrations of intermediaries. Hypothesis: Absent changes of ATP and ADP preclude a primary action of these molecules on the activity of mitochondria. Instead, new maps of the location of the lactate receptor GPR81 in brain suggest that depolarization of neuronal membranes acutely and directly affects the rate of aerobic glycolysis and hence the local generation of lactate, with consequences for the local concentration of cAMP and hence of second messenger cascades.

Methods: We used different novel antibodies against GPR81 and confocal immunofluorescence and electron microscopy.

Results: The membrane locations of the GPR81 protein included the postsynaptic density of dendritic spines, and the endothelial part of the blood brain barrier in the mouse hippocampus.

Conclusions: This is the first identification of lactate receptors in cellular and subcellular structures of brain tissue. The location of receptors in endothelial membranes is consistent with a role of lactate in neurovascular coupling (Bergersen & Gjedde 2012; Gordon et al. 2008). The location of receptors in postsynaptic densities is consistent with a role in the regulation of brain energy metabolic mechanisms that also include the active movement of mitochondria to the necks of dendritic spines (Macaskill et al. 2009).

References

Bergersen LH and Gjedde A (2012) Is lactate a volume transmitter of metabolic states of the brain? Front Neuroenerg 4:5; doi: 10.3389/fnene.2012.00005.

Gordon GR, Choi HB, Rungta RL, Ellis-Davies GC, and MacVicar BA (2008) Brain metabolism dictates the polarity of astrocyte control over arterioles. Nature 456:745-749.

Macaskill AF, Rinholm JE, Twelvetrees AE, Arancibia-Carcamo IL, Muir J, Fransson A, Aspenstrom P, Attwell D, Kittler JT (2009) Miro1 is a calcium sensor for glutamate receptor-dependent localization of mitochondria at synapses. Neuron 26:541-555.

P161. Unbiased noninvasive quantification of [¹¹C](+)McN5652 and [¹¹C]MDL100,907 cerebral specific binding in human dynamic PET studies

Yun Zhou, James Brasic, Eram Zaidi, Boris Frolov, Noble George, Weiguo Ye, Andrew Crabb, Rebecca Mellinger Pilgrim and Dean F. Wong

Johns Hopkins University School of Medicine, Baltimore, MD, USA

Background: [¹¹C](+)McN5652 and [¹¹C]MDL100,907 human dynamic PET studies are used for imaging serotonin transporter and 5-HT_2A receptor, respectively. The need for non-invasive quantification of dynamic PET has long been recognized. Although cerebellum is identified as a reference tissue in both [¹¹C](+)McN5652 and [¹¹C]MDL100,907 PET studies, reference tissue models such as simplified reference tissue model (SRTM), result in significantly biased estimation in binding potential (BP_ND), especially for [¹¹C]MDL100,907. The objective of the study is to evaluate a novel population-based plasma input function (PIF) estimation method for unbiased noninvasive quantification of [¹¹C](+)McN5652 and [¹¹C]MDL100,907 specific binding.

Methods: Fifteen [¹¹C](+)McN5652 and twenty [¹¹C]MDL100,907 human 90-min dynamic PET studies with measured PIF were collected. A PIF template was generated as a mean of normalized PIFs. The estimated PIF (ePIF) was obtained by using few blood samples and interpolated by the PIF template for each study. In addition to cerebellum, 5 regions of interest (ROIs), caudate, putamen, thalamus, pons, and midbrain, and 10 ROIs, caudate, cingulate, occipital, obital frontal, parietal, prefrontal, putamen, superior frontal, temporal, and thalamus, were defined on the coregistered MRIs, and copied to dynamic PET images to obtain ROI time activity curves (TACs). The graphical analysis using the Logan plot with PIF was applied to ROI TACs to obtain distribution volume V_T, and BP_ND was then calculated as V_T(ROI)/V_T(cerebellum)-1. For comparison, SRTM and non-invasive Logan plot using reference tissue input were also applied to ROI TACs to estimate BP_ND.

