Impact of scatter correction on D2 receptor occupancy measurements using 123I-IBZM SPECT: Comparison to 11C-Raclopride PET

NeuroImage 50 (2010) 1511–1518

Contents lists available at ScienceDirect

NeuroImage j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / y n i m g

Impact of scatter correction on D2 receptor occupancy measurements using 123I-IBZM SPECT: Comparison to 11C-Raclopride PET S. Bullich a,⁎, A. Cot b,c, J. Gallego b,d, R.N. Gunn e,h, M. Suárez a, J. Pavía c,f,g, D. Ros b,c,g, M. Laruelle e,h, A.M. Catafau i a

Neurosciences Imaging Group, Molecular Imaging Centre (CRC-CIM), Barcelona Biomedical Research Park, C/ Dr. Aiguader 88, soterrani-1, 08003 Barcelona, Spain Biophysics and Bioengineering Unit, University of Barcelona, Barcelona, Spain Networking Research Center on Bioengineering, Biomaterials and Nanomedicine (CIBER-BBN), Barcelona, Spain d Institut de Tècniques Energètiques, Universitat Politècnica de Catalunya, Barcelona, Spain e Clinical Imaging Centre, GlaxoSmithKline, London, UK f Nuclear Medicine Dept., Hospital Clínic, Barcelona, Spain g Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Barcelona, Spain h Department of Neurosciences, Imperial College, London, UK i Discovery Medicine, Neurosciences Centre of Excellence for Drug Discovery, GlaxoSmithKline, Barcelona, Spain b c

a r t i c l e

i n f o

Article history: Received 29 June 2009 Revised 28 December 2009 Accepted 7 January 2010 Available online 18 January 2010 Keywords: 123 I-IBZM SPECT 11 C-Raclopride PET D2 receptor occupancy Schizophrenia Antipsychotic

a b s t r a c t Reported values of D2 receptor occupancy (RO) achieved by antipsychotic drugs tend to be lower when measured with 123I-IBZM SPECT than with 11C-Raclopride PET. Image degrading factors such as attenuation, distance-dependent collimator response and scatter could account for this difference. While attenuation correction is routinely applied to SPECT images, the other degradations are not usually accounted for. The aim of this work was to assess the impact of scatter correction on D2 RO quantification with 123I-IBZM SPECT, and to compare the results of both corrected and un-corrected SPECT values with 11C-Raclopride PET measurements. Phantom experiments as well as within-subject human data from a previous study were used for this purpose. SPECT images were reconstructed using filtered back-projection including attenuation correction (FBPA), ordered subsets expectation maximization including attenuation and point spread function corrections (OSEMA+PSF) and ordered subsets expectation maximization including attenuation, point spread function and scatter corrections (OSEMA+PSF+SCT). PET images were reconstructed using the FBP algorithm and corrected for attenuation, scatter, random coincidences and dead time. Quantification of receptor availability was performed using the tissue ratio at pseudoequilibrium for SPECT, and the simplified reference tissue model (SRTM) for PET. Analysis was performed using both occipital cortex (occ) and cerebellum (cer) as reference regions for both modalities. When images were reconstructed using FBPA, SPECT D2 RO values were significantly lower as compared with PET leading to a D2 RO difference of −20% (CI95%: −13, −27%) (occ) and −23% (CI95%: −14, −31%) (cer). When images were reconstructed using OSEMA+PSF, SPECT D2 RO values were also lower as compared with PET leading to a D2 RO difference of −21% (CI95%: −14, −27%) (occ) and −24% (CI95%: −18, −30%) (cer). When images were reconstructed using OSEMA+PSF+SCT, the D2 RO bias was reduced to −6% (CI95%: 0, −13%) (occ) and −11% (CI95%: −4, −18%) (cer). These data suggest that the scatter correction plays a major role in explaining the differences between D2 RO measurements using 123I-IBZM SPECT and 11C-Raclopride PET. Crown Copyright © 2010 Published by Elsevier Inc. All rights reserved.

Introduction In vivo dopamine D2 receptor availability and the degree of D2 receptor occupancy (D2 RO) induced by drugs or endogenous dopamine can be measured by Single Photon Emission Computed Tomography (SPECT) and Positron Emission Tomography (PET), using selective radioligands such as 123I-IBZM and 11C-Raclopride (Talbot

⁎ Corresponding author. Fax: +34932240359 E-mail address: [email protected] (S. Bullich).

and Laruelle, 2002; Laruelle, 2000a; Verhoeff et al., 1991; Kegeles et al., 1999, 2000). Antipsychotic-induced striatal D2 RO measured using these two ligands has been extensively reported in the literature, but for a given drug between studies comparison is difficult due to differences in patient sample characteristics and study design. Clarification of any systematic differences between SPECT and PET D2 RO measurements would be important for accurate application of these techniques in drug development, and would improve the understanding and interpretation of the antipsychotic D 2 RO literature. In previous work, our group has compared within-subject striatal D2 RO using 11C-Raclopride PET and 123I-IBZM SPECT

1053-8119/$ – see front matter. Crown Copyright © 2010 Published by Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2010.01.013

