Performance evaluation of the mouse version of the LabPET II PET scanner

Abstract

The LabPET II is a new PET technology platform designed to achieve submillimetric spatial resolution imaging using fully pixelated APD-based detectors and highly integrated parallel front-end processing electronics. The detector was designed as a generic building block to develop devices for preclinical imaging of small to mid-sized animals and for clinical imaging of the human brain. The aim of this work is to assess the physical characteristics and imaging performance of the mouse version of LabPET II scanner following the NEMA NU4–2008 standard and using high resolution phantoms and in vivo imaging applications. A reconstructed spatial resolution of 0.78 mm (0.5 μl) is measured close to the center of the radial field of view. With an energy window of 350–650 keV, the system absolute sensitivity is 1.2% and its maximum NECR reaches 61.1 kcps at 117 MBq. Submillimetric spatial resolution is achieved in a hot spot phantom and tiny bone structures were resolved with unprecedented contrast in the mouse. These results provide convincing evidence of the capabilities of the LabPET II technology for biomolecular imaging in preclinical research.

1. Introduction

High spatial resolution is critical in preclinical positron emission tomography (PET) imaging where the ultimate goal would be to obtain an image definition in small animal models similar to that obtained in humans with clinical imaging. Since mice are three orders of magnitude smaller than humans, submillimetric (or submicroliter) spatial resolution is essential to discriminate small regions involved in tumor heterogeneity (Buvat et al. 2015), inflammation processes (Rucher et al. 2018), neurodegenerative disease progression (Dubois et al. 2016) or cardiac gated studies (Croteau et al. 2003).

Several preclinical PET scanners were recently developed for high resolution imaging using pixelated scintillator arrays coupled to photomultiplier tubes (PMT) (Belcari et al. 2017, Pilleri et al. 2019), position sensitive PMT (Bao et al. 2009, Wang et al. 2015, Pei et al. 2019, Teuho et al. 2019, Amirrashedi et al. 2019, Son et al. 2020), avalanche photodiodes (APD) (Bergeron et al. 2009, Bergeron et al. 2014, Vrigneaud et al. 2017) and silicon photomultiplier (SiPM) (Weissler et al. 2015, Ko et al. 2016, Omidvari et al. 2017, Hallen et al. 2018, Vrigneaud et al. 2018, Stortz et al. 2018, Sajedi et al. 2019, Lee et al. 2019). Preclinical PET scanners based on a monolithic crystal were also developed for high sensitivity PET imaging (Krishnamoorthy et al. 2018, Xu et al. 2019). Many of those scanners reported reconstructed spatial resolution between 1 and 2 mm (Miyaoka & Lehnert 2020). Submillimetric spatial resolution has been the main focus to design mouse brain (Yang et al. 2016) and rat paw (Godinez et al. 2018) dedicated PET scanners. While spatial resolution is a key feature for PET imaging, following NEMA standard is quite challenging in newer systems where maximum-likelihood expectation-maximization (MLEM) reconstruction algorithms supersede filtered backprojection ones. To overcome this problem, it was proposed to perform point source measurements using a significant hot background to prevent convergence to unrealistic solutions (Gong et al. 2016). However, many studies still overlook this recommendation and report artificially enhanced spatial resolutions, that are directly dependent on the voxel size used during reconstruction. This observation seems especially obvious in commercial promotion materials, where submillimetric spatial resolution is advertised while millimeter-sized details can hardly be resolved in animal imaging. Consequently, very few general purpose preclinical PET scanners achieving truly submillimetric spatial resolution for whole body mouse imaging were reported up to now.

Submillimetric spatial resolution in PET imaging is physically within reach provided that all factors contributing to the accuracy have been optimized. This is why small diameter systems and radiotracers with low-energy positron emission, such as ¹⁸F and ⁶⁴Cu, will be preferred to mitigate the effects of the annihilation photon non-collinearity and the positron range on image resolution. However, the major contribution to the degradation of spatial resolution in preclinical PET generally remains the detector size. Highly pixelated detectors with individual readout electronics then become mandatory to achieve the targeted submillimetric resolution goal.