Results: With two blood samples at 5 and 50 min for [¹¹C](+)McN5652, two blood samples at 10 and 65 min for [¹¹C]MDL100,907, highly linear correlations between the BP_ND estimates from the Logan plot using the measured PIF (mPIF) and ePIF were obtained: BP_ND(ePIF)=1.00BP_ND(mPIF)-0.01, R²=0.97 for [¹¹C](+)McN5652, and BP_ND(ePIF)=1.02BP_ND(mPIF)-0.02, R²=0.99 for [¹¹C]MDL 100,907. The BP_ND estimates from ePIF with 3 or more blood samples and optimal sampling scheme were approximately same as those from mPIF. The estimates of BP_ND from ePIF is not biased relative to the estimates of BP_ND from mPIF. The estimates of BP_ND from SRTM were underestimated 40 to 50% in cortex ROIs for [¹¹C]MDL 100,907, and overestimated 12 to 50% for [¹¹C](+)McN5652. By using the means of intercept of Logan plot with mPIF for cerebellum, the BP_ND estimates obtained from the Logan plot with reference tissue input were not biased, and the linear correlations with the BP_ND estimates from the Logan plot with mPIF were also obtained with R²=0.92 and R²=0.96 for [¹¹C](+)McN5652 and [¹¹C]MDL 100,907, respectively.

Conclusions: The population-based PIF estimation method is reliable to non-invasively quantify [¹¹C](+)McN5652 and [¹¹C]MDL100,907 specific binding. The method is also applicable to quantify radioligand-receptor binding PET studies without reference tissue.

P162. VOI-based evaluations of the accuracy of spatial normalization and existing automated VOI methods for neuroimaging

Hyeyun Jung, Hiroto Kuwabara, Noble George, Boris Frolov, Eram Zaidi and Dean F. Wong

Radiology, Johns Hopkins University, Baltimore, MD, USA

Background: The usefulness of conventional structure-based data analyses using volumes of interest (VOIs) may appear diminishing as spatial normalization techniques and voxel-based statistical analyses improve. To address this possibility, we examined the agreement of automated VOIs across subjects in a standard space after confirming acceptable agreement of automated VOIs against manually defined VOIs in individual subjects' MRI space. The results of this study are expected to address whether VOI-based analysis is still justified.

Methods: VOIs were defined on a total of 487 MRI of consecutively studied subjects, by FSL (Patenaude et al., 2011) and Freesurfer (FS; Fischl et al., 2004) using default settings. On a subset of MRIs (n=107), VOIs were defined manually (MM) using MRI intensity thresholds with continuation of within-threshold voxels to surrounding structures eliminated (Tx-mode), and meticulous refinement in three orthogonal views. VOIs included putamen (Pu), caudate nucleus (CN), hippocampus (Hp), and cerebellum (Cb) for MM, thalamus (Th), globus pallidus, Hp, amygdala, and ventral striatum (vS) for FSL in addition to MM VOIs, and 66 cortical VOIs for FS in addition to FSL VOIs. Agreements of VOIs between the three methods were examined in native space using the agreement score (AS) given by the ratio of the number of intersect voxels to the number of union voxels. To evaluate the accuracy of spatial normalization of the unified segmentation method of SPM5 (Fristone and Ashburner 2003; using ICBM probabilistic maps), VOIs were transferred to a standard space according to the transformation matrix given by this method. In standard space, AS was determined by the ratio of the number of intersect voxels of more than 80% of subjects to the number of union voxels of all subjects. AS was reported separately for FSL subcortical VOIs (FSL and FS), and cortical VOIs (FS alone).

Results: Agreement scores (AS) in native and standard spaces are listed in Table 1. Cb VOIs were not compared in native space because of the different definitions among FSL, FS, and MM. In the standard space AS for Cb were 30.6% by FS and 18.5% by FSL.

Conclusions: Despite acceptable agreement in native MRI space, automated VOIs showed poor spatial agreement across subjects in standard space. These findings indicate that the current spatial normalization technique was not as accurate as to justify not performing VOI-based analysis for subcortical regions. Poorer agreements of cortical regions suggested the spatial normalization technique may be even less accurate for cortical regions, although the accuracy of FS cortical VOIs must be established to be certain.

Acknowledgements: NIMH, NIDA, NIAAA, NIBIB, and NIDNS.

P163. Quantification of ¹⁸F-PBR111 PET for TSPO imaging in humans

Qi Guo¹, Alessandro Colasanti¹, Mayca Onega², Aruloly Kamalakaran², Sabina Pampols-Maso², Graham Searle², Paul Matthews^1,3, Eugenii Rabiner^1,2, Federico Turkheimer¹ and Roger Gunn^1,2

¹Imperial College London, UK; ²Imanova Ltd, London, UK; ³GlaxoSmithKline, London, UK

Background: TSPO has been observed in higher density in activated macrophages and microglia across various brain diseases. Recently, it has been found that TSPO from human tissue samples binds 2nd generation TSPO radioligands with either high affinity (high affinity binders, HABs), low affinity (LABs) or expresses both HAB and LAB binding sites (mixed affinity binders, MABs), and the affinity is determined by the rs6971 polymorphism in the TSPO gene [1]. Here, we aim to evaluate the in vivo performance of ¹⁸F-PBR111, one of the 2nd generation TSPO tracers [2], in human brain and compare the in vivo signal with the genetics in a pilot study.