1512

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

measurements, and found D2 RO values to be consistently lower when measured by 123I-IBZM SPECT in comparison to those measured by 11C-Raclopride PET (Catafau et al., 2009). Moreover, it was shown that the assumptions made in the quantification of 123I-IBZM SPECT, when applying a tissue ratio method at pseudoequilibium after a single bolus injection, were unlikely to explain the differences found between 123I-IBZM SPECT and 11C-Raclopride PET D2 RO measurements. It was therefore hypothesized that the appropriate correction of degrading factors during image reconstruction could reduce the bias. The reconstruction of neurotransmission SPECT images is affected by a number of degrading factors such as statistical noise, attenuation, distance-dependent collimator response and scatter. SPECT images are usually reconstructed using the filtered backprojection algorithm (FBP) which is widely available in common software packages for SPECT reconstruction. Attenuation correction is usually included, but the other degradation corrections are not widely applied to 123I-IBZM SPECT images. Iterative algorithms, such as ordered-subsets expectation-maximization (OSEM) (Hudson and Larkin., 1994), have been proposed as an alternative to FBP, since they allow inclusion of models for the emission-detection process in the reconstruction algorithm. Previous studies have assessed the performance of OSEM in simulated studies (Pareto et al., 2003) and in clinical data (Koch et al., 2006; Catafau et al., 2008). 3D-OSEM reconstruction algorithms for SPECT which include attenuation correction, collimator-detector response and scattering correction have been developed and several of the methods have proved to be accurate tools for absolute quantification of brain studies with 99mTc-labelled tracers (Beekman et al., 2002; Cot et al., 2005). However, when 123I-labeled tracers are used, scatter correction is more complex because, in addition to the scattered photons of the main emission of 159 keV, this radioisotope also emits other photons (346, 440, 505, 529 and 539 keV mainly) which scatter in the collimator/detector system and may be detected within the 159 keV energy window (Dobbeleir et al., 1999; Cot et al., 2004; Staelens et al., 2007). Recently, the interaction of photons emitted by 123I with the collimator/detector system has been modeled (Cot et al., 2006; Small et al., 2006; Tanaka et al., 2007) and algorithms to correct scatter in 123I SPECT imaging have been developed (Larsson et al., 2006; Crespo et al., 2008). A more detailed description of scatter correction algorithms used in SPECT can be found elsewhere (Zaidi and Koral, 2004). The impact of scatter correction on SPECT regional tracer distribution (Ito et al., 1999) and kinetic parameters (Fujita et al., 2004; Kim et al., 2001b; Moore et al., 2001) has been studied. However, the impact of scatter correction on other parameters such us the D2 RO measurement remains unclear. The aim of this work was to assess the impact of the scatter correction on the striatal D2 RO quantification when 123I-IBZM SPECT is used, and to compare the results of both corrected and un-corrected SPECT values with the 11C-Raclopride PET measurements. Phantom experiments as well as within-subject human data acquired in our previous study (Catafau et al., 2009) were used for this purpose. Materials and methods

or long scanning sessions over time, such as the tissue ratio (TR) method at tracer equilibrium are most commonly applied when using SPECT (Laruelle, 2000b; Catafau et al., 2008). Moreover, even when the tracer is at equilibrium, TR-based quantification may be biased due to other factors deriving from reconstruction issues. The TR method uses the specific to non-displaceable uptake ratio (SUR) as a measure which is proportional to the receptor availability. The SUR can be calculated as follows: SUR =

CT −CREF CREF

ð1Þ

where CT and CREF are the mean counts per voxel in the target region with specific uptake and in a reference region devoid of receptors, respectively. The RO is a measure of the percentage of change in the receptor availability between a baseline condition and after receptor blockade by endogenous or exogenous compounds such as ongoing patient medication. However, it is often difficult to scan patients in a drug naive or free condition (baseline) and as a consequence, the values obtained in a normal control group are frequently used to study drug-induced D2 RO in patients (Knable et al., 1997; Dresel et al., 1999; Tauscher et al., 1999). Then, the percentage receptor occupancy (RO) can be estimated as B

RO = 100

P

SUR −SUR SURB

ð2Þ

where SURB and SURP are the specific to non-displaceable uptake ratios at baseline condition (either in a control group or in the same patient) and after receptor blockade in patients, respectively. Several papers have shown that the measured SUR (SURM) with SPECT can be quite different from the true SUR (SURT) when the correction for degrading factors is not included in the reconstruction algorithm. Nevertheless, it has been highlighted that there is a linear relation between SURT and SURM (Koch et al., 2005; Cot et al., 2005; Soret et al., 2003, 2006; Varrone et al., 2008; Crespo et al., 2008; Koch et al., 2006), which can be expressed as SURM = α· SURT + β

ð3Þ

where α and β are constants. The measured receptor occupancy (ROM) can then be expressed in terms of the true receptor occupancy (ROT) as ROM = ROT

1 β 1+ αSURBT

= ROT ð1−

β Þ SURBM

ð4Þ

where SURBM and SURBT are the measured and true SUR in the baseline condition (see Appendix A for derivation). The percentage of the bias in RO measurements (ROBIAS) can be expressed as   ROM − ROT −β ROBIAS = 100 ð5Þ = 100 ROT SURBM

A theoretical model of the influence of reconstruction degradation factors on the D2 RO bias was developed. Then, phantom data were used to characterize the model parameters expected in the SPECT system, and their dependence on scatter. Finally, the impact of scatter correction in the D2 RO bias was assessed using human data.

The RO bias equation shows that the ROM with SPECT will only be equal to ROT when β equals zero. Otherwise, when β N 0, the ROM will underestimate the ROT and, when β b 0, the ROM will overestimate the ROT. In both cases, the bias is reduced when β/SURBM or β/α approaches zero. Moreover, for certain ligands where the SURBT is high enough, this bias may be neglected.

