Imaging and quantification of presynaptic dopamine transporters (DATs) availability with SPECT and ligands like N-(3-fluoropropyl)-2β-carbomethoxy-3β-(4-iodophenyl)nortropane (123I-FP-CIT) is an important tool for the diagnosis and follow-up of neurodegenerative parkinsonian syndromes and dementia with Lewy bodies.1 The commonly used reconstruction method for 123I-FP-CIT SPECT is filtered back projection (FBP). Its most important benefit is its rapid calculation speed. However, FBP is prone to (streak) artifacts and image noise, necessitating imaging filtering, which degrades resolution. In turn, several iterative reconstruction methods that can incorporate models of the physical aspects of SPECT imaging (eg, resolution, collimator, scatter radiation, and attenuation) have been proposed, facilitating improved image quality, most notably ordered-subset expectation maximization (OSEM) reconstruction with 3-dimensional (3D) resolution recovery.2 In the present work, we contemplated an OSEM method that uses a measured 3D beam model for collimation for image reconstruction, which is performed simultaneously for all slices (3D-OSEM; Flash 3D Technology; Siemens, Erlangen, Germany). Three-dimensional OSEM was shown to effectively minimize data deformation in the z direction and to improve SPECT image quality compared with conventional 2-dimensional (2D) OSEM.3 In addition, 3D-OSEM was found to be superior to FBP in image noise for low count statistics.4 Consequently, an improved image quality may be yielded even with a considerably reduced (approximately −50%) injected activity dose and/or acquisition time.5,6 These features may be a particular advantage of 3D-OSEM in DAT or dopamine receptor SPECT studies: These studies are usually quantified by pseudo or peak equilibrium analyses, in which the estimate of binding site density (binding potential, BP ND)7 is calculated by the ratio of specific striatal ligand uptake to nonspecific ligand uptake in a reference region.8 Thus, an improved delineation of high-uptake striatal structures with enhanced signal recovery and an optimal image reconstruction of the low-uptake reference region may be provided by 3D-OSEM. Although previous studies explored the use of 2D-OSEM for DAT SPECT reconstruction using phantom and patient data,9,10 a systematic evaluation of 3D-OSEM for DAT SPECT imaging on actual patient data is still missing.
Thus, the present study was undertaken to investigate possible benefits of 3D-OSEM reconstruction for DAT SPECT imaging in image quality and data quantification in comparison to standard FBP and 2D-OSEM.
MATERIALS AND METHODS
The SPECT data of 18 patients [age ranging from 45 to 79 y; mean (SD), 70.5 (7.5) y] were included in this retrospective analysis. The patients were referred for routine clinical 123I-FP-CIT SPECT investigations to verify or exclude neurodegenerative parkinsonism. A broad spectrum of DAT pathological findings was intentionally selected ranging from normal to severely decreased DAT binding. The clinical diagnoses were as follows: nonneurodegenerative tremor in 3 patients, mild cognitive impairment/Alzheimer disease in 2 patients (ie, total n = 5 patients with normal DAT binding), Parkinson disease in 8 patients, Dementia with Lewy bodies in 3 patients, and progressive supranuclear palsy and multiple-system atrophy in 1 patient each (ie, total n = 13 patients with moderately to severely reduced DAT binding).
SPECT Data Acquisition and Reconstruction
All data sets were acquired on a dual-head, large-FOV rotating gamma camera (E.CAM; Siemens Gammasonics, Inc, Hoffman Estates, Ill) equipped with a standard LEHR parallel-hole collimator (energy window, 159 keV ± 15%) at exactly 180 minutes after bolus injection of 190.2 (11.6) MBq (range, 165–204 MBq) 123I-FP-CIT (GE Healthcare Buchler GmbH & Co. KG, Braunschweig, Germany). Sixty projections per camera head of 30 seconds each were acquired in step-and-shoot mode at a rotation radius of 13.5 cm using an acquisition matrix of 128 × 128 and a zoom factor of 1.23. The projection data were checked visually for patient motion using the cine display and sinograms (Syngo Software; Siemens).
