Introduction

Molecular imaging is the visualization, characterization and measurement of biological processes at the molecular and cellular levels in humans and other living beings [1]. Molecular imaging instrumentation consists of a variety of modalities that are nowadays often combined in multimodal imaging systems: SPECT (single photon emission computed tomography), PET (positron emission tomography), optical imaging, MRI (magnetic resonance imaging), MRS (magnetic resonance spectroscopy) and US (ultrasonography). Compared to techniques such as autoradiography and microscopy, the possibility of studying small animals longitudinally in vivo justifies the need for molecular imaging. Functional molecular imaging studies usually assess the spatial distribution of administered exogenous molecules and expression levels of their target (mostly enzymes and receptors). These imaging biomarkers can provide a certain degree of contrast by specifically binding to a target at an exquisite sensitivity in the picomolar range [2].

The use of extrinsic collimation to derive the direction of the photons hampers the overall sensitivity of SPECT compared to that of PET, which is based on electronic coincidence counting to gather spatial information. Therefore in SPECT, one needs to find an optimum between imaging time, injected dose and image noise. On the other hand, the spatial resolution of μSPECT is much higher since there is no physical lower limit caused by positron range (which can be reasonably high for some positron emitters, e.g. mean 0.6 mm for 18F in water [3]) and photon acolinearity as is the case in μPET. Also, parallax (depth-of-interaction effects) in the detector, which is the dominant factor in the resolution loss of PET, is much smaller in SPECT due to its lower photon energy. Moreover in μSPECT, these depth-of-interaction effects are usually reduced by pinhole magnification (typically by a factor of 3 to 12). While PET is able to follow the distribution of radiolabelled synthetic molecules with exquisite sensitivity, the relatively short half-lives of the common positron emitters 11C (20 min) and 18F (109 min) make them less suited to radiolabelling endogenous biomolecules. Due to their relatively large size, peptides and antibodies diffuse slowly into tissue, particularly if obstacles such as the blood–brain barrier reduce the delivery rate, and have relatively slow clearance from blood. In imaging studies, this may require hours or days for localization and washout from blood to achieve acceptable target to background levels. The time required for localization and blood clearance favours isotopes with longer half-lives such as the single photon emitters 99mTc (6.02 h), 123I (13.2 h) and 111In (2.8 days). Technetium, indium and iodine also have good chemical properties in binding biological compounds and do not require a cyclotron close by, which reduces costs. Although clinical PET imaging nowadays often outperforms SPECT in terms of image quality, the contrary is true for the preclinical arena. Here, the significantly higher resolution, although in a smaller field-of-view (FOV), of multipinhole SPECT compared to μPET is in many cases essential when imaging small animals, especially mice.

Small-animal SPECT systems are not merely scaled down versions of their clinical counterparts, but make use of dedicated multipinhole collimators. As a consequence of the pinhole magnification, measuring with a pinhole collimator can yield a reconstructed spatial resolution that is better than the detector’s intrinsic spatial resolution. However, a small pinhole results in reduced the sensitivity, which has to be counteracted to avoid high injected activities or excessive acquisition times. While the first generation of systems were still manufactured with a single pinhole [4–7] in combination with a conventional gamma camera requiring long scan times (about 1 h) and high doses (>1 mCi), systems are now built with multiple pinhole collimators [8–13]. Current small-animal systems have detectors that rotate combined with axial bed translation, or have stationary detectors and bed translation in XYZ directions to extend the FOV up to the entire animal’s body. Examples of such designs are, amongst others, the A-SPECT [6], the HiSPECT [14], the T-SPECT [15], the SemiSPECT [16], the FAST-SPECT [17], the U-SPECT-II, the NanoSPECT and the X-SPECT. A more extensive overview of pinhole imaging has been provided by Beekman and van der Have [18].

To provide multimodality imaging, SPECT systems are nowadays combined with an integrated CT scanner, which is placed behind or within the gantry of the SPECT imager. The most important application is to localize activity in the anatomical framework provided by CT. The CT information can also be used to perform partial volume, scatter and attenuation correction for improved tracer quantification [19]. SPECT has already been used as a tool in a broad range of applications: cardiovascular imaging [20, 21], imaging gene expression [22], oncology [23, 24], bone metabolism [25], neuroimaging [26] and inflammation [27], amongst other fields.

Imaging techniques are increasingly being applied to more challenging questions that relate to multiple molecular pathways in the body. Thus, the ability of SPECT to simultaneously acquire separate images of different molecules, enabling the resolution of the temporal relationship between different biological processes has become more important. This cannot be ensured with sequential studies when there is a rapidly changing pathophysiology. Imaging multiple molecular pathways at the same time can be solved by multi-isotope imaging in SPECT or the use of another collimator for simultaneous μPET and μSPECT [28].

We evaluated and compared the performance of the three most widely used state-of-the-art μSPECT systems for small-animal imaging: the U-SPECT-II (MILabs), the NanoSPECT (Bioscan) and the X-SPECT (GE). The evaluation criteria used in our comparison were reconstructed spatial resolution, sensitivity, contrast recovery and image uniformity for different isotopes (99mTc, 125I and 111In). These evaluations were performed for high-resolution (HR) and general purpose (GP) collimators, and involved mouse-sized phantoms.

