Adaptation of a Brain-Dedicated Multipinhole SPECT System for Imaging Higher-Energy Photons of Theranostic Agents—A Simulation Study

Abstract

Multi-pinhole SPECT of the brain is capable of producing high-resolution images. However, pinhole imaging of high-energy photons remains challenging due to the trade-off of greatly sacrificing sensitivity to limit penetration through the edges of the pinholes and system shielding.

The impact of photon penetration on spatial resolution, image contrast, and lesion signal-to-noise was assessed through simulation for the brain-dedicated multi-pinhole system AdaptiSPECT-C. In addition, minor changes to the system design to reduce penetration were considered; these were updating the materials of the pinhole aperture edges and shutters to a higher-density tungsten alloy and including platinum in the shielding around the pinhole apertures.

A significant degradation in spatial resolution, image contrast, and signal-to-noise above approximately 200 keV was found with the current system design. It was shown that adapting the pinhole and shutter material to a higher-density tungsten alloy would greatly reduce this degradation and permit multi-pinhole imaging with ¹⁷⁷Lu and ¹¹¹In in the human brain. Incorporating a small amount of platinum around each aperture further improved the system point-spread function. However, results indicate high-resolution imaging of ¹³¹I would require a larger modification to the system design.

I. Introduction

Single-Photon Emission Computed Tomography (SPECT) of the brain is often performed for diagnostic purposes to assess blood perfusion or the concentration of dopamine transporters, using agents labeled with ^99mTc (140.5 keV photon) or ¹²³I (159.0 keV) such as ^99mTc-HMPAO and ¹²³I-ioflupane [1]. Recent advancements and interest in theranostics for neurooncology have shown demand for brain imaging of therapeutic agents for the treatment of glioblastoma, meningioma, or brain metastases from other primary cancers [2]–[5]. Many of the radiopharmaceuticals for theranostics in the brain are still in clinical or pre-clinical trials [2], but the surge of theranostics is expected to spread into neurooncology [5], [6]. One radionuclide of promise is ¹⁷⁷Lu (112.9 and 208.4 keV), which can be bound to the somatostatin receptor type 2 (SSTR2) used in the treatment of meningiomas as ¹⁷⁷Lu DOTATATE or used in the treatment of glioblastomas as ¹⁷⁷Lu-PSMA [7]–[9]. Radiotracers labeled with ¹¹¹In (171.3 and 245.4 keV) are in clinical use for visualisation of the somatostain receptors of brain tumours as ¹¹¹In-indium-pentetreotide, and more are under investigation for imaging of angiogenesis in glioma (¹¹¹In-hnTf-VEGF and ¹¹¹In-abergrin™) and epidermal growth factor receptors (¹¹¹In-hEGF) [2], [10]–[12]. ¹¹¹In also has therapeutic potential through its emission of Auger electrons, and has been labeled to antibodies in glioblastoma therapy trials [13]. ¹³¹I (364.5 keV) emits therapeutic β particles with ¹³¹I-phenylalanine having shown efficacy in the therapy of glioblastoma and ¹³¹I-mIBG in the treatment of neuroblastoma [3], [14]. In addition, ²⁰³Pb (279.2 keV), although not therapeutic itself, has been proposed as an imaging analogue for ²¹²Pb used in targeted alpha therapy, including for brain tumours [15], [16].

The use of pinhole collimation for brain SPECT improves the trade-off between sensitivity and spatial resolution compared to the more-common parallel hole collimation [17]–[19]. However, it can be challenging to shield for higher-energy photons without sacrificing sensitivity and inducing significant material costs [20]. Pre-clinical multi-pinhole systems with tungsten collimation have been successfully used for PET imaging at 511 keV [21], [22], but the application of multi-pinhole collimators for high-energy clinical imaging remains limited. AdaptiSPECT-C is a clinical brain-dedicated multi-pinhole SPECT system that is currently under construction at the University of Arizona [23]–[28]. It was designed for diagnostic brain tracers with photon energies below 200 keV. This work aims to assess the feasibility of imaging therapeutic agents with this system, and explores small design changes to improve imaging performance for these agents.