Theoretical background

Phantom acquisitions

The gold standard method to measure receptor availability is the tracer kinetic compartmental analysis based on dynamic time-course data and an arterial plasma input function (Laruelle et al., 2002). However, simpler methodologies which avoid either blood sampling

Experimental measurements were carried out with the anthropomorphic striatal brain phantom (Radiology Support Devices, Inc., Long Beach, CA, USA) in the SPECT camera in order to estimate the slope (α) and y-intercept (β) in Eq. (3). This phantom consists of 5 realistic

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

compartments corresponding to the left and right caudate nuclei, left and right putamen, and the rest of the brain. All these structures have a normal human brain shape and size and are surrounded by two different materials that act as soft tissue and skull. The phantom was filled with different concentrations of 123I in the striatum (caudate and putamen) and brain in order to reproduce a range of SUR values, covering normal values as well as a range of occupancy values. A total of 9 acquisitions were performed, three for each of the following SUR values of 0.00, 1.97 and 3.73. Experimental projections were acquired on a three-headed Prism 3000S camera (Picker) fitted with ultra-high-resolution fan-beam collimators in a 128 × 128 matrix. The projection data set consisted of 120 projections over 360° with a bin size of 3.56 mm and a 20% energy window centered on 159 keV (143–175 keV). A radius of rotation of 145 mm was selected. All the experimental studies had over 2 million counts. Human data acquisition Measurements from 9 healthy volunteers (5 males/4 females, 26.89 ± 6.39 years) and 17 patients (9 males/8 females 37.16 ± 10.18 years) on antipsychotic treatment (risperidone, olanzapine, or clozapine) taken from our previously reported study (Catafau et al., 2009) were used to assess the impact of scatter correction on the measured D2 RO. All subjects underwent both a 123I-IBZM SPECT and a 11 C-Raclopride PET in a randomized order on different days within a single week. The time elapsed after last antipsychotic dose administration until the scanning session was the same for the SPECT and PET scans. Each subject showed comparable antipsychotic plasma levels during both SPECT and PET procedures as described in Catafau et al. (2009). Moreover, each subject underwent a magnetic resonance (MR) scan for delineating anatomical regions of interest (ROIs) and coregistration with SPECT and PET images. The study was approved by the local Ethics Committees and Ministry of Health, and written informed consent was obtained in all cases prior to inclusion in the study. 123

I-IBZM SPECT 123 I-IBZM (GE Healthcare, Netherlands) was administered as a single i.v. injection flushed with 10 mL of saline. Subjects received a mean 123I-IBZM dose of 213.3 ± 10.9 MBq. To decrease radiation exposure to the thyroid gland, potassium perchlorate (8 mg/kg) was administered p.o. 20 min before 123I-IBZM injection. For SPECT and MR coregistration, four multi-modality SPECT/MR markers (MM3003, IZI Medical Products Corp., Baltimore, USA), each filled with 0.074 MBq of 99mTc, were attached to the subject's head at the level of the forehead (temples) and behind the ears (mastoids) to allow for accurate SPECT-MR registration. All the subjects underwent a 30 min. scan that started during the pseudoequilibrium phase (120 min p.i.). The radius of rotation was chosen to be the minimum required for each subject. The remaining acquisition parameters were the same as the ones used for phantoms. 11

C-Raclopride PET PET was performed using a Siemens ECAT-EXACT HR+camera. Prior to 11C-Raclopride injection, transmission scans were obtained to define each attenuation map. After injection of a mean 11C-Raclopride dose of 257.7 ± 28.6 MBq, a series of 26 emission scans were obtained from time of injection up to 90 min (8 × 15 s, 3 × 60 s, 5 × 120 s, 5 × 300 s, 5 × 600 s). MR A T1 weighted 3D MR scan was performed using a superconductive 1.9 T system (Prestige 2T, General Electric) equipped with a head coil. An axial three-dimensional spoiled gradient-echo slab was positioned to include the entire head, and images were acquired

1513

with the following parameters: repetition time, 25 ms; echo time, 6 ms; flip angle, 28°; field of view, 25 × 25 cm2; matrix size, 256 × 256; section thickness, 2 mm with no interslice gap; and number of excitations, 1. Reconstruction Phantom data SPECT acquisitions were reconstructed using different methods. Firstly, images were reconstructed using the filtered-backprojection (FBP) algorithm implemented in the Odyssey-FX software (Philips), and by applying a two-dimensional Butterworth filter (fifth order, 0.4 cycle/pixel). Attenuation was corrected using the Chang algorithm with a manually fitted elliptic attenuation map on each slice. The attenuation coefficient of the soft tissue for the main emission of 123I was used (μ = 0.14 cm−1) (Shiga et al., 2002; Soret et al., 2003). SPECT phantom acquisitions were also reconstructed with a 3D iterative reconstruction algorithm based on the OSEM algorithm (Hudson and Larkin., 1994) and described elsewhere (Crespo et al., 2008). First, OSEM included the attenuation and point spread function modeling (PSF) of the gamma-camera incorporated into the transition matrix (OSEMA+PSF). Second, in addition to the attenuation and PSF corrections, the algorithm incorporated scatter correction (OSEMA+PSF+SCT). The scatter contribution was calculated by using the SimSET Monte Carlo code (Harrison et al., 1993) with the corresponding probability density function of the gamma camera for all the photons of the 123I spectra with the method described in Crespo et al. (2008). The scatter estimate was calculated using the reconstructed image of the original projections without any scatter compensation. This reconstructed image was the input activity distribution of the SimSET simulator weighted by the emission yields of the main 123I ray. Nevertheless, the original projections had to be masked before reconstruction because of scatter contamination. Thus, the values located outside of the phantom were discarded to avoid them interfering in the reconstruction process. Masking was performed by applying a geometric projector to the attenuation map and giving a value of 1 inside the resulting projections and 0 outside their limits. All the OSEM reconstructions were performed using 30 subsets and 8 iterations. Human data SPECT. SPECT acquisitions were reconstructed using the same algorithms as the phantom data. Uniform attenuation maps (μ = 0.14 cm−1) used for OSEM reconstruction were obtained by thresholding PET transmission maps previously coregistered to the SPECT images. PET. PET frames were reconstructed using the FBP algorithm and a Hanning filter (kernel FWHM = 4.9 mm) on a 128 × 128 matrix size with pixel sizes of 2.57 mm within slice and 2.43 mm in the axial direction. Images were corrected for attenuation, scatter, random coincidences and dead time with the software provided by the manufacturer. Image analysis and quantification Phantom data quantification A numerical phantom obtained from the CT scan (256 × 256 × 196 voxels with a size of 0.89 × 0.89 × 0.89 mm) of the anthropomorphic striatal brain phantom was used to quantify the reconstructed phantom images of the gamma-camera. To this end, a 12-parameter affine transformation was performed to coregister CT and SPECT images. SPM2 software (Wellcome Trust Centre for Neuroimaging, London, UK) allows to perform this affine transformation by using a normalization procedure with an extremely heavy regularization. This