Materials and methods

To obtain objective and representative data samples, measurements were performed in five different imaging facilities: the University of Ghent, Belgium (X-SPECT), the University of Florence, Italy (X-SPECT), the University Medical Center Utrecht, The Netherlands (U-SPECT-II), Radboud University Nijmegen, The Netherlands (U-SPECT-II) and Queen Mary University London, UK (NanoSPECT-NSO).

System descriptions

The main difference among the systems under evaluation was that in one camera (U-SPECT-II) the detectors are stationary, while in the others (NanoSPECT-NSO and X-SPECT) a gantry rotates around the object. A stationary system does not need rotation of heavy detectors and the only moving part is an XYZ stage that is also used for system matrix measurement [29, 30] obviating the need to perform geometric parameter calibration. The X-SPECT and the NanoSPECT-NSO have an adjustable radius of rotation (ROR) to adjust the magnification and a FOV for each specific imaging task. The U-SPECT-II on the other hand uses cylindrical collimators with different sizes and imaging FOV for rats and mice in order to maximize the count yield for the task at hand [31]. Furthermore, there is also a difference in the overlap of the projections. The U-SPECT-II makes use of detectors for which the projections do not overlap, while the NanoSPECT-NSO and the X-SPECT make use of projection multiplexing. While multiplexing increases the sensitivity, it also creates ambiguity during image reconstruction [32–34]. It has been reported that artifacts leading to, for example, image nonuniformities and ‘ghost activity’ can be attributed to this ambiguity [32]. The effects of multiplexing depend on the activity distribution, and also on the pinhole design, detector size and imaging distance.

U-SPECT-II

The U-SPECT-II system (Fig. 1a) has three detectors similar to a clinical triple headed SPECT system, resulting in a triangular shape. Each detector has a 9.5-mm thick crystal (NaI(Tl)) with an active detector area of 50.8 × 38.1 cm optically coupled through a light guide to 55 photomultiplier tubes (PMTs). The energy resolution is 10 % for 99mTc at 140 keV. The large detectors allow high pinhole magnification factors, which reduce the effects of low intrinsic detector resolution (3–4 mm) on the total system resolution. The total detector surface area is 5,806 cm2. Projections are discretized using a pixel size of 1 × 1 mm. Before reaching the detectors the photons first need to pass the 75-pinhole collimator in a configuration of five rings with 15 pinholes per ring. This provides sufficient sampling in a small region such that there is no need for rotation of either the object or the detector. However, the small FOV requires the animal bed to be translated in three dimensions for whole-body (WB) acquisitions, which is called the scanning focus method and combines the scanning of multiple focus positions with the simultaneous reconstruction of all the projection data [35]. Also, the number of bed positions can be reduced when using spiral trajectories on the U-SPECT-II [36]. Around the pinholes of the mouse collimators there is a tungsten tube with 75 rectangular holes to prevent overlap of the projections. We used the 0.35, 0.6 and 1 mm aperture size collimator tubes. A more detailed description of the system has been provided by van der Have et al. [37], and a selection of user applications with the U-SPECT-II have been described [38–46].

Fig. 1
figure 1

Commercially available small-animal SPECT systems evaluated in this study: a U-SPECT-II, b X-SPECT, c NanoSPECT-NSO

Full size image

X-SPECT

The X-SPECT system (Fig. 1b), as part of the Triumph (SPECT/PET/CT) system in its most complete configuration, has four rotating gamma camera heads (a configuration with one camera head was used in this study with compensation in phantom acquisition times) and is mounted on the same axial location of the gantry as the CT tube and X-ray detector. The camera consists of 5 × 5 CZT (CdZnTe) modules each made up of 16 × 16 pixel arrays of 1.5 mm square giving a total of 80 × 80 pixels and an active detector area of 12.7 × 12.7 cm2. The pixelated detector thus has a 1.5-mm intrinsic (discrete) resolution and 5 % energy resolution at 140 keV. Each gamma camera head can be equipped with interchangeable single, multipinhole (five) [47] or parallel-hole collimators. In this study, we used only the 0.5-mm and 1.0-mm multipinhole collimators. A selection of user applications with the X-SPECT can be found in the literature [48, 49].

NanoSPECT-NSO

The NanoSPECT-NSO system (Fig. 1c) consists in its most complete form of four rotating heads each with a 215 × 230 mm2 detector. The crystal thickness is 6.35 mm and the material NaI(Tl) covers 33 PMTs per detector and has an intrinsic resolution of 3.5 mm for 99mTc [50] and an energy resolution of 9.5 %. These detectors feature multiplexed multipinhole collimation with 9 up to 16 (optional) pinholes per detector. In this study, we used the 0.6-mm and 1-mm pinhole collimators. More information can be found in in the literature [50–53], and a selection of user applications with the NanoSPECT-NSO have also been described [54, 55].

