Targeted multi-pinhole SPECT
- First Online:
- 626 Downloads
Small-animal single photon emission computed tomography (SPECT) with focused multi-pinhole collimation geometries allows scanning modes in which large amounts of photons can be collected from specific volumes of interest. Here we present new tools that improve targeted imaging of specific organs and tumours, and validate the effects of improved targeting of the pinhole focus.
A SPECT system with 75 pinholes and stationary detectors was used (U-SPECT-II). An XYZ stage automatically translates the animal bed with a specific sequence in order to scan a selected volume of interest. Prior to stepping the animal through the collimator, integrated webcams acquire images of the animal. Using sliders, the user designates the desired volume to be scanned (e.g. a xenograft or specific organ) on these optical images. Optionally projections of an atlas are overlaid semiautomatically to locate specific organs. In order to assess the effects of more targeted imaging, scans of a resolution phantom and a mouse myocardial phantom, as well as in vivo mouse cardiac and tumour scans, were acquired with increased levels of targeting. Differences were evaluated in terms of count yield, hot rod visibility and contrast-to-noise ratio.
By restricting focused SPECT scans to a 1.13-ml resolution phantom, count yield was increased by a factor 3.6, and visibility of small structures was significantly enhanced. At equal noise levels, the small-lesion contrast measured in the myocardial phantom was increased by 42%. Noise in in vivo images of a tumour and the mouse heart was significantly reduced.
Targeted pinhole SPECT improves images and can be used to shorten scan times. Scan planning with optical cameras provides an effective tool to exploit this principle without the necessity for additional X-ray CT imaging.
KeywordsSPECT Multi-pinhole Small-animal imaging Cardiac Tumour Focusing pinholes
Molecular imaging has proven to be extremely valuable in studying animal models of human disease and in the development of new pharmaceuticals and tracers. Many molecular mechanisms can be assessed quantitatively in vivo using radionuclide techniques such as single photon emission computed tomography (SPECT) and positron emission tomography (PET). In the past, SPECT lacked the resolution necessary to accurately image organs of small animals such as mice and rats. Several newly developed dedicated small-animal SPECT systems have overcome this limitation [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12].
Recently, sub-half-millimetre image resolution has been achieved in SPECT, using multi-pinhole collimators combined with high pinhole magnification factors [1, 3, 13, 14, 15]. As these systems are equipped with collimators that have pinholes that focus on a central area in the imaging cavity, the fraction of detected photons from specific organs or tissue of interest is very high. This can result in improved noise resolution trade-offs over systems with a lower level of focusing, and the possibility of reducing the tracer dose or the acquisition time. Using focused pinhole geometries, detailed images of mouse and rat organs (e.g. beating heart, kidney and the brain) and tumours have been acquired [1, 3, 13, 15, 16, 17, 18].
Multi-pinhole collimators with focused geometries are also able to scan larger volumes – up to the total body of mice and rats. This is accomplished by translating the animal through the collimator in concert with specially adapted reconstruction methods that use projections from all bed positions simultaneously [3, 14]. To increase count yield from a specific organ or tumour, the field-of-view (FOV) of a system with a highly focused geometry and an XYZ stage can be confined to a region that mainly contains the tissue of interest (“sensitivity painting”). This requires making an estimate of the location of these tissues. Disadvantages to performing this estimation based on X-ray CT images include additional radiation dose, hardware, and scan time. Localization based on MRI, which is currently an area of active research, also requires additional hardware and scan time. Furthermore, using pinhole projection images combined with a persistence scope, accurate localization is difficult to achieve since there is a small FOV, few gamma photons can be detected in a limited time and the tissue being localized may have a very low uptake. The aim of the present study was to explore the alternative possibility of using low-cost optical cameras for tissue localization and FOV selection, and to empirically investigate the effects of targeting on sensitivity and the quality of reconstructed images.
Materials and methods
SPECT system with optical cameras
The FOV outside the CFOV also contributes to the projection data, but in order to correctly reconstruct volumes significantly larger than the CFOV, the system must move the focus over the region of interest. This scanning focus method  enables sensitivity painting similar to dose painting in radiotherapy. The bed is mounted on a motor-controlled XYZ stage, which allows accurate positioning of the animal. During SPECT acquisition, the XYZ stage automatically moves the animal stepwise through the collimator, thereby effectively moving the CFOV over the animal in order to obtain complete data for any part of the animal that is selected by the user. The bed is transparent and has a half-cylindrical shape. It contains a transparent heater pad to control the temperature of the animal (Fig. 1a).