II. Materials and Methods

Simulations of AdaptiSPECT-C under several design variations were conducted, as detailed in Section II-A. Simulations of point sources were used to assess the point-spread function (PSF) of the system (discussed in Section II-B), and three-dimensional voxelised phantoms were evaluated with several image metrics (discussed in Section II-C).

A. AdaptiSPECT-C models and materials

Two simulation models of AdaptiSPECT-C were created in the Geant4 Application for Emission Tomography (GATE) [29]: a single modular detector and a full detector system model (24 modules). Each module of AdaptiSPECT-C is comprised of a (18.4 × 18.4 × 1) cm³ NaI crystal and a 2 cm thick aperture plate with 5 pinholes. A shuttering mechanism allows each keel-edge pinhole to be independently shuttered or set to a 1.2, 2.6 or 4 mm aperture diameter. The system images a spherical volume of interest (VOI) with diameter of 21 cm. An illustration of the AdaptiSPECT-C system is shown in Figure 1. The aperture plates consist of two parts: a base plate with cylindrical cavities around each pinhole, and press-fit fillings for the pinholes which are affixed into the cavities. The aperture base plates were machined from a high-density tungsten (W) alloy, known as EF17, with a density of 17.0 g/cm³. The pinhole press-fit fillings and the shutters were 3-D printed from a tungsten-bronze alloy with a density of (13.60 ± 0.14) g/cm³. An aluminium (Al) detector box houses the crystal and electronics and a lead (Pb) shroud surrounds the aperture plate and extends over the detector box to provide shielding from scattered photons. In this work, the geometry and design of the AdaptiSPECT-C modules was assumed to be fixed, but the composition of the pinhole press-fit fillings and shutters was varied. As shown in Figure 1, these components are the thinnest part of the shielding, so increasing the density of these components would likely reduce penetration from higher-energy photons. Selective laser melting 3-D printing technologies have been shown to achieve W densities between 80 and 96 % that of pure W [30] or 15.4 to 18.5 g/cm³. EF17 W (17.0 g/cm³) is at the average of this range, so simulations were conducted with all components of the aperture plate and shutters constructed from EF17 W. Platinum (Pt, density 21.4 g/cm³) is one of few elements with a higher density than W. 3-D printing of platinum alloys through wax casings or selective laser melting is possible with varying density depending if Ru, Rh or Ir is used in the alloy and their quantities [31]–[33]. For this initial study, pure Pt at 21.4 g/cm³ was assumed, but slightly higher-density Pt-based alloys have been used in another imaging system [34]. The cost of Pt can be prohibitive; the idea of using small inserts of denser material around the pinhole aperture has been previously investigated in animal systems using ^99mTc [35], [36] and ex-vivo SPECT for several isotopes [37]. Thus, we investigated if using Pt for just a small cylindrical insert at the aperture would reduce penetration sufficiently. To that end, five variations of AdaptiSPECT-C were considered:

Fig. 1.

The AdaptiSPECT-C model. Top left: A cross-sectional schematic of a single camera module. Bottom: front (patient-facing), back (detector-facing), and cross-sectional view of the aperture plate. Top right: The GATE model of the full system with 24 modules and the 21 cm spherical VOI in green.

1. Ideal: All shielding components (aperture plate base and press-fit fillings, shutters and shroud) were set to a theoretical material with density of 10000 g/cm³.

2. ASC: AdaptiSPECT-C as being built with EF17 W base plate, 3-D printed W plate press-fit fillings and shutters, and Pb shrouds.

3. All_EF17: ASC modified so plate press-fit fillings and shutters are also EF-17 W.

4. All_Pt: ASC modified so plate press-fit fillings and shutters are Pt.

5. PtIns: All_EF17 but with a small Pt insert inside each shutter covering the smallest part of the aperture.

The following radionuclides were simulated, with the photon energies of interest given in parentheses: ^99mTc (140.5 keV), ¹²³I (159.0 keV),¹⁷⁷Lu (208.4 keV), ¹¹¹In (171.3 and 245.4 keV), ²⁰³Pb (279.2 keV) and ¹³¹I (364.5 keV). All γ emissions and intensities are summarised in Table I. The source definition for each radionuclide was created using the GATE histpoint function for all γ emissions (β, Augers and X-rays were not included). In accordance with initial experiments of the system, the intrinsic spatial resolution was set to 1.5 mm at full-width half-maximum and the energy resolution of the NaI crystal was modelled as 8 % at 159.0 keV [38]–[40]. The inverse square digitizer module in GATE was used to extend the resolution modelling to other energies.

