Introduction

Alzheimer's disease (AD) is a neurodegenerative disease associated with activation of normally dormant microglia cells (Cagnin et al, 2001a). These microglia cells are the brain's intrinsic macrophages (Kreutzberg, 1996). Activated microglia express the peripheral benzodiazepine receptor, but expression of this receptor on other constituents of the brain is low (Banati et al, 2000). Thus compounds that bind to the peripheral benzodiazepine receptor can be used to image-activated microglia.

PK11195 is a specific ligand for the peripheral benzodiazepine receptor, which has been used to investigate microglia activation using either positron emission tomography (PET) or single photon emission computed tomography. Unfortunately, analysis of (R)-[¹¹C]PK11195 PET scans has proven to be difficult because of its low specific to nonspecific binding ratio and its very low extraction fraction (Kropholler et al, 2005). Consequently, attention should be paid toward optimizing methods for extracting the specific signal. Currently, more promising peripheral benzodiazepine receptor tracers are under investigation, such as DAA1106 (Maeda et al, 2004; Ikoma et al, 2007) and vinpocetine (Gulyas et al, 2002a, 2002b, 2002c, 2005).

Several (R)-[¹¹C]PK11195 studies investigating AD have been undertaken. Two studies used PET (Groom et al, 1995; Cagnin et al, 2001a) and one single photon emission computed tomography (Versijpt et al, 2003). Only Cagnin et al, (2001a) used the pure (R)-enantiomer, which has higher binding than the (S)-enantiomer (Shah et al, 1994), whereas the other studies used a racemic mixture. To determine the degree of binding in the regions of interest (ROI), all three studies used a tracer kinetic model depending on a reference region either by using the simplified reference tissue model (SRTM) (Lammertsma and Hume, 1996) or by normalizing activities. Two groups (Groom et al, 1995; Versijpt et al, 2003) used cerebellum as reference region, whereas a different group used a cluster analysis method to derive the reference region (Ashburner et al, 1996; Cagnin et al, 2001a). This cluster analysis approach is commonly used to analyze (R)-[¹¹C]PK11195 scans (Banati et al, 2000, 2001; Cagnin et al, 2004, 2001b; Gerhard et al, 2003; Turner et al, 2004, 2005; Henkel et al, 2004). Recently, a comparison of parametric methods for analysis of (R)-[¹¹C]PK11195 was published by our group (Schuitemaker et al, 2007).

For analysis of (R)-[¹¹C]PK11195 PET scans in AD and mild cognitive impairment (MCI), it has been argued that the cerebellum may be inadequate as a reference region, as activated microglia could be present (Braak and Braak, 1991). Therefore, cluster analysis, which defines groups of voxels with a 'normal' (R)-[¹¹C]PK11195 binding, could be an alternative method, as this method does not depend on predefined regions. However, this method has not been fully validated yet. It is of interest to note that the two PET studies of Groom (Groom et al, 1995) and Cagnin et al (2001a) produced conflicting results. It is likely that these different results are, at least in part, because of differences in methodology. Clearly, there is a need to evaluate which method is optimal for defining a reference region for (R)-[¹¹C]PK11195 studies in AD. This is especially important for multicentre studies on MCI.

The purpose of this study was to compare various reference region definitions and to evaluate their applicability for analysing differences in microglia activation in MCI and AD subjects. Moreover, simulations were performed to evaluate the performance of the cluster analysis approach under the influence of varying blood volume fractions (V_b) and number of extracted clusters.

Materials and methods

Clinical Evaluation

Subjects

Four groups were included, namely younger healthy subjects (n=10, 7 male subjects, age 3010 years), elderly healthy subjects (n=10, 6 male subjects, age 706 years), patients diagnosed with amnestic MCI (n=10, 6 male subjects, age 746 years) according to Petersen criteria (Petersen et al, 1999) and subjects diagnosed with probable AD (n=9, 6 male subjects, age 716 years), meeting NINCDS-ADRDA criteria (McKhann et al, 1984). Average (s.d.) mini mental state examination scores (Folstein et al, 1975) were 291 for elderly healthy subjects, 261 for MCI patients and 223 for AD patients, respectively. Control subjects had a normal physical and neurologic examination, including screening laboratory assessments, and had a normal magnetic resonance imaging (MRI) scan, which was evaluated by a neuroradiologist. The study was approved by the medical ethics committee of the VU University Medical Centre and all experiments were performed in line with the recommendations of the Helsinki declaration. All subjects gave written informed consent before inclusion. Data were only used for comparison of methods. Clinical results are beyond the scope of the present study and will be reported elsewhere.

