A High-Performance Self-Collimation SPECT for Small Animal Imaging

Abstract

Stemmed from our novel single-photon imaging concept of detector self-collimation—which leverages detectors themselves as collimators to overcome the inherent resolution-sensitivity trade-off in conventional SPECT—this study presents the design and evaluation of the first full-ring self-collimation SPECT (SC-SPECT) scanner for small animal imaging. The system features four concentric detector rings and two interchangeable high-aperture-ratio tungsten collimator rings optimized for high-resolution (HR) and general-purpose (GP) imaging applications. Detector rings contain 480, 720, 960, and 1,200 evenly distributed GAGG(Ce) scintillators, each measuring 0.84 mm (tangential) × 6 mm (radial) × 20 mm (axial) and separated by 0.84-mm gaps to enable effective photon collimation. Inner detector rings and the collimator ring collectively provide collimation for photons reaching subsequent outer rings. Dual-end SiPM readouts facilitate axial depth-of-interaction measurements. Phantom and mouse studies are performed to assess the system’s resolution, sensitivity, and field-of-view volume, and SC-SPECT demonstrates generally superior performance compared with state-of-the-art small-animal SPECT systems. Mouse bone images using ^99mTc-MDP show CT-like resolution, clearly delineating detailed tracer uptake distributions within small structures such as mouse paws and skulls, indicating a significant technological advancement in small-animal SPECT imaging.

I. Introduction

Single photon emission computed tomography (SPECT) is an essential molecular imaging technique that facilitates the visualization and measurement of biological processes in vivo and non-invasively with radioisotope-labeled tracers [1–3]. By imaging murine models, it helps elucidate molecular interactions crucial to the onset and progression of disease, as well as monitor the kinetics and therapeutic effectiveness of pharmaceuticals, thereby playing an important role in fundamental medical research and drug development [4–6].

In small animal imaging (SAI), achieving high spatial resolution is critical due to the small size of in vivo tissues and organs. Early attempts at SPECT animal scanning involved the use of rotating gamma cameras equipped with parallel-hole collimators, which provided a spatial resolution of approximately 15 mm [7, 8]. While this setup was suitable for imaging larger animals (e.g., dogs), it proved insufficient for SAI applications. Although dedicated parallel-hole SAI SPECT allowed for sub-millimeter imaging within a smaller field of view (FOV), its sensitivity was extremely low (~0.002%) [9]. Subsequent advancements incorporated single-pinhole collimators attached to rotating gamma cameras [10–12], providing reasonable SAI images with a resolution of approximately 2 mm. However, these systems still required long acquisition times due to their relatively low sensitivity (~0.01%). The introduction of dedicated multi-pinhole collimators in the early 21^th century further enhanced sensitivity, which, in turn, facilitated sub-millimeter resolution by enabling the use of smaller pinholes [13–15].

Up to now, multi-pinhole collimation remains the dominant collimation strategy in SPECT SAI. For instance, the nanoScan SPECT/CT system (Mediso, Hungary) with four rotating detector heads, each featuring a NaI(Tl) detector and a 9-pinhole collimator, resolves 0.3-mm-diameter hot rods with 0.031% central sensitivity (MU collimator, 0.5-mm-diameter pinholes, FOV: 16 mm (diameter) × 14 mm (length)) and 0.7-mm-diameter hot rods with 0.082% central sensitivity (MH collimator, 1.0-mm-diameter pinholes, FOV: 30 mm (diameter) × 14 mm (length)) [16, 17]. The U-SPECT⁺ system (MILabs, Netherlands), containing three stationary NaI(Tl) detectors and a 75-pinhole collimator, resolves 0.25-mm-diameter hot rods with 0.038% central sensitivity (XUHR-M collimator, 0.25-mm-diameter pinholes, FOV: 12 mm (diameter) × 7 mm (length)) and 0.4-mm-diameter hot rods with 0.171% central sensitivity (GP-M collimator, 0.6-mm-diameter pinholes, FOV: 12 mm (diameter) × 7 mm (length)) [18, 19], with the former achieving state-of-the-art SPECT resolution, to the best of our knowledge.

The imaging performance of current SPECT systems remains suboptimal, primarily due to the inherent trade-off between resolution and sensitivity imposed by the mechanical collimator [20]. As shown in Fig. 1a, mechanical collimator restricts photons to specific paths, enabling detectors to capture photon directional information necessary for image reconstruction [21, 22]. However, this collimation strategy results in a significant absorption of photons at the collimator, leading to a considerable reduction in sensitivity. In the absence of viable alternatives, mechanical collimator has remained a crucial component of SPECT systems for generations [23], posing a fundamental constraint on SPECT imaging.

Fig. 1.