TABLE I.

Energy and intensity of γ emissions of radionuclides used in this work. Emissions with intensity > 0.1 % are shown, but all emissions were included in simulation.

Radionuclide	γ energy, keV (Emission intensity, %)
99mTc	140.5*(89)
123I	159.0*(83.6), 529.0(1.3), 440.0(0.39), 538.5(0.31), 346.4(0.12)
177Lu	208.4*(10.4), 112.9(6.2), 321.3(0.22), 249.7(0.20), 71.64(0.16)
111In	245.35*(94.1),171.28*(90.7)
203Pb	279.2*(80.9), 401.32(3.4), 680.5(0.75)
131I	364.5*(81.5), 637.0(7.2), 284.3(6.1), 80.2(2.6), 722.9(1.8), 503.0(0.36), 325.8(0.27), 177.2(0.27), 642.7(0.22)

^*denotes a photopeak used to generate projections. Data were taken from [43].

B. Point source simulations

The single module simulation was used to assess the point-spread function (PSF) of a point source of activity at the centre of the field of view for each of the radionuclides of interest. Initially, impact of photon penetration was evaluated for each of the three aperture sizes through simulations of a point source of ^99mTc and ¹³¹I with the ASC or Ideal material cases. For each, a 1 µm-diameter spherical 500 MBq source in air was defined and a 30 second simulation was acquired. Simulations were run with the 4 mm-, 2.6 mm- and 1.2 mm-diameter apertures, referred to as HighSens, StdRes and High-Res, respectively. Projections were created from all detected photons within a 15 % symmetric photopeak window. Each projection was collapsed to a one-dimensional profile of pixel counts. The projection full-width half maximum $\left({\text{FWHM}}_p\right)$ and full-width tenth maximum $\left({\text{FWTM}}_p\right)$ were determined for each peak in the profile using the SciPy interp1d function in Python [41] to interpolate between pixels. Since pinhole collimation results in a magnification of projections, the resolution must be converted back from projection to object space through

$$\text{FWHM}=\frac{{\text{FWHM}}_p}{M}\text{and FWTM}=\frac{{\text{FWTM}}_p}{M}$$ (1)

where $M$ is the magnification given by the aperture to image distance divided by the object to aperture distance [42]. For AdaptiSPECT-C, $M=0.503$. An average FWHM and FWTM was calculated for each radionuclide.

Further simulations were then conducted for the smallest, HighRes, apertures using the above procedure for each of the material cases and radionuclides discussed in Section II-A. The sensitivity, $S$, was calculated (in kBq/s) for each simulation as

$$S=\frac{\text{Projection counts}}{\text{Activity × Scan time}\times ϵ}$$ (2)

where $ϵ$ is the fractional emission intensity for the photopeak of interest, quoted in Table I.

C. Phantom simulations

The full system model was used for the analysis of three-dimensional source distributions. A voxelised Derenzo-like phantom of 20 cm diameter and 16 cm length with circular rods of diameter 11.1, 9.5, 7.9, 6.4, 4.8 and 3.2 mm was used to assess tomographic spatial resolution and contrast [44], [45]. No background activity or attenuation medium was used. A male XCAT [46] brain phantom was used as an approximation of a clinical acquisition. The head and brain regions were scaled to the 99^th percentile size using data from DoD tables [47] to maximise the irradiation angles inside the FOV and model penetration similar to a patient acquisition. An approximation of a glioma study was created, using data from a study of 70 patients which found a mean glioma volume of 34416.5 mm³ [48]. Here, spherical tumours were created for these volumes: a average tumour with a diameter of 40.4 mm and a small tumour with a diameter of 12.6 mm based on 95 % of the published volume range. Five tumours of each size were randomly distributed throughout the brain, under the conditions that they were wholly contained within the brain and did not overlap. A tumour-to-background uptake ratio of 3.4:1 was assigned [49] and the background was assumed to be uniform. Twelve material definitions for the components of the XCAT phantom were used in GATE, based on ICRP publication 110 values [50]; these components were air, compressed air inside body, muscle, skin, brain, blood, cartilage, eyes, cranium spongiosa, cervical spongiosa, mineral bone, and teeth.

