Performance evaluation of the ACTIVE 7 MAX benchtop preclinical PET scanner in accordance with the NEMA NU 4-2008 standard

Abstract

Introduction

ACTIVE 7 MAX is a compact benchtop preclinical PET scanner dedicated to high sensitivity and high-resolution imaging of small animals. This study evaluated the performance of the ACTIVE 7 MAX system using the National Electrical Manufacturers Association NU 4-2008 standard protocol.

Methods

The scanner consists of four rings, each containing 14 detector modules. Each detector module is made up of a 16 × 16 array of lutetium yttrium orthosilicate (LYSO) scintillation crystals, with each crystal measuring 1.47 × 1.47 × 10.0 mm³. The crystal array was coupled to a novel 6 × 6 epitaxial-quenching-resistor silicon photomultiplier (EQR SiPM) array. Flood images and energy resolution were obtained by irradiating each detector module with a ¹⁸F source.

Results

The average energy resolution for the 56 detector modules in the system was found to be 11.46% in Full Width at Half Maximum (FWHM). The filtered back-projection (FBP) image spatial resolution of a point source varied from 1.57 to 1.81 mm in the radial direction and from 1.58 to 1.71 mm in the tangential direction within the radius of 25 mm. For the Derenzo phantom imaging, the hot rod with a diameter of 1.25 mm was identified. With an energy window of 350–650 keV, the sensitivity at the center of the scanner was 4.18%. The peak noise equivalent count rate (NECR) of 103 kcps was achieved at 22 MBq, and the scatter fraction (SF) is 12%. The reconstructed images of the NEMA image quality phantom show a uniformity of 5.0%, and recovery coefficients for rods with diameters of 1 mm and 5 mm ranging from 0.14 to 0.93. The spillover rates for air-filled and water-filled cold regions were 0.08 and 0.03, respectively.

Conclusion

This study evaluated the performance of the ACTIVE 7 MAX preclinical PET system. The results demonstrated excellent imaging performance for molecular imaging in biomedical studies.

Introduction

Positron emission tomography (PET) is a sophisticated non-invasive molecular imaging modality with widespread applications across various fields, including diagnosis of malignancies, neurological disorders, and cardiovascular diseases, as well as in radiopharmaceutical research [1–5]. The advancement of preclinical PET technology plays a critical role in supporting these disciplines, while increasing demands from these applications, in turn, drive continuous innovation in preclinical PET systems [6–11]. In recent years, preclinical imaging has gained prominence in biological research and has become a focal point for preclinical studies, prompting the development of novel PET systems to accommodate diverse research demands [12–16].

In response to these evolving demands, Insight Imaging Healthcare, Inc. has introduced ACTIVE 7 MAX, a benchtop preclinical PET system designed for high-resolution small-animal imaging. The system features a field of view (FOV) with a transaxial diameter of 120 mm and an axial length of 106 mm, enabling whole-body imaging of mice in a single bed position.

A key innovation in ACTIVE 7 MAX system is the incorporation of a novel epitaxial-quenching-resistor silicon photomultiplier (EQR SiPM) array [17–19], developed at our institute. Unlike conventional SiPMs with discrete resistors per microcell for quenching, EQR SiPMs achieve quenching through the bulk resistance of an epitaxial silicon layer. This design enables direct signal collection at a common anode plane, allowing for the fabrication of large SiPMs while maintaining a fast recovery time. Each EQR SiPM consists of microcells with a 20 µm pitch, offering a photon detection efficiency (PDE) exceeding 50%, a gain of 10⁶ at 5 V overvoltage, a low breakdown voltage of 27.5 V and enhanced radiation resistance. In addition to this advanced detector technology, ACTIVE 7 MAX features a GPU-accelerated, list-mode PET reconstruction platform [20, 21]. The reconstruction software integrates comprehensive correction algorithms, including attenuation, normalization, randoms, scatter, and point spread function (PSF) modeling, significantly improving computational efficiency and enabling accurate, high-quality quantitative image reconstruction [22, 23].