Comparison of conventional mechanical collimation and novel self-collimation. (a) Mechanical collimation: a mechanical collimator restricts photons to specific paths by absorbing those traveling in other directions. The choice of the aperture size on the collimator determines the trade-off of sensitivity and resolution. (b) Self-collimation: active detectors replace part of the mechanical collimators to guide the photons’ trajectories toward the subsequent layers. Resolution is not solely determined by the aperture size.

To achieve a significant breakthrough, we recently introduced a novel single-photon imaging approach, self-collimation (SC), in which active detectors serve dual functions: photon detection and collimation [24]. As shown in Fig. 1b, we employ a multi-layer sparse-scintillator detector architecture to implement SC approach, where each layer consists of scintillators sparsely distributed with air gaps [25]. In this configuration, the front layers not only detect incoming photons but also act as collimators for the layers behind them. The detected photon events from all detector layers are treated equally and jointly used for image reconstruction. To further enhance photon collimation—particularly for the foremost detector layer—a high-aperture-ratio metal plate is positioned between the imaging target and the detector. This hybrid collimation strategy, which integrates the multi-layer sparse-scintillator detectors with the high-aperture-ratio metal plate (offering a photon transmission ratio an order of magnitude higher than that of conventional mechanical collimators), results in a high-resolution system response [25]. By incorporating enhanced sampling, specifically object movements during the imaging process, we have demonstrated—via Monte Carlo simulations and prototype experiments—that this design effectively improves both resolution and sensitivity simultaneously [24, 25].

Based on the SC imaging approach, we have designed a SC-SPECT system for SAI with its geometry thoroughly optimized using Cramér-Rao Lower Bound (CRLB) theory [26, 27], demonstrating promising imaging performance with Monte Carlo simulations [25, 28]. Additionally, we have found that randomly arranging apertures, rather than distributing them evenly, on the high-aperture-ratio metal plate can effectively enhance the signal-to-noise ratio of the reconstructed image [29, 30].

Building on our previous work, we have constructed a full-ring SC-SPECT system for high-resolution SAI. The system consists of four concentric sparse-scintillator detector rings and a high-aperture-ratio metal ring featuring a randomized aperture arrangement. Compared with our earlier SC-SPECT prototype [24]—which comprises seven detector modules aligned toward the FOV and exhibits large gaps between adjacent modules—this full-ring configuration provides more complete angular sampling and improved sensitivity. As an initial step, we implement a single-axial-ring configuration with an axial length of 20 mm, targeting a FOV comparable to that of commercial SAI SPECT systems [16–19] and allowing for further improvements in sensitivity and FOV extension by stacking multiple rings along the axial direction.

In this work, we provide a comprehensive description of the system’s overall structure, detectors, electronics, and the techniques employed for signal readout and system calibration. We conduct phantom studies and animal scans to assess the system’s sensitivity, resolution, and animal imaging performance.

II. Methods

A. System structure

The small animal SC-SPECT system (Fig. 2) consists of four concentric detector rings and a high-aperture-ratio metal ring. The four detector rings, denoted as Ring-I, Ring-II, Ring-III, and Ring-IV, have inner diameters of 260 mm, 390 mm, 520 mm, and 650 mm, respectively. These four rings are constructed with 40 detector cassettes, each covering a 9° angular region.

Fig. 2.

(a) Photographs of the small animal SC-SPECT system. (b) A schematic diagram and a cross-sectional view of the four detector rings.

As shown in Fig. 3a, each detector cassette comprises four detector blocks, with each block forming 1/40 of the corresponding ring. The four detector blocks are composed of 6, 9, 12, and 15 detector cells, respectively. Each detector cell (Fig. 3b) contains two GAGG(Ce) (Gadolinium Aluminum Gallium Garnet doped with Cerium) scintillators, measuring 0.84 mm (tangential) × 6 mm (radial) × 20 mm (axial), separated by a 0.84 mm gap in the tangential direction. GAGG(Ce), as scintillator for SPECT detector, possesses many desired properties: high density (6.6 g/cm³), high atomic number (Z=54.4), high light yield (50,000 ~ 54,000 photons/MeV), no intrinsic background radiation, and cost-effective.

Fig. 3.

(a) A schematic diagram and a photograph of the detector cassette. The coordinate system follows the same convention as in Fig. 2B. (b) A schematic diagram of the detector cell. (C) The CovB and DPB used for signal readout.