Scanning Protocol

All subjects underwent a 1-h dynamic brain scan in three-dimensional mode performed on a Siemens/CTI ECAT EXACT HR+ scanner (CTI/Siemens, Knoxville, TN, USA). Characteristics of this scanner have been reported elsewhere (Brix et al, 1997; Adam et al, 1997). Frame durations of the scan were, in order, 1 15, 1 5, 1 10, 2 15, 2 30, 3 60, 4 150, 5 300, and 2 600 secs. A bolus of 40060 MBq of (R)-[¹¹C]PK11195 with a specific activity of 6532 GBq/mol at time of injection was injected at the start of the scan. A neuro-insert (CTI, Knoxville, TN, USA) was used to decrease scatter from outside the field of view. Before tracer injection, a two-dimensional transmission scan was performed, which was used to correct the subsequent emission scan for attenuation. Arterial blood was withdrawn continuously using a blood-sampling device described previously (Boellaard et al, 2001). At set times (3, 5, 10, 20, 30, 40, and 60 mins after injection), continuous blood sampling was interrupted briefly for withdrawal of manual samples followed by flushing of the arterial line with heparinized saline. Radioactivity concentrations of (R)-[¹¹C]PK11195 in plasma and whole blood of the manual samples were measured using a cross-calibrated well counter, and metabolites in plasma using high-pressure liquid chromatography analysis, as described previously (Greuter et al, 2005).

In addition, all subjects underwent a T1 weighted MRI scan on a 1T IMPACT scanner (Siemens Medical Solutions, Erlangen, Germany). This scan was used for gray/white matter segmentation and for ROI definition.

Image Reconstruction

All emission scans were normalized and corrected for attenuation, randoms, dead time, scatter (Watson, 2004), and decay. Emission scans were reconstructed using filtered back projection with a Hanning filter with cutoff at 0.5 times the Nyquist frequency. A zoom factor of 2 and a matrix size of 256 256 63 were used, resulting in a voxel size of 1.2 1.2 2.4 mm and a reconstructed spatial resolution of approximately 7 mm full-width at half-maximum at the center of the field of view.

Region of Interest Definition and Extraction of Time–Activity Curves

The skull was removed from the MRI scans using a brain extraction tool (Smith, 2002), which is part of the FMRIB software library (www.fmrib.ox.ac.uk/fsl, Image analysis group, Oxford, UK). Deskulled brain and corresponding PET scans were aligned using a mutual information algorithm included in the image registration package MIRIT (Maes et al, 1997; West et al, 1997). Correct alignment was checked visually. Both total cerebellum and cerebrum ROI were drawn manually on the coregistered MRI scans.

The MRI was segmented into gray and white matter (Zhang et al, 2001) and separate gray and white matter cerebellum and cerebrum ROI were defined. The software used for this segmentation was also part of the FMRIB software library. To evaluate performance of the various reference regions, parametric binding potential (BP) images (Ichise et al, 1996) were used to draw a data-driven (target) ROI around the thalamus. In agreement with a previous observation in elderly healthy subjects and AD patients (Cagnin et al, 2001a), highest binding was observed in this region. In addition, by including young healthy subjects, the thalamus ROI provided the largest range of possible BP values for the present comparison of reference regions. The cluster region was extracted as described elsewhere (Banati et al, 2000). This region represents a group of voxels with similar kinetics and does not correspond to an actual 'anatomic' reference region, such as cerebellum or pons. Nevertheless, as the corresponding time–activity curve (TAC) and those of cerebellum and cerebrum were compared as input curves in SRTM, the term reference region will also be used for the cluster region. Software to perform this clustering, written in Matlab (The Mathworks, Natick, MA, USA), was kindly provided by the MRC Clinical Sciences Centre (Faculty of Medicine, Imperial College, London, UK). To produce results comparable with previous studies, where no plasma input was used, the cluster with visually the lowest BP was selected as reference region. Because at several occasions clustered regions did not correspond to anatomic structures, they were evaluated visually to ensure that tissue was selected. Ten clusters were extracted from each scan (Banati et al, 2000).

Kinetic Analyses

Kinetic modeling was performed using dedicated software developed within Matlab. A metabolite-corrected plasma input function was obtained as follows. Plasma/whole-blood concentration ratios obtained from the manual blood samples were fitted to an exponential function. The whole-blood (online sampler) curve was multiplied with this exponential function to obtain the corresponding (total) plasma curve. The (R)-[¹¹C]PK11195 parent fraction in the plasma samples was fitted to the function f=1-a((2-exp(-bt)-exp(-ct)), where f is parent fraction and a, b, and c are positive fitted parameters. The plasma curve was multiplied with this function to obtain the metabolite-corrected plasma input function. Delay between arrival of the (R)-[¹¹C]PK11195 bolus in the brain and at the online blood sampler was estimated by fitting a reversible two-tissue compartment model including blood volume and delay to the subjects' whole cortex TAC. Delay was then fixed to this value for all other ROI of the subject.