Methods: 10 healthy subjects (4 HABs, 4 MABs and 2 LABs), who were genotyped for the rs6971 polymorphism, were scanned for 120 min following bolus injection of 166.0±11.5 MBq of ¹⁸F-PBR111. The arterial plasma activity was measured and corrected for the presence of metabolites. 1-tissue (1TC) and 2-tissue (2TC) plasma input compartmental models were used to estimate the regional total volume of distribution (V_T).

Results: In all subjects, the unchanged ¹⁸F-PBR111 in plasma decreased rapidly in the first 30 min and reached between 10% to 30% at 120 min. The tissue kinetics were well characterized by a 2TC model, which provided significantly better fits than 1TC as indicated by a smaller AIC (2TC: -38.9±10.3, 1TC: 19.9±6.6), and V_T was estimated with good identifiability (COV<5%). Time stability analysis of V_T indicated that there was a bias with shorter scan duration: ∼10% reduction in V_T across regions for 90min versus 120min. Relatively high V_T was observed in brain stem and thalamus, and lowest in caudate in all the subjects (see Table), which is consistent with previous literature data from ¹¹C-PBR28 [3] (R²=0.93). Whilst caudate cannot be assumed to be a true reference region, it has been previously shown to have low displaceable fraction of TSPO in rats [5]. In order to explore the association between specific binding and genetics independent of plasma variability, we estimated a pseudo binding potential (BP_ref) [4] using grey matter masked caudate as a reference region. The genotypic influence was not observed on the V_T, but was seen on BP_ref and particularly evident in cortical regions.

Conclusions: Binding of ¹⁸F-PBR111 can be quantified with a 2TC model in healthy subjects. When normalized by a pseudo reference region, there was evidence that the cortical binding was influenced by the status of the rs6971 polymorphism in the TSPO gene.

References

[1] Owen et al. J Cereb Blood Flow Metab, Vol. 32, 2012.

[2] Owen et al. J Nucl Med, Vol. 52, 2011.

[3] Fujita et al. NeuroImage, Vol. 40, 2008.

[4] Gunn et al. Synapse, Vol. 65, 2011.

[5] Jones et al. WMIC Meeting, 2011.

P164. Investigation of noise-induced correlations in dual-biomarker parametric imaging from dynamic [¹¹C]PiB PET

Hassan Mohy-Ud-Din¹, Nicolas A. Karakatsanis¹, Julie C. Price², Yun Zhou¹, Susan M. Resnick³, Christopher J. Endres¹, William E. Klunk², Chester A. Mathis², Dean F. Wong¹ and Arman Rahmim¹

¹Johns Hopkins University, Baltimore, MD, USA; ²University of Pittsburgh, Pittsburgh, PA, USA; ³Intramural Research Program, National Institute on Aging, NIH, Baltimore, MD, USA

Background: To investigate the impact of noise on correlations between R₁ (relative flow) and BP (binding potential) parametric images as obtained from dynamic [¹¹C]PiB PET. Motivation and Significance: Generation of R₁ and BP parametric images from dynamic [¹¹C]PiB PET has been proposed as an approach for dual-biomarker imaging of dementia from a single acquisition. This approach has also been considered to evaluate correlations between flow and β-amyloid depositions. We demonstrate that noise-induced correlations pose significant confounds to interpretation.

Methods: For validation, kinetic parameters (K₁, k₂, k₃, k₄) from [¹¹C]PiB studies of normal controls were estimated, averaged and assigned to a mathematical brain phantom, from which time-activity-curves (TACs) were generated using a 2-tissue compartmental model (blood-volume fraction was set to 3%). This was followed by realistic simulations of dynamic frames for the geometry of the HRRT scanner (including 20 noise realizations in the sinogram-space of the same subject; i.e. simulating no biological correlations). OSEM reconstructions (1-10 iterations) for each dataset were generated. Parametric Images of R₁ and DVR (=BP+1) were obtained via the Simplified Reference Tissue Model (SRTM) using (1) the Basis Function Method (BFM) of Gunn et al. (1997), and (2) linearized formulation and regression (LR) of Zhou et al. (2007) to estimate the parameters. Noise-Bias trade-off curves were obtained for R₁ and DVR images, demonstrating reduced bias with increasing iterations at the cost of enhanced noise levels, as expected. Next, Pearson correlation coefficients were determined for each voxel-pair of (R₁, DVR) vectors (across the 20 noise realizations) with lower and upper bounds for a 95% confidence interval, and quantitatively evaluated for each OSEM reconstruction. A hypothesis test for no correlation with a probability value (p-value) of 0.05 was also conducted, and the resulting correlation images were analyzed qualitatively and quantitatively for all OSEM iterations.