In each detector cell, the two scintillators, labeled GAGG-90 and GAGG-200, have decay time constants of 90 ns and 200 ns, respectively, enabling crystal identification through the pulse shape discrimination (PSD) method [31, 32]. The two scintillators are optically coupled to four SiPMs (ONSEMI FJ30035, USA) – two on each axial end, and then connected to two front-end boards (FEBs). This dual-end readout configuration facilitates the determination of photon’s depth-of-interaction (DOI) along the axial direction [33–35]. Each scintillator is divided into 10 DOI bins, each 2 mm in length. Consequently, the total number of projection bins is given by:

$$N_{\mathit{\text{Proj}}}=40\left(\text{detector}\text{cassette}\right)\times 42\left(\text{detector}\text{cell}\right)\times 2\left(\text{scintillator}\right)\times 10\left(\text{DOI}\text{bin}\right)=33,600$$ (1)

To further enhance photon collimation, especially for the innermost Ring-I, we design two high-aperture-ratio metal rings aimed for high-resolution (HR) and general-purpose (GP) SAI, respectively. The HR metal ring is made of tungsten, with an inner diameter of 67.5 mm, an axial length of 290 mm, and a thickness of 2.5 mm. It incorporates 180 circular apertures, each with a diameter of 0.4 mm. These apertures are randomly distributed across the central transverse plane, with a minimum center-to-center spacing of 0.8 mm to ensure structural integrity. The intended FOV is a 10 mm (diameter) × 6 mm (length) cylindrical volume.

The GP metal ring, also made of tungsten, has an inner diameter of 117.5 mm, an axial length of 290 mm, and a thickness of 2.5 mm. It contains 180 circular apertures, each 0.7 mm in diameter, randomly distributed across the central transverse plane with a minimum center-to-center spacing of 1.1 mm. The intended FOV is a 24 mm (diameter) × 10 mm (length) cylindrical volume.

The aperture ratio—defined as the ratio of the total aperture area to the collimation-effective metal ring area—is significantly higher for both the HR and GP metal rings compared to conventional multi-pinhole collimators, where the collimation-effective metal ring area is defined as the minimum area on the metal ring required to block all photons emitted from the FOV that would otherwise reach the detectors without passing through the intended aperture. In our system, each metal ring incorporates 180 apertures, resulting in aperture ratios of 1.0% and 1.2% for the HR and GP rings, respectively. For comparison, the U-SPECT⁺ system [18, 19] utilizes five aperture rings, each comprising 15 apertures, yielding aperture ratios of 0.029% and 0.16% for the XUHR-M and GP-M collimators, respectively.

B. Readout electronics

For each of the 40 detector cassettes, the 168 SiPM signals (42 detector cells × 4 SiPMs) are amplified and subsequently multiplexed to eight position signals, namely $X_1^{+},X_1^{-},Y_1^{+},Y_1^{-},X_2^{+},X_2^{-},Y_2^{+}$, and $Y_2^{-}$ on a pair of FEBs, where $X_i^{+},X_i^{-},Y_i^{+},$ and $Y_i^{-}(i=\mathrm{1,2}$) are the four position signals from the SiPMs coupled to one end of the 42 detector cells. Four pairs of FEBs are then connected to a conversion board (CovB), as shown in Fig. 3c, where the Anger position signals ($X$ and $Y$) and dual-end energy signals ($E_1$ and $E_2$) are calculated by Anger logic [34]:

$$X=\frac{X_1^{+}+X_2^{+}-X_1^{-}-X_2^{-}}{X_1^{+}+X_2^{+}+X_1^{-}+X_2^{-}},Y=\frac{Y_1^{+}+Y_2^{+}-Y_1^{-}-Y_2^{-}}{Y_1^{+}+Y_2^{+}+Y_1^{-}+Y_2^{-}}$$ (2)

$$E_1=X_1^{+}+X_1^{-}+Y_1^{+}+Y_1^{-},E_2=X_2^{+}+X_2^{-}+Y_2^{+}+Y_2^{-}$$ (3)

Each pair of CovBs receives signals from eight detector cassettes, which are then connected to a digital processing board (DPB). The DPB digitizes the waveform signals and calculates the PSD factor ($r_{\mathit{\text{PSD}}}$), a peak-width metric used to differentiate between GAGG-90 and GAGG-200 scintillators [31, 32]:

$$r_{\mathit{\text{PSD}}}=\frac{\int E\left(t\right)\mathit{\text{dt}}}{E\left(t_{\mathit{\text{peak}}}\right)},E\left(t\right)=E_1\left(t\right)+E_2\left(t\right)$$ (4)

where $t_{\mathit{\text{peak}}}$ is the peak time of the energy signal, and $\int E\left(t\right)\mathit{\text{dt}}$ is the integral of the energy signal. All the five DPBs are connected to a switch, which transmits the output signals to a computer for further processing. In the uploaded raw data, each single-photon event is represented by {${\mathit{\text{ID}}}_{\mathit{\text{cassette}}},X,Y,E_1,E_2,r_{\mathit{\text{PSD}}}$}, where ${ID}_{\mathit{\text{cassette}}}$ is the detector cassette index.