These advancements position ACTIVE 7 MAX as a compact yet highly capable preclinical PET system, offering improved imaging performance to support cutting-edge research in molecular imaging and biomedical sciences. This study aims to evaluate the overall performance of the ACTIVE 7 MAX system. Where applicable, the National Electrical Manufacturers Association (NEMA) NU 4–2008 protocol was referenced [24], and studies were conducted on energy resolution, spatial resolution, scatter fraction and count-rate performance, sensitivity, and image quality based on the specific conditions. Furthermore, in vivo imaging studies of a mouse labeled with radiotracers were conducted and representative animal images were shown to demonstrate the exceptional capability of the ACTIVE 7 MAX for high-sensitivity, high-resolution molecular imaging.

Materials and methods

System description

ACTIVE 7 MAX PET scanner consists of four rings, each containing 14 detector modules. The system has a radial distance of 120 mm from crystal surface to surface and an axial length of 106 mm, as shown in Fig. 1, and the Characteristics of the scanner are summarized in Table 1. The system adopts an integrated design of the mechanical structure, wherein the detector module is precisely aligned and securely mounted to the system frame. The LYSO crystals were polished and optically isolated via 0.075 mm thick enhanced specular reflectors (ESR) and the outside of the crystal array was also covered using ESR and BaSO₄ reflectors.

Fig. 1.

Schematic of the ACTIVE 7 MAX preclinical PET system. The detector module, composed of LYSO crystal coupled with EQR SiPM array, is precisely fitted to the system frame

Table 1.

Characteristics of ACTIVE 7 MAX system

Characteristic	Data
Crystal material	LYSO
Crystal size	1.47 × 1.47 × 10.0 mm3
Crystal pitch	1.54 mm
Number of rings	4
Number of modules per ring	14
Number of crystals per module	256
Total LORs	41,287,680

A novel SiPM device (EQR20-11-3030D-S), developed at our institute, was employed in the PET system. Each SiPM has a physical size of 3.75 × 3.55 mm² and an active area of 3.00 × 3.00 mm². A 6 × 6 SiPM array was assembled with horizontal and vertical pitches of 0.58 mm and 0.82 mm, yielding a total array dimension of 25.4 × 25.4 mm². Crystal arrays were optically coupled to SiPM arrays using optical glue. Light guides of 1–4 mm thickness were evaluated in combination with 16 × 16 and 24 × 24 crystal arrays. Flood image decoding tests, assessed by peak-to-valley ratio, showed that light guides thicker than 1 mm did not improve decoding performance and instead degraded spatial resolution. Optimized results were obtained with the 16 × 16 array and 1 mm light guides, which provided accurate decoding and preserved resolution, whereas the 24 × 24 array consistently exhibited poor edge decoding. Considering decoding quality, resolution, and cost, the 16 × 16 crystal array with 1 mm light guides was adopted for system integration. Furthermore, leakage currents of the novel EQR SiPMs were characterized at different operating temperatures (Table 2). The results showed that the leakage current remained below 2 μA at temperatures lower than 30 °C, confirming the thermal stability of the devices and ensuring reliable performance under practical operating conditions.

Table 2.

Real-time leakage current data of EQR SiPM

Bias voltage /V	Temperature/℃	Voltage /mv	Resistance to ground /kΩ	Leakage current /μA
32.44	28	15	10	1.5
	30	16	10	1.6
	35	21	10	2.1
	40	27	10	2.7
	45	33	10	3.3
	50	39	10	3.9
	55	55	10	5.5

Front-end readout electronics were implemented on a printed circuit board directly connected to each SiPM array. A capacitive multiplexing circuit reduced the 36 SiPM channels to four position-encoded outputs. Signals from 56 detector modules were acquired using a 256-channel DAQ system (TOFTEK, Wuxi, China), supporting both single-photon and coincidence acquisition. This centralized DAQ architecture, while effective for our system, contrasts with fully digital, distributed data acquisition strategies employing Field-Programmable Gate Arrays (FPGAs) for individual detector processing, which can offer enhanced count-rate performance and flexibility [25]. In coincidence mode, the time window was set to 3 ns, and a 30 ns delay circuit enabled random correction via delayed coincidences. Additionally, a geometric window was applied: coincidences were accepted only when crystals belonged to modules separated by at least two intervening modules, excluding interactions with four immediate neighbors. Under these conditions, the system supported a total of 41,287,680 valid lines of response (LORs).