Evaluation strategy

The highest achievable resolutions with the systems were measured with 99mTc using the HR collimators and by scanning both a phantom with three hot rod inserts (Fig. 4a) as well as a line source. In addition, to mimic GP use (higher throughput because of the higher sensitivity but with reduced resolution) of the systems, the GP collimators were used to measure spatial resolution, sensitivity, uniformity and contrast recovery. A matched pinhole diameter of 1 mm for the GP collimators was used for all three systems in this study.

HR mode measurements

The collimators used were the respective vendors’ highest resolution option, being 0.35-mm ultrahigh resolution (UHR) WB/focused mouse (75 pinholes) for the U-SPECT-II, the 0.5-mm low-energy (LE) mouse (5 pinholes/plate) for the X-SPECT, and the 0.6-mm UHR/focused mouse (9 pinholes/plate) for NanoSPECT-NSO (Table 1).

Table 1 HR and GP collimators
Full size table

To obtain a qualitative measure of the resolution over the entire transaxial FOV, we scanned a mouse-sized phantom containing three hot rod inserts (outer diameter of one insert 1 cm, length 0.85 cm) with capillary diameters ranging from 0.35 mm to 0.75 mm (Fig. 4a, Table 2) for 1 h on all systems. The minimum distance between the capillaries in the phantom within a certain segment was equal to the capillary diameter in that segment. These mouse hot rod phantoms were filled with a 99mTc solution at a concentration of 500 MBq/ml to avoid noise as a confounding factor in this HR experiment. Circular scans were performed for the X-SPECT (all the scans in the study were circular with the X-SPECT) and the NanoSPECT-NSO as the phantom fitted the FOV of one bed position. All the scans in the study for the X-SPECT had 64 detector positions and 24 detector positions for the NanoSPECT-NSO. Depending on the acceleration of the motor and the maximum speed of rotation, 64 detector positions result in about 30 s of dead time for the X-SPECT while 24 detector positions result in about 48 s of dead time for the NanoSPECT-NSO with an additional 1 s for changing the bed position. With the U-SPECT-II, 17 bed translations (3 min 32 s per position + 36 s total overhead due to bed travel and detector initialization) with overlapping FOVs were automatically performed.

Besides this qualitative evaluation of the reconstructed spatial resolution, we also measured the full-width at half-maximum (FWHM) of two line sources (polyethylene tubing filled with 370 MBq/ml 99mTc) with an inner diameter of 0.28 mm for 1 h with one line source (2.5 cm length) axially oriented and the other (1.5 cm length) transaxially positioned (Fig. 2). With the NanoSPECT-NSO, a spiral scan of three ‘bed positions’ was used. With the U-SPECT-II, 36 bed positions were needed with 1 min 40 s per position (+1 min 12 s). The FWHMs were determined from profiles taken over several reconstructed cross-sectional image slices, and these values were averaged to obtain one value, which was recorded as the FWHM (± SD). The axial and transaxial resolution was then defined, and the average of these two resolutions was also determined as (transaxial + axial)/2. SPECT integral and differential uniformities were measured for a region containing 75 % of the FOV (CFOV) of uniformly filled cylinders. The uniform phantom was a 20-ml syringe (internal diameter 19 mm) filled with 8 ml 99mTc solution (78 MBq) and was scanned for 2 h. With the NanoSPECT-NSO, a spiral scan of two ‘bed positions’ was used. With the U-SPECT-II, 54 positions (2 min 15 s per position + total overhead of 1 min 35 s) were needed. With the NanoSPECT-NSO, a 5-ml syringe (internal diameter 12 mm) was also additionally scanned because with the 20-ml syringe scan severe artifacts were observed in the NanoSPECT-NSO images. Integral and differential uniformities were then calculated using the NEMA (National Electrical Manufacturers Association) formula [56]:

$$ \mathrm{Uniformity}\,\left( \% \right)=100\times \frac{{\mathrm{Max}\,\mathrm{count}-\mathrm{Min}\,\mathrm{count}}}{{\mathrm{Max}\,\mathrm{count}+\mathrm{Min}\,\mathrm{count}}} $$
(1)
Table 2 The dimensions of the hot rod capillaries and scan durations used for the different isotopes at the same concentration
Full size table
Fig. 2
figure 2

Schematic drawing of the line sources, their positioning and dimensions. The black double-headed arrows show the length of the capillaries. The white part in the black arrows indicates the part used to extract the profiles

Full size image

The integral uniformity indicates the uniformity calculated over the CFOV, whereas the differential uniformity is calculated for all sets of three contiguous pixels separately. The maximum over these sets is then recorded as the differential uniformity [57]. Uniformity measures are strongly affected by the voxel size and image resolution. In order to prevent differences in uniformity between the various systems solely due to the resolution effect, the images were smoothed with a gaussian filter complementing the resolution of each scanner. The gaussian filter kernel widths to result in an equal resolution of 1 mm were 0.92, 0.66 and 0.80 mm FWHM for the U-SPECT-II, the X-SPECT and the NanoSPECT-NSO, respectively.

The average object-to-collimator distance for the U-SPECT-II was 22 mm while the ROR for both the X-SPECT and the NanoSPECT-NSO was 30 mm. However, to encompass the phantom with the three hot rod inserts a ROR of 35 mm was also needed for the X-SPECT. These RORs were the closest possible to each system’s hardware and software.

