Active-PET: a multifunctional PET scanner with dynamic gantry size featuring high-resolution and high-sensitivity imaging: a Monte Carlo simulation study

The bulk of research in positron emission tomography (PET) instrumentation during the last few years focused on refining the overall performance of PET scanners through improving the hardware and software (Jones and Townsend 2017). An ideal PET scanner should simultaneously achieve high sensitivity and high spatial resolution imaging. A high-sensitivity PET scanner should enable pinpointing lesions presenting with low contrast resolution and signal-to-noise (SNR) ratio. Furthermore, a high sensitivity PET scanner allows decreasing the injected activity and/or scanning time. In some research applications, such as cell tracking, it is essential to have a high-sensitivity scanner (Aide et al 2021). The spatial resolution is a crucial factor dictating the capability of the PET scanner to image small structures and separate two closely located sources. A high-resolution scanner can distinguish the border between healthy and cancer cells and accurately localize small abnormalities (Bal et al 2014). The need for high spatial resolution imaging is more prominent in preclinical studies and scanning of pediatric patients (Amirrashedi et al 2020).

Although the quest for the ideal PET scanner continues to be a goal of the molecular imaging community, the tradeoff between sensitivity and spatial resolution, along with the remaining intrinsic technical limitations hinders the realization of simultaneous high-sensitivity and high-resolution imaging. As a rule of thumb, the sensitivity improves by (i) increasing the number of detected photons via enlarging the pixel size and thickness of crystal elements in pixelated scintillators, (ii) decreasing the diameter of the gantry, (iii) increasing the number of rings, and (iv) using scintillators with high atomic number. Conversely, the spatial resolution improves by (i) decreasing the parallax error by reducing the pixel size and thickness of crystal segments in pixelated scintillators and (ii) decreasing the acollinearity factor by using a small diameter gantry while keeping it large enough to cover the targeted subject being imaged (Arabi et al 2010). Acollinearity errors take place when the annihilation photons are not emitted exactly in opposite directions (i.e. 180 degrees apart) owing to a small amount of initial kinetic energy of positrons during the annihilation. Conversely, the parallax error occurs when the LORs do not enter a detector perpendicularly, which causes inaccuracy in the estimated LOR's coordinates. The Parallax error increases by increasing the crystal thickness and reducing the gantry's diameter.

A number of studies focused on the improvement of PET's performance through enhancing the software and hardware components. Organ-specific PET scanners dedicated for brain, breast (Hsu et al 2016), prostate (Cañizares et al 2020), and small-animal (Amirrashedi et al 2020) imaging have been developed and deployed in clinical and research settings. The design of large axial field-of-view PET scanners to enhance the sensitivity is also well underway and the number of installed systems continues to increase at a healthy pace (Nadig et al 2022).

Deep learning (DL)-based methods aiming at resolution and sensitivity enhancement focused mainly on improving the overall performance of PET scanners (Gong et al 2020, Sanaat and Zaidi 2020, Sanaat et al 2021a, 2021b). Some studies improved PET's performance by DL-based positioning in monolithic crystals, clearly outperforming conventional event positioning algorithms (Sanaat et al 2020b). To optimize the coincidence timing resolution in time-of-flight (TOF) PET, experimental measurements consisting of scanning a ⁶⁸Ga point source shifted repetitively with steps of 5 mm (over a 15 cm range between the two detector modules) across the PET scanner's field-of-view was employed to train a convolutional neural network (Berg and Cherry 2018). This technique enabled to improve significantly the TOF coincidence time resolution from 527 ps to approximately 185 ps. In another work, Hong et al used Monte Carlo (MC) simulations and DL models to synthesize images that could be produced by a scanner with a small crystal bin size artificially from a scanner with large crystal size (Hong et al 2018). The EndoTOFPET-US is a scanner consisting of an external detector plate and a PET detector extension to an endoscopic ultrasound probe designed for prostate and pancreatic cancer imaging (Aubry et al 2013). A number of studies attempted to improve PET scanners' performance by introducing novel configurations for PET scanners. For instance, Zein et al proposed a new model of the Siemens Biograph Vision PET scanner with an extended axial FOV and sparse detector module rings (Zein et al 2021). Their model consisted of a non-static PET scanner, wherein the rings move in the axial direction.

Micro insert is a full-ring small animal PET scanner whose scintillators are connected through optical fibers to photosensors that provides high spatial resolution within a reduced imaging FOV (Wu et al 2008). Another study reported that a conventional PET scanner equipped with a Silicon detector probe in coincidence with the main scanner can produce images with higher spatial resolution compared to the scanner alone (Brzeziński et al 2014). Zoom-in PET combines a high-resolution detector with a conventional PET scanner to generate high-resolution images (Qi et al 2011). Another novel configuration was also introduced for plant imaging, which extends with a vertical FOV. In order to make imaging possible for different size plants, they assumed two moveable half-cylinders that can be separated at a distance up to 40 mm (Antonecchia et al 2021).