To avoid any artefacts due to multiplexing (overlapping projections), all phantom simulations were conducted for three consecutive projection frames: first using only the central pinholes for 40 % of the acquisition, then using each diagonal pair of pinholes for 30% consecutively. This arrangement of frames was previously shown to give the best image fidelity for multiplex-free acquisitions [51], [52]. Projections with matrices of 184 × 184 × 24 (1 mm)³ voxels were generated from the GATE output ROOT files. For both phantom simulations, all photons which scattered outside the crystal were removed; an assessment of the impact of these scattered photons is discussed in Section III-D. In clinical practice, the desired number of projection counts will vary depending on the specific application and tracer used. To focus our comparison on the penetration of gammas through the pinholes and detector shielding, an equal number of counts was acquired to avoid any degradation in imaging due to reduced photon emission or detection probability for the different radionuclides. Thus, our comparisons are solely dependent on the penetrability of the photons. Simulations were run such that all scatter-free projections had counts within one standard deviation (assuming Poisson statistics) for all isotopes and system configurations. The projections for the brain phantom all had 5.0 M counts (the minimum recommended projection counts for clinical brain perfusion imaging [53]) and this was doubled to 10.0 M for the Derenzo phantom to reflect the higher counts used for image-quality assessment scans. The projections were reconstructed into 120 × 120 × 120 (2 mm)³ images using an in-house GPU-based maximum-likelihood expectation-maximization (MLEM) package with attenuation correction [54], [55]. Attenuation was modeled in the reconstruction of the XCAT phantom from the map of linear attenuation coefficients. The reconstruction code models the detector and collimator configuration and response within its iterative algorithm. It also assigns an effective diameter to the pinholes to account for some penetration, based on the linear attenuation of the photon through the the pinhole edges [56]. To determine linear attenuation coefficients (in cm⁻¹), mass attenuation coefficients were taken from the NIST XCOM database [57], linearly-extrapolated to the energies of interest, and multiplied by the density of the different pinhole materials. No regularisation or post-reconstruction smoothing was applied.

The quality of the Derenzo phantom images was assessed using the contrast recovery factor (CRF), defined such that

$$CRF=\frac{\sum \left(I\odot I_{Hot}\right)/N_{Hot}-\sum \left(I\odot I_{Cold}\right)/N_{Cold}}{\sum \left(I\odot I_{Hot}\right)/N_{Hot}+\sum \left(I\odot I_{Cold}\right)/N_{Cold}}$$ (3)

where $I$ is the reconstructed image, $I_{Hot}$ and $I_{Cold}$ are binary masks of the hot and cold voxels in the ground truth image of the Derenzo phantom respectively, $N_{Hot}$ and $N_{Cold}$ are the number of hot and cold voxels respectively, and ⊙ denotes the Hadamard product for element-wise matrix multiplication [23].

For the XCAT phantom, the signal-to-noise ratio (SNR) between tumours and backgound was calculated. Masks of the hot (spherical tumours) and cold (background of the brain) regions were created from the phantom and down-sampled to match the reconstructed image volume. Signal to noise was calculated through

$$\text{SNR}=\frac{C_t/N_t-C_{bg}/N_{bg}}{{\sigma }_{bg}}$$ (4)

where $C_t$ and $C_{bg}$ are the total counts and $N_t$ and $N_{bg}$ are the number of voxels in the tumour and background, respectively, and ${\sigma }_{bg}$ is the standard deviation of counts in the background. To avoid edge effects such as spill-out and Gibbs artefacts, the background mask was eroded by 3 pixels at each boundary and the tumour masks were eroded by one pixel. SNR was calculated for each tumour individually and a mean and standard deviation was obtained for the five average and small tumours independently.

The MLEM iteration used in comparisons was defined based on the metric of interest (SNR or CRF). For SNR, this was the iteration where the sum of average SNR for the two tumour sizes was maximised. Since CRF asymptotically approaches a maximum, the iteration of interest for the Derenzo phantom was defined as the one in which the variation in CRF between it and the previous fell below 0.5 %.

