Visual Abstract
Abstract
The uMI Panorama is a novel PET/CT system using silicon photomultiplier and application-specific integrated circuit technologies and providing exceptional spatial and time-of-flight (TOF) resolutions. The objective of this study was to assess the physical performance of the uMI Panorama in accordance with the National Electrical Manufacturers Association (NEMA) NU 2-2018 standard. Methods: Spatial resolution, sensitivity, count rate performance, accuracy, image quality, and TOF resolution were evaluated in accordance with the guidelines outlined in the NEMA NU 2-2018 standard. Energy resolution was determined using the same dataset acquired for the count rate performance evaluation. Images from a Hoffman brain phantom, a mini-Derenzo phantom, and 3 patient studies were evaluated to demonstrate system performance. Results: The transaxial spatial resolution at full width at half maximum was measured as 2.88 mm with a 1-cm offset from the center axial field of view. The sensitivity at the center axial field of view was 20.1 kcps/MBq. At an activity concentration of 73.0 kBq/mL, the peak noise-equivalent count rate (NECR) reached 576 kcps with a scatter fraction of approximately 33.2%. For activity concentrations at or below the peak NECR, the maximum relative count rate error among all slices remained consistently below 3%. When assessed using the NEMA image quality phantom, overall image contrast recovery ranged from 63.2% to 88.4%, whereas background variability ranged from 4.2% to 1.1%. TOF resolution was 189 ps at 5.3 kBq/mL and was consistently lower than 200 ps for activity concentrations at or below the peak NECR. The patient studies demonstrated that scans at 2 min/bed produced images characterized by low noise and high contrast. Clear delineation of nuclei, spinal cords, and other substructures of the brain was observed in the brain PET images. Conclusion: uMI Panorama, the world’s first commercial PET system with sub–200-ps TOF resolution, demonstrated fine spatial and fast TOF resolutions, robust count rate performance, and high quantification accuracy across a wide range of activity levels. This advanced technology offers enhanced diagnostic capability for detecting small and low-contrast lesions while showing promising potential under high-count-rate imaging scenarios.
The concept of using time-of-flight (TOF) information to better determine the location of positron annihilation in PET was first introduced during the early stages of PET development (1). By measuring the difference in arrival time between the detection of 2 coincident photons, it is possible to determine the annihilation position along a line of response (LOR) within a spatial uncertainty of cΔt/2, where c and Δt represent the speed of light and the timing resolution, respectively. The incorporation of TOF not only expedites PET reconstruction convergence but also reduces image noise and enhances the image signal-to-noise ratio by constraining the LORs within a smaller image volume during the image reconstruction process (2,3). In analytic image reconstruction algorithms, compared with non-TOF PET, TOF PET can achieve a relative gain in image signal-to-noise ratio equal to the square root of the object size divided by the product of speed of light and the system’s timing resolution (4–8).
The first TOF PET scanner was developed in the 1980s by coupling cesium fluoride or barium fluoride scintillators with photomultiplier tubes, achieving a TOF resolution ranging from 400 to 600 ps (9,10). However, factors such as poor stopping power and low light yield limited the use of cesium fluoride– or barium fluoride–based TOF PET systems to primarily high-count-rate brain or cardiac studies. The introduction of lutetium-based scintillation materials such as lutetium oxyorthosilicate and lutetium-yttrium oxyorthosilicate (LYSO) significantly improved sensitivity while maintaining TOF resolution ranging from 450 to 600 ps (11–16).
In recent years, silicon photomultipliers (SiPMs) have made remarkable progress as novel photodetectors with enhanced photodetection efficiency, reduced detector noise/cross talk, and improved timing performance (17). As new photodetectors with linear arrays or matrices of various sizes, SiPMs are progressively preferred as the photodetectors for the next generation of commercial TOF PET scanners. Furthermore, advancements in photodetector technology have led to the development of PET detectors with expanded coverage and reduced signal multiplexing (i.e., the number of crystals per photosensor channel), aiming to improve both timing and spatial resolutions. These detectors require low power consumption and highly integrated readout electronics that allow for the individual acquisition of charge and timing information from tens of thousands of SiPM channels. Consequently, dedicated TOF PET application-specific integrated circuit (ASIC) chips have been developed and implemented in various commercial TOF PET scanners (18–20). These new scanners have achieved improved TOF resolution, sensitivity, and spatial resolution compared with photomultiplier tube–based systems, resulting in improved lesion detectability, decreased scan time, and reduced injected radioactivity (21–24).