1514

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

transformation rather than a rigid-body registration was used to account for minor mismatch that may occur between image modalities. Quantification of phantom data was done by using regions-of-interest (ROIs) manually drawn on the striatal cavities and the background (reference region) over the corresponding slices of the segmented CT image of the striatal brain phantom. The specific to non-displaceable uptake ratio (SUR) was then calculated according to Eq. (1). Human data quantification SPECT. The corresponding external markers were manually identified on the MR and the FBP reconstructed SPECT scans. Then, rigid-body coregistration was estimated automatically using in-house software which minimized the sum of the squared distances between the corresponding marker positions (Bullich et al., 2009). Regions of interest (ROIs) were manually drawn by the same investigator on multiple slices of the MR image covering the whole striatum, occipital cortex and the cerebellum. The mean volumes for the ROIs were 10.52, 29.85 and 30.14 mL, respectively. These ROIs were then placed on the coregistered SPECT and mean counts from the ROIs were calculated. The TR method at pseudoequilibrium was used for quantification in all subjects. The SUR and RO were calculated as described in Eqs. (1) and (2), respectively, and two sets of data were obtained, one using the occipital cortex (occ) and the other using the cerebellum (cer) as the reference region. 3D-OSEM reconstructed SPECT images were coregistered to the MR image using the same transformation matrix used to coregister the FBP reconstructed SPECT images. The same ROIs were used to quantify the images reconstructed using the 3DOSEM algorithm and the SURs and ROs were calculated. PET. PET frames were realigned to correct for patient motion between frames and the mean of the realigned frames was obtained. Then, the mean of the realigned frames was coregistered to the MR image and the same transformation was applied to all the individual realigned frames. Realignment and coregistration were performed using mutual information as the objective function as implemented in the SPM5 software (Wellcome Trust Centre for Neuroimaging, London, UK). The same ROIs that had been delineated on the MR and previously applied to the SPECT data were applied to the PET images to obtain time-activity curves for the striatum, occipital and cerebellum. The simplified reference tissue model (SRTM) was used for quantification (Lammertsma and Hume, 1996). Time–activity curves from the occipital cortex and cerebellum were used to provide the input function for the kinetic model, and the estimated outcome measure was the binding potential (BPND). The receptor occupancy was calculated as RO = 100·(BPBND − BPPND)/BPBND, where BPBND and BPPND are the binding potential in the baseline condition (in normal controls) and after receptor blockade in patients, respectively.

with y-intercept fixed to 0 was used to fit SPECT and PET D2 RO as follows: ROSPECT = c  ROPET

ð7Þ

The slope (c), 95% confidence interval of the slope and the correlation coefficient were obtained for each of the different image reconstructions under study. The percentage of difference in RO between SPECT and PET (RODIF) was calculated as   ROSPECT −ROPET = 100ðc−1Þ RODIF = 100 ROPET

ð8Þ

Results Phantom data A linear relationship was found between true and measured SUR. When images were reconstructed using FBPA, the relationship was SURM = 0.42 · SURT+ 0.11; R2 = 0.99 (Fig. 1). The ratio β/α was different from zero (β/α = 0.268) which, according to Eq. (5), indicates that the RO measured from these data will underestimate the true RO. When images were reconstructed using OSEMA+PSF, the relationship was SURM = 0.46 · SURT+ 0.14; R2 = 0.99 (Fig. 1). In this case, both the y-intercept and slope of the regression line increased, but the ratio β/α was maintained (β/α = 0.295) which indicated that the inclusion of the PSF in addition to the attenuation will not improve RO estimates. When images were reconstructed using OSEMA+PSF+SCT, the relation between calculated and measured SUR was still linear but the y-intercept of the regression line was substantially reduced and the slope increased (SURM = 0.60 · SURT+ 0.03; R2 = 0.99) (Fig. 1). Thus, when scatter correction was included in the algorithm, the ratio β/α was significantly reduced (β/α = 0.045) which indicates that the application of scatter correction reduced the difference between ROT and ROM for SPECT. Human data SPECT images displayed larger noise and lower specific uptake than PET images. The specific uptake in the striatum slightly increased when scatter correction was applied to the images as shown in Fig. 2.

Statistical analysis Linear regression was used to compare SUR and BPND values obtained with SPECT and PET, respectively. SUR ¼a  BPND + b

ð6Þ

Then, the slope (a), y-intercept (b), and correlation coefficient were obtained for each reconstruction technique. Linear regression was also used to compare D2 RO values obtained with PET and SPECT. Test–retest experiments with both PET and SPECT in healthy volunteers (Volkow et al., 1993; Hietala et al., 1999; Catafau et al., 2008) have demonstrated that, in the absence of D2 receptors blockade, the mean D2 RO values are not significantly different from zero in both techniques. Therefore, linear regression

Fig. 1. Measured SUR (SURM) versus True SUR (SURT) obtained from phantom data for different reconstruction strategies. FBPA (SURM = 0.42 · SURT+ 0.11; R2 = 0.99) (black circles, solid line); OSEMA+PSF (SURM = 0.46 · SURT+ 0.14; R2 = 0.99) (crosses, dotted line); OSEMA+PSF+SCT (SURM = 0.60 · SURT+ 0.03; R2 = 0.99) (white circles, dashed line); line of identity (dash dot line).