The mouse version of the LabPET II scanner is an APD-based PET scanner designed to achieve submillimetric spatial resolution. This new generation of scanners relies on pixelated APD/LYSO detector arrays at a 1.2 mm pitch implementing individual readout and highly integrated parallel electronic architecture based on a dual-threshold time-over-threshold (ToT) signal digitization scheme (Grant & Levin 2014, Gaudin et al. 2020). The detection front-end was designed to accommodate various scanner configurations, from small animals to the human brain. The aim of this study is to evaluate the physical and imaging performance of the mouse version of the LabPET II using the NEMA NU4–2008 (NEMA 2008) standard, high resolution phantoms and in vivo studies.

2. Materials and methods

2.1. System description

The LabPET II detector array for mouse imaging is based on a 4 × 8 crystal block made of 1.12 × 1.12 × 10.6 mm³ Lu_1.9Y_0.1SiO₅:Ce (LYSO) optically isolated scintillator pixels glued to a monolithic 4 × 8 pixelated APD array (Excelitas Inc., Canada), to achieve true one-to-one coupling of individual pixels. APD pixels have an active area of 1.1 × 1.1 mm² at a 1.2 mm pitch. The basic components of the detector are displayed in Fig. 1a. Four of these detector arrays are mounted on a printed circuit board (PCB) carrying two flip-chip, 64-channel, mixed-signal application-specific integrated circuits (ASIC) on the backside, each interfacing to two 32-pixel detector arrays. The ASIC, designed in TSMC CMOS 0.18 μm technology, allows independent analog parallel processing and digitization of the signal from every detector pixel by a dual threshold ToT scheme to extract the required energy and timing information. A complex architecture of finite-state machines clocked at 100 MHz ensures the ASIC real time ToT calculation operations (Arpin et al. 2011). The ToT data are serialized into fully digital output with a maximum rate of 1.8 Mevents/s and then transferred to a FPGA, part of the embedded signal processing unit (ESPU), through differential high-speed links. The FPGA corrects, merges and sorts all detected events prior to transfer to a coincidence engine ensuring proper clocking scheme to all ESPUs and ASICs and supporting all data communication through an embedded Ethernet network (Fontaine et al. 2016).

Figure 1.

Pictures of: (a) a 32-pixel LabPET II detector array with its two main components, a 4 × 8 LYSO scintillator array and a monolithic 4 × 8 APD array; (b) mouse version of the LabPET II front-end electronics; (c) mouse version of the LabPET II detection ring.

The LabPET II front-end was designed as a modular free-standing detector that can be readily assembled to form scanners of various geometries (Fig. 1b). The mouse version of the LabPET II is based on 48 of these front-end detection modules arranged in a 12 × 4 cylinder to form a 78.8 mm diameter ring with an axial length of 50.4 mm. The scanner is depicted in Fig. 1c and its design characteristics are summarized in Table 1.

Table 1.

Design characteristics of the mouse version of the LabPET II scanner.

Parameters
Scintillator material	LYSO
Number of scintillators	6,144
Scintillator size (mm3)	1.12 × 1.12 × 10.6
Scintillator array dimensions	4 × 8
Photodetector	APD
Scintillator-to-photodetector ratio	1 : 1
Scintillators per ring	192
Number of rings	32
Axial length (mm)	50.4
Ring diameter (mm)	78.8
Packing fraction (%)	62
Reconstructed transaxial FOV (mm)	60
Coincidence time window (ns)	10
Energy window (keV)	350–650

2.2. System calibration

A complete optimization of the scanner was performed prior to the measurements including temperature stabilization, APD bias optimization, timing alignment and energy calibration. An automated procedure seeking the APD breakdown voltage while monitoring the signal baseline was used to operate the detector arrays at optimal bias maximizing the signal/noise ratio. The inherent non-linearity of the ToT signal (Orita et al. 2011) was corrected using average corrections factors based on an energy calibration procedure using a single γ-ray source and a variation of the ASIC internal electronic gain (Gaudin et al. 2020). Unless otherwise stated, measurements were performed using an effective energy window of 350 to 650 keV and a coincidence time window of 10 ns.