The uMI Panorama (United Imaging Health Care) is a novel digital PET/CT system that uses SiPM-based detectors coupled with LYSO crystals of 2.76 × 2.76 × 18.1 mm3. The equipped modular uExcel detector platform provides flexibility to vary the axial field of view (FOV). To enhance signal quality and minimize the effects of dead time, an intricately designed multichannel ASIC capable of precise time and amplitude measurements is used.
The aim of this study was to assess the physical performance of the uMI Panorama system in accordance with the National Electrical Manufacturers Association (NEMA) NU 2-2018 standard (25). Furthermore, an initial evaluation of its capabilities was conducted by assessing images acquired using a Hoffman brain phantom, a mini-Derenzo phantom, and 3 patient studies.
MATERIALS AND METHODS
PET/CT System
The uMI Panorama PET/CT system consists of an SiPM-based PET scanner with LYSO crystals and a 160-slice clinical CT scanner. The CT scanner is equipped with a liquid metal bearing x-ray tube, capable of rotating at a maximum speed of 0.25 s/360° and achieving a minimum slice thickness of 0.5 mm. The PET system has a transaxial FOV of 76.0 cm and an axial FOV of 35.1 cm. The system comprises 34 detector modules, with each module consisting of 5 axial “cells” with an intercell gap of 2.2 mm. The PET system’s extendable design allows for the extension of the axial FOV of the scanner by concatenating different numbers of axial “cells.”
Each “cell” comprises a 2 × 2 matrix of miniblocks, and each miniblock consists of a 4 × 4 matrix of microblocks. Each microblock has a 3 × 3 array with a pitch size of 2.85 mm. The total number of crystals is 816 × 120 (transaxial × axial). Additionally, each microblock is coupled to a 2 × 2 SiPM array, achieving an Anger multiplexing ratio of 9:4 for crystal identification. Two multichannel ASIC chips are integrated into each miniblock to obtain the charge and arrival times from SiPMs in each array.
Raw PET coincidence data are stored in list-mode format. The TOF measurement is discretized into 6.1-ps bins, and the energy measurement is discretized into 1-keV bins within an energy window of 430–650 keV. Detectors within the same miniblock share energy information, enabling the recovery of intercrystal scattered events from 1 microblock to another within the same miniblock (26).
Measurements
Spatial Resolution
Spatial resolution was assessed by imaging an 18F-FDG point source (0.77 MBq) that had an inner diameter of 0.5 mm and an axial length of less than 1 mm and that was encapsulated at the end of a capillary tube. The source was positioned at 2 axial planes: 1/2 axial FOV and 1/8 axial FOV from the end of the scanner. At each axial plane, resolution measurements were obtained at 3 transaxial positions (x, y): (0, 1 cm), (0, 10 cm), and (0, 20 cm). List-mode data of LORs with oblique angles of less than 3.7° were rebinned into 2-dimensional sinograms using the Fourier rebinning method (27). The data were subsequently reconstructed using the 2-dimensional filtered backprojection algorithm without attenuation correction, scatter correction, or postreconstruction smoothing. The voxel size was set to 0.6 × 0.6 × 0.8 mm3. The axial, radial, and tangential resolutions were evaluated in accordance with the NEMA NU 2-2018 standard. Additionally, PET resolution performance was further assessed using a mini-Derenzo phantom. This phantom consisted of hot rods with various diameters (1.6, 2.0, 2.4, 3.6, 4.0, and 4.8 mm) distributed across 6 sectors. The center-to-center spacing between rods within a given sector was twice the rod diameter. Initially filled with 16.3 MBq of 18F-FDG, the phantom was imaged for 10 min with the hot rods oriented in line with the axial direction. The same phantom, with an initial activity of 8.6 MBq, was then imaged for 20 min with the hot rods oriented perpendicular to the axial direction. Image reconstruction was performed using the TOF ordered-subset expectation maximization (OSEM) algorithm with the point spread function (PSF), 7 iterations, 10 subsets, and a 1-mm gaussian postreconstruction filter. The voxel size of the image was 0.6 × 0.6 × 0.8 mm3.