Energy resolution

A cylindrical phantom (diameter: 100 mm, length: 120 mm) filled with approximately 3.7 MBq ¹⁸F solution was positioned at the center of the field of view (CFOV) of the PET scanner to acquire a two-dimensional flood irradiation for each detector module. The four-channel readout signals from each detector module were processed using Anger logic to determine the crystal positions.

Detector look-up tables (LUT) were generated through a custom automatic peak-finding algorithm, followed by manual correction. These look-up tables were then used to delineate the regions corresponding to individual crystals within the flood image. Energy spectra of each crystal pixel was extracted based on the look-up table, and a Gaussian function was fitted to the full-energy peak to determine the photopeak position. The energy resolution of each crystal was calculated as the full width at half maximum (FWHM) of the Gaussian function, normalized to the photopeak energy at 511 keV, and expressed as a percentage. Channel–energy calibration was performed individually for all crystals using three isotopes (⁹⁹ᵐTc, ¹³³Ba, and ⁶⁸Ga), as illustrated in Fig. 2. The results demonstrated that within the energy range below 511 keV, which is the focus of this study, the response exhibited sufficient linearity to enable accurate calibration and subsequent analyses.

Fig. 2.

Channel-energy relationship

Spatial resolution

In the absence of a ²²Na point source with a nominal size of 0.3 mm embedded in a 1-cm³ piece of acrylic as recommended by NEMA NU 4 standard [24], a sub-0.5 mm anion-exchange resin bead was agitated in an ¹⁸F-containing solution for spatial resolution measurements. The source was positioned at two axial locations: the axial CFOV and at one-fourth of the axial FOV. At each axial position, the source was placed at radial offsets of 0, 5, 10, 15, and 25 mm from the center. Five-minute acquisitions were performed per position, with each measurement accumulating more than 10⁵ prompt counts. The acquired data were sorted into sinograms using single-slice rebinning (Maximum number of axial cross-planes = 6) and reconstructed via filtered back-projection (FBP) algorithm.

This methodology primarily accounts for hardware-level limitations and facilitates benchmark comparison of different systems. However, from a statistical perspective, iterative reconstruction algorithms extract more information from raw data does FBP [14]. To address the potential overestimation of air-filled point-source resolution by iterative methods [26], actual image quality was further validated using a micro-Derenzo phantom. This phantom contains six sets of hot rods with different sizes (diameters of 0.8, 1.0, 1.25, 1.5, 2.0, and 2.5 mm), with a center-to-center rod spacing of twice the rod diameter. An 18.5 kBq solution of ¹⁸F was injected, and data were collected for 30 min. The acquired data were reconstructed using the list-mode OSEM algorithm [27, 28].

Sensitivity

To measure the sensitivity, a ²²Na point-like source with a diameter of 1.0 mm and a thickness of 1.0 mm, was positioned axially at the center of the FOV. The point source was stepped through the axial FOV in 1 mm increments from one end to the other to obtain a comprehensive sensitivity profile. The coincidence counts on each axial position were calculated to obtain the axial sensitivity distribution at the center of the FOV. At the same time, the activity was sufficiently low to ensure that the counting losses were less than 1%, and the random event rate was less than 5% of the true event rate. Data was acquired for a total of 300 s within 350–650 keV, and background rate were also collected over the same energy window and duration. Sensitivity calculation took into account the decay branching ratio of ²²Na, which is 90.6%.

Scatter and count-rate performance

The scatter and count-rate performance were evaluated using the NEMA NU 4 mouse-like scatter phantom. A flexible tube filled with ¹⁸F solution at an activity of 37 MBq was inserted into the hole, radially offset by 10 mm.