2.3. Image reconstruction

Images were reconstructed using a 3D-MLEM algorithm relying on an analytical system matrix that models the photon attenuation and interaction probabilities within the detector arrays. The analytical matrix is computed using techniques similar to those described in (Leroux 2014), which are based on cylindrical image representation allowing the same matrix coefficients to be reused among the symmetric lines of response. The system matrix is pre-computed once with high precision and can be fully loaded in memory to achieve fast and accurate reconstruction. Corrections for detector efficiency and random events were included in the reconstruction model. Unless otherwise specified, images were reconstructed using 32 MLEM iterations with a voxel size of 0.3 × 0.3 × 0.3 mm³.

2.4. Physical performance

The spatial resolution, the count rate performance and the sensitivity of the mouse version of the LabPET II scanner were evaluated following the NEMA NU4–2008 standard (NEMA 2008).

2.4.1. Spatial resolution

The spatial resolution was measured using a 0.18 MBq ²²Na point source of 0.25 mm diameter enclosed in a 1 cm³ acrylic cube (Eckert & Ziegler 2010). Measurements were performed with the source located at the axial center of the FOV for several radial positions from 0 to 20 mm from the radial center in 5 mm increments. Measurements were repeated with the source placed at one fourth of the axial FOV, or 12.6 mm from the axial FOV center. Acquisition of a 50 mm diameter flood phantom filled with ¹⁸F-FDG was also performed and the flood data was summed to the point source data to add a uniform background before reconstruction, with an average contrast of 10%, in order to minimize the bias induced by the iterative reconstruction of such point sources (Gong et al. 2016). The spatial resolution was measured as the full width at half maximum (FWHM) and full width at tenth maximum (FWTM) in the radial, tangential and axial directions of the point source profiles in the reconstructed image.

2.4.2. Count rate performance

The count rate performance was measured using the NEMA mouse-like phantom consisting of a 70 mm long by 25 mm diameter polyethylene cylinder with a 3.2 mm diameter line source located at 10 mm from the cylinder center and parallel to the central axis. Measurements were performed using a 3.2 mm I.D. capillary tube filled with ¹¹C (τ_1/2 = 20 min) having an initial activity of 200 MBq. Four acquisitions were performed every half-life with a minimum of 500,000 prompt coincidences recorded for each measurement. Counts from lines of response (LORs) located further away than 8 mm from the edges of the phantom in the sinogram were set to zero for the analysis. For each acquisition, the total count rate (R_TOT) and the true (R_true) coincidence count rate were measured as a function of the activity as

$$R_{\text{TOT}}=\frac{C_{\text{TOT}}}{T_{\text{acquisition}}},$$ (1)

$$R_{\text{true}}=\frac{C_{\text{TOT }}-C_{\text{scatter }+\text{random}}}{T_{\text{acquisition }}},$$ (2)

where C_TOT is the total recorded coincidences, C_{scatter+random} the number of scatter plus random counts and T_acquisition the acquisition time. The system scatter fraction (SF) was measured using the phantom described above. The SF data was obtained using a source with a low activity (0.1 MBq) to avoid the randoms and only consider the trues and scatters. The scatter counts were evaluated as the sum of all LOR counts located 7 mm away from the source and the fractional values of the linear interpolation obtained between the edges of a central 14 mm wide strip. The SF was computed from the ratio

$$\text{SF}=\frac{R_{\text{scatter }}}{R_{\text{true }}+R_{\text{scatter }}}.$$ (3)

The random rate (R_random), scatter rate (R_scatter) and noise equivalent count rate (NECR) were computed as

$$R_{\text{random }}=R_{\text{TOT}}-\frac{R_{\text{true }}}{1-SF},$$ (4)

$$R_{\text{scatter }}=R_{\text{TOT }}-R_{\text{true }}-R_{\text{random }}.$$ (5)

$$\text{NECR}=\frac{R_{\text{true }}^2}{R_{\text{true }}+R_{\text{scatter }}+R_{\text{random }}}.$$ (6)