Sensitivity
The sensitivity of the system was evaluated using a NEMA NU 2 sensitivity phantom, which consists of 5 concentric aluminum sleeves, each 70 cm long. Polyethylene tubing (70 cm long) filled with 12.8 MBq of 18F-FDG solution was inserted into the aluminum sleeves and positioned at the center of the transaxial FOV. Five 300-s acquisitions were conducted by successively removing the outermost sleeve to extrapolate attenuation-free sensitivity. Then, the entire process was repeated with the line source positioned at 10 cm off-center in the transaxial FOV. List-mode data at both positions were processed using the single-slice rebinning method with a slice thickness of 1.46 mm, and random coincidences were subtracted using the delayed coincidence time window technique.
Scatter Fraction, Count Losses, and Randoms
In accordance with the NEMA NU 2-2018 standard, count rate performance was measured by inserting a 70-cm line source into the NEMA NU 2 scatter phantom, a 20-cm-diameter, 70-cm-long polyethylene cylinder. The phantom was positioned at the center of the FOV and was axially aligned with the scanner axis. The line source contained 1.82 GBq of 18F-FDG solution at the beginning of the acquisition. A total of 34 acquisitions were performed throughout the PET scan protocol described in Supplemental Table 1 (supplemental materials are available at http://jnm.snmjournals.org). Random estimation was obtained using raw data from the delayed coincidence channel. Subsequently, the scatter fraction, count rates for different types of coincident events, and the NECR were determined.
Timing Resolution and Energy Resolution
The data acquired from the count rate performance measurement were used to evaluate the timing resolution and energy resolution (28). To determine the source position, data acquired at activity concentrations below the peak NECR were reconstructed using non-TOF OSEM with all corrections applied. Only coincidences within the ±20-mm region of interest centered on the line source were used to obtain the timing histogram. The timing error was calculated for the data for each coincidence by finding the difference between the measured TOF data and the expected TOF offset on the basis of the point closest to the line source on the corresponding LOR. Scatter and random coincidences were subtracted using the tails of the TOF histograms. The timing resolution was obtained by measuring the full width at half maximum (FWHM) of the final time distribution histogram. For all 34 acquisitions, crystal-level energy spectra with a 1-keV energy bin width were combined into a single spectrum by aligning the 511-keV peaks. The maximum value of the 511-keV peak was determined by fitting a parabolic curve based on 3 points: 1 at the peak and 2 at its nearest neighboring points. The FWHM was determined through linear interpolation between adjacent pixels at half of the maximum value of the peak.
Accuracy: Correction for Count Losses and Randoms
The quantitative accuracy was measured using the same dataset acquired during the count rate performance measurement. Images under various activity concentrations were reconstructed with all PET data corrections. Random coincidences were corrected by using data from the delayed coincidence channel. Dead time correction was based on a nonparalyzable model–based adaptive dead time correction method (29,30). The standard whole-body image reconstruction protocol used at our institution was applied, that is, a PSF-based TOF OSEM algorithm with 3 iterations, 10 subsets, a voxel size of 1.8 × 1.8 × 2.14 mm3, and a 2-mm FWHM gaussian postreconstruction filter. For each reconstruction, 12 slices at each end of the axial FOV were excluded from the analysis.
Image Quality and Accuracy of Correction
To conform to the NEMA NU 2-2018 standard, a NEMA NU 2 image quality phantom was used to evaluate image quality. The NEMA NU 2 scatter phantom, used for the count rate performance measurement, was positioned adjacent to the NEMA image quality phantom. At the start of image acquisition, the background activity concentration of 18F-FDG was 5.3 kBq/mL, and the line source activity in the scatter phantom was 116 MBq. The duration of the phantom scan was set to 6.8 min with a 35% bed overlap, which is the overlap used in our clinical routines. Six spheres with diameters of 10, 13, 17, 22, 28, and 37 mm were filled with an 18F-FDG solution containing an activity concentration ratio of 4.0:1 relative to the background. In accordance with the NEMA NU 2-2018 standard, duplicate scans were performed sequentially 3 times to evaluate the reproducibility of the results. The durations of the second and third scans were adjusted to compensate for the decay of 18F. The resulting image was reconstructed using our institutional standard whole-body protocol with all data corrections applied.