III. Results

A. Planar spatial resolution and sensitivity

Projections for simulations of point sources of ^99mTc and ¹³¹I imaged with the three aperture sizes are shown in Figure 2. The full-width half maxima (FWHM) calculated from the projections with Equation 1 are also shown. The Ideal and ASC projections for ^99mTc are visually indistinguishable. However the projections for ¹³¹I show blurring and defects due to penetration of the higher-energy photons. The ratio of the FWHM for ASC/Ideal was calculated for the two radionuclides for each aperture size. There is a (16.13 ± 0.12) % increase in FWHM for the HighRes ASC simulation of ^99mTc compared to Ideal, suggesting penetration is degrading resolution even for 140.5 keV photons. The FWHM increases by over a factor of two for ¹³¹I HighRes when ASC is compared to Ideal. In terms of sensitivity, the bottom plot in Figure 2 shows the ratios of ASC and Ideal for the HighSens and StdRes pinholes are equivalent to 1 but equal to 1.350 ± 0.091 for HighRes for ^99mTc, again demonstrating that even 140.5 keV photons are penetrating the aperture shielding. The sensitivity ratios are much larger for ¹³¹I, ranging from 3.47 ± 0.13 for HighSens to 29.6 ± 3.4 for HighRes. The HighRes apertures have a much lower sensitivity than the others (around 4 and 9 times lower than StdRes and HighSens, respectively, for ^99mTc), so even a small number of penetrating photons can become comparable to the direct photons passing through the aperture as intended. The results presented in Figure 2 showed the HighRes apertures are the most impacted by penetration. Thus, the subsequent analyses considered only the HighRes apertures.

Fig. 2.

Projections for point sources of ^99mTc and ¹³¹I imaged with the three aperture sizes for the Ideal and ASC simulations. The full-width half and maxima (FWHM) of profiles over each projection was calculated and the ratio (ASC/Ideal) is shown for each case. The ratio of the ASC and Ideal sensitivity (S) is shown in the bottom plot - a broken y-axis is used due to the large difference in values.

The sensitivity was calculated for each of the HighRes point source simulations, according to Equation 2. For easier comparison, all sensitivities were normalised relative to that of the Ideal case for ^99mTc; these are shown in Figure 3. The sensitivity decreases with energy for the Ideal simulations as the 1 cm-thick NaI crystal becomes less efficient at detecting the photons. The simulation does not model the electronic signal-processing components; other work has shown this can lead to an energy-dependent deviation between experimental and simulated sensitivities [58]. In contrast, the sensitivity increases rapidly with energy for the ASC simulations demonstrating that penetration through the shielding (pinhole edges or aperture plate) becomes more prevalent.

Fig. 3.

The sensitivity for each of the simulations for a point source of activity at the focal spot imaged with the HighRes apertures, calculated with Equation 2. All values are relative to the Ideal case for ^99mTc.

The FWHM and FWTM were determined for each radionuclide for each combination of shielding materials for these smallest apertures and adjusted for system magnification, according to Equation 1. A comparison of the profile fits, the FWHMs and FWTMs, and the sensitivity ratios are shown in Figure 4. The FWHM for the ideal system increased from (3.140 ± 0.020) mm for ^99mTc to (3.485 ± 0.087) mm for ¹³¹I. For all radionuclides, both FWHM and FWTM for ASC are significantly larger than the Ideal case. FWTM increases more rapidly than FWHM as the photon energy increases, suggesting that the tails of the PSF are more impacted by penetration than the peak. This is likely because, as energy increases, photons are able to penetrate the aperture shielding from directions which were previously inhibited, spreading the photon detection to a greater area of the crystal. The FWHM and FWTM for the higher-density W in All_EF17 agree within 2 standard deviations of Ideal for ^99mTc, and no significant difference is seen if platinum is used. For all other radionuclides, using Pt significantly improves the FWHM compared to All_EF17. However, the FWHM agrees within 2 standard deviations for ¹³¹I or 1 standard deviation for all other radionuclides regardless of if the full shutter and insert is Pt (All_Pt) or just a small insert of Pt inside the shutter is used (PtIns). The FWTM between All_Pt and PtIns also agree within 1 standard deviation for all emissions below the 279 keV peak of ²⁰³Pb. The bottom of Figure 4 shows the ratio of sensitivity (calculated with Eq. 2) acquired with each material type compared to Ideal. A sensitivity ratio of 2, indicated as a black dashed line on the figure, indicates that there are twice as many counts being detected in that case compared to Ideal; i.e. that as many photons are penetrating the shielding somewhere as pass through the aperture. All sensitivity ratios for the tested materials are below this threshold for ^99mTc and ¹²³I at 141 and 159 keV, respectively. For the ASC material simulations, the 171 keV simulations of ¹¹¹In cross this sensitivity threshold. When All_EF17 is used, the threshold is not reached until the 245 keV peak of ¹¹¹In. The sensitivity ratio for the 279 keV peak of ²⁰³Pb is within one standard deviation of the threshold for both simulations including platinum.