2.4.3. Sensitivity

The absolute sensitivity of the scanner was measured using the ²²Na point source described earlier. Data were acquired with the source centered in the axial and radial FOV of the scanner and then moved in 1 mm axial increments to cover the entire axial FOV of the scanner. Acquisitions of 3 min were performed to collect at least 100,000 prompt coincidences at each source location with a random rate of less than 5%. An acquisition of 6 min without any source was performed to estimate the background count rate due to the natural radioactivity of ¹⁷⁶Lu contained in the LYSO scintillators. For the sensitivity measurement, all pixels further than 1 cm of the highest pixel value in the sinogram were set to zero and the counts in all pixels were then summed. The absolute sensitivity was calculated as

$$S=\frac{R_{\text{true }}-R_B}{A_{cal}}\times \frac{100}{\beta },$$ (7)

where R_true is the true event rate, R_B the background coincidence rate due to the LYSO intrinsic radioactivity, A_cal the ²²Na source activity and β the branching ratio of 0.906 for positron decay of the ²²Na source.

2.5. Imaging performance

2.5.1. NEMA image quality phantom

The NEMA NU4 image quality phantom was used to assess the uniformity, partial volume effect (PVE) and spill-over ratios (SOR). The phantom is a 30 mm diameter cylinder consisting of three parts including a 15 mm long uniform region, five 20 mm long fillable cylinders having internal diameters from 1 to 5 mm and two cold cylindrical chambers (15 mm long and 8 mm in diameter) filled respectively with water and air. The phantom was filled with ¹⁸F-FDG calibrated at 6.3 MBq (303 kBq/cc) at the beginning of the acquisition. The phantom was placed in the center of the transaxial FOV of the scanner and data were acquired for 60 min. The relative mean, maximum, minimum and percentage standard deviation (%STD) values were measured in the uniform region using a 22.5 mm diameter by 10 mm long cylindrical volume of interest (VOI), drawn at the center of the uniform region. The PVE was evaluated using cylindrical VOIs around each cylinder with a diameter twice as large as the physical rod diameter for a 10 mm slice. For each rod size, the maximum value was measured from the reconstructed image and divided by the mean value obtained in the uniform region to obtain the corresponding recovery coefficient. The spill-over ratios were calculated using the ratio of the mean values in the air and water regions, measured using a 4 mm diameter and 7 mm long VOI, to the mean of the uniform region. Standard deviation values for PVE and SOR were calculated using the properties of uncertainty propagation for linear combinations.

2.5.2. Ultra-Micro Hot Spot phantom

A Ultra-Micro Hot Spot phantom (Data Spectrum Corporation 2006) was used to evaluate the reconstructed spatial resolution of the scanner. The phantom consists of six sectors, with rods of 0.75, 1.0, 1.35, 1.7, 2.0 and 2.4 mm diameter, enclosed in a 28 mm diameter cylinder. The phantom was filled with ¹⁸F-FDG with an activity concentration of 5.2 MBq/cc (17.3 MBq) at the start of a 2-hour acquisition. After reconstruction, the smallest resolved region was found and documented. The spatial resolution was assessed quantitatively as the smallest sector where the average valley-to-peak ratios of the line profiles through the spots were below 0.735, following the Rayleigh criterion (McKechnie 2016, Hallen et al. 2020).