A. Detector calibration and data processing

To identify the projection bin index ${\mathit{\text{ID}}}_{\mathit{\text{Proj}}}$ of each single-photon event in the raw data, we calibrate the 40 detector cassettes using a ring phantom. The phantom has an inner diameter of 90 mm, an outer diameter of 120 mm, and an axial length of 20 mm. It is filled with Na^99mTcO4 solution with an activity of 5 μCi and scanned for 10 minutes, with the metal ring removed during the calibration process. The count rate is approximately 1 × 10⁴ cps. The acquired raw data is then used for detector calibration with the following steps:

1). Detector cell identification

We generate the flood histogram of detector cells for each of the 40 detector cassettes to identify the cell where photon interactions occur.

2). Scintillator identification

We then generate the PSD factor histogram for each detector cell to identify the scintillator (GAGG-90 or GAGG-200) the photons interact with.

3). DOI bin identification

We then generate the DOI ratio histogram for each scintillator to determine the DOI bin in which the photon interaction occurs, where the DOI ratio is calculated by:

$${\alpha }_{\mathit{\text{DOI}}}=\frac{E_1-E_2}{E_1+E_2}$$ (5)

Using the flood histogram, PSD factor histogram, and DOI ratio histogram, the detector cell index ${\mathit{\text{ID}}}_{\mathit{\text{cell}}}$, scintillator index ${\mathit{\text{ID}}}_{\mathit{\text{scin}}}$, and DOI bin index ${\mathit{\text{ID}}}_{\mathit{\text{DOI}}}$ for each event can be determined sequentially, as detailed in Section III.A along with the corresponding results.

The projection bin index is calculated as:

$${\mathit{\text{ID}}}_{\mathit{\text{Proj}}}=42\times 2\times 10\times {\mathit{\text{ID}}}_{\mathit{\text{cassette}}}+2\times 10\times {\mathit{\text{ID}}}_{\mathit{\text{cell}}}+10\times {\mathit{\text{ID}}}_{\mathit{\text{scin}}}+{\mathit{\text{ID}}}_{\mathit{\text{DOI}}}$$ (6)

The energy of an event is calculated as:

$$E=E_1+E_2$$ (7)

As a result, the list-mode event dataset can be generated from the raw data, with each event represented as {${\mathit{\text{ID}}}_{\mathit{\text{Proj}}},E$}. The list-mode event dataset is then reduced to projection data after applying an energy window of 120 ~ 160 keV, which is used for image reconstruction.

B. Image reconstruction

We perform image reconstruction using the ordered-subset expectation-maximization (OS-EM) algorithm [36]. The system matrix is derived from experimental measurements using 0.36-mm-diameter ^99mTc point sources, each with an activity of 4.0 mCi [37]. The point source is translated across the FOV in a crisscross pattern with uniform step sizes of 0.2 mm and 0.3 mm for the HR and GP configurations, respectively. Data acquisition is performed over five days for the HR configuration and ten days for the GP configuration, with a new point source employed each day. For both configurations, the system matrix is generated by B-spline interpolation. The voxel size and reconstruction parameters are summarized in Table 1, with the iteration number empirically determined for varying scan durations.

Table 1.

Parameters used for image reconstruction.

Parameters	HR	GP
Voxel size	0.05 mm	0.133 mm
Subset number	4	4
Iteration number	500 – 2000	500 – 2000
Post-filter	Gaussian (0.16 mm FWHM)	Gaussian (0.2 mm FWHM)

C. Phantom Studies

We measure the system sensitivity map for both GP and HR metal ring by scanning a point source placed across the FOV in a crisscross pattern, using a step size of 0.5 mm and 0.2 mm, respectively. The diameter of the point source is 0.36 mm, and the activity of the point source is 4.0 mCi. The total measurement time for each metal ring is 12 hours.

We evaluate the spatial resolution using a hot-rod phantom. The phantom is made of plastic and consists of six sections, each containing fillable cylinders with lengths of 1 mm and with diameters of 0.2, 0.25, 0.3, 0.35, 0.4, and 0.45 mm, respectively. The center-to-center distances between the cylinders are twice their diameters. The phantom is filled with 0.18 mCi of Na^99mTcO₄ solution, and is scanned at four positions during the imaging session for sampling enhancement. This procedure, referred to as the T4 protocol, involves scanning at relative positions of (−0.4 mm, −0.4 mm, −0.4 mm), (0.4 mm, 0.4 mm, −0.4 mm), (−0.4 mm, 0.4 mm, 0.4 mm), and (0.4 mm, −0.4 mm, 0.4 mm). The total scan time is 4 hours, and projection data from all positions are combined for image reconstruction. Additionally, projection data of 48-minute and 12-minute scans is extracted and used for image reconstruction. For comparison, the phantom is also scanned for 48 minutes without sampling enhancement (T1 protocol).

To quantitively determine whether a rod group is resolvable, we extract a 1-D profile across the rods and compute the peak-to-valley ratio (PVR), defined as the ratio of the mean intensities of the rod peaks and inter-rod valleys in the profile [38]. A group is deemed distinguishable if PVR > 2.