Description of the FOV selection tool
Optical image-based positioning
Prior to acquiring SPECT data, three optical cameras, which are integrated with the U-SPECT-II system, take photographs of the animal from the left, top and right (Fig. 1a). The cameras (DFK 21F04; The Imaging Source, Germany) are equipped with a quarter-inch CCD detector with a resolution of 640 × 480 pixels. The photographs are displayed on a graphical user interface, on which the user can define a box to be scanned (Fig. 1a, b). Next, the software calculates a sequence of bed positions in such a way that the volume within the box will be sampled by the CFOV in at least one of the bed positions. The user can optionally check and fine-tune the position of the FOV using real-time gamma photon projection images of the centre of the selected FOV, obtained through pinholes that provide views of the animal at approximately the same angles as the optical cameras. Depending on the size of the selection, the number of positions can range from one or two positions (for organs such as the heart, the brain or a tumour) up to tens of positions for total-body scanning. Since changing the bed position takes only 0.7 s on average, even for total-body studies fast dynamic acquisitions are possible using the scanning focus method [15, 21].
Aligning optical images to SPECT images
The mapping of the optical images (and consequently also the FOV selection) to the SPECT reconstructed voxel grid is determined by calibration. Before calibration, the optical cameras were corrected for rotation, scaling and barrel distortions based on optical images of a millimetre grid. In order to reduce perspective errors and approximate parallel projections, the optical images were acquired as a set of small image strips that were stitched together. The calibration was performed by scanning a phantom containing several point sources and registering its optical images to maximum intensity projections of the SPECT volume. To create the point sources, ion exchange resin beads with a diameter of approximately 0.2 mm were dipped in a mixture of 99mTc-pertechnetate and ink, to make them visible in both modalities.
The optical-to-SPECT registration was performed by applying a rigid transformation that minimizes the mean distance between the optical point sources and their corresponding point sources in the maximum intensity projection of the SPECT volume. After applying this transformation to our system, the maximum distance between any of the point sources in the optical and SPECT images was 0.25 mm.
Improvements due to restricting the scan area, in terms of sensitivity and resolution, were determined with a resolution phantom, whereas improvements in contrast-to-noise ratio were measured in scans of a mouse myocardial phantom containing a cold lesion. Finally, in vivo studies illustrated the effects of scan area size in mouse tumour and myocardial perfusion imaging.
All scans discussed here were performed three times, each time employing a different FOV selection: (1) a nontargeted scan, (2) a 1D-targeted scan (with the FOV only restricted in the z-dimension), emulating a system that can only target in the axial direction, and (3) a 3D-targeted scan with the FOV restricted in the x-, y- and z-dimensions (Fig. 2).
Resolution phantom study
In order to assess the visibility of small details, a Jaszczak-style resolution phantom (ultra-high-resolution micro-phantom 850.100; VANDERWILT techniques, Boxtel, The Netherlands) with six sections containing capillaries with diameters of 0.35, 0.40, 0.45, 0.50, 0.60 and 0.75 mm was imaged. In this phantom, the distance between the capillaries in each section equals the capillary diameter in that section. The phantom was filled with 145 MBq 99mTc-pertechnetate. The resolution phantom and the selected volumes for the different protocols are shown in Fig. 2a.
A 10-min acquisition was performed for each of the protocols (ultra-high-resolution study) and a second acquisition series with the same phantom was performed with lower activity by repeating the three scans after 20 h (high-resolution study). The ultra-high-resolution and high-resolution studies were performed with, respectively, the 0.35-mm and 0.6-mm diameter pinhole mouse collimators . Images were reconstructed on a 0.1875 mm isotropic voxel grid with ten iterations pixel-based ordered subset expectation maximization (POSEM) with 16 subsets . Furthermore, the total number of detected photons was determined for each scan in a 20% energy window around 140 keV to estimate the sensitivity gain achieved by restricting the scan volume. The count totals were corrected for background radiation by subtracting the number of counts detected in a separate background acquisition.