Hoffman Brain Phantom
A Hoffman brain phantom was filled with 48.8 MBq of 18F-FDG and scanned for 10 min. The acquired images were reconstructed using the PSF-based TOF OSEM algorithm, with a voxel size of 0.6 × 0.6 × 0.8 mm3, 7 iterations, 10 subsets, and a 1-mm gaussian postreconstruction filter—our institutional standard brain protocol. In accordance with the method described by Leemans et al. (31), a central slice of the Hoffman brain phantom at the level of the basal ganglia and lateral ventricle was selected to visualize gray matter, white matter, and cerebrospinal fluid. The radioactivity concentration ratios between the gray matter and the white matter, as well as the contrast recovery of gray matter relative to cerebrospinal fluid, were calculated.
Patient Study
The clinical performance of the system was assessed through a study conducted on 3 patients. The study was approved by the Institutional Review Board of Xijing Hospital (KY20212145-F-1), and informed consent was obtained from all patients (informed consent for the publication of related images was obtained from all human research participants). The first patient, a 56-y-old woman (160.3 cm, 58.2 kg), was diagnosed with plantar squamous cell carcinoma and metastases. This patient received an injection of 314 MBq of 18F-FDG. A whole-body PET/CT scan was performed at 69 min after injection, with a scan duration of 2 min/bed, a bed overlap of 40%, and a total of 7 beds. Reconstructions with and without TOF were performed using the PSF-based TOF OSEM algorithm with 3 iterations, 10 subsets, a voxel size of 1.8 × 1.8 × 2.14 mm3, and a 2-mm FWHM gaussian postreconstruction filter for comparison. Three additional TOF reconstructions were performed using portions of the original 2-min/bed–position list-mode file to evaluate the impact of scan duration. The second patient, a 63-y-old man (180 cm, 59 kg), was diagnosed with cognitive impairment and received an injection of 303 MBq of 18F-FDG. At 52 min after injection, a 10-min brain scan was conducted and reconstructed using our standard brain protocol. Again, the non-TOF and TOF images were compared. The third patient, a 72-y-old man (172 cm, 55 kg), presented with non–small cell lung cancer for staging. This patient received an injection of 203 MBq of 18F-FDG. A whole-body PET/CT scan consisting of 4 beds (2 min/bed with a bed overlap of 45%) was performed at 65 min after injection. The same whole-body PET image reconstruction protocol was used.
The datasets generated during the present study are available from the corresponding authors on reasonable request.
RESULTS
Spatial Resolution
The spatial resolution performance of the system is summarized in Table 1. The transverse resolution, averaged over both tangential and radial directions at a distance of 1 cm off-center, was 2.88 mm. Figure 1 displays image slices of the mini-Derenzo phantom positioned at the center of the axial FOV in 2 orientations. Single-slice images (with slice thicknesses of 0.8 mm for the transaxial orientation and 0.6 mm for the coronal orientation) as well as averaged multiple-slice images (over 30 slices for the transaxial orientation and over 40 slices for the coronal orientation) are shown. In the transaxial orientation, the smallest resolved rod size was 2.0 mm, and the shape of the hot rods remained visually undistorted. In the coronal orientation, most of the 2-mm rods were resolved, but some were slightly distorted along their axial direction.
Measured NEMA NU 2-2018 Spatial Resolutions
Illustrations of single slice (left) and average of multiple slices (right) for mini-Derenzo phantom at center of FOV in 2 orientations: with rod perpendicular to axial plane (initially filled with 16.3 MBq of 18F-FDG and imaged for 10 min) and in coronal plane (same phantom imaged for 20 min with initial activity of 8.6 MBq). Images were reconstructed using PSF-based TOF OSEM algorithm with 7 iterations, 10 subsets, voxel size of 0.6 × 0.6 × 0.8 mm3, and 1 mm FWHM gaussian postreconstruction filter. Gray scale was adjusted to cover range from minimum value to 40% of maximum value for each respective image.