Results: Specifically, 13 regions-of-interest (ROIs) were considered. Overall, R₁ and DVR showed statistically significant correlations, across the OSEM iterations, for all ROIs using the BFM approach, while this was also the case for the LR method with the exception of the parietal cortex, pons and occipital pole.

Conclusions: Significant correlations attributed purely to noise were observed between R₁ and DVR parametric images in parametric quantification of dynamic [¹¹C]PiB PET. Caution should be exercised when performing R₁ and DVR analysis, where the detection of biological correlations may be confounded by noise-induced correlations. Future work should explore potential approaches that quantify and account for these associations to provide more accurate estimates of correlations between DVR and R₁ estimates from a single acquisition.

P165. In vivo amyloid deposition in the aging brain: methodological considerations for partial volume correction

Olivier Rousset¹, Pierre-Louis Bazin², Aaron Carass³, Christopher Endres⁴, Amith Harsha⁴, Dzung Pham⁵, Susan Resnick⁶ and Dean F. Wong¹

¹Department of Radiology and Radiological Sciences, Division of Nuclear Medicine, Section of High Resolution Brain PET imaging, Johns Hopkins University, Baltimore, MD, USA; ²Max Planck Institute for Human and Cognitive Brain Sciences, Leipzig, Germany; ³Image Analysis and Communication Lab, Johns Hopkins University, Baltimore, MD, USA; ⁴Department of Radiology and Radiological Sciences, Division of Neuroradiology, Johns Hopkins University, Baltimore, MD, USA; ⁵Center for Neuroscience and Regenerative Medicine, Bethesda, MD, USA; ⁶Gerontology Research Center, National Institute on Aging, National Institutes of Health, Baltimore, MD, USA

Background: PET imaging with amyloid tracers such as [11C]-PiB is contributing to differential diagnosis of Alzheimer's disease (AD) by demonstrating the presence of beta-amyloid plaques. While partial volume correction (PVC) in emission tomography is essential for accurate assessment of radioactivity concentration, this is of even greater significance in the aging and diseased brain where the amount of tissue atrophy further modulates the reduction in apparent tracer uptake. While PVC can be successfully carried out in the normal human brain using anatomy-based probabilistic atlases built upon healthy controls (Rousset et al., 2008), this approach is not optimal in the presence of large anatomical deformations such as enlarged ventricles and cortical atrophy as seen in AD.

Methods: The proposed PVC implementation uses an open-source medical image analysis and batch processing framework, the Java Imaging Science Toolkit (JIST) [1], developed for the Medical Image Processing, Analysis, and Visualization (MIPAV) program [2]. JIST was applied to a sample data set of 6 participants from the Baltimore Longitudinal Study of Aging (BLSA) who underwent a [¹¹C]PiB-PET scan, together with MPRAGE and FLAIR magnetic resonance imaging (MRI). Image processing driven by JIST included isolating the brain from the skull using a joint registration-segmentation approach [3], multispectral image segmentation using the Lesion-Toads plug-in [4], and multi-atlas statistical analysis and registration using the OASIS (Open Access Series of Imaging Studies) database [5]. MRI data were co-registered to the time-averaged dynamic 70-min PET, and segmented into 21 distinct brain regions, including major cortices (bilaterally), striata, thalami, brainstem, cerebellar cortex, as well as white matter lesions commonly seen in older brains. The GTM-PVC approach [6] was applied using the 50% axial recovery coefficient threshold for region-of-interest (ROI) definition [7].

Results: Preliminary results showed significant increase of PiB brain uptake in all major cortices after PVC (Figure), with associated distribution volume ratios (DVRs) underestimated by ≈10-30%. Regions least affected included brainstem and thalami, followed by striatum, while white matter and lesions were found to both be overestimated.

Conclusions: Single time-point calculation of DVRs performed towards the end of the study seems to introduce the least amount of bias (Figure), but at the expense of increased variance due to low counts. The proposed approach allows quantification of the brain time-activity curves in their entirety, hence preserving the kinetics of PiB uptake and its potential clinical significance. Application of the GTM-PVC using the JIST pipeline and associated MIPAV software to larger cohorts might provide better correlates between regional levels of beta-amyloid deposits and the diagnosis, stage, progression, and treatment of Alzheimer's disease.