Data reconstruction was performed using FBP, 2D-OSEM, and 3D-OSEM with 2 different reconstruction parameter settings for each method to yield 3 pairs of reconstructions with relatively lower (FBPlow, 2D-OSEMlow, and 3D-OSEMlow) and higher (FBPhigh, 2D-OSEMhigh, and 3D-OSEMhigh) spatial resolution and noise: FBPlow uses a seventh-order Butterworth filter and a cutoff frequency of 0.36 Nyquist. It represents the method currently used in clinical routine and will thus be used as reference method. Although image reconstruction is somewhat degraded by the relatively low filter cutoff frequency, the resulting low image noise is judged to be favorable for the definition of the reference region and for avoidance of artificial uptake asymmetries in the posterior putamen, which may be mistaken to indicate neurodegeneration. In a previous work, we demonstrated that these reconstruction settings enable optimal diagnostic accuracy for visual readings of parametric 123I-FP-CIT SPECT maps.11 Filtered back projection was also used with a higher cutoff frequency of 0.45 Nyquist, allowing somewhat higher image resolution at the cost of increased noise (FBPhigh).
Iterative reconstruction parameters were chosen to represent an optimal combination of high BP ND recovery at low reference region noise (variability of counts, expressed as the coefficient of variation of counts in the occipital reference region). To this end, we systematically analyzed the dependence of BP ND and of the variability of counts in the occipital cortex on the number of expectation maximization (EM)–equivalent iterations ranging from 16 to 150 with various combinations of iterations and subsets (with additional 8-mm Gaussian postprocessing filtering). Figure 1 shows a representative example of such an analysis for a patient with normal DAT binding. As depicted (Fig. 1A), BP ND (given here for caudate nucleus) shows a saturation-like behavior with an increasing number of EM-equivalent iterations (=number of iterations × number of subsets in case of 2D/3D-OSEM), being close to convergence at approximately 100 EM-equivalent iterations (96; ie, 6 iterations with 16 subsets in the present example), in line with the recent phantom study by Dickson et al.10 The variability of counts in the occipital cortex, in turn, increases almost linearly with the number of EM-equivalent iterations (Fig. 1B), so that we decided to use 8 iterations with 8 subsets for 2D-OSEMlow and 3D-OSEMlow reconstructions and 6 iterations with 16 subsets for 2D-OSEMhigh and 3D-OSEMhigh reconstructions. An 8-mm Gaussian postprocessing filtering was applied to all iterative reconstructions.
All data sets were corrected for photon attenuation using Chang’s first-order correction (μ = 0.12/cm, automated ellipse fitting using identical edge-detection settings for all reconstructions). All reconstructions were carried out on the Syngo MultiModality Workplace (Siemens).
SPECT Data Analysis
The summed image of all 6 reconstructions from 1 selected patient without nigrostriatal neurodegeneration was reoriented to the anterior commissure–posterior commissure (AC/PC) orientation (approximately, given SPECT resolution) and was used as a reference image. A standard volume-of-interest (VOI) template, consisting of symmetric VOIs for left and right caudate nuclei (2.1 mL) and putamen (3.2 mL) and a single occipital cortex reference VOI (23.8 mL), was defined using this reference image on 3 consecutive transaxial planes showing maximum striatal uptake.
All other data sets were coregistrated to the reference image using a rigid matching algorithm (normalized mutual information) to align all data sets to the AC/PC orientation. This was performed by summing all data sets of each subject, coregistering the summed image to the aforementioned reference image and applying individual coregistration parameters to each individual data set. Subsequently, the VOI template was loaded into the individual summed, AC/PC–aligned data set. The position and rotation of VOI were manually adjusted to the individual’s anatomy without changing the VOI size. The individually adjusted VOI template was then used to analyze all individual data sets (ie, 6 reconstructions). Volume-of-interest analyses were performed using a commercial software package (PMOD, Version 3.0; PMOD Technologies Ltd, Adliswil, Switzerland).