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

1515

Fig. 2. PET and SPECT images of a healthy volunteer coregistered and overlaid with the MR image. SPECT images were reconstructed with the filtered backprojection algorithm (FBP) (middle row) and with the 3D-OSEM algorithm including scatter correction (bottom row).

Human data exhibited a linear relationship between the measurement of the receptor availability using SPECT (SUR) and PET (BPND), independently of the reference region used (Fig. 3, Table 1). The inclusion of scatter correction led to an increase in the SUR. In particular, the baseline SUR increased from 1.21 ± 0.14 (mean ± SD) (occipital cortex) and 1.59 ± 0.21 (cerebellum) without scatter correction to 1.55 ± 0.28 (occipital cortex) and 1.99 ± 0.25 (cerebellum) with scatter correction. Simultaneously, scatter correction produced a decrease in the y-intercept in the regression line (β).

The ratio −β/SURBM (Eq. (5)) that measures the degree of RO difference between SPECT and PET was −0.20 (occipital cortex) and −0.23 (cerebellum) without scatter correction and was substantially reduced when scatter was applied (−0.07 (occipital cortex) and −0.12 (cerebellum)). D2 RO using SPECT and PET showed a linear relationship, independently of the reference region (Fig. 4). The degree of discrepancy between SPECT and PET was reduced when scatter correction was applied (Table 1). When images were reconstructed

Fig. 3. Scatter plots of calculated SUR versus the BPND values for the two reference regions. SUR values correspond to FBPA reconstruction (SUR = 0.51·BPND+0.24 (R2 = 0.92) (occipital); SUR = 0.49·BPND+0.36 (R2 = 0.86) (cerebellum)) (black circles, solid line), OSEMA+PSF reconstruction (SUR = 0.57·BPND+0.26 (R2 = 0.88) (occipital); SUR = 0.54·BPND +0.41 (R2 = 0.91) (cerebellum)) (crosses, dotted line) and OSEMA+PSF+SCT reconstruction (SUR = 0.75·BPND+0.11 (R2 = 0.89) (occipital); SUR = 0.69·BPND+0.24 (R2 = 0.91) (cerebellum)) (white circles, dashed line). Dash dot line corresponds to the line of identity.

1516

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

Table 1 PET-SPECT relationship of D2 receptor measurements for each of the reference regions (occipital cortex and cerebellum) and reconstruction algorithms without and with scatter correction.

FBPA

3D-OSEMA+PSF

3D-OSEMA+PSF+SCT

Occipital cortex

Cerebellum

SUR = 0.51·BPND + 0.24 (R2 = 0.92) ROSPECT = 0.80·ROPET (R2 = 0.80) RODIF = −20 (−13, −27) % SUR = 0.57·BPND + 0.26 (R2 = 0.88) ROSPECT = 0.79·ROPET (R2 = 0.79) RODIF = −21 (−14, −27) % SUR = 0.75·BPND + 0.11 (R2 = 0.89) ROSPECT = 0.94·ROPET (R2 = 0.81) RODIF = −6 (−0, −13) %

SUR = 0.49·BPND + 0.36 (R2 = 0.86) ROSPECT = 0.77·ROPET (R2 = 0.66) RODIF = −23 (−14, −31) % SUR = 0.54·BPND + 0.41 (R2 = 0.91) ROSPECT = 0.76·ROPET (R2 = 0.67) RODIF = −24 (−18, −30) % SUR = 0.69·BPND + 0.24 (R2 = 0.91) ROSPECT = 0.89·ROPET (R2 = 0.71) RODIF = −11 (−4, −18) %

SUR, specific uptake ratio; BPND, binding potential; ROSPECT, percentage of D2 receptor occupancy measured using SPECT; ROPET, percentage D2 receptor occupancy measured using PET; RODIF, percentage of D2 RO difference between SPECT and PET and 95% confidence interval (between brackets); FBP, filtered back-projection; OSEM, ordered subsets expectation maximization; A, attenuation correction; PSF, point spread function correction; SCT, scatter correction.

using FBPA, SPECT D2 RO values were clearly lower as compared with PET leading to a D2 RO difference of −20% (CI95%: −13, −27%) (occ) and −23% (CI95%: −14, −31%) (cer). When images were reconstructed using OSEMA+PSF, SPECT D2 RO values were also lower as compared with PET leading to a D2 RO difference of −21% (CI95%: −14, −27%) (occ) and −24% (CI95%: −18, −30%) (cer). When images were reconstructed using OSEMA+PSF+SCT, the D2 RO bias was reduced to −6% (CI95%: 0, −13%) (occ) and −11% (CI95%: −4, −18%) (cer). Discussion This study confirms our previous hypothesis that the contribution of 123I scattered photons to the SPECT image degradation may account for the SPECT underestimation of D2 RO measurements (Catafau et al., 2009). Phantom and clinical data have shown that the D2 RO difference between 123I-IBZM SPECT and 11C-Raclopride PET was substantially reduced when scatter correction was applied. The present study has implications for the design of 123I-IBZM SPECT studies aiming to quantify striatal D2 SPECT studies. Scatter should be corrected in 123I-IBZM SPECT scans in order to obtain more comparable SPECT and PET measurements, or alternatively the bias needs to be taken into account to interpret and compare results from different PET and SPECT studies. To obtain the aforementioned results, an accurate scatter correction method was included in the reconstruction. The algorithm can