References

[1]Lucas BC et al., Neuroinformatics 8:5-17 (2010).

[2]McAuliffe MJ et al., IEEE-CBMS 381-386 (2001).

[3]Carass A et al., 4th IEEE ISBI 656–659 (2007).

[4]Bazin PL, Pham DL, J Cogn Neurosci 19:1498–1507 (2007).

[5]Marcus DS et al., J Cogn Neurosci 19(9):1498-1507 (2007).

[6]Rousset OG et al., J Nucl Med 39(5):904–911 (1998).

[7]Rousset OG et al., J Nucl Med 49(7):1097–1106 (2008).

P166. Evaluation of noninvasive estimation of distribution volume and distribution volume ratio method for unbiased quantification of multi-center [¹¹C]PiB human dynamic PET studies

Yun Zhou¹, Susan Resnick², Jitka Sojkova², Julie Price³, Chester Mathis³, William Klunk³, Brian Lopresti³, Gwenn Smith¹ and Dean F. Wong¹

¹Johns Hopkins University School of Medicine, Baltimore, Maryland, USA; ²Laboratory of Behavioral Neuroscience, Intramural Research Program, National Institute on Aging, National Institutes of Health, Baltimore, Maryland, USA; ³University of Pittsburgh School of Medicine, Pennsylvania, USA

Background: Distribution volume (V_T) ratio (DVR) is commonly used as a measure of specific [¹¹C]PiB specific tissue binding. A standard method to measure V_T is to fit a kinetic model with plasma input function (PIF) to measured tissue time activity curves (TAC), where the PIF is tracer concentration in plasma obtained by arterial blood sampling procedure. Although defined as the ratio of tissue V_T to reference tissue cerebellum, DVR is often determined by reference tissue-based kinetic modeling methods without estimation of V_T. This avoids the need for arterial blood sampling for V_T estimation. The objective of the study is to evaluate a noninvasive method for V_T and DVR estimation that utilizes multi-center dynamic [¹¹C]PiB PET data.

Methods: Twenty-four (8 healthy controls, 10 mild cognitive impairment (MCI) subjects, and 6 Alzheimer's disease patients) 90-min dynamic PET studies with measured PIF (mPIF) by arterial blood sampling were performed at the University of Pittsburgh. Seventy-eight (66 controls, 12 MCI subjects) 90-min [¹¹C]PiB dynamic PET studies without PIF measurements were collected from the Baltimore Longitudinal Study on Aging, and twenty-seven dynamic PET scans (10 controls, 12 patients of depressive disorder, 5 MCI subjects) from the geriatric depression imaging project performed at Johns Hopkins. Regions-of-interest (ROIs) were defined on the coregistered MRIs, and copied to dynamic PET images to generate TACs. A PIF template was generated as a mean of the normalized PIFs. The estimated PIF (ePIF) was obtained by using a few blood samples and interpolated based on the PIF template for each study. The ROI V_T was estimated by applying the invasive Logan plot (t^*=50 min) with either measured PIF (mPIF), PIF template, or estimated PIF (PIF). The DVR was then calculated as: ROI V_T/Cerebellum V_T. The mPIF V_T and DVR were used for reference comparison.

Results: The linear correlation between the ROI V_T (PIF template) and V_T from mPIF (V_T(mPIF)) increased from R²=0.52 to 0.83 when V_T (PIF template) was scaled by injected dose (mCi)/body weight (kg). Highly linear correlations between the V_T using ePIF and V_T(mPIF) were obtained, with the following relationships: V_T(ePIF, 1 blood sample, 7 min)=1.03^*V_T(mPIF) -0.07, R²=0.95; V_T(ePIF, 2 blood samples, 7 and 60 min)=1.03^*V_T(mPIF) -0.10, R²=0.98. There were no significant difference between V_T(ePIF) and V_T(mPIF) (p>0.30). The DVR obtained by V_T(PIF template) was highly correlated with DVR obtained by mPIF as: DVR(PIF template)=0.99^*DVR(mPIF) +0.01, R²=0.94; and DVR(ePIF, blood samples, 7 and 60 min)=0.98^*DVR(mPIF) +0.02, R²=0.97. There were no significant differences between DVR(PIF template), DVR(ePIF, blood samples at 7 and 60 min) and DVR(mPIF) (p>0.50).

Conclusions: Noninvasive unbiased V_T and DVR estimates can be obtained with a minimum of one blood sample. Future studies will investigate whether this method is feasible using a single venous plasma sample with population metabolite correction, the suitability of such a method for voxel level analyses, and their applications in multi-center [¹¹C]PiB dynamic PET studies.