Reference region BP (i.e., nonspecific binding) was obtained using a reversible two-tissue compartment model (Kropholler et al, 2005), containing a blood volume term with separate inputs for metabolite-corrected plasma (input to tissue) and whole blood (fractional blood volume) and is denoted as BP^PLASMA. To allow equal comparison between reference regions the K₁/k₂ ratio was not fixed. Binding potential in a target region was derived using SRTM (Lammertsma and Hume, 1996) and is denoted as BP^SRTM. This BP^SRTM was also corrected for nonspecific binding by multiplying it with 1+BP^PLASMA (Kropholler et al, 2006), which is then denoted as BP^SRTM_CORR.

Simulations

Simulations were performed to evaluate the ability of the cluster analysis algorithm to separate voxels with high and low BP. Because (R)-[¹¹C]PK11195 uptake in brain is very low (K₁0.05 ml/min ml) and the concentration in blood decreases rather slowly (50% parent fraction after 1 h), the shape of (R)-[¹¹C]PK11195 TAC are, to a large extent, determined by the radioactivity concentration in blood. Therefore, the sensitivity of cluster analysis BP^PLASMA to physiologic variation in V_b was also determined. Volumes with simulated TAC and either 5%, 10%, or 15% coefficient of variation (COV) noise level were generated and grouped into clusters. Normal tissue was simulated using a two-tissue reversible compartment model and kinetic parameters reported previously (Kropholler et al, 2005), namely K₁=0.05 ml/min ml, k₂=0.15 min^-1, k₄=0.04 min^-1, BP=1.5, and V_b=0.06. In the subject population, a variation in BP of 0.7 and in V_b of 0.03 was found. Volumes containing both an ROI with increased BP (2.2) and an ROI with increased V_b (V_b=0.09) were simulated. Simulated volume size was 32 32 15 voxels. Regions of interest with different kinetics were 9 9 9 voxels large. In clinical studies, 10 clusters were used to represent structures such as tissue, blood vessels, skull and ventricles. In the simulations only three regions with different characteristics were present. To account for this smaller number two, three, or four clusters were extracted from the simulated volumes, representing a lower, the same or a higher number of clusters than present in the data. The extracted cluster ROI with lowest BP^PLASMA was used as reference region cluster. BP^SRTM using the cluster analysis ROI as reference region and BP^PLASMA of the cluster analysis ROI were evaluated. A noiseless TAC with the same kinetic parameters as the high BP ROI was used as target ROI and the effects of noise level and number of clusters on BP^SRTM values were evaluated.

Results

Clinical Evaluation

The clustering algorithm resulted in a rough separation between gray and white matter, blood vessels, and cerebrospinal fluid (Figure 1A and B). In total, 10 clusters were extracted and the TAC with the lowest BP^PLASMA was used as reference region TAC. For four out of 39 subjects clustering did not result in physiologically meaningful clusters (Figure 1C). Results from these subjects were not included in the comparative analyses. No correlation was observed between unidentifiable clusters and injected dose. The distribution of voxels over the 10 clusters decreased from 274% of voxels in the largest cluster to 204% in the second largest and 10.3% in the smallest cluster. The cluster containing the largest number of voxels (partially) represented skull in more than 80% of the cases. For several subjects fitting a plasma input model to the cluster TAC provided poor results (very high BP or high COV associated with fitted BP). The BP^PLASMA values were excluded from analysis when COV was larger than 60% or when BP^PLASMA itself was >5. One subject was rejected because of inconsistent BP^PLASMA results between ROI, which was not observed for the other subjects. Rejection of fits was assessed independently for different kinetic models, that is subjects that gave unrealistic results for one model were included when analyzing other models. The number of excluded scans is reported in Table 1. It can be seen that a substantial number of fits were rejected, with the highest number for BP^SRTM (and consequently also BP^SRTM_CORR) using cerebrum white matter as reference region input. This high rejection rate indicates that this white matter region is not a suitable reference tissue input for the gray matter thalamus region. For most reference regions, the rejection rate of BP^SRTM fits was <10%. The rejection rate of BP^SRTM_CORR was higher, as this analysis also required an acceptable plasma input fit of the reference region itself. Often the same fit is rejected for several reference regions. Rejection of a fit for several reference regions results in only one scan excluded in the total comparison, because it is the same subject for which several TACs result in rejected fits.

Figure 1.

Cluster images of (A) one elderly, (B) one AD, and (C) one young subject. Each color corresponds to a cluster index and contains voxels having similar TAC shapes, for example red corresponds with large blood vessels, light blue with gray matter, dark blue with skull, and green with white matter. A and B show that clustering results in a rough separation into gray and white matter. Occasionally clustering did not provide meaningful clusters (C).

Full figure and legend (174K)

Table 1 - Number of rejected fits. Number of rejected subjects for a specific ROI and model can be higher than the total number of rejected subjects as not all ROI were included in all comparisons.