Sensitivity
The sensitivity profiles are shown in Figure 2. The system sensitivities were 20.1 kcps/MBq at the transaxial center and 20.9 kcps/MBq at 10 cm off-center.
Axial sensitivity profiles measured with 70-cm line source positioned at transaxial center and 10 cm off-center. Sinogram slice thickness was 1.46 mm.
Count Rate Performance, Timing and Energy Resolution, and Accuracy
Figure 3A shows a plot of count rates for different types of coincident events and NECR at various activity concentrations. The peak NECR was 576.0 kcps at 73.0 kBq/mL. Even at the highest activity concentration of 82.9 kBq/mL, the true count rate reached 3.6 Mcps and continued to increase. Figure 3B depicts the scatter fraction, TOF resolution, energy resolution, and accuracy as a function of activity. The scatter fractions were 34.7% at 5.3 kBq/mL and 33.2% at the peak NECR. The TOF resolutions observed were 189 and 197 ps at 5.3 kBq/mL and at the peak NECR, respectively. The energy resolution was 9.6% FWHM at the 511-keV photopeak with 5.3 kBq/mL. The maximum absolute error, at or below the peak NECR, was 2.81%.
(A) Results for true, scatter, and random count rates and NECR as function of activity concentration. (B) Scatter fraction, TOF resolution, energy resolution, and accuracy measured as function of activity concentration. NECR curve was plotted using double y-axes. Vertical dashed lines denote peak NECR.
Image Quality
Table 2 summarizes the contrast recovery, background variability, and lung residual error for 3 duplicate scans.
Mean, Maximum, and Minimum Percentages of Contrast Recovery, Background Variability, and Lung Residual for 4.0:1 Sphere-to-Background Ratio*
Hoffman Brain Phantom
Four transverse slices of the Hoffman brain phantom are shown in Figure 4. The radioactivity concentration ratio between the gray matter and the white matter was 3.27:1, compared with the ground truth of 4:1; the contrast recovery of gray matter relative to cerebrospinal fluid was 0.95, in comparison to the true value of 1.
Four transverse image slices of Hoffman brain phantom. Phantom was filled with 48.8 MBq of 18F-FDG and scanned for 10 min. Gray scale was scaled to same range from minimum to maximum for each slice. Bottom left slice was used to calculate radioactivity concentration ratio and contrast recovery.
Patient Study
Figure 5A shows the maximum-intensity projections and axial slices across the liver of the first patient. There was no obvious increase in false-positive lesions or absence of suspected lesions for all scan durations. Figures 5B and 5C show axial images containing 2 subcentimeter pulmonary nodules. In the TOF image, 2 nodules in very close proximity could be clearly distinguished. The comparison of the brain images of the second patient with and without TOF correction is shown in Figure 6. The TOF images clearly revealed more detailed substructures of the brain, demonstrating significant benefits from the improved TOF resolution of the system. Figure 7 displays the primary lesion and multiple suspected metastatic lesions in the lung, including both hilar and right lower-lobe areas. In both PET and CT images, several nodules with a diameter of approximately 4 mm and a suggestive hypodense cyst with a diameter of approximately 6 mm could be observed. No motion correction was applied to any of the patient studies.
Images from 56-y-old female patient (body mass index, 22.6) diagnosed with plantar squamous cell carcinoma; injected activity was 314 MBq, and uptake time was 69 min. Images were reconstructed using PSF-based TOF OSEM algorithm with 3 iterations, 10 subsets, 2 mm FWHM gaussian postreconstruction filter, and voxel dimensions of 1.8 × 1.8 × 2.14 mm3. (A) Sequential maximum-intensity projections and transaxial image slices with acquisition times of 14, 10.5, 7, and 3.5 min from left to right, corresponding to 2, 1.5, 1, and 0.5 min/bed, respectively. Liver coefficients of variability were 12.8%, 15.3%, 19.7%, and 25.6% from left to right, respectively. (B) Comparison of TOF (upper row) and non-TOF (lower row) images; from left to right, images represent CT, PET, and PET/CT fusion images, respectively. (C) Zoomed-in images from B, highlighting high resolution of closely located pulmonary nodules in TOF images. Images from left to right represent CT, TOF fusion, and non-TOF fusion images, respectively. For fusion images in C, lower window of intensity scale bar was set to SUV of 1.