Data were collected in 31 sets over a period of 14 h, until the single event rate reached twice the intrinsic single event rate. To determine the intrinsic true count rate , we performed a 30-min scan was performed within the scanner's FOV in the absence of any radioactive source [29]. The data were processed according to NEMA NU 4. Data processing was performed in accordance with the NEMA NU 4 standard, and the scatter count rate () can be calculated using Eq. 1:

1

where , , and are the prompt, true, scatter and random event rates, respectively. The scatter fraction was measured using the last three data sets, where the activity was low enough that the effects of dead time and random events could be neglected. The scatter fraction (SF) was calculated by Eq. 2:

Random coincidence events were measured using the delayed coincidence method. All data were reconstructed into two-dimensional sinograms using the single-slice rebinning (SSRB) method [30]. The noise-equivalent count rate (NECR) for each prompt sinogram was determined using the following equation:

3

Imaging studies

Image-quality studies were conducted using the NEMA NU 4 image-quality phantom. The phantom was injected with 100 μCi of ¹⁸F solution, and data were collected for 20 min. The percentage standard deviation (%STD), recovery coefficient (RC), and spill-over ratio (SOR) were calculated according to the NEMA NU 4 standards. The known anatomical structure of the phantom was drawn for attenuation correction and scatter correction of the PET data.

To further demonstrate the imaging capabilities of this scanner, a tumor-bearing mouse weighing 21.4 g was scanned. The mouse was anesthetized with 2% isoflurane in oxygen and injected with 12.14 MBq of ¹⁸F-FDG via the tail vein. Thirty minutes after the injection of the tracer, PET data were acquired for 20 min with an energy window of 350–650 keV. The data were reconstructed using the OSEM method. Since no registered CT images were available, attenuation and scatter corrections were not applied.

Image reconstruction

A general PET reconstruction framework that incorporates various data corrections and supports accurate quantitative reconstruction was developed for PET systems. The image reconstruction software is based on the list-mode OSEM algorithm running on multiple GPUs [20], enabling real-time computation of the system matrix. Attenuation correction was performed using a forward projection of the attenuation coefficient map, based on the known geometry of the object. A component-based normalization correction was applied to account for intrinsic crystal efficiency, count rate variations, and geometric factors influencing the sensitivity of lines of response [31]. For random coincidence correction, delayed coincidence events were acquired using a delayed time window, and the variance reduction technique was implemented to mitigate statistical noise, thereby improving the quality of the reconstructed image. Scatter correction was performed using the single scatter simulation algorithm [32, 33], which is widely adopted in clinical PET imaging.

Results

Energy resolution and flood image

Figure 3A shows the flood image of one representative detector module, where each crystal pixel in the 16 × 16 array can be clearly identified. Based on the flood image, the energy resolution for each crystal pixel was calculated. Figure 3B shows the energy spectrum of a representative crystal pixel from a ²²Na source. The average energy resolution for the 56 detector modules in the system was found to be 11.46% (FWHM) at 511 keV.

Fig. 3.

Flood image A acquired from a single detector module and the ²²Na energy spectrum B measured from one crystal within this module

Spatial resolution

Figure 4A and B show the FBP reconstructed image spatial resolution in the radial, tangential, and axial directions measured in the central and one fourth axial transverse plane. The spatial resolution of the point source varied from 1.57 to 1.81 mm in the radial direction and from 1.58 to 1.71 mm in the tangential direction within the radius of 25 mm.

Fig. 4.

FBP reconstructed image spatial resolution of the point source, showing FWHM and full width at tenth of maximum (FWTM) of radial, tangential, and axial image resolution at axial CFOV (A) and at one-fourth of the axial FOV (B)

Figure 5A shows the transverse axis image of the Derenzo hot rod phantom. A set of rods with a diameter of 1.25 mm and a center-to-center spacing of 2.5 mm is clearly identified when the data were reconstructed with 3D ordered-subset expectation maximization. For the 1 mm diameter rods, complete resolution remains challenging due to the absence of depth-of-interaction correction, point spread function modeling, and mechanical calibration. We plotted the line profiles of three 1.0 mm-diameter rods (Fig. 5), obtaining an average peak-to-valley ratio of 2.18.

Fig. 5.

A Image of the Derenzo phantom. The set of rods with a diameter of 1.0 mm is clearly identified. B Line profile of three sources with a diameter of 1.0 mm along the yellow line

Sensitivity

Figure 6 presents the axial absolute sensitivity profile measured with a point-like source positioned at the center of the system axially. Using an energy window of 350–650 keV, the peak absolute sensitivity at the scanner's center was 4.18%. Due to the relatively small diameter characteristic of preclinical PET systems, the effects of individual detector rings and inter-ring gaps become more pronounced, leading to the appearance of multiple secondary peaks in the sensitivity curve.