2.5.3. Animal imaging

Whole-body images of mice were obtained to assess the image quality of the scanner using different radiotracers. A 19.2 g healthy mouse was initially injected with 27 MBq of Na¹⁸F and imaged at 75 min post injection during 45 min using three overlapping bed positions. A ¹⁸F-FDG whole-body study was then performed using a 17 g healthy mouse imaged with an activity of 6.7 MBq at 150 min post injection and after a 30-min awake period. A cardiac-gated study was performed with a 19.5 g healthy mouse injected with 20 MBq of ¹⁸F-FDG and imaged at 30 min post injection during 15 min. Finally, images of the mouse brain were obtained using ¹⁸F-FMPEP-d2, a radiotracer for PET imaging of the cannabinoid CB1 receptors (Terry et al. 2009). A 23.4 g mouse was imaged for 30 minutes 2 hours post injection. The activity at the start of the scan was 11.0 MBq. All animal procedures were performed in compliance with the policies and directives of the Canadian Council on Animal Care and were approved by the Ethics Committee for Animal Care of the Université de Sherbrooke under Protocol 199–18R.

3. Results

3.1. Spatial resolution

Figure 2 shows the reconstructed image after 32 MLEM iterations, of point sources placed at the axial center of the FOV and five radial positions, along with their radial line profiles. Spatial resolution results measured along the radial, tangential and axial profiles in the reconstructed images are reported in Table 2. The volumetric resolution, calculated as the product of radial, tangential and axial FWHM resolution, is displayed as a function of the radial offset in Figure 3. The data show that a submicroliter volumetric resolution is achieved within a 30 mm diameter FOV, which is adequate for whole-body mouse imaging.

Figure 2.

Reconstructed image of a point source placed at different radial positions with 32 MLEM iterations (top) and line profiles through the point sources (bottom).

Table 2.

Radial, tangential and axial reconstructed resolutions (FWHM and FWTM) at 0, 5, 10, 15 and 20 mm from the center of the FOV and for two axial positions.

At axial center
	0 mm		5 mm		10 mm		15 mm		20 mm
	FWHM	FWTM	FWHM	FWTM	FWHM	FWTM	FWHM	FWTM	FWHM	FWTM
Radial	0.78	1.41	0.84	1.53	1.08	1.98	1.41	2.58	1.69	3.09
Tangential	0.88	1.61	0.83	1.52	0.87	1.59	0.92	1.68	1.08	1.97
Axial	0.73	1.32	0.72	1.31	0.77	1.41	0.79	1.44	0.81	1.47
At ¼ axial FOV from center
Radial	0.74	1.35	0.82	1.49	1.02	1.86	1.31	2.39	1.59	2.90
Tangential	0.87	1.59	0.86	1.58	0.94	1.71	0.95	1.74	1.07	1.95
Axial	0.60	1.09	0.61	1.11	0.70	1.28	0.70	1.28	0.74	1.35

Figure 3.

Volumetric resoluction (μl) as a function of radial offset.

3.2. Count rate performance

The count rates as a function of the activity are presented in Figure 4. A maximum NECR value of 61.1 kcps is obtained at 117 MBq. The SF was evaluated from equation 3 at 22.4%. The true coincidence rate stays higher than the random coincidence rate up to 40 MBq, which is more than what is typically injected in small animals.

Figure 4.

Count rates as a function of the activity.

3.3. Sensitivity

Analyzes were performed following the NEMA NU4 standard, hence both the background activity induced by the LYSO intrinsic radioactivity and the events located further away than 1 cm of the highest pixel intensity were removed. The scanner maximum sensitivity was measured at 1.22% with an energy window of 350–650 keV. When enlarging the energy window to 250–650 keV, the maximum sensitivity increases to 2.14%. Figure 5 shows the axial sensitivity profile for the two energy windows.

Figure 5.

Axial sensitivity profiles for energy windows of 250–650 keV and 350–650 keV.

3.4. NEMA image quality phantom

The reconstructed uniformity, recovery coefficients and spill-over regions are shown in Figure 6. Recovery coefficients as a function of the object size are displayed in Figure 7 Quantitative data are reported in Table 3. Recovery coefficients of more than 75% are obtained for object size as small as 2 mm.

Figure 6.

Reconstructed regions of the NEMA NU4 image quality phantom: (a) transverse slice of the recovery coefficients region; (b) transverse slice of the uniform region; (c) transverse slice of the water and air spill-over region; (d) coronal slice through the center of the phantom and (e) radial line profile across the uniform region.

Figure 7.