Full table

Reasonable correlations were observed between total binding of anatomic and cluster ROI obtained using a plasma input model (BP^PLASMA). Scatter plots between total cerebellum and cluster analysis BP^PLASMA and between cerebrum gray matter and cluster analysis BP^PLASMA are shown in Figures 2A and B. Correlations were highest between BP^PLASMA from cerebrum and cluster analysis (R²=0.93, 0.92, and 0.94 for cerebrum gray matter, white matter, and total cerebrum, respectively) and considerably lower between cerebellum and cluster analysis (R²=0.52, 0.68, and 0.69, respectively). Measured BP^PLASMA values for the different subject groups are shown in Figure 3A. No statistical significant differences were observed between groups.

Figure 2.

Scatter plots of BP^PLASMA (A and B) and BP^SRTM (C and D) between cluster analysis and total cerebellum (A and C) or cerebrum gray matter (B and D). Data are shown for young (), elderly (), MCI (), and AD () subjects. For BP^PLASMA the regions mentioned are the fitted regions. For BP^SRTM they are the reference regions used for fitting the thalamus TAC.

Full figure and legend (40K)

Figure 3.

(A) BP^PLASMA, (B) BP^SRTM and (C) BP^SRTM_CORR for young (black), elderly (hatched), MCI (gray), and AD (dotted) subjects. For BP^PLASMA, the regions mentioned are the fitted regions, whereas for BP^SRTM and BP^SRTM_CORR they are the reference regions used for fitting the target TAC (thalamus).

Full figure and legend (109K)

The BP^SRTM values were rejected if COV of the fitted BP was larger than 70% or when BP itself was either larger than 1 or smaller than 0. A high number of unrealistic BP^SRTM results were observed when cerebellum white matter, total cerebrum, or cerebrum white matter was used as reference region (6, 6, and 13, respectively; Table 1). This suggests that these ROI are not suitable as reference regions. To increase statistical power, these ROI were excluded from BP^SRTM analysis. Correlations were observed between target BP^SRTM calculated using the various ROI as reference region. Correlation was calculated between target BP^SRTM values using total cerebellum and cluster analysis as reference region (Figure 2C, R²=0.76). Lower correlations were observed between the use of cluster analysis reference region and cerebrum gray matter (Figure 2D, R²=0.71) and between the use of cluster analysis reference region and cerebellum gray matter (R²=0.59). Measured BP^SRTM values for the different subject groups are shown in Figure 3B. No statistically significant differences were observed between groups. Little variation was observed in the COV between ROI or between subject groups.

High correlations were observed between BP^SRTM_CORR values (i.e. BP^SRTM corrected for nonspecific binding) of the target region measured using the various reference region methods. Correlation coefficients between the use of cluster analysis as reference region and the use of either cerebellum gray matter, total cerebellum, or cerebrum gray matter were R²=0.86, 0.86, and 0.83, respectively. Measured BP^SRTM_CORR values for the different subject groups are shown in Figure 3C. No statistically significant differences were observed between groups.

To investigate possible effects of intravascular activity on extracted cluster TAC, the correlation between cluster analysis BP^PLASMA and V_b was investigated. In contrast to other results, all extracted clusters, except those representing tissue or cerebral spinal fluid, were included in this analysis. A scatter plot of cluster analysis BP^PLASMA versus V_b is shown in Figure 4. A large number of clusters (201 out of 390) were excluded because they resulted in unrealistic or unreliable BP or V_b values (BP^PLASMA>15 or its COV>70%; V_b<0, V_b>0.12 or its COV>70%). Cluster TAC with low BP^PLASMA tended to have high V_b and cluster TAC with high BP^PLASMA tended to have low V_b.

Figure 4.

Scatter plot of cluster analysis BP^PLASMA and V_b. Results are shown for all extracted clusters and for eight young (), nine elderly (), 10 MCI (), and eight AD subjects ().

Full figure and legend (12K)

Simulations

Cluster analysis resulted in separation of areas with different BP and V_b (Figure 5A–F). The algorithm was more sensitive to BP than to V_b changes (Figure 5A). When the number of clusters used was higher than the number of areas with different kinetics, the simulated area with the lowest BP was separated into more than one cluster (Figure 5C and F). Similar ROI were extracted independent of noise level.

Figure 5.

Images of various clustered simulated volumes containing a region with increased BP (2.2, left inserted square) and a region with increased V_b (0.09, right inserted square) in a homogeneous background (BP=1.5; V_b=0.06). Images are shown for data with 10 (A–C) and 15% (D–F) noise levels, clustered into two (A and D), three (B and E), and four (C and F) clusters.

Full figure and legend (98K)

Extraction of the volume into two clusters resulted in the background cluster (i.e., the volume without the high BP and V_b region) corresponding to the cluster ROI with the lowest BP^PLASMA.

When the volume was clustered into three clusters, the reference region cluster corresponded to the simulated ROI containing the high V_b voxels. When the volume was extracted into four clusters, the smallest cluster in the background, which did not contain the high BP or high V_b voxels, was the cluster with lowest BP^PLASMA.