Fig. 6.

Axial absolute system sensitivity as function of axial position with an energy window of 350 – 650 keV

Scatter and count-rate performance

Figure 7 plots the count rate and NECRs as functions of the activity in the scatter phantom, including prompt, true, random, and scatter rates. The prompt and true rates peaked at around 28 MBq of total activity. With this activity, the random counts accounted for approximately 5.7% of the total prompt counts. The peak noise equivalent count rate (NECR) of 103 kcps was achieved at 22 MBq. The resultant scatter fraction was 12%.

Fig. 7.

Count rates of ACTIVE 7 MAX as a function of radioactivity concentration for the mouse-like phantom

Imaging studies

Figure 8 shows views of the reconstructed NEMA image quality phantom, including the transverse and coronal views of the uniform region, as well as the transverse view of the five sets of hot rods. Reconstruction was performed using the OSEM method, with an image voxel size of 0.5 × 0.5 × 0.5 mm³. Random, attenuation, scatter correction, and normalization were applied. Figure 8D shows a profile across the uniform area. The average activity concentration, the maximum and minimum values were 0.788, 0.610, and 0.987, respectively, with the %STD was 6.0%. The RCs for the five sets of rods with diameters ranging from 1 to 5 mm were 0.14 ± 0.02, 0.59 ± 0.05, 0.91 ± 0.07, 0.89 ± 0.06, and 0.93 ± 0.08, respectively. The SORs for the water-filled and air-filled cold regions were 0.08 and 0.03, respectively, with %STDs of 11.2% and 14.9%.

Fig. 8.

Reconstructed images of the small-animal NEMA image quality phantom scanned for 20 min: A transverse slice of the uniform region, where the yellow line corresponds to the profile position plotted in (D); B coronal slice; C transverse slice of the rod region; D profile across the uniform area

The reconstructed images of the mouse are shown in Fig. 9, clearly showing the distribution of the radioactive tracer within the body. Since only PET images were available and no CT images were used, attenuation and scatter corrections were not applied. Figure 9A and B show two views of the results of mouse, where organs and tissues such as the heart, spine, and bladder are clearly identifiable. The mouse has a tumor implanted in the axilla, approximately 4 mm in diameter. The reconstructed image of the tumor's transverse view is shown in Fig. 9C, with the tumor's edge clearly delineated.

Fig. 9.

Reconstructed images of tumor-bearing mouse injected with ¹⁸F-FDG. A transverse view; B sagittal view; C transverse view of the tumor

Discussion

This study evaluated the system performance of the ACTIVE 7 MAX preclinical PET scanner under both practical imaging conditions and in accordance with the NEMA NU 4 standards. Key performance metrics assessed included energy resolution, count rate performance, system sensitivity, and image quality, providing a comprehensive characterization of the system’s capabilities for preclinical molecular imaging applications.

The ACTIVE 7 MAX system incorporates a novel SiPM technology developed by our research group and manufactured by Capital Photonics. This SiPM adopts an innovative epitaxial-quenching-resistor design, which provides high PDE, robust radiation resistance, and excellent signal consistency [17–19]. The device also features a low breakdown voltage of 27.5 V, contributing to improved operational efficiency and stability. Each SiPM measures 3.75 × 3.55 mm², with an active area of 3.00 × 3.00 mm². To ensure precise alignment with the crystal array, a 6 × 6 SiPM configuration was implemented, featuring horizontal and vertical inter-SiPM spacings of 0.58 mm and 0.82 mm, respectively, resulting in a total array size of 25.4 × 25.4 mm². Given the relatively large gaps between adjacent SiPMs, the crystal size was selected to optimize the balance between spatial resolution, system sensitivity, and crystal decoding performance. To further reduce the dead area and enable the use of smaller crystal elements, SiPMs employing through-silicon via (TSV) packaging technology are currently under development at our institute. This advancement is expected to improve spatial resolution while preserving crystal decoding quality and overall detection efficiency.