Ordered subset expectation maximization (OSEM) image reconstruction was used for all systems and also the more specific POSEM (pixel-based subsets [58], [29]) for the U-SPECT-II. For the NanoSPECT-NSO the raw projections were smoothed first (1.25 mm gaussian kernel) to suppress the noise prior to reconstruction. The software-recommended settings were used for the number of iterations and subsets (nine iterations with 16 subsets per iteration for U-SPECT-II, five iterations with 8 subsets per iteration for X-SPECT, and three iterations with 8 subsets per iteration for NanoSPECT-NSO) with the lowest image voxel sizes possible (0.125, 0.25 and 0.13 mm, respectively). An energy window of 20 % was set around 140 keV for all three systems.

GP mode measurements

For the evaluation of the GP mode, collimators with a 1-mm pinhole size were used for all systems; i.e. the 75-multipinhole tube for the MILabs U-SPECT-II, and a 5- and 9-multipinhole plate per head for the X-SPECT and the NanoSPECT-NSO, respectively (Table 1). Note that our definition of GP collimator (i.e. 1-mm pinhole diameter) does not correspond to the names the different vendors use to market their collimators. Therefore, we also included the U-SPECT-II measurements with the 0.6-mm pinhole collimator aperture in the GP mode data. Hence, the collimator of choice for imaging the mouse-sized phantoms in GP mode was also based on the manufacturers’ recommendations. The resolution with these collimators was measured again for 1 h using the two line sources discussed in the previous section and a hot rod phantom with capillary diameters in the range 0.7 mm to 1.5 mm (Table 2) and filled with a 99mTc solution at a concentration of 500 MBq/ml. We used an additional hot rod resolution phantom with smaller capillaries (0.35 to 0.75 mm) for the GP U-SPECT-II 0.6-mm collimator as this setup was able to achieve a higher resolution.

With the NanoSPECT-NSO, a spiral scan with three ‘bed positions’ and a circular scan were used for the line sources and the hot rod phantom, respectively. With the U-SPECT-II, 12 positions (5 min per position + total overhead of 24 s) and 18 positions (3 min 20 s per position + total overhead of 24 s) were needed for the line sources and the hot rod phantom, respectively. Sensitivity (in counts per second per megabecquerel) was measured using a 99mTc point source with known activity (2.96 MBq) positioned in the centre of the FOV and scanned for 1 h. SPECT integral and differential uniformities were measured as described for the HR mode measurements with a syringe of the same size (20 ml). With the NanoSPECT-NSO, a spiral scan of four ‘bed positions’ was used. With the U-SPECT-II, 72 positions (1 min 40 s per position + total overhead time of 2 min 24 s) were needed. We again matched the resolution as for the HR mode uniformity measurements with a gaussian filter kernel of 0.98, 0.8 and 0.9 mm FWHM for the U-SPECT-II, the X-SPECT and the NanoSPECT-NSO, respectively, resulting in a common resolution of 1.2 mm.

To measure contrast recovery, we designed and measured a mouse-sized phantom with five capillaries (Table 3 and Fig. 3) for 20 min. A spiral scan of three ‘bed positions’ was used with the NanoSPECT-NSO. To measure the mouse-sized phantom with the U-SPECT-II system, a total of 63 bed positions, 19 s for each position (+ 2 min 6 s total overhead) were needed. The background (5 MBq/ml) and the four smallest capillaries (20 MBq/ml) were filled with a 99mTc solution to result in a capillary-to-background ratio of 4 to 1. The 2-mm capillary was left unfilled to create a cold region in a hot background. Capillary and background volumes of interest (VOIs) were delineated on the corresponding CT images. The VOIs were repeated in seven 1-mm thick transaxial slices 1.5 mm apart to obtain seven results. The contrast recovery coefficient (CRChot) was then calculated as follows:

$$ \mathrm{CR}{{\mathrm{C}}_{\mathrm{hot}}}=\frac{{\frac{{{{\mathrm{m}}_{\mathrm{hot}}}-{{\mathrm{m}}_{\mathrm{BG}}}}}{{{{\mathrm{m}}_{\mathrm{BG}}}}}}}{{{{\mathrm{C}}_{\mathrm{true}}}-1}} $$
(2)

where mhot and mBG are the mean concentrations measured in capillary and background VOIs averaged over the seven results and Ctrue is the real capillary-to-background ratio.