We evaluate the DOI resolution along the axial direction using a collimated planar photon beam with an energy of 140 keV and a FWHM of 1.5 mm, incident on the scintillators perpendicular to the DOI direction. The beam is translated to five DOI positions (−8 mm, −4 mm, 0 mm, 4 mm, and 8 mm) to assess the resolution at these positions. For each scintillator, the corresponding DOI ratio histograms are generated. The DOI FWHM, denoted as $R_{\mathit{\text{hist}}}$, is then determined by applying a Gaussian fit to the histogram and a linear approximation of the DOI ratios with respect to the five known DOI positions [39].

The DOI resolution is calculated as:

$$R_i=\sqrt{R_{\mathit{\text{hist}}}^2-R_0^2}$$ (8)

where $R_0$ is the FWHM of the planar photon beam. The average DOI resolution across all crystals at a given DOI position is taken as the resolution for that position, and the average DOI resolution across all crystals and all positions is taken as the overall DOI resolution.

Since self-collimation is implemented only in the transverse plane and not along the axial direction in this system, the axial resolution follows that of a conventional pinhole SPECT system. Accordingly, the axial spatial resolution is characterized using the standard pinhole geometrical model [21]:

$$R=\sqrt{{\left(\frac{1}{M}*R_i\right)}^2+{\left(\frac{M+1}{M}*a_{\text{eff}}\right)}^2}$$ (9)

$$M=d_2/d_1$$ (10)

$$a_{\text{eff}}=a_0+\frac{\text{ln}2}{\mu }\left(\text{tan}\frac{\omega }{2}\right)$$ (11)

where $M$ is the magnification number, $R_i$ is the detector’s axial intrinsic resolution (i.e., the DOI resolution in SC-SPECT), $a_{\text{eff}}$ is the resolution-effective pinhole diameter, $d_2$ is the distance from the pinhole to the detector (i.e., the distance from the metal ring to the outermost detector ring in SC-SPECT), $d_1$ is the distance from the pinhole to the center of the FOV, $a_0$ is the physical pinhole diameter, $\mu$ is the attenuation coefficient of the collimator material, and $\omega$ is the opening angle of the pinhole.

The axial resolutions of two commercial SPECT systems (nanoScan and U-SPECT⁺) are also calculated using this analytical model for comparison, with detector intrinsic resolutions of 3.5 mm [16] and 3.2 mm [40], respectively, and the specified geometric parameters [16, 18, 19].

D. FOV-Sensitivity-Resolution Quotient

To enable a comprehensive and quantitative comparison between SAI systems, we introduce a new metric—the FOV-Sensitivity-Resolution Quotient (FSRQ)—which serves as an integrated indicator of overall imaging performance:

$$\text{FSRQ}=\frac{\left(\text{FOV}\text{volume}\right)\times \left(\text{Average}\text{sensitivity}\right)}{\left(\text{Volume}\text{resolution}\right)}$$ (12)

$$\text{Volume}\text{resolution}={\left(r_{\mathit{\text{trans}}}\right)}^2*r_{\mathit{\text{axial}}}$$ (13)

where $r_{\mathit{\text{trans}}}$ is the diameter of the smallest distinguishable hot rod, a value that most SAI systems report, and $r_{\mathit{\text{axial}}}$ is the axial spatial resolution. A higher FSRQ value indicates an overall superior performance considering imaging coverage, sensitivity, and spatial resolution.

E. Animal scans

The study is approved by the Animal Experiments and Experimental Animal Welfare Committee of Capital Medical University, and all animal procedures are carried out in accordance with the guidelines established by the committee.

A 23 g mouse is injected with 4.8 mCi of ^99mTc-MDP and euthanized 90 minutes post-injection. A focused scan of the mouse’s back paw is performed using the GP metal ring. The total scan duration is 4 hours (T4 protocol, 1 hour per position), with projection data from all positions combined for image reconstruction. Additionally, two subsets of projection data—corresponding to a 1-hour scan (T4 protocol, 15 minutes per position) and a 20-minute scan (T4 protocol, 5 minutes per position)—are extracted and reconstructed separately for comparison.

A second 23 g mouse is injected with 5.0 mCi of ^99mTc-MDP and euthanized 90 minutes post-injection. The mouse head is scanned using the GP metal ring for a total of 6 hours. The scan includes six bed positions with a 4-mm step size, acquiring data for 1 hour at each position. At each bed position, sampling enhancement using the T4 protocol is applied, with 15 minutes per sampling-enhancement position. A subset of projection data—corresponding to a total scan time of 1 hour (6 bed position × 4 sampling-enhancement position, 2.5 minutes per position)—is extracted and reconstructed separately for comparison.