The reflective layer between adjacent crystals is 75 μm thick, yielding a stacking fraction of 0.91. For edge crystals, a 0.4 mm thick reflective layer composed of BaSO₄, ESR, and aluminum foil is used to minimize scintillation light loss. This design also ensures that the effective area of the edge crystals is slightly smaller than that of the SiPM array, thereby reducing the angular loss of incident photons and enhancing edge decoding accuracy. A light guide is placed between the crystal array and the SiPM to balance light transmission efficiency and decoding uniformity. The system underwent temperature coefficient calibration, and all detector modules demonstrated stable performance across the typical laboratory temperature range of 20–35 °C.

In addition to imaging performance, cost efficiency is a critical consideration for the commercial viability of the system. Achieving a balance between high-performance imaging and affordability is essential to facilitate widespread adoption. The system does not incorporate time-of-flight (TOF) technology and depth of interaction (DOI) correction. For preclinical PET systems, the imaging region in preclinical PET systems is typically only a few centimeters. The TOF capability imposes more stringent requirements on detectors while providing limited improvement in image quality. And existing DOI methods—such as dual-end readout and light sharing—necessitate additional photodetectors and a redesign of the readout electronics, significantly increasing costs [34–36]. Similarly, application-specific integrated circuit (ASIC) technology were not utilized [37–39]. In contrast, customized readout electronics, specifically designed and optimized according to the characteristics of the crystals and photodetectors, are expected to reduce costs and enhance performance. The resulting detector module demonstrated high performance in crystal decoding and energy resolution. The flood image exhibited ultra-high clarity, accurately resolving edge crystals, with an average peak-to-valley ratio of 50.5 for the central crystal row. The energy resolution at 511 keV ranged from 11.10 to 11.86%, with a mean value of 11.46%.

In this study, a consistent energy window of 350–650 keV was applied to ensure comprehensive photopeak acquisition for system evaluation. Given the relatively narrow energy resolution (~ 11%), employing a narrower energy window could enhance the true-to-scatter ratio and potentially improve image quality. Future work will systematically investigate the influence of different energy window settings on overall system performance, with an emphasis on optimizing this parameter for specific imaging tasks. Notably, our reconstruction algorithm integrates a robust scatter correction module, thereby permitting the use of a wider energy window to maximize data utilization without compromising image fidelity.

Due to the unavailability of NEMA NU-4 compliant point sources (i.e., ²²Na embedded in an acrylic cube), we did not strictly follow the NEMA NU-4 standard for spatial resolution measurements. Ideally, spatial resolution should be assessed using a point source in which all positrons annihilate within the confined volume of the source. However, if significant radioactivity is distributed on the surface of the source—as is often the case with ¹⁸F-labeled anionic resins—positrons may travel through air before annihilation, thereby increasing the positron range. This effect leads to an intrinsically non–point-like distribution of the resulting back-to-back annihilation photons, which introduces additional blurring and degrades the measured spatial resolution. Therefore, the spatial resolution obtained under these conditions represents a conservative estimate of the system’s performance. While the NEMA NU 4 standard recommends evaluating system spatial resolution using point sources reconstructed with FBP algorithm, this method tends to underestimate the practical resolution achievable with modern iterative reconstruction algorithms [40]. Furthermore, advanced reconstruction techniques utilizing non-Gaussian kernels have been shown to better model the physical processes in PET, potentially leading to further improvements in spatial resolution recovery [41]. To provide a more representative assessment, we further evaluated the imaging performance of the system using a Derenzo phantom, which features six different hole sizes arranged in triangular patterns around the phantom center and contains no background activity. Utilizing a list-mode OSEM-based PET reconstruction software, the PET scanner was able to clearly resolve rods with 1.25 mm diameter and 2.5 mm center-to-center spacing, demonstrating its excellent spatial resolution under realistic imaging conditions.

The system features an integrated design of the mechanical structure and detector module, with the detector module precisely fitted to the frame. This design ensures the stable positioning of the detector module under various conditions and reduces geometric errors caused by the mechanical design. The interior of the ring scanner consists of a cylindrical aluminum alloy structure, which forms a thin layer of uneven thickness with the surface of the detector module. Although this results in minor variations in photon attenuation, with thickness deviations of less than 1 mm, GATE-based Monte Carlo simulations confirmed that the associated attenuation effects are negligible and do not significantly impact image quality or quantitative accuracy.