Because of nonspecific binding (BP^PLASMA1.5), all BP^SRTM values were scaled down by a factor of approximately 2.5. Considerable variation was observed in BP^SRTM depending on the number of extracted clusters (Figure 6A). BP^SRTM appeared to be nearly independent of noise level when the number of extracted clusters was correct or too low, but variation up to 40% was observed when the number of extracted clusters was higher than the number of different structures in the scan. BP^SRTM was approximately 40% lower when three clusters were extracted instead of two. The BP^PLASMA results showed a more consistent trend, with increasing negative bias as the number of extracted clusters and the noise level increased (Figure 6B).

Figure 6.

Results from simulations showing target ROI BP^SRTM using the cluster analysis ROI as reference region (A) and cluster analysis ROI BP^PLASMA (B) as function of noise level (5%, 10%, and 15%) for two (black), three (hatched) or four (gray) extracted clusters.

Full figure and legend (65K)

Discussion

The most important finding of this study is that BP^SRTM and BP^SRTM_CORR obtained with anatomic reference regions are linearly related to BP^SRTM (R²=0.59 to 0.76) and BP^SRTM_CORR (R²=0.83 to 0.86) obtained with the reference region defined by cluster analysis in controls and patients with MCI and AD (Figure 2). On the basis of these relationships, the validity of using anatomic ROI and cluster analysis as reference region is similar, although performance differences exist which are reflected in different correlation coefficients for observed BP^PLASMA and BP^SRTM values (Figures 2 and 3). These results confirm those of Cagnin et al, (2001b), who found high BP^SRTM correlations (R²=0.94 to 0.96) when using either total cerebellum or cluster analysis as reference region in a healthy subject and a subject suffering from herpes encephalitis.

Correlation coefficients observed between cluster analysis and cerebrum ROI BP^PLASMA were high (R²=0.92 to 0.94) compared with those obtained between cluster analysis and cerebellum ROI BP^PLASMA (R²=0.52 to 0.68). Because all evaluated ROI were relatively large, differences in noise level are not responsible for this observed difference. Because the cluster analysis ROI will have more voxels in common with cerebrum ROI than with cerebellum ROI, their TAC shapes will be more similar and therefore the higher correlation with cerebrum BP^PLASMA is logical. This increased correlation, however, does not imply increased accuracy or precision.

Differences in COV of BP^PLASMA, BP^SRTM, and BP^SRTM_CORR for the various reference regions were small. Because noise levels in TAC of all reference regions evaluated were also low, no distinction between ROI could be made in terms of precision. Better precision of BP^SRTM was observed by Turkheimer et al (2007), who showed that use of anatomically defined reference regions resulted in better test–retest variabilities of BP^SRTM in target areas than those based on cluster analysis, even when using a cluster analysis with improved precision at the cost of decreased accuracy.

Observed BP^SRTM results using simulated clusters as reference region resulted in lower BP^SRTM values for three clusters than for either two or four clusters. For three clusters, the extracted reference region cluster represented voxels with high V_b, which resulted in a negative bias in BP^SRTM values (simulations, data not shown). Selection of voxels on the basis of V_b differences may occur, which can lead to a cluster analysis reference region with biased BP, also because a correlation was observed between V_b and BP (Figure 4).

The cluster analysis resulted in rejection of 10% of the scans evaluated. Banati and coworkers (Banati et al, 2000; Cagnin et al, 2001a) used comparison of extracted cluster TAC with a database of normal cluster TAC and rejected TAC according to dissimilarity. It is not clear whether this criterion would have selected the same scans for rejection. Unfortunately, Banati and coworkers did not mention the number of rejected scans. Furthermore, the distribution of voxels over the various clusters in the present study was significantly different from that previously reported. Banati and coworkers (Banati et al, 2000) found that the two largest clusters contained 90% of the voxels. In the present study, the cluster volume decreased from 274% of the voxels toward 10.3% of the voxels, with the largest cluster representing skull in most of the cases. These differences may be related to differences in scanner characteristics, reconstruction algorithm, scatter correction, or differences in shielding for out of field of view scatter.

From a physiologic point of view, the cerebellum is affected in a more advanced stage of AD than the cerebrum and should therefore be preferred (Braak and Braak, 1991). In general (R)-[¹¹C]PK11195 binding is assumed to be absent in normal brain (Banati et al, 2000), except in the thalamus of healthy elderly controls (Cagnin et al, 2001a). In addition, it is assumed that increased BP reflects activation of microglia. Clearly, any dissociation between total (R)-[¹¹C]PK11195 uptake and binding to activated microglia between regions and/or subject groups would complicate physiologic interpretation of BP results. This is, however, unlikely a histologic study (Banati et al, 2000) has shown that (R)-[¹¹C]PK11195 binds to activated microglia and not to other cells. Finally, it is assumed that labelled metabolites do not cross the blood–brain barrier. At least for mice this has been confirmed (Dumont et al, 1999).