Additionally, the mouse is scanned using a commercial PET/SPECT/CT system (InliView-3000B, Novel Medical, China) to obtain a high-resolution CT image for comparison. The CT system features a focal spot size of 33 μm, a pixel size of 74.8 μm, an isocenter size of 34.6 μm, a source-to-detector distance of 284 mm, and a source-to-isocenter of 188 mm, offering an extreme modulation transfer function of 14.4 lp/mm. The CT scan duration is 20 minutes, and the voxel size for image reconstruction is 0.14 mm × 0.14 mm × 0.10 mm.

To quantitively evaluate the structural similarity between the SPECT and CT images, we calculate the structural similarity index measure (SSIM) [41], defined as:

$$\text{SSIM}(x_1,x_2)=\frac{(2{\mu }_{x_1}{\mu }_{x_2}+c_1)(2{\sigma }_{x_1{x}_2}+c_2)}{({\mu }_{x_1}^2+{\mu }_{x_2}^2+c_1)({\sigma }_{x_1}^2+{\sigma }_{x_2}^2+c_2)}$$ (14)

where $x_1$ is the CT image, $x_2$ is the SPECT image, ${\mu }_{x_1}$ is the mean value of $x_1,{\mu }_{x_2}$ is the mean value of $x_2,{\sigma }_{x_1}^2$ is the variance of $x_1,{\sigma }_{x_2}^2$ is the variance of $x_2$, and ${\sigma }_{x_1{x}_2}$ is the covariance between $x_1$ and $x_2$. The constants $c_1$ and $c_2$ are used to stabilize the SSIM computation, defined as $c_1={\left(0.01L_1\right)}^2$ and $c_2={\left(0.03L_2\right)}^2$, where $L_1$ and $L_2$ are the dynamic ranges of the voxel values in the CT and SPECT images, respectively.

III. Results

A. Detector calibration and data processing

Fig. 4a presents the flood histogram for a sample detector cassette with the same detector block orientation as that of Fig. 3a, that is, block I to IV lines from right to left. All 42 cells in the cassette are clearly distinguishable. Fig. 4b shows the PSD factor histogram of a detector cell, with two identifiable peaks corresponding to GAGG-90 and GAGG-200. We set a threshold of 10.4, corresponding to the minimum point between the two peaks, to distinguish events in GAGG-90 (<10.4) from that in GAGG-200. Fig. 4c displays the DOI ratio histogram of a scintillator, with the half-maximum positions of the left and right peaks denoted as ${\alpha }_1$ and ${\alpha }_2$. The interval [${\alpha }_1,{\alpha }_2$] is divided into 10 equally spaced subintervals, each corresponding to a DOI bin. Any value of $\alpha$ outside [${\alpha }_1,{\alpha }_2$] is assigned to the nearest subinterval.

Fig. 4.

(a) Detector cells’ flood histogram in a cassette. (b) PSD factor histogram of a detector cell. The threshold used to separate GAGG-90 from GAGG-200 is set to 10.4, corresponding to the minimum point between the two peaks. (c) DOI ratio histogram of a scintillator.

B. Sensitivity

Fig. 5 presents the sensitivity map of the small animal SC-SPECT. With the HR metal ring, the sensitivity ranges from 0.018% to 0.064% in the FOV (Φ = 10 mm, L = 6 mm), with an average of 0.033% and a standard deviation of 0.011%. With the GP metal ring, the sensitivity ranges from 0.032% to 0.080% in the FOV (Φ = 24 mm, L = 10 mm), with an average of 0.054% and a standard deviation of 0.010%.

Fig. 5.

Sensitivity map of the small animal SC-SPECT with the HR and GP metal rings in the three orthogonal planes, transverse, coronal, and sagittal, passing the origin.

C. Spatial resolution

Fig. 6 displays the reconstructed hot-rod images. With the HR metal ring, the smallest distinguishable hot rod diameter (based on the criterion of PVR > 2) is 0.3 mm using the T1 protocol (48-minute total scan time), which improves to 0.2 mm with the T4 protocol (also 48 minutes). With the GP metal ring, the smallest distinguishable hot rod diameter is 0.45 mm using the T1 protocol (48 minutes), which improves to 0.3 mm with the T4 protocol (48 minutes). This demonstrates a significant enhancement in spatial resolution for SC-SPECT through sampling enhancement.

Fig. 6.

(a) Reconstructed images of the hot rod phantom acquired using the HR and GP rings. For each scan, the phantom contains a total activity of 0.18 mCi. Under the T1 scanning protocol, the total scan time is 48 minutes. For the T4 protocol, reconstructions are performed using total scan times of both 48 minutes and 12 minutes. (b) Line profiles of the smallest distinguishable hot rods, evaluated using a criteria of PVR >2. The corresponding PVR values are annotated above each line profile.

When the scan time is reduced to 12 minutes under the T4 protocol, the smallest distinguishable hot rod diameter increases to 0.25 mm and 0.35 mm with the HR and GP metal rings, respectively, reflecting a slight loss in resolution due to increased noise.