Fig. 4.

The results of the single-module simulations of a point source in air with the HighRes apertures. Top: the profile through the projections of the central pinhole for each radionuclide. Middle: the full-width half maximum (FWHM) and full-width tenth maximum (FWTM) of each of the profiles. Uncertainties are shown in black on top of each bar. Bottom: The ratio of sensitivities to the Ideal case for each of the other four material definitions; a broken y-axis is used due to the large difference in values. A sensitivity ratio of 2 is highlighted with the dashed black line.

B. Contrast recovery in rod phantom

Simulations of the Derenzo phantom were run for the full set of radionuclides. Since the results of Section III-A suggest no significant difference in resolution between All_Pt and PtIns, only the smaller platinum inserts (PtIns) were simulated and compared to ASC and All_EF17. The 20 central transaxial slices of the iteration of interest for the Derenzo phantom reconstructions were summed (a total range of 40 mm) and are shown in Figure 5. All counts outside the spherical FOV have been removed. The loss of contrast can clearly be seen due to increased penetration as the photon energy increases. When the material is changed from ASC to All_EF17, a stark difference is seen in several images up to the 279 keV peak of ²⁰³Pb. However, there is not a clear visual difference for ¹³¹I. This is reflected in the CRF plot where a clear improvement is seen for the majority of the radionuclides when transitioning from ASC to All_EF17, but the improvement for ¹³¹I is minor. The CRF for All_EF17 and PtIns is similar for all radionuclides.

Fig. 5.

The sum of 20 transaxial slices through the center of the reconstructed simulations of the Derenzo rod phantom. All images are thresholded to 85% of their maximum pixel value. The rod size in mm is given on the True image. The numbers in the bottom right show the MLEM iteration of interest (where CRF converged). The bottom figure shows the contrast recovery factor calculated from each image.

C. SNR in brain phantom

Reconstructed images of the brain phantom along with plots of SNR are given in Figure 6. A single transaxial slice through the brain showing two of the average-diameter and one small-diameter lesions is shown for each of the system models and radionuclides. As with the Derenzo phantom, a clear degradation in image quality is seen as the photon energy increases but this is improved in most cases with the introduction of higher-density W or Pt inserts. Again, no visual improvement is seen for ¹³¹I. The plots of SNR reflect the visual interpretation: SNR is degraded for ASC with increased photon energy and this effect is especially impactful in the small lesions. The three system designs agree within one standard deviation for ^99mTc and ¹²³I. For the other photopeaks, All_EF17 shows a visual improvement and superior SNR compared to ASC for both tumour sizes, apart from for ¹³¹I. Incorporating Pt into the shielding improves SNR for some average-diameter tumours, but the PtIns and All_EF17 results are equivalent within one standard deviation for the smaller-diameter tumours.

Fig. 6.

Simulations of the XCAT brain phantom with two sizes of spherical tumours. The top two rows show a series of transaxial slices through the true activity map and the ASC simulation with ^99mTc. Transaxial slices go from caudal to cranial. The subsequent images are the same transaxial slice through the brain for each of the isotope and material combinations; a threshold of 85% the maximum pixel value has been used for all and a mask of the head was used to trim extra-FOV counts. The numbers at the bottom right of each image show the MLEM iteration of interest (where SNR reached its maximum). The bottom figure shows the signal-to-noise ratio (SNR) for the two tumour diameters as a function of photopeak energy.