manage patient individualized source and attenuation distributions, and a detailed tracking of photons in the object is included. Collimator/detector response for each energy ray, position and direction is included. Furthermore, as scatter correction is included as an additive term in the ML-EM algorithm, this maintains the noise distribution of the projections. Such an approach, which is essential for the investigation of the origin of the D2 RO bias, is not widely available because of the difficult implementation since a full, tailored characterization of the interaction of the photons emitted by 123I with each collimator is required (Cot et al., 2006). Moreover, scatter correction including the whole energy spectra of 123I is very time consuming due to its extensive use of Monte Carlo techniques. Alternatively to scatter correction, our previous work (Catafau et al., 2009) showed that unbiased results with respect to PET can be obtained without using scatter correction but rather by applying an empirical uniform attenuation correction factor which provided the most homogeneous images of a uniform cylindrical phantom. Several uniform attenuation factors, ranging from 0.00 (no attenuation correction applied) to 0.14 cm−1, have been used in the literature. In our previously reported experiments (Catafau et al., 2009), the empirical uniform attenuation factor that provided better agreement between SPECT and PET D2 RO measurements was μ≈0.06 cm−1, which is far lower than the attenuation coefficient of the soft tissue for the main emission of 123I (μ ≈ 0.14–0.15 cm−1) (Shiga et al., 2002; Soret et al., 2003). Nevertheless, this uniform empirical attenuation

Fig. 4. Scatter plots of calculated D2 RO using SPECT versus the D2 RO using PET for the two reference regions. SPECT D2 RO values correspond to FBPA reconstruction (ROSPECT = 0.80·ROPET (R2 = 0.80) (occipital); ROSPECT = 0.77·ROPET (R2 = 0.66) (cerebellum)) (black circles, solid line), OSEMA+PSF reconstruction (ROSPECT = 0.79·ROPET (R2 = 0.79) (occipital); ROSPECT = 0.76·ROPET (R2 = 0.67) (cerebellum)) (crosses, dotted line) and OSEMA+PSF+SCT reconstruction (ROSPECT = 0.94·ROPET (R2 = 0.81) (occipital); ROSPECT = 0.89·ROPET (R2 = 0.71) (cerebellum)) (white circles, dashed line). Dash dot line corresponds to the line of identity.

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

factor may vary depending on the collimator used and should be determined for each SPECT camera. A more detailed discussion on the application of empirical attenuation correction factors can be found elsewhere (Zaidi and Montandon, 2002). Scatter correction of SPECT images contributes to the reduction in the difference between SPECT and PET D2 RO measurements irrespective of the reference region used (occipital cortex or cerebellum). However, a small but statistically significant bias remains when the cerebellum is used as reference region. Different bone structures surrounding cerebellum may make uniform attenuation correction suboptimal. Additionally, the photons emitted in regions of the body other than the brain can be measured in the lower parts of the field of view due to the large penetration of 123I high energy photons. This contribution may have an impact on the cerebellum activity measurements (Kim et al., 2001a). Our results indicate that scatter correction plays a major role in explaining the differences between D2 RO measurements using SPECT and PET. Other factors such as the use of different radioligands in SPECT and PET and the assumptions made by a tissue ratio method at pseudoequilibium after a single bolus injection of 123I-IBZM SPECT cannot be completely excluded. Nevertheless, these factors have been previously studied and were not found to have a substantial impact on explaining differences between SPECT and PET measurements of D2 RO (Catafau et al., 2009). In addition to attenuation and scatter corrections, the effect of the point spread function (PSF) degradation was also studied. Previous studies have shown an improvement in the SUR recovery factors (α) when PSF correction was included (Crespo et al., 2008). However, for the RO recovery, the parameter that should be minimized to reduce the bias between the calculated and theoretical RO value is β/α. This work has shown that the larger reduction of the β/α ratio was found when scatter was corrected. Moreover, the PSF correction did not introduce a significant reduction of the β/α ratio for a fan-beam collimator. This fact supports the idea that scatter compensation plays a larger role in the 123I-IBZM measured D2 RO bias than PSF compensation. In conclusion, the scatter correction plays a major role in explaining the differences between D2 RO measurements using 123IIBZM SPECT and 11C-Raclopride PET. Scatter correction is recommended to obtain values that are consistent with the 11C-Raclopride PET literature. This fact should be taken into account when interpreting D2 RO results measured by 123I-IBZM SPECT. Acknowledgments The authors would like to thank the PET teams of the Institut d'Alta Tecnologia (IAT) and CRC Molecular Imaging Centre (CRC-CIM) (Barcelona, Spain) for their contribution to the human PET scans performance and all the volunteers who participated in the study. The work of D. Ros and J Pavía was supported in part by a GlaxoSmithKline consultancy contract and the Spanish Ministry of Science and Innovation (SAF2009-08076). Appendix A The measured receptor occupancy (ROM) by SPECT and the true receptor occupancy (ROT) are given by the following expressions: ROT = 100

SURBT −SURPT SURBT

ROM = 100

SURM −SURM SURBM

B

ðA1Þ P

ðA2Þ

where SURBT and SURPT are the true specific to non-displaceable uptake ratios at baseline condition and after receptor blockade; and SURBM

1517

and SURPM are the measured specific to non-displaceable uptake ratios at baseline condition and after receptor blockade. Given SURM = α· SURT + β

ðA3Þ

where α and β are the slope and the y-intercept, respectively. Then, substituting Eq. (A3) into Eq. (A2) and rearranging it, yields, ROM  B

P

SURM − SURM ROM = 100 = 100 SURBM

   B P α·SURT + β − α·SURT + β α·SURBT + β

=

SURBT −SURPT SURBT −SURPT SURBT = 100 · B β β SURT SURBT + SURBT + α α 1 = ROT · β 1+ α·SURBT = 100

1

ROM = ROT · 1+

β α·SURBT

ðA4Þ

Alternatively, Eq. (A4) can be expressed in terms of SURBM by substituting again Eq. (A3) in Eq. (A4) to obtain, ROM = ROT ·ð1−