Fig. 7a shows the DOI ratio histograms of a scintillator for collimated planar photon beam incidences at −8 mm, −4 mm, 0 mm, 4 mm, and 8 mm, and Fig. 7b shows the corresponding DOI resolutions, averaged across all crystals, at these positions. The DOI resolution exhibits little variation across different DOI positions. The overall DOI resolution (detector intrinsic resolution), averaged across all crystals and all positions, is 3.8 mm. The axial image spatial resolution, calculated using the standard pinhole geometrical model, is 0.64 mm for the HR metal ring and 1.22 mm for the GP metal ring.

Fig. 7.

(a) DOI ratio histograms of a scintillator for collimated planar photon beam incidences at −8 mm, −4 mm, 0 mm, 4 mm, and 8 mm. (b) DOI resolutions at the corresponding five DOI positions.

D. FSRQ

Table 2 presents the metrics of the small animal SC-SPECT with the HR and GP rings compared to those of commercial peers by Mediso and MILabs [16–19]. Despite having less than one-fourth of the peers’ detection areas—with only one ring in axial direction—SC-SPECT achieves about two-thirds of the FSRQ with the HR metal ring, and twice the FSRQ with the GP metal ring, representing a notable breakthrough of the trade-off among imaging FOV, sensitivity, and spatial resolution.

Table 2.

Metrics of the small animal SC-SPECT compared with that of commercial peers by nanoScan and MILabs.

Metric	nanoScan (MU)	nanoScan (MH)	U-SPECT+ (XUHR-M)	U-SPECT+ (GP-M)	SC-SPECT (HR)	SC-SPECT (GP)
Detection area	267,240 mm2	267,240 mm2	842,520 mm2	842,520 mm2	56,448 mm2	56,448 mm2
FOV volume	2,815 mm3	9,896 mm3	792 mm3	792 mm3	471 mm3	4,524 mm3
Average sensitivity	<0.031%	<0.082%	<0.038%	<0.171%	0.033%	0.054%
Hot-rod resolution (transverse)	0.30 mm	0.70 mm	0.25 mm	0.40 mm	0.20 mm	0.30 mm
Spatial resolution (axial)	1.00 mm	1.57 mm	0.62 mm	0.91 mm	0.64 mm	1.22 mm
FSRQ	<9.7	<10.5	<7.8	<9.3	6.1	22.2

E. Animal scans

Fig. 8a shows maximum intensity projections (MIPs) of the mouse paw acquired from 4-hour, 1-hour, and 20-minute scans. All scans clearly depict tracer distribution within fine bone structures, with higher uptake observed in the joint region. The 1-hour scan demonstrates image quality comparable to the 4-hour scan, whereas the 20-minute scan exhibits increased noise.

Fig. 8.

(a) MIPs of the mouse paw acquired with SC-SPECT using the GP metal ring. (b) MIPs of the mouse skull acquired with SC-SPECT using the GP metal ring. (c) MIPs of the mouse skull acquired with micro-CT. (d) Transverse images of the mouse skull acquired with SC-SPECT (6-hour scan) and micro-CT.

Fig. 8b shows MIPs of the mouse skull acquired from 6-hour and 1-hour scans using SC-SPECT, and Fig. 8c displays MIPs of the mouse skull using micro-CT. Fig. 8d shows the corresponding transverse images with a slice thickness of 0.3 mm. The fine bone structures of the mouse skull and the MDP tracer uptake within the skull are clearly resolved in both the CT and SPECT images, respectively. Fig. 8b through Fig. 8d demonstrate a clear correspondence between CT and SPECT images. As expected, ^99mTc-MDP accumulates more in regions with higher bone metabolic activity, such as the temporomandibular joint, malar bone, premaxilla, occipital bone, mandible, and alveolar bone, and shows reduced uptake in the incisors and molars with lower metabolic activity.

The SSIM between the SPECT and CT images are 0.88 and 0.86 for the 6-hour and 1-hour scans, indicating high structural similarity. Note that the differences are partly attributable to the higher noise levels and lower spatial resolution of SPECT relative to CT. More importantly, they reflect the inherent disparity between the tracer uptake visualized by SPECT and the bone density information provided by CT.

IV. Discussion

In this work, we present our first full-ring SC-SPECT system dedicated for high-resolution SAI. The system comprises four concentric sparse-scintillator detector rings and a high-aperture-ratio metal ring, where each inner ring collimates photons for all outer rings. The combined collimation of multiple detector rings and the metal ring facilitates high-resolution imaging. Additionally, the collimation function of the detector rings substantially reduces reliance on mechanical collimation, allowing for a greater number of apertures on the metal ring and significantly boosting sensitivity. The four detector rings are constructed with 40 detector cassettes, each covering a 9° angular region, and two metal ring configurations are designed for high-resolution and general-purpose SAI applications. Photon interaction positions within the three-dimensional sparse-scintillator detector architecture are determined using a combination of Anger logic, dual-end readout technique, and PSD method.