P167. Improving microPET quantification: semi-automated volumes of interest

Ayon Nandi¹, Hiroto Kuwabara¹, Weiguo Ye¹, Anil Kumar¹, Kyungah Chun¹, Heather Valentine¹, Mary Blue² and Dean F. Wong¹

¹Johns Hopkins University, Baltimore, Maryland, USA; ²Kennedy Krieger Institute, Baltimore, Maryland, USA

Background: Genetically manipulated animal models have rendered microPET a vital tool for enhancing our understanding of pathophysiological mechanisms of neuropsychiatric disorders. We propose and evaluate practical solutions for two major confounding factors of mice microPET, namely multi-animal scans (to be efficient) which often cause crooked head positions, and resolution differences between PET and MRI (used for defining volumes of interest (VOIs) or VOI templates) which may cause over- or under-estimation of PET outcome variables.

Methods: Seven mice were studied: 4 wildtype CD1 mice, and 3 mice which carried mutation of the MeCP₂ gene (the Adrian Bird mouse model of Rett's syndrome, provided by Mary Blue). Animals were scanned (1-2 at a time) with the GE Vista microPET for 60 min following a bolus injection (∼0.3 μCi) of [11C]methylphenidate (MP), a dopamine transporter radioligand, under isoflurane anesthesia. CT scans were obtained with a Gamma Medica X-SPECT/CT scanner. A template of volumes of interest (VOIs), prepared on a MRI was used as follows: 1) the brain outlines of the standard MRI were manually aligned to the animal's skull seen on upright-oriented CT (i.e., the mid-plane is vertical). 2) VOIs were transferred to the upright CT's space. 3) CT and averaged PET were coregistered using SPM coregistration routine. 4) PET was resampled at the upright CT's space. 5) time-activity curves of regions were obtained by applying VOIs obtained in Step 2 (Method 1 or M1, using standard VOIs). In Method 2 or M2, VOIs were modified on averaged PET with the volume kept constant. Four raters performed M1 and 2 to obtain binding potential (BP_ND) in left and right striatum using multilinear reference tissue method with 2 parameters (MRTM2; Ichise et al., 2002) and reference tissue graphical analysis (RTGA; Logan et al., 1996). Inter-rater reliability was assessed by average absolute deviation (AAD; =given by abs(v-mv)/mv^*100 where v and mv stand for BP_ND of individual raters and the mean across raters), and intra-class correlation coefficient (ICC) which was obtained with program R.

Results: The Table lists ADD (mean +/- SD) and ICC. M1 underestimated BP_ND by 50.9+/-39.1% (t=-14.4; p<0.0001) for MRTM2, and 43.0+/-30.9% for RTGA (t=-13.48; p<0.0001) compared to M2.

Conclusions: The finding that M1 was internally consistent amongst raters suggested that the proposed steps 1 through 4 yielded consistent VOI placement among raters. However, the underestimation of BP_ND by M1 indicated M1 could result in slight but influential misplacement of VOIs which could be corrected by manual adjustments of M2. Therefore, this study suggested that manual adjustment of initial automated VOIs could be useful for brain studies of single or multiple animals with microPET.

P168. Evaluation of inter-scanner differences in effect of radioactivity from outside FOV

Takahiro Shiraishi, Yasuyuki Kimura, Iwao Kanno, Tetsuya Suhara and Hiroshi Ito

Molecular Imaging Center, National Institute of Radiological Sciences, Chiba, Japan

Background: We found inter-scanner differences in dopamine receptor imaging. In order to evaluate different effects of physical performances on tracer kinetics images, we tested accuracy in deadtime correction and scatter correction using phantoms mimicked the head and whole body of human subject.

Methods: The two PET scanners used were Eminence Sophia (Shimadzu, Corp) and EXACT HR+ (Siemens). Each scanner was tested the following examinations. The deadtime correction was tested using Hoffman phantom filled by 3mCi C-11 radioactivity. The effects of the scatter emitted from the whole bogy were evaluated using three uniform pool phantoms, one inside of FOV mimicking the head and two outside of FOV mimicking the whole body. These pools are filled by 3 mCi of [C-11] radioactivity for the inside of FOV and two of 12 mCi for the outside FOV. Dynamic scan were performed over 4 hours. Images were reconstructed by the filtered back-projection method. Images were analyzed using PMOD.

Results: Both PET scanners did not show any significant systematic errors in the deadtime correction. The existence of a massive (8 times) radioactivity in the outside the FOV was overestimated image concentrations of the pool at the inside FOV by 10% in Eminence, and this overestimation was continued till the radioactivity was decayed. However, the overestimation was reduced less than a few percent after software correction. In HR+ scanner, the same experiment resulted in the underestimation of the image concentration by 10-20% for the early phase but the underestimation was become negligible in late phase.