The following parameters were assessed: mean VOI counts, variability of counts in the occipital cortex (see above), and binding potential BP ND (=mean striatal counts / mean occipital counts − 1). BP ND is the primary outcome measure of DAT SPECT studies, which is proportional to the density of striatal DAT available for 123 I-FP-CIT binding.7
All data are expressed as mean (SD). FBPlow served as the reference reconstruction method. Thus, the results of all other analyses were compared with FBPlow using the paired Student t test. An adjusted 2-tailed P < 0.00167 (= 0.05 / 30; 30 comparisons) was considered to indicate a significant difference. In addition, linear regression analyses were performed to assess the association between BP ND values gained from the different reconstruction methods. Statistical analyses were carried out using PASW 18 (SPSS, Inc, Chicago, Ill).
Figure 2 shows the 123I-FP-CIT SPECT scans (transaxial slices with maximum striatal uptake) of 5 representative patients with normal to severe DAT loss, each reconstructed with the 6 reconstruction methods. Data sets are given as parametric uptake ratio (ie, distribution volume ratio, DVR) images with identical display settings for optimal comparability.11 When visually rated by 2 experienced raters, FBPlow, 2D-OSEMlow, and 3D-OSEMlow were judged to be inferior to FBPhigh, 2D-OSEMhigh, and 3D-OSEMhigh, respectively, in delineating striatal structures (spatial resolution) in all patients (n = 18), although this difference was sometimes only marginal with OSEM reconstructions. Also as expected, FBPlow, 2D-OSEMlow, and 3D-OSEMlow gave a more homogeneous (less noisy) nonspecific cortical 123I-FP-CIT uptake than FBPhigh, 2D-OSEMhigh, and 3D-OSEMhigh, respectively.
Comparing different reconstruction methods, delineation of striatal structures and homogeneity of nonspecific uptake were found to be comparable between 2D-OSEMlow and 3D-OSEMlow on one hand and 2D-OSEMhigh and 3D-OSEMhigh on the other. All OSEM reconstructions were superior to FBPlow concerning delineation of the striatum, which resulted in a better separation between the caudate nucleus and the putamen by OSEM reconstructions, although this was not possible with FBPlow. In this regard, OSEM reconstructions were also superior to FBPhigh in most cases, although the difference was less pronounced. Finally, FBPlow resulted in the overall most homogeneous nonspecific uptake (least noisy) and, thus, distinguishability between the cortex and cerebral spinal fluid (CSF) spaces. Nonspecific cortical uptake was of intermediate and relatively high (but still acceptable) heterogeneity in case of OSEMlow and OSEMhigh reconstructions, respectively.
Results of VOI analyses are summarized in Figures 3 to 5: In line with the aforementioned visual impression (ie, higher spatial resolution and, thus, less spill-out), mean striatal counts were higher for FBPhigh and all OSEM reconstructions than for FBPlow (Table 1). Compared with FBPlow (reference method), FBPhigh resulted in significantly (P corrected < 0.05) higher counts in the caudate nucleus (mean difference, +3.4%) and the putamen (+1.2%), whereas the counts of the occipital cortex did not change between FBPlow and FBPhigh (+0.1%). This translated into significantly higher BP ND values for the caudate nucleus (+5.2) and the putamen (+1.8%).
With 2D-OSEM (compared with FBPlow; Table 1), a slight, although not statistically significant, increase in counts was found in the putamen for 2D-OSEMlow (+1.4%; P corrected = 0.07; P uncorrected < 0.05). In contrast, counts increased significantly in the putamen for 2D-OSEMhigh (+2.1%) and in the caudate nucleus for both 2D-OSEM reconstructions (+3.7%/+5.4% for 2D-OSEMlow/high; all P corrected < 0.05). Counts in the occipital cortex also showed a trend toward an increase with 2D-OSEM reconstructions (+1.3%/+1.4% for 2D-OSEMlow/high; P uncorrected < 0.05). Consequently, mean BP ND in the putamen remained unchanged (−0.2%/+0.8 for 2D-OSEMlow/high), whereas mean BP ND in the caudate nucleus increased (+3.7%/+6.2% for 2D-OSEMlow/high; P corrected < 0.05).