Mouse myocardial phantom study
The phantom was scanned using each of the three FOV selection protocols. The duration of the first scan was 60 min, and the next scans were slightly extended to correct for decay. Each of the three list-mode datasets was split into 60 noise realizations each containing the same number of list-mode events, which were spread out regularly over the entire scan time. This emulates 3 × 60 acquisitions with a left ventricular uptake of 0.12 MBq. Because the projections obtained each contained only 1/60 of the usual amount of background counts, projections from 60 separate 60-min background acquisitions without a phantom were added to the noisy phantom projections before reconstruction. The uptake value of 0.12 MBq was the average reconstructed left ventricular uptake measured in two mouse 99mTc-tetrofosmin scans after applying attenuation correction as described previously . For each targeting level, ten noise realizations were reconstructed on a 0.1875-mm isotropic voxel grid. Maximum likelihood expectation maximization instead of POSEM was used for reconstruction because the former updates the image with smaller increments, and therefore allows for constructing plots of contrast-to-noise ratios based on a higher number of stages of convergence.
In vivo animal studies
Figure 2c, d shows the three FOV selection protocols applied to, respectively, a mouse tumour scan and a myocardial perfusion scan. All procedures employed in these studies were approved by the local ethics committee and were performed in accordance with international guidelines on handling laboratory animals.
For the tumour scan, a 21-g female mouse (CB17/SCID) was used which had a 0.5–1.0 ml A431 human carcinoma on its right flank. The scans were acquired under isoflurane anaesthesia 3 days after injection of 49 MBq 111In-labelled Unibody (Genmab, The Netherlands) using the 0.6-mm diameter pinhole mouse collimator tube . The duration of each acquisition was 45 min. The images were reconstructed on a 0.1875-mm isotropic voxel grid employing six iterations POSEM with 16 subsets . The reconstructed images were postfiltered using a gaussian filter with σ = 0.1875 mm.
For the mouse cardiac perfusion study, a 29-g male mouse (C57BL/6J) was anaesthetized with isoflurane and injected with 134 MBq 99mTc-tetrofosmin. At 30 min after injection, the first SPECT scan of 45 min was acquired. The other two scans were corrected for decay of the isotope by adjusting the duration of each acquisition. The scans were acquired using the 0.6-mm diameter pinhole mouse collimator tube  and reconstructed on a 0.1875-mm isotropic voxel grid using four iterations POSEM with 16 subsets , employing cardiac gating with 16 intervals. The reconstructed images were postfiltered with a gaussian filter in time (σ = 1.27 time intervals) and space (σ = 0.75 mm).
Data analysis of the mouse myocardial phantom
Reconstructed images of the mouse myocardial phantom were postfiltered with a gaussian filter (σ = 0.1875 mm) and profiles were generated by dividing the profile range (depicted in Fig. 3c) into 36 segments and calculating the mean voxel value for each segment. In addition to plotting the profiles separately, the profiles for all ten noise realizations were also averaged into a mean profile. The variation from the mean was visualized by including two profiles denoting the mean plus and minus one standard deviation.
Resolution phantom study
Sensitivity estimates for phantom scans using various FOV selection protocols and activity levels. Increases in sensitivity are expressed as a sensitivity increase factor, defined as the number of counts divided by the number of counts measured in the nontargeted scan
Number of counts
Sensitivity increase factor
Number of counts
Sensitivity increase factor
Mouse myocardial phantom study
In vivo animal studies
A navigation and selection tool for SPECT acquisition based on optical imaging was developed to increase count yield from specific organs and tissues of interest. The results reported here show that targeting in three dimensions is important, because it improves count yield and contrast-to-noise trade-off.
When a large FOV is required, such as in total-body scanning, the level of targeting may become low. This results in reduced sensitivity as the total scan time is distributed over a large number of CFOV positions. Previously we and others have shown that in such situations the count rate can still be high enough to obtain excellent images  even for gated total-body SPECT or for studies which employ low doses . Furthermore, although the overall sensitivity in total body scanning may then become similar to non-focused geometries, the combination of focusing and moving the bed in three dimensions effectively translates the axis of rotation and may therefore have the advantage of acquiring more angle information compared to detectors rotating around one longitudinal axis. In the recently launched U-SPECT-II/CT and VECTor/CT (a combined SPECT/PET/CT device with sub-millimetre resolution [25, 26]), FOV selection can alternatively be performed based on X-ray planar images, with or without the option of registering projections of an atlas to them. It should, however, be noted that these devices are also equipped with webcams for, for example, FOV selection.