Recovery coefficients extracted from the reconstructed image of the NEMA NU4 image quality phantom.

Table 3.

Uniformity, recovery coefficients and spill-over ratios using the NEMA NU4 image quality phantom.

Uniformity		STD (%)	2.4
Recovery	Object size (mm)	1	0.24 ± 0.01
coefficients		2	0.75 ± 0.02
		3	0.91 ± 0.02
		4	0.91 ± 0.02
		5	0.91 ± 0.02
Spill-over		Water	9.9 ± 0.2
		Air	19.1 ± 0.5

3.5. Ultra-Micro Hot Spot phantom

Figure 8 shows the reconstructed image of the Ultra-Micro Hot Spot phantom using 16 iterations and 8 subsets OSEM. Results of the valley-to-peak ratios for the phantom are reported in Table 4. The 0.75 mm hot spots are resolved with a mean valley-to-peak ratio of 0.672 ± 0.091, slighly below the Rayleigh criterion of 0.735 (McKechnie 2016).

Figure 8.

Transverse view of a 0.3 mm slice (36×10⁶ events) obtained from the Ultra-Micro Hot Spot phantom reconstructed using 16 iterations and 8 subsets. The line profiles used for computing the valley-to-peak ratios are overlaid on the right image.

Table 4.

Mean valley-to-peak ratios for the six regions of the Ultra-Micro Hot Spot phantom obtained from line profiles through the hot spots of the reconstructed phantom.

Object size (mm)	Mean valley-to-peak ratios
0.75	0.672 ± 0.091
1.0	0.329 ± 0.069
1.35	0.203 ± 0.040
1.7	0.159 ± 0.048
2.0	0.107 ± 0.038
2.4	0.068 ± 0.009

3.6. Animal imaging

Figures 9a and 9b display coronal and sagittal views of the reconstructed images of the 19.2 g mouse injected with Na¹⁸F using 32 MLEM iterations. Submillimetric details such as sternocostal joints, vertebrae and small bone structures of the skull and knee articulations can be differentiated from adjacent structures with a high level of detail. Reconstructed ¹⁸F-FDG whole-body images of the 17 g mouse using 30 iterations are displayed in figures 9c and 9d. The bladder, heart, brown fat behind the vertebra and harderian glands can be clearly seen.

Figure 9.

Volume-rendered images of mice. Measurements were performed using (a-b) a 19.2 g mouse injected with 27 MBq of Na¹⁸F and imaged during 45 min at 75 min post injection, and (c-d) a 17 g mouse imaged with a ¹⁸F-FDG activity of 6.7 MBq at 150 min post injection and after a 30-min awake period.

Figure 10 demonstrates the scanner capabilities for ¹⁸F-FDG gated cardiac imaging of a 19.5 g mouse where both the right and left ventricles can be resolved. End-diastolic and end-systolic ventricular volumes of 25.3 and 4.6 μl were measured, as well as an ejection fraction of 82% in this healthy mouse. Figure 11 shows reconstructed views, obtained after 32 MLEM iterations, of the head of a 23.4 g mouse injected with ¹⁸F-FMPEP-d2. Structures of the brain can be easily discriminated such as the lobes and cortex (transverse view), the hemispheres (coronal view) and the cerebellum (sagittal view).

Figure 10.

Reconstructed views of a gated mouse cardiac study at end diastole (top) and end systole (bottom). The 19.5 g mouse was administered with 20 MBq of ¹⁸F-FDG and imaged during 15 min starting 30 min post injection.

Figure 11.

Reconstructed views of the brain of a 23.4 g mouse administered with the cannabinoid CB1 receptor radiotracer ¹⁸F-FMPEP-d2. The static image was recorded for 30 min starting 2 hours post injection when the the injected activity was 11 MBq.

4. Discussion

In this paper we present a complete evaluation of the mouse version of the LabPET II scanner. The LabPET II was designed to achieve submillimetric and submicroliter volumetric spatial resolution using a unique detector design that allows truly individual scintillator-to-detector coupling interfaced to highly integrated parallel electronics.