D. Assessment of scatter

The impact of scattered photons was assessed with the XCAT brain with ¹¹¹In, since it includes both a photopeak which experiences down-scatter from higher energies (EM1 at 171 keV) and one which does not (EM2 at 245 keV). The fraction of photons in the two 15 % energy windows which scattered outside the crystal was recorded for the three system variations. For EM1, the fraction of scattered photons reduced from (43.343 ± 0.027) % for ASC, to (39.076 ± 0.026) % for All_EF17 and (39.126 ± 0.026) % for PtIns. EM2 had a lower total scatter fraction, reflecting the lack of down-scattered photons, but still showed a reduction from (22.569 ± 0.011) % for ASC to (20.302 ± 0.016) % for All_EF17 and (20.373 ± 0.017) % for PtIns. The SNR was calculated for reconstructions including scatter and compared to the scatter-free data presented in Section III-C. For both energy windows and tumour sizes, SNR degraded when scatter was included in the projections but the ratio of SNR for the projections with and without scatter were consistent within one standard deviation for the system variations; this suggests scatter degrades images but is not significantly influenced by the material in the shutter. Many methods for accurate correction for scatter for ¹¹¹In have been proposed [59], [60], and their comparison and optimisation is outside the scope of this work.

IV. Discussion

The PSF analysis in Figure 4 demonstrates planar spatial resolution was improved when higher-density W was used for the press-fit inserts and shutters, and further improved when Pt was introduced. However, it also demonstrated that there was no significant difference in FWHM or FWTM if either the entire volume of the press-fit inserts and shutters were Pt, or if just a small insert surrounding the aperture was Pt. Using component volumes given by SolidWorks, this suggests the total mass of Pt could be reduced from 29.2 kg to 493 g whilst achieving the same improvement to planar resolution. The FWHM for the ASC design for a point source of ^99mTc at the focal spot imaged with the HighSens, StdRes and HighRes apertures was found to be (10.30 ± 0.10) mm, (6.672 ± 0.063) mm and (3.647 ± 0.031) mm, respectively. The HighRes FWHM was improved slightly further to (3.259 ± 0.017) mm for the All_EF17W design and (3.191 ± 0.041) mm for PtIns. This demonstrates the potential for imaging with a superior resolution than that seen in clinical systems which employ parallel hole collimators; for example the FWHM of a point source of ^99mTc imaged with the low-energy high-resolution collimators of a Mediso AnyScan triple-headed system was measured to be (10.14 ± 0.22) mm at a detector radius of 106.5 mm (the minimum permitted on the system) [58]. A theoretical evaluation of parallel-hole collimators predicts a system resolution limit of 8 mm at the 13.5 cm radius of rotation commonly used for brain SPECT with the Philips Forte’s VXHR collimator [61]. Other commercial systems designed for high-resolution brain imaging are available, such as the SMARTZoom^™ converging collimator by Siemens Healthineers with extended magnification, stated to have a spatial resolution at 100 mm ≤7.5 mm [62], [63]. Thus, the HighSens apertures of AdaptiSPECT-C have comparable resolution to current parallel-hole clinical systems, but the HighRes apertures can provide much higher-resolution brain imaging than the current clinical practice.

For the Derenzo phantom shown in Figure 5, there is a clear improvement in image contrast when All_EF17 is compared to ASC, but further improvement with PtIns is negligible. A clear ring artefact is seen for the higher-energy radionuclides. This work used only multiplex-free projections but, nonetheless, similar artefacts were seen for the ASC system when overlapping (multiplexed) projections were acquired [51], [52], [64]. It is possible that photon penetration through the shielding leads to an ambiguity in the system matrix of the reconstruction, in a similar way to multiplexing.

The commonly-used Rose criterion states that a signal-to-noise ratio must be at least 5 for an object to be detectable by human observers [65]. Following this criterion, the average-sized tumours were detectable other than for ¹³¹I with the ASC system, but the small tumours were not detectable at photon energies higher than the 171 keV peak of the radionuclides considered. Changing the design to All_EF17 means the small tumours are also detectable at the 208 and 245 keV emissions, and the SNR at 279 keV is within 2 standard deviations of 5. We, again, begin to see multiplex-like artefacts at higher energies, with a cold ring visible under the two average-sized tumours and a cooling seen at the center of the brain for ¹³¹I.