Finally, with 3D-OSEM, counts increased moderately in the putamen (+5.0%/+5.9% for 3D-OSEMlow/high) and strongly in the caudate nucleus (+8.9%/+10.8% for 3D-OSEMlow/high; P corrected < 0.05) (Table 1). A trend toward increased counts was also detected in the occipital cortex (+1.6%/+1.7% for 3D-OSEMlow/high; P uncorrected < 0.05). This translated into a moderate to a strong increase in BP ND in the putamen (+5.6%/+6.8% for 3D-OSEMlow/high) and the caudate nucleus (+11.1%/+14.0% for 3D-OSEMlow/high; all P corrected < 0.05).
Results of linear regression analyses between BP ND estimates gained from the different reconstruction methods are given in Table 2. There was an excellent correlation between the reference method FBPlow and all other methods (R 2 > 0.97). Notably, all slope estimates were 1.0 or higher, with comparable slopes for FBPhigh (1.06 and 1.04 for the caudate nucleus and the putamen, respectively) and 2D-OSEM (2D-OSEMlow/high: 1.04/1.03 and 1.01/1.04 for the caudate nucleus and the putamen, respectively) and highest slopes for 3D-OSEM (3D-OSEMlow/high: 1.15/1.16 and 1.12/1.15 for the caudate nucleus and the putamen, respectively). Intercepts were significantly different from zero only for the putamen with 3D-OSEM (3D-OSEMlow/high: −0.06/−0.08).
The effect of the reconstruction method on the variability of counts in the occipital cortex is summarized in Figure 5: In line with the visual impression and in comparison to the reference method FBPlow, the variability of counts in the occipital cortex was higher for FBPhigh [15.1% (4.0%)] than FBPlow [12.5% (4.3%); mean relative difference, +23.3%]. With 2D/3D-OSEMlow reconstructions, the variability of counts increased slightly less to 13.9% (3.8%) and 14.2% (4.3%) for 2D-OSEMlow and 3D-OSEMlow, respectively (mean relative difference, +13.8% and +15.2%, respectively). In contrast, the increase in the variability of counts in the occipital cortex was relatively pronounced with 2D/3D-OSEMhigh reconstructions [variability, 17.0% (3.8%) and 17.6% (4.1%) for 2D-OSEMhigh and 3D-OSEMhigh, respectively; mean relative difference, +40.9% and +46.4%, respectively; all P corrected < 0.05].