Table 3 Dimensions of the capillary phantom. Units are millimetres except volume in millilitres
Full size table
Fig. 3
figure 3

a Drawing of the contrast phantom with five capillaries inside surrounded by a hot background. b Photograph of the capillaries and the rings. For dimensions, see Table 3

Full size image

The cold-to-background ratio (CBRair) was defined as the activity measured in the cold region (mcold) divided by the mean of the background concentration (mBG), which we represent as CRCcold:

$$ \begin{array}{*{20}c} {\mathrm{C}\mathrm{B}{{\mathrm{R}}_{\mathrm{air}}}=\frac{{{{\mathrm{m}}_{\mathrm{cold}}}}}{{{{\mathrm{m}}_{\mathrm{BG}}}}}} \hfill \\ {\mathrm{C}\mathrm{R}{{\mathrm{C}}_{\mathrm{cold}}}=1-\mathrm{CB}{{\mathrm{R}}_{\mathrm{air}}}} \hfill \\ \end{array} $$
(3)

The images were repeatedly smoothed with a 0.3-mm gaussian filter to obtain the contrast recovery at different levels of background standard deviation. This noise coefficient (NC) was calculated as follows [59]:

$$ \mathrm{NC}\left( \% \right)=100\times \frac{1}{\mathrm{P}}\sum\limits_{\mathrm{p}}^{\mathrm{p}} {\frac{{{\sigma_{\mathrm{p}}}}}{{{{\mathrm{m}}_{\mathrm{p}}}}}} $$
(4)

where P is the total number of pixels in the background VOI and for each background pixel p, σ p is the standard deviation and m p the mean calculated from the seven slices. The average object-to-collimator distance in these GP measurements for the U-SPECT-II was 22 mm while the RORs for the X-SPECT and the NanoSPECT were 35 mm and 30 mm, respectively. However, for the uniform cylinder a ROR of 45 mm was needed with the X-SPECT. The same number of iterations and subsets were used as in the HR measurements, and the software selected voxel sizes for the 1-mm collimators for the U-SPECT-II, X-SPECT and NanoSPECT-NSO were 0.2, 0.5 and 0.2 mm, respectively.

Other isotopes

Besides 99mTc, we also use 111In and 125I in our experiments. Different scan times were set to have the same number of decays (Table 2). For these extra isotopes we measured GP spatial resolution (line source and hot rod phantom) and uniformity as described in the previous paragraph. As with 99mTc, OSEM reconstruction was applied with a 20 % energy window set around the main peaks for 111In and a 100 % window around the 27 keV peak for 125I.

Results

System measurements

Spatial resolution – HR collimators

Figure 4 shows the mouse-sized phantom with the three hot rod inserts measured with the HR apertures and shows qualitatively the spatial resolution in the entire FOV. The U-SPECT-II was able to resolve rods as small as 0.4 mm (Fig. 4b), the X-SPECT (Fig. 4c) was able to resolve rods of 0.75 mm, and the NanoSPECT-NSO was able to resolve rods as small as 0.6 mm, although with the NanoSPECT-NSO the transaxial FOV was only 20 mm leaving one hot rod phantom truncated (Fig. 4d). The resolutions with the line sources in the centre of the FOV (Table 4) and with these HR apertures (average of axial and transaxial) were as small as 0.38 mm with the U-SPECT-II, 0.49 mm with the X-SPECT and 0.66 mm with the NanoSPECT-NSO.

Fig. 4
figure 4

Mouse-sized phantom with three hot rod inserts (capillaries 0.35, 0.4, 0.45, 0.5, 0.6 and 0.75 mm) measured with the three systems using their HR collimators: a the phantom, b U-SPECT-II 0.35-mm collimator, c X-SPECT 0.5-mm collimator, and d NanoSPECT-NSO 0.6-mm collimator

Full size image
Table 4 Spatial resolutions: line sources
Full size table

Spatial resolution – GP collimators

Figure 5 shows the hot rod phantoms measured with the U-SPECT-II 0.6-mm apertures and with the 1-mm apertures of all the scanners. The U-SPECT-II 0.6-mm collimator resolved rods of 0.45 mm with 99mTc and between 0.5 and 0.6 mm with 111In and 125I. The FWHM of the line sources acquired with this collimator (Table 4) gave a quantitative result for the spatial resolution. The average resolutions were 0.63, 0.71 and 0.66 mm for 99mTc, 111In and 125I, respectively. For the 1-mm collimators and 99mTc, the U-SPECT-II resolved rods as small as 0.7 mm, the X-SPECT resolved rods of 0.9 mm, and the NanoSPECT-NSO resolved rods of 0.8 mm. The average resolutions for the 99mTc line sources were 0.76 mm for the U-SPECT-II, 0.58 mm for the X-SPECT, and 0.69 mm for the NanoSPECT-NSO. For 111In, the U-SPECT-II resolved rods of 0.7 to 0.8, and the X-SPECT and the NanoSPECT-NSO resolved rods of 0.9 to 1 mm. The average resolutions of the 111In line sources for the U-SPECT-II were 0.85 mm, for the X-SPECT 0.80 mm, and for the NanoSPECT-NSO 0.78 mm. For 125I, the U-SPECT-II resolved rods from 0.8 to 0.9 mm, and the X-SPECT and the NanoSPECT-NSO resolved rods of 0.9 mm. The average resolutions of the line sources for the U-SPECT-II were 0.79 mm, for the X-SPECT 0.68 mm, and for the NanoSPECT-NSO 0.92 mm.