Using optical cameras with the option of registering projections of an atlas to the optical images is a cost-effective method and has many advantages over scan planning using an additional CT scan. First, the limited tissue contrast in a CT scan may hamper accurate localization for many organs, whereas our atlas images are already presegmented. Furthermore, X-ray imaging exposes the animal to an extra radiation dose [27, 28, 29]. Another advantage of recording optical images is that, after correcting for differences in positioning and distortion of the lens, they can be related directly to images obtained using other optical modalities such as bioluminescence or fluorescence. Additional CT scans or scans from other anatomic modalities can be useful in particular for the localization of activity in unknown internal structures such as a tumour.
In this study we used a basic atlas containing brain, heart, lungs, liver, spleen and kidneys, based on a mouse MRI scan. In practice, the localization of the organs worked well in most scans, especially since the posture of the animal was chosen to be close to that of the mouse in the MRI scan. More extensive in vivo position verification may be an interesting subject for future studies. In addition, atlases and anatomical images based on MRI or other modalities are being created, in which the animals can be placed in many different postures. Furthermore, other atlases of mice and rats are available [30, 31, 32, 33, 34, 35]. Even when using accurate atlases, small localization errors may remain as a result of individual differences or pathologies. An atlas that can be automatically deformed to match the optical image, based on thin-plate spline deformations, is currently being investigated [36, 37].
Focused multi-pinhole geometries combined with an XYZ translation stage give the opportunity to acquire a high number of counts from the organ of interest. As a result, much smaller pinholes can be used to obtain ultra-high-resolution SPECT images or images with reduced noise. The results of the present study show that count yield increases dramatically when targeting is applied, which results in new opportunities for fast dynamic imaging of tumours or organs. This new method could also be applied to increasing throughput or to reducing radiation doses. We have developed a fast and user-friendly tool for estimating organ positions based on optical images and optional atlases. This tool allows maximal benefit to be obtained from the unique advantages of SPECT with focused multi-pinhole collimators.
We thank Ruud Ramakers (UMC Utrecht) and Jeroen van den Brakel (Genmab) for their technical support and help with measurements, and Jan van Ewijk, Erwin Bakker and Jesse Bosma (UMC Utrecht) for their help in designing and manufacturing the mouse myocardial phantom.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
- 18.Van Steenkiste C, Staelens S, Deleye S, De Vos F, Vandenberghe S, Geerts A, et al. Measurement of porto-systemic shunting in mice by novel three-dimensional micro-single photon emission computed tomography imaging enabling longitudinal follow-up. Liver Int 2010;30(8):1211–20.CrossRefPubMedGoogle Scholar
- 19.Beekman FJ. Radiation detection device, scintillation device and detection method, as well as multiple image-forming device. PCT/NL2006/000513. April 19, 2007.Google Scholar
- 21.Vastenhouw B, Ramakers R, Beekman F. High resolution dynamic total-body animal imaging with U-SPECT-II. J Nucl Med 2009;50(Suppl 2):526.Google Scholar
- 26.Beekman FJ, van der Have F, Kreuger R, Goorden MC. Simultaneous sub-millimetre PET and SPECT with a dedicated multi-pinhole geometry. 2010 World Molecular Imaging Congress. Kyoto, Japan.Google Scholar
- 36.Baiker M, Dijkstra J, Que I, Löwik CWGM, Reiber JHC, Lelieveldt BPF. Organ approximation in micro-CT data with low soft tissue contrast using an articulated whole-body atlas. Proceedings of the 5th IEEE International Symposium on Biomedical Imaging. Paris, France. 2008;1267–70.Google Scholar
- 37.Baiker M, Vastenhouw B, Branderhorst W, Reiber JHC, Beekman F, Lelieveldt BPF. Atlas-driven scan planning for high-resolution μSPECT scanning from multi-view photographs: a pilot study. Proceedings of SPIE Medical Imaging 2009. Lake Buena Vista (Orlando), Florida. 2009; paper #72611L, 8 pages.Google Scholar