The present study was undertaken to investigate whether 3D-OSEM reconstruction offers a relevant advantage over conventional FBP and 2D-OSEM reconstructions in case of DAT SPECT. Earlier studies on other applications (ie, renal and bone SPECT in pediatric patients)5,6 demonstrated that 3D-OSEM not only provides an improved image quality and resolution but also needs fewer gamma counts for satisfactory image reconstruction than FBP does. These properties are of crucial importance in DAT SPECT because accurate data interpretation is based on the reconstruction and subsequent comparison of tracer concentrations in a relatively small target region with high tracer uptake and a larger reference region with low uptake. Thus, 3D-OSEM seems to be particularly promising for DAT SPECT reconstruction. With FBP, both regions can hardly be assessed satisfactorily because optimal delineation of the striatum necessitates a higher filter cutoff frequency that leads to noisy and hardly definable reference region uptake and vice versa. Choosing filter parameters to primarily optimize delineation of striatal structures (ie, target region) may be intuitively correct, but it ignores the importance of properly assessing reference region uptake for accurate BP ND estimation. Precise delineation of the reference region is of crucial importance in clinical routine to avoid inclusion of CSF spaces, which causes artificially high BP ND estimates (eg, in case of atrophy, hydrocephalus, or ischemic defects). Moreover, a higher noise level may also mimic asymmetric striatal DAT binding caused by neurodegeneration. At our institution, we therefore used a relatively low filter cutoff frequency in clinical routine (FBPlow) so far. Despite a loss of resolution, it offered a very high diagnostic accuracy on visual readings of standardized parametric DVR images.11
On visual inspection, we found that 2D-OSEM and 3D-OSEM (with slightly superior resolution for 2D/3D-OSEMhigh compared with 2D/3D-OSEMlow) offer a clearly improved delineation of striatal structures compared with FBPlow and, less so, FBPhigh. At the same time, image noise in the occipital cortex was only slightly higher for OSEMlow than for FBPlow and brain parenchyma, hence surrounding CSF spaces and extracerebral tissue could still be very well differentiated with OSEMlow reconstructions. With respect to image noise in the reference region, FBPhigh and 2D/3D-OSEMhigh reconstructions performed less favorable, but image quality seemed still sufficient for accurate reference region VOI definition. This observation is in line with a previous study by Koch et al9 comparing FBP and 2D-OSEM for 123I-FP-CIT SPECT reconstruction. They found a better separation between the caudate nucleus and the putamen and a more homogeneous nonspecific tracer uptake with 2D-OSEM compared with FBP (being closer to FBPhigh than FBPlow of the present study).
In line with an improved spatial resolution (ie, less spill-out from striatum), we showed that 2D- and 3D-OSEM reconstruction yielded significantly higher count values in the caudate nucleus and the putamen (not significant for 2D-OSEMlow) compared with FBPlow, whereas the counts in the occipital cortex were not significantly changed. This led to a significant BP ND increase in the caudate nucleus (+3.7% to 6.2% for 2D-OSEM and +11.1% to 14% for 3D-OSEM) and the putamen (not significant for 2D-OSEM; +5.6% to 6.8% for 3D-OSEM). Given that the caudate nucleus is smaller than the putamen, it is expected that the aforementioned effect is more pronounced in the caudate nucleus. Furthermore, whereas 3D-OSEM was performed with recovery correction, this was not the case for 2D-OSEM, which explains that the increase in counts and BP ND was considerably larger for 3D-OSEM than 2D-OSEM. Interestingly, Koch et al9 found that 2D-OSEM provided approximately 6% lower BP ND estimates than FBP, although their 2D-OSEM algorithm involved a depth-independent resolution modeling (2D-OSEMRM). However, this discrepancy may be explained by the fact that they performed considerably fewer EM-equivalent iterations (a total of 24)9 than would be needed for convergence (approximately 100; Fig. 1).10 Furthermore, Dickson et al10 showed in their phantom study that the same 2D-OSEMRM algorithm performed with 200 EM-equivalent iterations gave 4% to 17% higher BP ND estimates than FBP did. This is in line with the present results. Thus, 2D-OSEMRM and 3D-OSEM probably provide very similar results in BP ND estimation. However, as a unique feature, 3D-OSEM reconstructs all slices simultaneously and models the collimator beams at all depths, which provides isotropic spatial resolution and minimizes localization errors and shape distortions in comparison to 2D-OSEM.3
Regression analyses revealed an excellent correlation between FBPlow and all other methods (R 2 ≥ 0.97). In line with optimal signal and BP ND recovery, 3D-OSEM showed the highest regression slope (1.12–1.16) in comparison with FBPlow, whereas regression slopes for FBPhigh and 2D-OSEM were consistently lower (2D-OSEM, 1.01–1.04; FBPhigh, 1.04–1.06). This implies that 3D-OSEM (and 2D-OSEMRM) significantly improves the insufficient recovery of the true target-to-background ratio provided by common SPECT systems and FBP.12,13 Thus, a given change in DAT availability can be expected to lead to slightly larger BP ND change with 3D-OSEM compared with the other included methods. This, in turn, may result in an improved diagnostic sensitivity. Future studies are warranted to investigate in how far the aforementioned advantages of 3D-OSEM reconstructions lead to a gain of diagnostic information.