Fig. 5
figure 5

a GP scans of the hot rod phantom with 0.35, 0.4, 0.45, 0.5, 0.6, 0.75-mm capillaries with the 0.6-mm pinhole collimator of the U-SPECT-II (for 125I the hot rod phantom with 0.25, 0.3, 0.35, 0.4, 0.5, 0.6-mm capillaries was scanned). b–d GP scans of the hot rod phantom with 0.7, 0.8, 0.9, 1.0, 1.2, 1.5-mm capillaries with the 1-mm pinhole collimator of the U-SPECT-II, X-SPECT and NanoSPECT-NSO, respectively (500 MBq/ml, scan time 1 h)

Full size image

Sensitivity

The point source sensitivities measured with 99mTc and the 1-mm pinhole apertures for the U-SPECT-II was 3,984 cps/MBq or 0.39 % and 1,500 cps/MBq or 0.15 % for the 0.6-mm collimator, for the X-SPECT 620 cps/MBq or 0.06 % (= 4 × 155 cps/MBq of the one-head system), and for the NanoSPECT 751 cps/MBq or 0.07 %.

Uniformity

The HR/GP integral and differential uniformities are summarized in Table 5. The HR integral and differential uniformities measured with 99mTc were 31 %/15 % for the U-SPECT-II, 56 %/38 % for the X-SPECT and 93 %/64 % for the NanoSPECT-NSO. Since the 20-ml syringe with the NanoSPECT-NSO produced severe artifacts, a 5-ml syringe was also scanned and resulted in better uniformities of 33 %/21 %. The GP 99mTc uniformities were similar to the HR values. The GP uniformity measured with the higher energy isotope 111In was only slightly different while the uniformity values with 125I were even worse (Table 5).

Table 5 GP integral and differential uniformities for the HR and GP collimators (a lower number represents better uniformity)
Full size table

Contrast recovery and cold-to-background ratio

Figure 6 shows the CRChot and CRCcold curves for the different capillary diameters and scanners as a function of the standard deviation of the background. At a noise level of 8 %, the U-SPECT-II achieved a CRC ranging from 0.05 for the 0.6-mm rod to 0.42 for the 1.5-mm rod. For this largest rod, the U-SPECT-II achieved a CRC of 0.34 at a noise level of 5 % and 0.45 at a noise level of 10 %. At this latter noise level the CRCcold was 0.70 for the U-SPECT-II. At a noise level of 8 %, the X-SPECT achieved a CRC ranging from 0.07 for the 0.6-mm rod to 0.37 for the 1.5-mm rod. For this largest rod, the X-SPECT achieved a CRC of 0.26 at a noise level of 5 % and 0.42 at a noise level of 10 %. The CRCcold at the 10 % noise level was 0.52. Finally, at a noise level of 8 %, the NanoSPECT-NSO achieved a CRC ranging from 0.09 for the 0.6-mm rod to 0.34 for the 1.5-mm rod. For this 1.5 mm rod, the NanoSPECT-NSO achieved a CRC of 0.27 at a noise level of 5 % and 0.36 at a noise level of 10 %. At this 10 % noise level, the CRCcold was 0.42 for the NanoSPECT-NSO. Cross-sections of the phantom are shown in Fig. 7 at a noise level of 10 %.

Fig. 6
figure 6

a–d CRChot curves for the different capillary diameters as a function of the standard deviation of the background. e CRCcold curves for the 2-mm cold capillary (only air) as a function of the standard deviation of the background

Full size image
Fig. 7
figure 7

a Photograph of the phantom. The white arrows indicate the 1.5-mm and 1-mm capillaries. The capillaries are filled with a red colouring liquid for visualization, except for the 2-mm capillary. b–d Transverse and sagittal cross-sections of the contrast phantom with 10 % background standard deviation measured with (b) the U-SPECT-II, (c) the X-SPECT and (d) the NanoSPECT-NSO

Full size image

Discussion

The performance of three state-of-the-art multipinhole μSPECT systems, all configured for mouse imaging, was evaluated. Although each system has been evaluated previously [37, 47, 51–53], there was no comparative study testing the systems with the same set of standardized experiments. We report here on the performance metrics for the objective characterization of these widely used μSPECT systems. We kept all acquisition parameters for all measurements as equal as possible between the scanners. The reconstruction parameters were chosen each time based on the recommended parameters of the software and the vendors’ experience with the systems.