Conclusions: The phantom studies mimicking the head and whole body for the receptor imaging did not explain the inter-scanner difference in dopamine imaging.

P169. Inter-scanner differences in dopamine receptor imaging

Yasuyuki Kimura, Hiroshi Ito, Takahiro Shiraishi, Makiko Yamada, Fumitoshi Kodaka, Harumasa Takano, Hironobu Fujiwara, Hitoshi Shimada, Iwao Kanno and Tetsuya Suhara

Molecular Imaging Center, National Institute of Radiological Sciences, Chiba, Japan

Background: Positron emission tomography is a useful tool to investigate the effects of treatments in drug development and to diagnose diseases in their early stages without clinical symptoms. To perform multi-center studies or compare the findings with previously reported ones, inter-scanner differences in the quantification need to be considered. The causes of the differences could be more complicated in the dynamic imaging in the quantification of neuroreceptor in the brain than in the static imaging. We investigated inter-scanner differences between two clinical PET scanners, EXACT HR+ (Siemens/CTI, Knoxville, TN) and Eminence Sophia (Shimadzu Corp, Kyoto, Japan), in the quantification of dopamine D2 receptors using two radioligands with different affinity to dopamine D2 receptors, 11C-raclopride and 11C-FLB 457.

Methods: Twelve healthy male volunteers were scanned twice after injections of 11C-raclopride and 11C-FLB 457 with HR+ or Eminence (6 males per scanner, age matched). 60 min-dynamic scan was performed after an injection of 222 MBq of 11C-raclopride, then 90 min-dynamic scan was performed after an injection of 222 MBq of 11C-FLB 457 with an interval of 60 min. Post-filter was adjusted to match the spatial resolution of reconstructed images. Binding potential of several regions was calculated by the simplified reference tissue model using cerebellum as a reference region.

Results: The binding potential of striatal regions measured with 11C-raclopride showed no significant differences between the two scanners, while that of the cortical regions measured with 11C-FLB 457 showed significant differences (Eminence showed ∼30% lower BPND than HR+). The time activity curves of 11C-FLB 457 measured with Eminence showed gradually higher concentration of radioactivity than those measured with HR+ in all regions.

Conclusions: In the quantification of dopamine D2 receptor in the brain, inter-scanner differences varied between 11C-raclopride and 11C-FLB 457.

P170. Direct high yield no-carrier added radiosynthesis of [F-18]catecholamines

Zheng Miao¹, Bin Shen¹, Linlin Qin², Kiel Neumann², Stephen DiMagno² and Frederick T. Chin¹

¹Molecular Imaging Program at Stanford (MIPS), Departments of Radiology and Bioengineering, Bio-X Program, Stanford University School of Medicine, Stanford, CA 94305-5484, USA; ²Department of Chemistry, University of Nebraska, Lincoln, Nebraska, USA

Background: 6-[¹⁸F]-L-FDOPA and 6-[¹⁸F]fluorodopamine (6-[¹⁸F]FDA) are potentially valuable radiotracers for imaging chromaffin-cell derived tumors, such as neuroblastoma and pheochromoctyoma, and imaging pancreatic cancer. Since 6-[¹⁸F]-L-FDOPA readily crosses the blood-brain barrier, it has found widespread use in imaging neurological disorders (e.g., Parkinson's disease). Despite the potential utility of these compounds, their difficult syntheses have hindered widespread use. Although some nucleophilic routes exist, 6-[¹⁸F]-L-FDOPA is still generally prepared by electrophilic fluorination despite the relatively low specific radioactivity (SR) of the final product. Current multistep, nucleophilic, no-carrier added (n.c.a.) syntheses rely on an approach in which an activating group on the aryl ring is eventually transformed into the amino acid functionality. In contrast, here we describe a two-step, n.c.a. synthesis using diaryliodonium salt precursors to provide 6-[¹⁸F]-L-FDOPA and 6-[¹⁸F]FDA with high SR for routine clinical use.