In summary, apart from the high number of unrealistic BP^SRTM results using the cerebellum white matter, total cerebellum, or cerebrum white matter as reference region, only small differences in reference region performance were observed. Total cerebellum, however, is the best choice, as it correlates well with other reference region BP, results in low numbers of unrealistic BP^SRTM values, and is likely to have specific binding only in a late state of the disease process. In addition, in contrast to cluster analysis, it does not have noise-induced bias and rejection of 10% of the scans.

At present there is increasing interest in in vivo imaging of inflammatory processes. A number of alternative ligands for the peripheral benzodiazepine receptor are under investigation, such as DAA1106 (Maeda et al, 2004; Ikoma et al, 2007) and vinpocetine (Gulyas et al, 2002a, 2002b and 2002c, 2005). Although initial results are promising in that uptake of these ligands in brain is higher than that of (R)-[¹¹C]PK11195, there is a need for full quantitative comparative studies. It is likely that results of the present study are also relevant for these new ligands, as they target the same receptor.

Conclusion

BP values of (R)-[¹¹C]PK11195 scans in normal subjects and MCI and AD patients are similar when total cerebellum or cluster analysis are used as reference regions. Total cerebellum is the reference region of choice, because cluster analysis fails in approximately 10% of cases.

References

1. Adam LE, Jears J, Ostertag H, Trojan H, Bellemann ME, Brix G (1997) Performance of the Whole-Body PET scanner ECAT EXACT HR+ following the IEC standard. IEEE Trans Nucl Sci 44:1172–1179 | Article | ISI |
2. Ashburner J et al. (1996) A cluster analysis approach for the characterization of dynamic pet data. In: Myers R et al. (eds) Quantification of brain function using PET. San Diego, Academic Press pp 301–306
3. Banati RB, Cagnin A, Brooks DJ, Gunn RN, Myers R, Jones T, Birch R, Anand P (2001) Long-term trans-synaptic glial responses in the human thalamus after peripheral nerve injury. Neuroreport 12:3439–3442 | Article | PubMed | ISI | ChemPort |
4. Banati RB, Newcombe J, Gunn RN, Cagnin A, Turkheimer F, Heppner F, Price G, Wegner F, Giovannoni G, Miller DH, Perkin GD, Smith T, Hewson AK, Bydder G, Kreutzberg GW, Jones T, Cuzner ML, Myers R (2000) The peripheral benzodiazepine binding site in the brain in multiple sclerosis: quantitative in vivo imaging of microglia as a measure of disease activity. Brain 123(Part 11):2321–2337 | Article | PubMed | ISI |
5. Boellaard R, van Lingen A, van Balen SC, Hoving BG, Lammertsma AA (2001) Characteristics of a new fully programmable blood sampling device for monitoring blood radioactivity during PET. Eur J Nucl Med 28:81–89 | Article | PubMed | ISI | ChemPort |
6. Braak H, Braak E (1991) Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol (Berl) 82:239–259 | Article | PubMed | ChemPort |
7. Brix G, Zaers J, Adam LE, Bellemann ME, Ostertag H, Trojan H, Haberkorn U, Doll J, Oberdorfer F, Lorenz WJ (1997) Performance evaluation of a whole-body PET scanner using the NEMA protocol. National Electrical Manufacturers Association. J Nucl Med 38:1614–1623 | PubMed | ISI | ChemPort |
8. Cagnin A, Brooks DJ, Kennedy AM, Gunn RN, Myers R, Turkheimer FE, Jones T, Banati RB (2001a) In-vivo measurement of activated microglia in dementia. Lancet 358:461–467 | Article | PubMed | ISI | ChemPort |
9. Cagnin A, Myers R, Gunn RN, Lawrence AD, Stevens T, Kreutzberg GW, Jones T, Banati RB (2001b) In vivo visualization of activated glia by [¹¹C] (R)-PK11195-PET following herpes encephalitis reveals projected neuronal damage beyond the primary focal lesion. Brain 124:2014–2027 | Article | PubMed | ISI | ChemPort |
10. Cagnin A, Rossor M, Sampson EL, Mackinnon T, Banati RB (2004) In vivo detection of microglial activation in frontotemporal dementia. Ann Neurol 56:894–897 | Article | PubMed | ISI |
11. Dumont F, De Vos F, Versijpt J, Jansen HM, Korf J, Dierckx RA, Slegers G (1999) In vivo evaluation in mice and metabolism in blood of human volunteers of [¹²³I]iodo-PK11195: a possible single-photon emission tomography tracer for visualization of inflammation. Eur J Nucl Med 26:194–200 | Article | PubMed | ChemPort |