For both the Derenzo and brain phantom, we find no significant improvement for ¹³¹I when the aperture and shutter material is changed. ¹³¹I represents a challenging case for imaging with the current AdaptiSPECT-C system design. Using linear attenuation data from XCOM [57], the transmitted fraction of 364.5 keV photons through the 2 cm-thick aperture base plate is 0.029 %, which is larger than the sensitivity of our system with all the HighRes pinholes open (0.01409 ± 0.00003) %; this suggests more 364.5 keV photons likely penetrate directly through the plate than pass through a pinhole. Increasing the aperture base plate to 3 cm of EF17 W would reduce the transmitted fraction to 0.0005 % which is still 3 % of the primary sensitivity. Thus, imaging ¹³¹I with AdaptiSPECT-C, at least with the HighRes pinholes, likely remains suboptimal unless the aperture plates are significantly thickened, but this would require changes in the aperture-to-origin or detector-to-origin distances and hence changes in projection magnification. In addition, the 1 cm NaI crystal is likely not the optimal thickness for imaging at this energy, as evidenced by the drop in Ideal sensitivity in Figure 3. An increase in crystal thickness or transferring from NaI to a higher-density scintillator, such as novel Tl-based crystals [66], would be required to optimise imaging.

This work investigated reduction of penetration prior to detection, and did not consider methods for reducing the impact of penetration post acquisition. It may be that post-processing techniques could improve resolution further (even if penetration occurred), such as post-reconstruction de-convolution methods proposed in other work [67]. In addition, this work considered only an analytic reconstruction code which is limited in its approximation of the system PSF. Applying reconstruction based on Monte Carlo modelling of the PSF for each radionuclide directly may improve imaging metrics [68], especially for the higher-energy emissions where the increased FWTMs imply non-Gaussian tails.

The quantitative assessment of the impact on spatial resolution and sensitivity presented here are specific to the AdaptiSPECT-C design, but the principles may be extrapolated to a general case for pinhole imaging systems. That is to say, the advent of novel laser melting 3-D printing technologies permitting higher-density printed material can be harnessed to reduce photon penetration and permit high-resolution pinhole imaging of medium-to-high energy photons in addition to those around 150 keV. This opens the door to high-resolution multi-pinhole imaging of theranostic agents in the human brain. The trade-off of material cost to improvement in imaging will depend on the design of each system individually. For AdaptiSPECT-C, a clear improvement in resolution was seen when the insert and shutter material was changed from the lower-density W in the ASC model, to the laser melting higher-density W in the All_EF17 model, suggesting this upgrade to be worthwhile. While an improvement in resolution was seen for the PSFs of point sources when a small amount of platinum was incorporated around the apertures, the testing with the Derenzo phantom and small lesions in XCAT brain phantoms did not show a statistically significant improvement between the All_EF17 and PtIns models. These results suggest that, at least for the current AdaptiSPECT-C design, the use of laser melting 3-D printing technology to increase the density of W surrounding the apertures may be enough to permit high-resolution brain imaging up to at least the 245 keV peak of ¹¹¹In and perhaps the 279 keV peak of ²⁰³Pb.

V. Conclusion

The impact of penetration of photons from several radionuclides was assessed for the multi-pinhole AdaptiSPECT-C system. It was demonstrated that there is significant degradation in spatial resolution, image contrast, and signal-to-noise at energies ≥ 200 keV with the current high-resolution system design. However, remodeling the press-fit pinhole fillings and shutters with a higher-density tungsten alloy would greatly reduce that degradation and permit detection of even small tumours with ¹⁷⁷Lu, both peaks of ¹¹¹In and perhaps ²⁰³Pb. Moreover, incorporating a small amount of platinum around each aperture would further improve the system PSF. These small modifications would result in a multi-pinhole SPECT system capable of performing high-resolution imaging of the human brain with theranostic agents. However, penetration from high-energy radionuclides such as ¹³¹I remains a challenge without significant re-design of the system.