Neuroimaging of 63-y-old patient (body mass index, 18.2) who had cognitive impairment with 303 MBq of 18F-FDG PET. Images were acquired at 52 min after injection with single-bed acquisition lasting 10 min. TOF and non-TOF images were reconstructed using PSF-based TOF OSEM algorithm with 7 iterations, 10 subsets, 1 mm FWHM gaussian postreconstruction filter, and voxel size of 0.6 × 0.6 × 0.8 mm3. (A) TOF images compared with non-TOF images, showing better recovery of cold regions in ventricles, gyri, sulci, and nucleus with TOF. (B) Multiple transaxial planes of TOF images. (C) Sagittal view showcasing single-bed axial FOV (35.1 cm) of uMI Panorama. (D) Zoomed-in comparison of TOF with non-TOF images from B, revealing enhanced clarity of insula gyri, nuclei, and spinal cords in TOF images. Grayscale consistency was maintained across PET images in A–D. DN = dentate nucleus; FN = fastigial nucleus; IC = inferior colliculus; LGB = lateral geniculate body; MB = mammillary body; MGB = medial geniculate body; RN = red nucleus; SC = superior colliculus; STN = subthalamic nucleus.
72-y-old male patient (body mass index, 18.6) presenting with non–small cell lung cancer for staging. Patient received injection of 203 MBq of 18F-FDG. Four-bed (2 min/bed, 45% bed overlap) whole-body scan was performed at 65 min after injection. Image was reconstructed using PSF-based TOF OSEM algorithm with 3 iterations, 10 subsets, 2 mm FWHM gaussian postreconstruction filter, and voxel size of 1.8 × 1.8 × 2.14 mm3. (A) Primary lesion (green arrowhead) and suspected metastatic lesion (red arrowhead). (B) CT transverse image reveals nodules of approximately 4-mm diameter in both lungs, corresponding to abnormal uptake on PET image. Long-axis measurement of 6.1 mm corresponds to suspected lesion highlighted by red arrowhead in A. Numeric values indicate lesion SUVmean. (C) Subcentimeter cold region suggestive of hepatic cyst, consistent with hypodense appearance on CT scan.
DISCUSSION
In the present study, we assessed the performance of uMI Panorama in accordance with the NEMA NU 2-2018 standard. The aims of this evaluation were to provide a reference for the system’s performance characteristics and to facilitate comparisons with other commercial PET/CT scanners.
Spatial Resolution
Compared with most commercially available PET/CT scanners today, uMI Panorama exhibited improved spatial resolution, primarily due to the smaller crystal size and a higher multiplexing ratio between the crystal and SiPM. The radial spatial resolution results were better than the tangential resolution results at a 1-cm offset position; this finding may be attributable to the rebinning method used in the filtered backprojection reconstruction. In the coronal images of the mini-Derenzo phantom, the hot rods exhibited distortion along the axial direction, possibly because of the slower convergence caused by PSF modeling for LORs with larger axial oblique angles.
Sensitivity
The sensitivity of uMI Panorama was 20.1 kcps/MBq with an axial FOV of 35.1 cm. Although at first glance it may seem that the system was not as sensitive (in terms of sensitivity per scanner length) as other commercially available systems (18,20), this relative difference in sensitivity per scanner length may be attributable to the slightly lower absorption efficiency of annihilation photons because of the use of shorter and smaller transaxial crystals to provide improved trade-offs among sensitivity, spatial resolution, TOF performance, and axial coverage. Also notable was that the sensitivity at 10 cm off-center was greater than that at the center, a finding that may be explained by the shorter crystal length and longer axial FOV.
Count Rate Performance and Accuracy of Corrections
The combination of high peak NECR and minimal error enabled the system to maintain high efficiency and quantitative accuracy across a wide dynamic range. These characteristics make the system suitable for clinical scenarios involving high count rates as well as for dynamic imaging. Of note was that as the activity concentration increased, the scatter fraction decreased. This phenomenon may be attributable to the energy peak drifting toward the lower end as the count rate rose because of pulse pileup and temperature increase in the detector. The energy drift effect resulted in more scattered events being excluded by the low-level discriminator (430 keV) than by true events.