12. Folstein MF, Folstein SE, McHugh PR (1975) 'Mini-mental state'. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 2:189–198 | Article |
13. Gerhard A, Banati RB, Goerres GB, Cagnin A, Myers R, Gunn RN, Turkheimer F, Good CD, Mathias CJ, Quinn N, Schwarz J, Brooks DJ (2003) [¹¹C](R)-PK11195 PET imaging of microglial activation in multiple system atrophy. Neurology 61:686–689 | PubMed | ISI | ChemPort |
14. Greuter HN, van Ophemert PL, Luurtsema G, van Berckel BN, Franssen EJ, Windhorst BD, Lammertsma AA (2005) Optimizing an online SPE-HPLC method for analysis of (R)-[¹¹C]1-(2-chlorophenyl)-N-methyl-N-(1-methylpropyl)-3-isoquinolinecarb oxamide [(R)-[¹¹C]PK11195] and its metabolites in humans. Nucl Med Biol 32:307–312 | Article | PubMed | ChemPort |
15. Groom GN, Junck L, Foster NL, Frey KA, Kuhl DE (1995) PET of peripheral benzodiazepine binding sites in the microgliosis of Alzheimer's disease. J Nucl Med 36:2207–2210 | PubMed | ISI | ChemPort |
16. Gulyas B, Halldin C, Sandell J, Karlsson P, Sovago J, Karpati E, Kiss B, Vas A, Cselenyi Z, Farde L (2002a) PET studies on the brain uptake and regional distribution of [¹¹C]vinpocetine in human subjects. Acta Neurol Scand 106:325–332 | Article | PubMed | ChemPort |
17. Gulyas B, Halldin C, Sovago J, Sandell J, Cselenyi Z, Vas A, Kiss B, Karpati E, Farde L (2002b) Drug distribution in man: a positron emission tomography study after oral administration of the labelled neuroprotective drug vinpocetine. Eur J Nucl Med Mol Imaging 29:1031–1038 | Article | PubMed | ChemPort |
18. Gulyas B, Halldin C, Vas A, Banati RB, Shchukin E, Finnema S, Tarkainen J, Tihanyi K, Szilagyi G, Farde L (2005) [¹¹C]vinpocetine: a prospective peripheral benzodiazepine receptor ligand for primate PET studies. J Neurol Sci 229–230:219–223 | Article |
19. Gulyas B, Vas A, Halldin C, Sovago J, Sandell J, Olsson H, Fredriksson A, Stone-Elander S, Farde L (2002c) Cerebral uptake of [ethyl-¹¹C]vinpocetine and 1-[¹¹C]ethanol in cynomolgous monkeys: a comparative preclinical PET study. Nucl Med Biol 29:753–759 | Article | PubMed | ChemPort |
20. Henkel K, Karitzky J, Schmid M, Mader I, Glatting G, Unger JW, Neumaier B, Ludolph AC, Reske SN, Landwehrmeyer GB (2004) Imaging of activated microglia with PET and [¹¹C]PK 11195 in corticobasal degeneration. Mov Disord 19:817–821 | Article | PubMed | ISI |
21. Ichise M, Ballinger JR, Golan H, Vines D, Luong A, Tsai S, Kung HF (1996) Noninvasive quantification of dopamine D2 receptors with iodine-123-IBF SPECT. J Nucl Med 37:513–520 | PubMed | ISI | ChemPort |
22. Ikoma Y, Yasuno F, Ito H, Suhara T, Ota M, Toyama H, Fujimura Y, Takano A, Maeda J, Zhang MR, Nakao R, Suzuki K (2007) Quantitative analysis for estimating binding potential of the peripheral benzodiazepine receptor with [(11)C]DAA1106. J Cereb Blood Flow Metab 27:173–184 | Article | PubMed | ChemPort |
23. Kreutzberg GW (1996) Microglia: a sensor for pathological events in the CNS. Trends Neurosci 19:312–318 | Article | PubMed | ISI | ChemPort |
24. Kropholler MA, Boellaard R, Schuitemaker A, Folkersma H, Berckel BNM, Lammertsma AA (2006) Evaluation of reference tissue models for the analysis of [¹¹C](R)-PK11195 studies. J Cereb Blood Flow Metab 26:1431–1441 | Article | PubMed | ChemPort |
25. Kropholler MA, Boellaard R, Schuitemaker A, van Berckel BN, Luurtsema G, Windhorst AD, Lammertsma AA (2005) Development of a tracer kinetic plasma input model for (R)-[¹¹C]PK11195 brain studies. J Cereb Blood Flow Metab 25:842–851 | Article | PubMed | ISI | ChemPort |
26. Lammertsma AA, Hume SP (1996) Simplified reference tissue model for PET receptor studies. Neuroimage 4:153–158 | Article | PubMed | ISI | ChemPort |
27. Maeda J, Suhara T, Zhang MR, Okauchi T, Yasuno F, Ikoma Y, Inaji M, Nagai Y, Takano A, Obayashi S, Suzuki K (2004) Novel peripheral benzodiazepine receptor ligand [¹¹C]DAA1106 for PET: an imaging tool for glial cells in the brain. Synapse 52:283–291 | Article | PubMed | ISI | ChemPort |