Three-dimensional OSEM considerably improves DAT SPECT reconstruction by offering a high-resolution delineation of striatal structures with superior recovery of signal and BP ND. At the same time, nonspecific cortical 123I-FP-CIT uptake is still sufficiently homogeneous (relatively low noise) as a prerequisite for accurate definition of the reference region.
1. Darcourt J, Booij J, Tatsch K, et al.. EANM procedure guidelines for brain neurotransmission SPECT
using (123)I-labelled dopamine transporter
ligands, version 2. Eur J Nucl Med Mol Imaging
. 2010; 37: 443–450.
2. Yokoi T, Shinohara H, Onishi H. Performance evaluation of OSEM reconstruction algorithm incorporating three-dimensional distance-dependent resolution compensation for brain SPECT
: a simulation study. Ann Nucl Med
. 2002; 16: 11–18.
3. Römer W, Reichel N, Vija HA, et al.. Isotropic reconstruction of SPECT
data using OSEM3D: correlation with CT. Acad Radiol
. 2006; 13: 496–502.
4. Alzimami K, Sassi S, Spyrou NM. An evaluation of 3D OSEM, and a comparison with FBP in 99m
images. In: Proceedings of the World Congress on Engineering, July 2–4, 2008
. Vol. I. London, UK: WCE; 2008.
5. Sheehy N, Tetrault TA, Zurakowski D, et al.. Pediatric 99m
performed by using iterative reconstruction
with isotropic resolution recovery: improved image quality and reduced radiopharmaceutical activity. Radiology
. 2009; 251: 511–516.
6. Stansfield EC, Sheehy N, Zurakowski D, et al.. Pediatric 99m
Tc-MDP bone SPECT
with ordered subset expectation maximization iterative reconstruction
with isotropic 3D resolution recovery. Radiology
. 2010; 257: 793–801.
7. Innis RB, Cunningham VJ, Delforge J, et al.. Consensus nomenclature for in vivo imaging of reversibly binding radioligands. J Cereb Blood Flow Metab
. 2007; 27: 1533–1539.
8. Meyer PT, Sattler B, Winz OH, et al.. Kinetic analyses of [123
for quantification of striatal dopamine D2
receptor binding: a critical evaluation of the single-scan approach. Neuroimage
. 2008; 42: 548–558.
9. Koch W, Hamann C, Welsch J, et al.. Is iterative reconstruction
an alternative to filtered backprojection in routine processing of dopamine transporter SPECT
studies? J Nucl Med
. 2005; 46: 1804–1811.
10. Dickson JC, Tossici-Bolt L, Sera T, et al.. The impact of reconstruction method on the quantification of DaTSCAN images. Eur J Nucl Med Mol Imaging
. 2010; 37: 23–35.
11. Meyer PT, Winz OH, Dafotakis M, et al.. Improved visual [123
interpretation for evaluation of parkinsonism by visual rating of parametric distribution volume ratio images. Q J Nucl Med Mol Imaging
. 2011; 55: 301–308.
12. Meyer PT, Sattler B, Lincke T, et al.. Investigating dopaminergic neurotransmission with 123
: comparability of modern SPECT
systems. J Nucl Med
. 2003; 44: 839–845.
13. Koch W, Radau PE, Münzing W, et al.. Cross-camera comparison of SPECT
measurements of a 3-D anthropomorphic basal ganglia phantom. Eur J Nucl Med Mol Imaging
. 2006; 33: 495–502.
Keywords:© 2012 Lippincott Williams & Wilkins, Inc.
SPECT; dopamine transporter; 123I-FP-CIT; iterative reconstruction