Through phantom studies, we demonstrate that the system, when equipped with the HR metal ring, achieves a sensitivity of 0.064% at the FOV center, and is capable of resolving hot rods with diameters of 0.2 mm (T4 protocol) and 0.3 mm (T1 protocol). With the GP metal ring, the system achieves a central sensitivity of 0.080%, and can resolve hot rods with diameters of 0.3 mm (T4 protocol) and 0.45 mm (T1 protocol). The sensitivity ratio from Ring-I to Ring-IV is approximately 8:4:2:1 with both HR and GP rings, as each detector ring absorbs about half of the incident photons. In mouse scans, the system clearly delineates fine bone structures of the mouse paw and skull. These results highlight the exceptional imaging performance of the small animal SC-SPECT system. We postulate that 1) the combination of the low-resolution, multiplexed (overlapping) projections from the inner rings and the high-resolution, non-multiplexed projections from the outer rings result in a sensitivity gain, and 2) the multi-layer sparse-scintillator detector architecture, coupled with enhanced sampling, results in better resolution.

Super spatial resolution achieved through enhanced sampling—including translational and rotational movements of the imaging object during acquisition has been shown to improve SPECT resolution [42–48]. In our prior work, we further demonstrated that SC-SPECT benefits more from enhanced sampling than conventional multi-pinhole SPECT [25]. Therefore, enhanced sampling is a critical factor for SC-SPECT to achieve high-resolution imaging.

The resolution performance shown is limited by the physical size of point sources used. Although Monte Carlo simulations suggest that the system can resolve hot rods as small as 0.1 mm [24, 25], the experimentally observed resolution is degraded—primarily due to the size of the point source (0.36 mm) used for system matrix measurement. Employing a smaller point source could improve resolution while would introduce significant image noise due to reduced activity. Future work will explore alternative methods for system matrix generation, including analytical calibration and deep learning.

The resolution in the axial direction is constrained by the lack of self-collimation and detector’s limited intrinsic resolution in this direction. To enhance detector’s axial intrinsic (DOI) resolution and implement axial self-collimation in the next-generation SC-SPECT system, we plan to replace axially continuous GAGG(Ce) scintillators with axially interleaved GAGG(Ce) scintillators and K9 glass elements, where the glass allows most photons to pass through and the scintillators both detects photons and provides collimation for outer rings. Our previous work has demonstrated such an interleaved scintillator bar design—each containing five 2-mm-long GAGG(Ce) segments alternating with five 2-mm-long K9 glass elements—can achieve 2-mm DOI resolution [49]. An alternative approach under consideration is constructing detector rings using pixelated GAGG(Ce) detector blocks (Fig. 3a) with SiPM readouts on the far-side from the FOV. We have constructed such detector design by coupling a 31 × 31 array of 0.8 mm × 0.8 mm × 6 mm GAGG(Ce) scintillators with 8 × 8 SiPM arrays, using a light guide with optical barrier slits, and demonstrated clear identification of all scintillators in the flood histogram [50]. This approach will reduce overall system complexity but add approximately 15% more photon scatter per ring from SiPMs and electronics.

In this work, we propose the FSRQ as a screening and comparative index that summarizes the intertwined effects of FOV size, sensitivity, and spatial resolution in SPECT SAI system design. Its calculation is straightforward, relying directly on established performance metrics. However, the absence of an uncertainty analysis may somewhat limit its interpretability, especially when comparing systems with substantially different geometries. The FSRQ is also closely related to the detection area, which reflects the system’s cost and complexity. For instance, when a system is extended along the axial direction, the FSRQ is approximately proportional to the solid angle subtended by the detectors at the FOV center. Consequently, the FSRQ should be analyzed alongside the detection area to assess the system’s technical value and cost-effectiveness.

The sensitivity and FOV of this system are limited by the short axial length (20 mm) of the current detector setup, which is less than one-tenth of other small-animal SPECT systems (250 ~ 500 mm) [16–19]. Nevertheless, SC-SPECT achieves about two-thirds of the FSRQ with the HR ring and twice the FSRQ with the GP ring compared with those systems. Next we will focus on extending the system’s axial length to around 320 mm, which is expected to substantially enhance sensitivity, as well as both FOV length and diameter, resulting in an approximately tenfold increase in the FSRQ. This would allow scanning larger regions with fewer bed positions and shorter scan times, potentially enabling faster, ultra-high-resolution whole-body dynamic imaging in small-animal studies.

V. Conclusion

The SC-SPECT system achieves high spatial resolution and enhanced overall imaging capability, representing a significant advancement in small animal SPECT imaging.