Timing Resolution
The timing resolution increased to 197 ps at peak NECR because of pileup affecting the detector signal when the count rate increased. The improved timing resolution led to higher image signal-to-noise ratio and contrast compared with the results provided by conventional PET scanners with a slower TOF resolution. Considering that OSEM images converge differently for different TOF resolutions at the same number of iterations and that convergence could be influenced by patient and lesion sizes, a more comprehensive method for comparing the benefits of various TOF resolutions may be used in future studies.
Image Quality
Contrast recovery is highly dependent on reconstruction parameters, such as voxel size, number of iterations/subsets, and postreconstruction smoothing methods (32). Increasing the number of iterations and decreasing the voxel size can improve contrast recovery but may also amplify noise and increase reconstruction time. Using a protocol similar to that used in clinical scenarios, we did not observe obvious Gibbs artifacts, which may lead to a higher contrast recovery for smaller spheres (18). In addition, the low residual of the cold area could be explained by the fine spatial and fast timing resolutions of the scanner.
Hoffman Phantom Study
Brain imaging is often affected by involuntary movement of the patient and variations in individual metabolic characteristics. In contrast, the Hoffman brain phantom offers quantitative and qualitative analyses of a normal brain structure with ground truth. The fine spatial resolution contributes to a high radioactivity concentration ratio between the gray matter and the white matter as well as accurate delineation of anatomic structures. The improved contrast recovery of the ventricle, which lacks radioactivity, is consistent with the low residual observed in the lung region of the NEMA image quality phantom.
Patient Study
The patient study provided initial insights into clinical patient images acquired by the system. The brain and whole-body images presented in this study illustrate the feasibility of fine-resolution imaging with the system using our standard scan duration and injection dose. In addition, there is potential for further reduction in either injection dose or scan time. For example, as illustrated in Figure 5A, clinically acceptable images could still be obtained when the scan time was reduced to 1 min/bed for this case, even when the injection activity was less than the minimum administered 18F-FDG activity according to European Association of Nuclear Medicine guidelines (33). The uMI Panorama’s high sensitivity and fast TOF performance primarily contributed to the low noise level and high contrast of the reconstructed images, even with a small voxel size.
CONCLUSION
The uMI Panorama system achieved a fine spatial resolution of 2.88 mm at a 1-cm offset from the FOV center. Demonstrating an average sensitivity of 20.1 kcps/MBq and a peak NECR of 576 kcps at 73.0 kBq/mL and maintaining less than 3% quantitative bias with activity concentrations below the peak NECR, the system showed robust performance. The incorporation of LYSO crystals, SiPM-based detectors, and ASIC chips led to the system’s fast timing resolution of 189 ps. Results from patient studies underscored the uMI Panorama system’s clinical potential in enhancing diagnostic capabilities, particularly for smaller or lower-contrast lesions, leveraging its exceptional spatial resolution.
DISCLOSURE
This work was supported by the National Natural Science Foundation of China (grant numbers 92259304, 91959208, 82122033, and 81971646) and the Key Science and Technology Program of Shaanxi Province (grant number 2022ZDLSF04-12). No other potential conflict of interest relevant to this article was reported.
KEY POINTS
QUESTION: What are the performance characteristics of the new SiPM- and ASIC-based uMI Panorama PET/CT system according to the NEMA NU 2-2018 standard?
PERTINENT FINDINGS: The uMI Panorama offers fine spatial (2.88 mm) and fast timing (189 ps) resolutions. It also shows improved count rate performance and quantitative accuracy over a wide range of activity levels (up to a peak NECR of 73 kBq/mL).
IMPLICATIONS FOR PATIENT CARE: The uMI Panorama offers improved diagnostic capabilities for small and low-contrast lesions and shows potential in applications involving high-count-rate scenarios.
Footnotes
↵* Contributed equally to this work.
Published online Feb. 22, 2024.
- © 2024 by the Society of Nuclear Medicine and Molecular Imaging.
REFERENCES
- Received for publication April 24, 2023.
- Revision received January 11, 2024.