β Þ SURBM

ðA5Þ

References Beekman, F.J., de Jong, H.W.A.M., van Geloven, S., 2002. Efficient fully 3-D iterative SPECT reconstruction with Monte Carlo-based scatter compensation. IEEE. Trans. Med. Imag. 21 (8), 867–877. Bullich, S., Ros, D., Pavía, J., Cot, A., López, N., Catafau, A.M., 2009. Neurotransmission SPECT and MR registration combining mutual and gradient information. Med. Phys. 36 (11), 4903–4910. Catafau, A.M., Bullich, S., Danús, M., Penengo, M.M., Cot, A., Abanades, S., et al., 2008. Test–retest variability and reliability of 123I-IBZM SPECT measurement of striatal dopamine D2 receptor availability in healthy volunteers and influence of iterative reconstruction algorithms. Synapse 62 (1), 62–69. Catafau, A.M., Suarez, M., Bullich, S., Llop, J., Nucci, G., Gunn, R., et al., 2009. Withinsubject comparison of striatal D2 receptor occupancy measurements using [123I] IBZM SPECT and [11C]Raclopride PET. NeuroImage 46, 447–458. Cot, A., Sempau, J., Pareto, D., Bullich, S., Pavía, J., Calviño, F., et al., 2004. Study of the point spread function (PSF) for 123I SPECT imaging using Monte Carlo simulation. Phys. Med. Biol. 9 (14), 3125–3136. Cot, A., Falcón, C., Crespo, C., Sempau, J., Pareto, D., Bullich, S., et al., 2005. Absolute quantification in dopaminergic neurotransmission SPECT using a Monte Carlobased scatter correction and fully 3-dimensional reconstruction. J. Nucl. Med. 46 (9), 1497–1504. Cot, A., Jane, E., Sempau, J., Falcon, C., Bullich, S., Pavia, J., et al., 2006. Modeling of highenergy contamination in SPECT imaging using Monte Carlo simulation. IEEE. Trans. Nucl. Sci. 53, 198–203. Crespo, C., Gallego, J., Cot, A., Falcón, C., Bullich, S., Pareto, D., et al., 2008. Quantification of dopaminergic neurotransmission SPECT studies with (123)I-labelled radioligands. A comparison between different imaging systems and data acquisition protocols using Monte Carlo simulation. Eur. J. Nucl. Med. Mol. Imaging. 35 (7), 1334–1342. Dobbeleir, A.A., Hambÿe, A.S., Franken, P.R., 1999. Influence of high-energy photons on the spectrum of iodine-123 with low-and medium-energy collimators: consequences for imaging with 123I-labelled compounds in clinical practice. Eur. J. Nucl. Med. 26 (6), 655–658. Dresel, S., Mager, T., Rossmüller, B., Meisenzahl, E., Hahn, K., Möller, H.J., et al., 1999. In vivo effects of olanzapine on striatal dopamine D(2)/D(3) receptor binding in schizophrenic patients: an iodine-123 iodobenzamide single-photon emission tomography study. Eur. J. Nucl. Med. 26 (8), 862–868. Fujita, M., Varrone, A., Kim, K.M., Watabe, H., Zoghbi, S.S., Seneca, N., Tipre, D., Seibyl, J.P., Innis, R.B., Iida, H., 2004. Effect of scatter correction on the compartmental measurement of striatal and extrastriatal dopamine D2 receptors using [123I] epidepride SPET. Eur. J. Nucl. Med. Mol. Imaging. 31, 644–654. Harrison, R.L., Vannoy, S.D., Haynor, D.R., Gillispie, S.B., Kaplan, M.S., Lewellen, T.K., 1993. Preliminary experience with the photon history generator module of a public-domain simulation systems for emission tomography. Records of IEEE Nuclear Science Symposium and Medical Imaging Conference, San Francisco, pp. 1154–1158. Hietala, J., Någren, K., Lehikoinen, P., Ruotsalainen, U., Syvälahti, E., 1999. Measurement of striatal D2 dopamine receptor density and affinity with [11C]-raclopride in vivo: a test–retest analysis. J. Cereb. Blood. Flow. Metab. 19 (2), 210–217.