Besides measuring the basic characteristics of these systems, the size of the FOV (and information on spiral pitch and number of bed positions needed) is an equally important aspect when using these μSPECT systems for molecular imaging in daily preclinical routine. These parameters will determine the injected dose and scan time, and may put the sensitivity versus resolution trade-off in another light. As discussed by Harteveld et al. in 2011 [60], a standard for μSPECT is still lacking, and they evaluated a μSPECT scanner (U-SPECT-II) using the NEMA NU 4 μPET phantom. This NEMA NU4 μPET IQ phantom has several disadvantages when used to evaluate μSPECT systems: (1) the diameters of the hot rods range from 1 to 5 mm which is above the current state-of-the-art μSPECT achievable submillimetre resolution, (2) the 30-mm diameter of the phantom does not allow μSPECT multipinhole scanners to be used in HR mode (small ROR and high magnification), and (3) the phantom does not offer the possibility of having hot rods on a warm background. To address some of these issues, Visser et al. [61] developed an alternative phantom dedicated to image quality evaluation with μSPECT having a smaller outer diameter of 23.45 mm and hot rods down to 0.35 mm diameter which is ideally suited to current μSPECT systems. These authors considered that having hot rods in a warm background would be more realistic. Their cold background configuration was based on the physical limitations that prevent the production of hot spheres with physical walls smaller than the spatial resolution [62]. As a solution to this limitation we constructed a μSPECT IQ phantom using ultrathin round borosilicate capillaries (internal diameter 0.6, 0.8, 1, 1.5 and 2 mm) aligned using two multijet-modelled (Shapeways, Eindhoven) rings (acrylic plastic, 1.8 × 1.8 × 1.0 cm). Our custom-made contrast phantom thus has a warm background and the capillary wall thickness ranged only between 0.2 mm for the smallest and 0.4 mm for the biggest capillary, which is smaller than half of the spatial resolution, measured with the 1-mm GP collimators (Table 4). An additional feature of our phantom is the longer capillaries (20 mm versus 6.5 mm in the IQ phantom) so that we can measure contrast/resolution over more slices and with a more realistic FOV size. This serves also as an indirect measure for the sensitivity over a larger FOV with many bed positions needed for the U-SPECT-II. Also the capillary diameters are optimized to evaluate the different scanners equipped with high-sensitivity 1-mm collimators for mice (rods ranging from 0.6 to 2.0 mm). This phantom may also be used to characterize different collimators and acquisition and reconstruction parameters, and the effect of scatter and attenuation correction.

When all systems were equipped with their HR collimators, the U-SPECT-II obtained the highest spatial resolution over the entire FOV for the multi-hot rod phantom as well as for the line sources. This can be attributed to the smaller diameter of the pinholes and a larger number of pinholes, the higher pinhole magnification factor and the fact that the pinholes are on average closer to the object. This larger multi-hot rod phantom was truncated with the NanoSPECT-NSO due to the smaller transaxial FOV of the 0.6-mm collimators of this system. When using the GP collimators (1-mm aperture, and also the 0.6-mm aperture for the U-SPECT-II) to increase the sensitivity, the highest overall resolution was also obtained with the U-SPECT-II based on the assessment of a hot rod phantom (0.7, 0.8, 0.9, 1.0, 1.2, 1.5 mm) placed in the centre of the FOV. This was the case for all isotopes imaged. Due to the higher energy of the 111In-emitted photons, the reconstructed resolution using this isotope was lower than for 99mTc as a result of increasing collimator scatter and penetration which was observed for all three systems tested. Similarly, the low energy of 125I resulted in more object scatter and poorer detector resolution. However, for all three systems tested, the image quality obtained remained high, even with these more challenging isotopes.

When additionally evaluating the spatial resolution with the line source measurements for these GP collimators, the X-SPECT and NanoSPECT-NSO provided a higher reconstructed resolution, which contradicted the apparent resolution of the hot rod phantom images. The use of the crossed capillaries for resolution measurements has often been debated when iterative reconstruction with resolution recovery is used [63–65]. A possible explanation is, as discussed by Mok et al. [32], that the sensitivity gained from multiplexing may result in a better resolution for sparse objects such as the line sources. When larger, nonsparse, objects such as the hot rod and the uniformity are scanned such resolution gains are lost due to the increased amount of overlap in the projections [32, 66]. The axial resolution of the NanoSPECT-NSO was inferior to its transaxial FWHM. A possible explanation could be that the pitch of the spiral SPECT acquisition mode slightly deteriorates the axial resolution or that the overall resolution is degraded more at the edge of the FOV for the NanoSPECT-NSO. Such an off-centre resolution degradation can also be seen in Fig. 4d: in the centre of the FOV rods as small as 0.6 mm are resolved while at the edge only the 0.75-mm rods are resolved.

Point source sensitivity was measured for all systems with the 1-mm collimators resulting in the highest sensitivity for the U-SPECT-II. This was mainly due to more (75) pinholes, and their strong focus to the same area in the FOV. As a result of this arrangement a smaller FOV was covered in one bed position and this point source measurement with the U-SPECT-II only needed a single position. The lower sensitivity of the X-SPECT (5 pinholes) and the NanoSPECT (9 pinholes) can be explained as these systems have fewer pinholes and also the NanoSPECT-NSO pinholes do not focus on the same area in the FOV, and therefore span a larger FOV. In addition, when measuring a point source, the X-SPECT and the NanoSPECT-NSO do not benefit, in terms of sensitivity, from their multiplexing capabilities as the point source projections do not overlap. To resolve the difficulties in comparing and interpreting point source sensitivity values of the different systems as a result of the diverse approaches (strong focusing pinholes, multiplexing, different FOV) amongst them, a CRC study was performed. Alternatively volume sensitivity could be considered for which the sensitivity of the NanoSPECT-NSO and the X-SPECT systems would benefit from multiplexing. However as less spatial information is available for photons when projections overlap the gain sensitivity gain is in a sense artificial. In addition, the sensitivity gained from multiplexing will depend on the activity distribution [32].