In summary, the ACTIVE 7 MAX is a commercial preclinical PET system incorporating advanced SiPM technology, offering a well-balanced combination of cost-effectiveness, system performance, and application versatility. Although it is less compact than the β-eye system, it provides notably improved spatial resolution (~ 1.0 mm versus ~ 2.0 mm for the β-eye) [42]. Its full 3D acquisition capability supports high-precision imaging demands, and its approximately 10 cm radial field of view is well-suited for rat imaging studies. Nevertheless, achieving submillimeter resolution in pixelated PET systems remains a considerable challenge, as further enhancement in intrinsic resolution typically entails increased costs, greater system complexity, and trade-offs in other performance metrics [11]. Looking ahead, the integration of artificial intelligence techniques may facilitate depth-of-interaction (DOI) capability without requiring additional hardware. For example, the MOLECUBES β-CUBE system demonstrated DOI resolution of 1.6 mm using monolithic scintillators and machine learning clustering algorithms trained on dedicated datasets, highlighting the potential of data-driven approaches in next-generation PET system design [43].

In molecular imaging applications, anatomic reference information is often required to accurately localize molecular signals and facilitate quantitative analysis of tracer concentration [44–46]. Computed tomography (CT) provides fully tomographic anatomical data, improving molecular signal localization. Additionally, volumetric CT data support accurate tracer quantification by enabling essential corrections for attenuation [47], gamma-ray scattering [48], and the partial-volume effect [49]. The current system is PET-only and does not include an anatomic imaging modality. To maintain the compact design of the benchtop platform, either a compact cone-beam CT or a deep learning-based approach for generating pseudo-CT images from PET data will be integrated.

To enable quantitative PET imaging with comprehensive data correction capabilities, we established a unified reconstruction framework applicable across PET systems. The reconstruction framework is designed to accommodate non-conventional system geometries, including Virtual-Pinhole PET (VP-PET) configurations. VP-PET enables enhanced spatial resolution and sensitivity in specific regions of interest by integrating high-resolution detectors in localized areas, without compromising the overall system performance [50–54]. Integration of VP-PET technology is planned for next-generation systems, aiming to further enhance imaging performance while maintaining the compact design and cost-efficiency essential for both preclinical research and commercial deployment.

Conclusion

This study systematically evaluated the performance of the ACTIVE 7 MAX preclinical PET system. The system integrates novel SiPM technology, an optimized mechanical structure, and cost-effective design while achieving high imaging performance. The average energy resolution was measured at 11.46%, and with an energy window of 400–650 keV, the system achieved a central line source sensitivity of 4.18%. The peak NECR for a mouse-sized phantom was observed at 22 MBq. Imaging studies demonstrated robust quantitative reconstruction capabilities and the effectiveness of the implemented correction algorithms. In preclinical molecular imaging experiments, the system consistently produced high-quality images, validating its suitability for advanced preclinical research applications.

Author Contribution

Haihao Wang and Kexin Wang contributed equally to this work. Haihao Wang was responsible for the study conception, experimental design, and major experimental implementation. Kexin Wang performed data processing and analysis. Ziquan Yuan, Chenxi Li, and Runze Liao assisted with system development and validation. Yucun Hou and Jianlang Hua contributed to data acquisition and experimental setup. Yi Tang and Qing Ruan participated in result discussion and manuscript revision. Dejun Han provided technical guidance. Jianyong Jiang supervised the overall project and reviewed the manuscript. All authors discussed the results and approved the final version of the paper.

Funding

This work was supported in part by the National Natural Science Foundation of China (No. 12475336), Beijing Nova Program (No. 20230484413).

Declarations

Ethics approval

Animal studies were conducted in accordance with the Regulation on Laboratory Animals of Beijing Municipality and the guidelines established by the Ethics Committee of Beijing Normal University (Permit No.BNUCC-EAW-2023-002).

Competing interests

The authors have no relevant conflicts of interest to disclose.