Methods: 6-[¹⁸F]-L-FDOPA and 6-[¹⁸F]FDA were synthesized by nucleophilic fluorination using a GE TRACERlab automated module equipped with a glassy carbon reactor. [¹⁸F]fluoride (n.c.a.) in [¹⁸O]water was added directly to the reactor, which was previously charged with Kryptofix (3 mg) and K₂CO₃ (0.5 mg) in 95:5 CH₃CN/water (2 mL). Residual water was removed by azeotropic distillation with an additional CH₃CN (2 mL). Diaryliodonium triflate and hexafluorophosphate (PF₆) salts precursors were prepared in-house. An CH₃CN solution of a diaryliodonium triflate or PF₆ salt precursor (10 mg in 0.5 mL) was added to the reactor, CH₃CN was removed under vacuum (50°C), toluene (1 mL) was added, and the vial was heated to 150°C for 5 m. The crude reaction mixture was passed through a silica Sep-pak Plus cartridge, and the [¹⁸F]fluorinated derivative was eluted using EtOAc (3.5 mL) and transferred to a glass vial in a secondary reactor. Following removal of the organic solvents, 47% aqueous HI (250 μL) was added and the resulting solution was heated at 155°C for 5 min, cooled, and quenched with 2 M citrate buffer (800 μL). The crude product was purified by semi-preparative HPLC and the isolated formulated product was confirmed by analytical HPLC.

Results: Radiochemical yields (RCYs) were decay-corrected to end of bombardment. Diaryliodonium triflate and PF₆ salts afforded RCYs ranging from ∼50-80% and ∼10-20% obtained respectively using this methodology. A stark difference was noted between the reactivities of the triflate and PF₆ salts of the catecholamine-derived diaryliodonium derivatives, with triflates providing better yields and higher SR. Radiochemical purities exceeded 99% with both salts but higher specific radioactivity could be achieved with the triflate versus the PF₆ salt. For example, 6-[¹⁸F]FDA was prepared in 71±9% yield (SR>2 Ci/μmol) from the triflate derivative, while 6-[¹⁸F]-L-DOPA was prepared in 15±3% RCY (SR>0.2 Ci/μmol) from the corresponding diaryliodonium-PF₆ salt. The purities and SRs of both 6-[¹⁸F]-L-FDOPA and 6-[¹⁸F]FDA from the corresponding triflate salts were sufficient for future clinical applications.

Conclusions: Clinical-grade 6-[¹⁸F]-L-FDOPA and 6-[¹⁸F]FDA with high SRs can now be made readily accessible to the clinic by employing a two-step method with diaryliodonium salt chemistry and n.c.a. [¹⁸F]fluoride in a GE TRACERlab module.

Table. Binding potential (BP_ND) estimates for clinical (high-IA) and control (low-IA) groups measured by 2TCM with plasma input function and reference tissue analyses, with corresponding statistics for group differences.

Table 1. ICC, mean absolute variability, and BP_ND for each session as mean±standard deviation.

Table 1. Comparison of linear and nonlinear registration methods on BP_ND estimates in small brain structures.

Table 1. Comparison of mean binding potentials estimated with 4 different reference regions in healthy controls and methamphetamine-dependent participants. MA;methamphetamine. Std; standarddeviation.

Table 1.

		Best Agreement		Worst Agreement
	Mean+/-stdv	Structure	AS	Structure	AS
Native space
FS vs. MM	68.3+/-8.3%	Pu	74.9+/-4.3%	Hp	56.7+/-7.8%
FSL vs. MM	70.5+/-9.7%	Pu	82.6+/-10%	Hp	58.2+/-11.3%
FS vs. FSL	71.0+/-6.8%	Th	77.5+/-38%	vS	58.1+/-20.4%
Standard space
FSL-subcortical	21.9+/-8.6%	Th	37.5%	vS	9.3%
FS-subcoritcal	18.2+/-7.4%	Th	30.2%	vS	7.8%
FS-cortical	0.6+/-1.0%	insula	4.9%	25 out of 66	0%

Table.

ROI	VT	BPref
	All (n=10)	All (n=10)	HABs (n=4)	MABs (n=4)	LABs (n=2)
Pons	4.11±1.92	0.43±0.16	0.48±0.20	0.34±0.12	0.51±0.10
Thalamus	3.86±1.54	0.38±0.12	0.47±0.12	0.32±0.07	0.32±0.13
Cortex	3.59±1.38	0.29±0.14	0.38±0.15	0.27±0.09	0.15±0.07
Cerebellum	3.58±1.67	0.25±0.13	0.34±0.15	0.16±0.09	0.25±0.03
Caudate	2.79±1.08	—	—	—	—

Table.

	MRTM2		RTGA
	AAD	ICC(1)	AAD	ICC(1)
Method 1	11.1% +/- 4.8	0.84*	10.1% +/- 4.1	0.86*
Method 2	3.5 +/- 2.8	0.95*	3.6% +/- 2.6	0.95*

p<0.001 for ICC(1) for all cases.