As can be concluded from Fig. 6, the U-SPECT-II maintains its higher sensitivity and resolution for a full-size FOV when acquiring this contrast recovery μSPECT image quality phantom, especially for the 1.5-mm and the 1.0 mm capillaries. For the 0.8-mm hot rods the performance of the U-SPECT-II and the NanoSPECT-NSO were comparable and the 0.6-mm capillaries were not resolved by any of the three systems when using the GP 1-mm collimator. In terms of the CRCcold, the U-SPECT-II clearly achieved better performance than the other two systems. The better CRChot and CRCcold values for the U-SPECT-II may be attributed to the lack of multiplexing and/or higher resolution obtainable over the entire phantom with the scanning focusing method [35].

The U-SPECT-II provided the best uniformity values (in both HR and GP mode). The X-SPECT values were higher (i.e. less uniform) but not significantly so than the U-SPECT-II values (Mann-Whitney test, p = 0.148). No artifacts were seen in the uniformity images (Fig. 8) for the U-SPECT-II or the X-SPECT. On the contrary, severe artifacts were seen on the NanoSPECT-NSO HR uniformity images which were probably a consequence of too much multiplexing (i.e. true overlap) [33] making the syringe look smaller (Fig. 8) with patterns inside. Replacing the 20-ml syringe with a 5-ml syringe resulted in normalization of the NanoSPECT-NSO HR image uniformity values (Table 5). In many small-animal imaging studies some of the activity is usually clustered in small volumes, which may render these artifacts less severe but may still result in erroneous quantification.

Fig. 8
figure 8

Transaxial and axial HR uniformity images measured with (a) the U-SPECT-II (b) the X-SPECT and (c) the NanoSPECT-NSO. Transaxial profiles (x and y directions through the centre) and axial profiles (in the centre and 3 mm from the syringe edge) are shown

Full size image

To make the performance comparison between the μSPECT scanners more complete we also refer to the results obtained by Magota et al. [67] and Boisson et al. [68] with the Inveon (dual head) SPECT system. In contrast to our study involving multipinhole collimators, Magota et al. performed an evaluation with a single-pinhole collimator. The 99mTc spatial resolution was measured with an ultramicro hot spot phantom (0.75, 1.00, 1.35, 1.7, 2.00 and 2.4 mm; Data Spectrum Corporation) filled with a concentration of 15.9 MBq/ml for 32 min. The phantom resolutions obtained with the 0.5-mm and 1-mm pinhole collimators (ROR 25 mm) were 0.84 mm and 1.20 mm, respectively. Boisson et al. obtained a spatial resolution of 1.0 mm by measuring line sources with the 1-mm multipinhole (five pinholes, mouse WB) collimators rotating at 30 mm ROR. Both studies showed resolutions (both in HR and GP mode) that were inferior to the obtained resolutions in this study. A cylindrical phantom (internal diameter 2.5 cm, length 9 cm) filled with 117 MBq of 99mTc (2.6 MBq/ml) was scanned by Magota et al. for 48 min with a ROR of 25 mm to obtain both volume sensitivity and integral uniformity. The sensitivity was 76 cps/MBq and the integral uniformity was 37 %. Sensitivity was evidently low for this single-pinhole system. The integral uniformity was comparable to that obtained with the scanners in our study. Boisson et al. obtained a system sensitivity of 403.6 cps/MBq with the 1-mm mouse WB collimators which is inferior to the values obtained in our study with the other three systems. One has to bear in mind that Magota et al. and Boisson et al. obtained their results with a different set of phantoms rendering direct comparison difficult.

Besides the system hardware, imaging performance also depends on maintenance, quality control, calibrations and environment (e.g. temperature). The CZT detectors in the X-SPECT, for example, may be prone to minute impurities associated with LE spectral tailing, pixel dropouts, hot spots and nonuniform response. Mechanical calibration is mandatory for rotating systems (X-SPECT and NanoSPECT-NSO) to accurately define the geometry of a detector rotating in a circular orbit. The more the system is reconfigured for different tasks (changing collimators, altering ROR), the more frequently will recalibration be required.

Future work will include comparisons using collimators designed for imaging rat-sized objects and in vivo experiments to complement our phantom-based findings. Also, the NEMA has now assembled a task force that has started designing a standard phantom for image quality evaluations in μSPECT.

All three scanners have made a large progress compared to the early human SPECTcameras that were modified for small animal imaging by using pinhole collimatorplates [8–10, 12, 13, 69] and evolve continuously through hardware upgrades such as the VECTor for Milabs and the NanoSPECT-Plus or NanoSPECT-II for Bioscan. The μSPECT system of choice depends on the applications for which it is used, the need for performance or the values placed on flexibility and cost.

Conclusion

We compared three state-of-the-art μSPECT systems based on image quality parameters including spatial resolution, reconstructed image uniformity, point source sensitivity and contrast recovery. To evaluate contrast recovery we designed and built a contrast-to-noise phantom which, to the best of our knowledge for the first time, provided the ability to measure hot contrast on a warm background in the submillimetre resolution range. We believe our evaluation realistically reflected the potential of each system to acquire mouse-sized objects.