28. Maes F, Collignon A, Vandermeulen D, Marchal G, Suetens P (1997) Multimodality image registration by maximization of mutual information. IEEE Trans Med Imaging 16:187–198 | Article | PubMed | ISI | ChemPort |
29. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of department of health and human services task force on Alzheimer's disease. Neurology 34:939–944 | PubMed | ISI | ChemPort |
30. Petersen RC, Smith GE, Waring SC, Ivnik RJ, Tangalos EG, Kokmen E (1999) Mild cognitive impairment: clinical characterization and outcome. Arch Neurol 56:303–308 | Article | PubMed | ISI | ChemPort |
31. Schuitemaker A, Berckel BNM, Kropholler MA, Kloet RW, Jonker C Scheltens P, Lammertsma AA. (2007) Evaluation of methods for generating parametric (R)-[¹¹ C]PK11195 binding images. J Cereb Blood Flow Metab (Epub ahead of print)
32. Shah F, Hume SP, Pike VW, Ashworth S, McDermott J (1994) Synthesis of the enantiomers of [N-methyl-¹¹C]PK 11195 and comparison of their behaviours as radioligands for PK binding sites in rats. Nucl Med Biol 21:573–581 | Article | PubMed | ChemPort |
33. Smith SM (2002) Fast robust automated brain extraction. Hum Brain Mapp 17:143–155 | Article | PubMed | ISI |
34. Turkheimer FE, Edison P, Pavese N, Roncaroli F, Anderson AN, Hammers A, Gerhard A, Hinz R, Tai YF, Brooks DJ (2007) Reference and target region modeling of [¹¹C]-(R)-PK11195 brain studies. J Nucl Med 48:158–167 | PubMed |
35. Turner MR, Cagnin A, Turkheimer FE, Miller CC, Shaw CE, Brooks DJ, Leigh PN, Banati RB (2004) Evidence of widespread cerebral microglial activation in amyotrophic lateral sclerosis: an [¹¹C] (R)-PK11195 positron emission tomography study. Neurobiol Dis 15:601–609 | Article | PubMed | ISI | ChemPort |
36. Turner MR, Gerhard A, Al Chalabi A, Shaw CE, Hughes RA, Banati RB, Brooks DJ, Leigh PN (2005) Mills' and other isolated upper motor neurone syndromes: in vivo study with ¹¹C-(R)-PK11195 PET. J Neurol Neurosurg Psychiatry 76:871–874 | Article | PubMed | ISI | ChemPort |
37. Versijpt JJ, Dumont F, Van Laere KJ, Decoo D, Santens P, Audenaert K, Achten E, Slegers G, Dierckx RA, Korf J (2003) Assessment of neuroinflammation and microglial activation in Alzheimer's disease with radiolabelled PK11195 and single photon emission computed tomography. A pilot study Eur Neurol 50:39–47 | ChemPort |
38. Watson CC (2004) New, faster, image-based scatter correction for 3D PET. IEEE Trans Nucl Sci 47:1587–1594 | Article |
39. West J, Fitzpatrick JM, Wang MY, Dawant BM, Maurer Jr CR, Kessler RM, Maciunas RJ, Barillot C, Lemoine D, Collignon A, Maes F, Suetens P, Vandermeulen D, van den Elsen PA, Napel S, Sumanaweera TS, Harkness B, Hemler PF, Hill DL, Hawkes DJ, Studholme C, Maintz JB, Viergever MA, Malandain G, Woods RP (1997) Comparison and evaluation of retrospective intermodality brain image registration techniques. J Comput Assist Tomogr 21:554–566 | Article | PubMed | ISI | ChemPort |
40. Zhang Y, Brady M, Smith S (2001) Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans Med Imaging 20:45–57 | Article | PubMed | ISI | ChemPort |

Acknowledgements

The authors thank personnel of the BV Cyclotron VU for the production of [¹¹C]CO₂ and staff of the Department of Nuclear Medicine & PET Research for the production of (R)-[¹¹C]PK11195, data acquisition, and blood metabolite analysis. In addition, Dr P Pouwels (Department of Radiology, VU University Medical Centre, The Netherlands) is acknowledged for help with the MRI sequences, Dr FE Turkheimer (MRC Clinical Sciences Centre, Faculty of Medicine, Imperial College, Hammersmith Hospital, United Kingdom) for useful discussions and exchange of data, Dr R Gunn (previously MRC Clinical Sciences Centre, Faculty of Medicine, Imperial College, Hammersmith Hospital, UK, currently Translational Medicine and Genetics, GlaxoSmithKline, Greenford, UK) for providing the cluster analysis software, and FHP. van Velden, MSc, for critical reading of the manuscript. This work was made possible by research grants from the European Union (NCI-MCI project, number QLK6-CT-2000-00502), the Netherlands Organisation for Scientific Research (NOW VIDI grant 016.066.309), and from the Dutch Brain Foundation (number 9F01.21). This study was funded by the VU University Medical Centre.