1518

S. Bullich et al. / NeuroImage 50 (2010) 1511–1518

Hudson, H.M., Larkin, R.S., 1994. Accelerated image reconstruction using ordered subsets of projection data. IEEE. Trans. Med. Imag. 13 (4), 601–609. Ito, H., Iida, H., Kinoshita, T., Hatazawa, J., Okudera, T., Uemura, K., 1999. Effects of scatter correction on regional distribution of cerebral blood flow using I-123-IMP and SPECT. Ann. Nucl. Med. 13, 331–336. Kegeles, L.S., Zea-Ponce, Y., Abi-Dargham, A., Rodenhiser, J., Wang, T., Weiss, R., et al., 1999. Stability of [123I]IBZM SPECT measurement of amphetamine-induced striatal dopamine release in humans. Synapse 31 (4), 302–308. Kegeles, L.S., Abi-Dargham, A., Zea-Ponce, Y., Rodenhiser-Hill, J., Mann, J.J., Van Heertum, R.L., et al., 2000. Modulation of amphetamine-induced striatal dopamine release by ketamine in humans: implications for schizophrenia. Biol. Psychiatry 48 (7), 627–640. Kim, K.M., Watabe, H., Shidahara, M., Ishida, Y., Iida, H., 2001a. SPECT collimator dependency of scatter and validation of transmission-dependent scatter compensation methodologies. IEEE Trans. Nucl. Sci. 48 (3), 689–696. Kim, K.M., Watabe, H., Shidahara, M., Onishi, Y., Yonekura, Y., Iida, H., 2001b. Impact of scatter correction in the kinetic analysis of a D2 receptor ligand SPECT study. IEEE Medical Imaging Conference, San Diego, pp. 1509–1512. Knable, M.B., Heinz, A., Raedler, T., Weinberger, D.R., 1997. Extrapyramidal side effects with risperidone and haloperidol at comparable D2 receptor occupancy levels. Psychiatry Res. 75 (2), 91–101. Koch, W., Hamann, C., Welsch, J., Pöpperl, G., Radau, P.E., Tatsch, K., 2005. Is iterative reconstruction an alternative to filtered backprojection in routine processing of dopamine transporter SPECT studies? J. Nucl. Med. 46 (11), 1804–1811. Koch, W., Radau, P.E., Münzing, W., Tatsch, K., 2006. Cross-camera comparison of SPECT measurements of a 3-D anthropomorphic basal ganglia phantom. Eur. J. Nucl. Med. Mol. Imaging 33 (4), 495–502. Lammertsma, A.A., Hume, S.P., 1996. Simplified reference tissue model for PET receptor studies. NeuroImage 4, 153–158. Larsson, A., Ljungberg, M., Jakobson Mo, S., Riklund, K., Johansson, L., 2006. Correction for scatter and septal penetration using convolution subtraction methods and model-based compensation in 123I brain SPECT imaging—a Monte Carlo study. Phys. Med. Biol. 51, 5753–5767. Laruelle, M., 2000a. Imaging synaptic neurotransmission with in vivo binding competition techniques: a critical review. J. Cereb. Blood. Flow. Metab. 20 (3), 423–451. Laruelle, M., 2000b. The role of model-based methods in the development of single scan techniques. Nucl. Med. Biol. 27 (7), 637–642. Laruelle, M., Slifstein, M., Huang, Y., 2002. Positron emission tomography: imaging and quantification of neurotransporter availability. Methods 27 (3), 287–299. Moore, S.C., Kijewski, M.F., Muller, S.P., Rybicki, F., Zimmerman, R.E., 2001. Evaluation of scatter compensation methods by their effects on parameter estimation from SPECT projections. Med. Phys. 28, 278–287. Pareto, D., Cot, A., Pavía, J., Falcón, C., Juvells, I., Lomeña, F., et al., 2003. Iterative reconstruction with correction of the spatially variant fan-beam collimator

response in neurotransmission SPET imaging. Eur. J. Nucl. Med. Mol. Imaging 30 (10), 1322–1329. Shiga, T., Kubo, N., Takano, A., Kobayashi, J., Takeda, Y., Nakamura, F., et al., 2002. The effect of scatter correction on 123I-IMP brain perfusion SPET with the triple energy window method in normal subjects using SPM analysis. Eur. J. Nucl. Med. 29, 342–345. Small, A.D., Prosser, J., Motherwell, D.W., McCurrach, G.M., Fletcher, A.M., Martin, W., 2006. Downscatter correction and choice of collimator in 123I imaging. Phys. Med. Biol. 51, N307–N311. Soret, M., Koulibaly, P.M., Darcourt, J., Hapdey, S., Buvat, I., 2003. Quantitative accuracy of dopaminergic neurotransmission imaging with (123)I SPECT. J. Nucl. Med. 44 (7), 1184–1193. Soret, M., Koulibaly, P.M., Darcourt, J., Buvat, I., 2006. Partial volume effect correction in SPECT for striatal uptake measurements in patients with neurodegenerative diseases: impact upon patient classification. Eur. J. Nucl. Med. Mol. Imaging 33 (9), 1062–1072. Staelens, S., de Wit, T., Beekman, F., 2007. Fast hybrid SPECT simulation including efficient septal penetration modelling (SP-PSF). Phys. Med. Biol. 52 (11), 3027–3043. Talbot, P.S., Laruelle, M., 2002. The role of in vivo molecular imaging with PET and SPECT in the elucidation of psychiatric drug action and new drug development. Eur. Neuropsychopharmacol. 12 (6), 503–511. Tanaka, M., Uehara, S., Kojima, A., Matsumoto, M., 2007. Monte Carlo simulation of energy spectra for 123I imaging. Phys. Med. Biol. 52, 4409–4425. Tauscher, J., Küfferle, B., Asenbaum, S., Fischer, P., Pezawas, L., Barnas, C., et al., 1999. In vivo 123I IBZM SPECT imaging of striatal dopamine-2 receptor occupancy in schizophrenic patients treated with olanzapine in comparison to clozapine and haloperidol. Psychopharmacology 141 (2), 175–181. Varrone, A., Sansone, V., Pellecchia, M.T., Amboni, M., Salvatore, E., De Michele, G., et al., 2008. Comparison between a dual-head and a brain-dedicated SPECT system in the measurement of the loss of dopamine transporters with [(123)I]FP-CIT. Eur. J. Nucl. Med. Mol. Imaging. 35 (7), 1343–1349. Verhoeff, N.P., Brücke, T., Podreka, I., Bobeldijk, M., Angelberger, P., Van Royen, E.A., 1991. Dynamic SPECT in two healthy volunteers to determine the optimal time for in vivo D2 dopamine receptor imaging with 123I-IBZM using the rotating gamma camera. Nucl. Med. Commun. 12 (8), 687–697. Volkow, N.D., Fowler, J.S., Wang, G.J., Dewey, S.L., Schlyer, D., MacGregor, R., et al., 1993. Reproducibility of repeated measures of carbon-11-raclopride binding in the human brain. J. Nucl. Med. 34 (4), 609–613. Zaidi, H., Koral, K.F., 2004. Scatter modelling and compensation in emission tomography. Eur. J. Nucl. Med. Mol. Imaging 31 (5), 761–782. Zaidi, H., Montandon, M.L., 2002. Which attenuation coefficient to use in combined attenuation and scatter corrections for quantitative brain SPECT? Eur. J. Nucl. Med. 29, 967–969.