The spatial resolution was reported using iterative reconstruction instead of the filtered backprojection as recommended by NEMA NU4, to avoid the artifacts due to aliasing and irregular sampling. However, following the recommendation of (Gong et al. 2016), a uniform background was added to minimize the bias induced by the iterative reconstruction. A submillimetric resolution and high contrast images were obtained for both phantom and animal studies, independently of the axial position. The reconstructed image of the Ultra-Micro Hot Spot phantom shows that 0.75 mm hot spots can be resolved with a valley-to-peak ratio of 0.672 ± 0.091, which is below the resolving threshold of 0.735 defined by the Rayleigh criterion. Excellent recovery coefficients were measured with the NEMA NU4 image quality phantom, providing evidence of the high contrast capabilities of the scanner. Since no attenuation and scatter corrections were performed during reconstruction, higher spill-over ratio was observed for the air compartment compared to the water compartment. Appropriate subtraction of the scatter events should help achieve better contrast, especially in the water region.

A sensitivity of 2.14% was measured with an energy window of 250–650 keV, while following GATE simulations, a sensitivity of 3% was expected for the same energy and time window. Some reasons can be given to explain this discrepancy. No corrections were performed to compensate for defective or desactivated pixels, which amounted to approximately 4% at the time of the sensitivity measurement. Moreover, since the LabPET II technology is based on time-over-threshold signal processing, the ToT to energy conversion is non-linear. In this study, the non-linearity calibration of energy was performed globally using average correction factors for the reported measurements. Hence the lower energy threshold may have been higher than expected for a number of detectors due to inaccurate energy assignment. Using individual ToT-energy calibrations per channel should slightly change the lower threshold of the energy window, thus impact the sensitivity and count rate performance of the scanner. No significant changes are expected for spatial resolution and image quality. However, at the time of the measurements this procedure was not implemented in the scanner and a global correction was deemed sufficient. Further investigations to recalibrate the scanner using more advanced non-linearity correction (Gaudin et al. 2020) are expected to reduce this discrepancy. Finally, the coincidence processor rejects all triple or higher order coincidences that cannot be assigned a single unambiguous line-of-response (LOR). This is a deliberate choice to maximise the resolution performance. However, alternate coincidence processing scheme can be implemented with a 1:1:1 pixelated detector front-end to recover those inter-crystal scatter events and maximise sensitivity (Rafecas et al. 2003, Pratx & Levin n.d., Michaud et al. 2015, Geoffroy et al. 2015).

Small animal studies were performed for whole-body, cardiac and brain imaging achieving high resolution and high contrast reconstructed images. The mouse version of the LabPET II is the first documented preclinical PET to resolve such high detail mouse structures (sternocostal joints, articulations…). Cardiac and brain studies showed promising results for future quantitative studies of small structures of the heart and the brain for applications such as myocardial blood flow or tumor heterogeneity evaluation. Dual modality imaging in combination with a computed tomography scan would facilitate in vivo image interpretation as well as provide a way to correct for attenuation and scatter coincidences.

In summary, with a true submillimetric spatial resolution, a sensitivity of 1.2% and a maximum NECR of 61.1 kcps obtained using an energy window of 350–650 keV, the LabPET II offers excellent physical and imaging capabilities for small animal imaging such as mice or small rodents.

5. Conclusion

A new generation of preclinical PET scanner featuring truly pixelated detectors was developed to achieve a spatial resolution approaching the physical limit of PET imaging. A complete assessment of the physical and imaging performance was performed based on the NEMA NU4 standard with extension to more quantitative measurements regarding the assessment of spatial resolution. Submillimetric radial spatial resolution was achieved within a diameter of almost 20 mm, while submicroliter volumetric resolution was demonstrated within a 30 mm diameter field-of-view, which is adequate for mouse imaging. While results were obtained for the mouse version, the LabPET II was designed as a modular technology; the reported results indicate strong evidence of the promising capabilities of the LabPET II platform for small-animal to human brain biomolecular imaging using the same detection technology.