A number of novel ideas for hardware development were proposed in the literature, focusing mostly on upgrading the detector module's physical performance. For instance, creating microscopic defects inside monolithic scintillators through laser-induced optical barriers used to combine the high resolution and high sensitive characteristics of pixelated and monolithic crystals has been reported (Sabet et al 2012). A similar technical approach was used to create a belt of reflector defects inside a monolithic crystal to change the direction of optical photons inside monolithic crystals toward the photosensor (Sanaat et al 2019). Labella et al demonstrated that if a prismatoid light guide array uses inter-crystal light sharing could lead to a better depth of interaction estimation and spatial resolution (LaBella et al 2020). Polaroid-PET is also a recently introduced concept that enhances PET scanner performance through adding a layer of Polaroid between the scintillator and photosensors to remove the internal-crystal reflection of optical photons and enhance the spatial resolution (Sanaat et al 2020b). Sanaat et al proposed a preclinical PET scanner where half of the detector blocks have thick monolithic crystals while the second half has thin crystal elements. The thick crystals increase the sensitivity whereas thin crystals improve the spatial resolution (Sanaat et al 2020a). Huber et al designed a compact PET scanner for prostate cancer imaging using a pair of external curved detector arrays placed above and below the patient (Huber et al 2006, 2011). The array at the bottom is fixed while the array at the top can move vertically to enable adjusting to patient size to increase the sensitivity.

The only transformable PET scanner reported in the literature can switch between whole-body scanning mode (83 cm detector ring diameter) and brain scanning mode (54 cm detector ring diameter) (Li et al 2007). The scanner has 12 rectangular detector modules equipped with bismuth germanate (BGO) scintillators with a crystal size of 2.7 × 2.7 × 18 mm³ and photomultiplier tubes as photodetectors.

In this work, we propose Active-PET concept, a novel multi-purpose and dynamic configuration of clinical PET scanners that can mechanically switch between different geometrical modes. Half of the detector modules provide high-sensitivity (thick crystals with large pixel size) whereas the second half consists of high-resolution detectors (thin crystals with small pixel size). Furthermore, the linear translational mechanism behind the detector modules repositions the detector modules forward and backward in the transverse direction to adapt the diameter of the gantry. This degree of freedom and the two high-sensitivity and high-resolution detector modules enable to change the geometry of the scanner according to the desired characteristics. We assessed four different configurations corresponding to various applications, including (i) large bore with thick and thin detectors for large subjects, (ii) small-bore with thick detectors for applications requiring high-sensitivity, such as cell tracking, (iii) small bore with thin detectors for high-resolution imaging to address the needs of preclinical and paediatric imaging, and (iv) non-cylindrical configuration for prostate and breast scanning and image guidance for therapy planning. The current study aims at evaluating prospectively the imaging performance of Active-PET to provide guidance prior to the actual construction of the prototype. The non-cylindrical configuration provides two major advantages, namely placing the detectors close to the breast and pelvis improves the spatial resolution and sensitivity and makes the device suitable for interventional procedures and treatment planning verification, including online dose monitoring in ion therapy.

We first describe the geometry of Active-PET and its switchable modes along with the specifications of the two different detector modules used in the simulation study. Then, the details of GATE simulations and image reconstruction are described. Finally, the validation and performance evaluation will be discused.

2.1. Geometrical configuration of Active-PET

Active-PET is intended to be a switchable scanner that can be geometrically adapted to respond to users' needs and requirements (see supplemental video (available online at stacks.iop.org/PMB/67/155021/mmedia)). Active-PET has 4 rings, each including two groups of high-resolution and high-sensitivity Lutetium Oxyorthoscilicate (LSO(Ce)) pixelated detector blocks (24 for each group, overall 48 detector modules per ring), one with large pixel pitch (4 × 4 mm²) and crystal thickness (20 mm), referred to as high-sensitivity detector modules (figure 1(f)) and another one with small pixel pitch (2 × 2 mm²) and crystal thickness (10 mm), referred to as high-resolution detector modules (figure 1(g)). Furthermore, each raw of detector modules is connected to a linear motor (actuator) that can displace the detectors forward and backward in the direction of the radial axis and vary the gantry diameter to fit the object to be scanned according to the targeted application. The diameter of the gantry can change from 150 mm to 842 mm to accommodate various subjects from mice to humans. The detectors can accommodate circular, oval, and semi-planar geometries. In this work, five configurations were simulated and investigated: Large gantry with high-resolution and high-sensitivity detectors (gantry diameter = 842 mm) (figure 1(b)), small gantry with high-resolution detectors (gantry diameter = 421 mm) (figure 1(c)), small gantry with high-sensitivity detectors (gantry diameter = 421 mm) (figure 1(d)), switchable partial oval shape gantry (figure 1(e)) where the scanner shape changes during scanning by replacing high-resolution detector modules with high sensitive ones (major axes = 350 mm). To evaluate the concept of online switchable PET scanner, partial PET was defined to be switchable during scanning. This implies that during the first half of the scan, the high-resolution detectors are close to the patient whereas during the second half of the scan, the high-sensitivity detectors get close. The ROOT files of these two modes get merged and reconstructed (Brun and Rademakers 1997). The detailed specifications of each configuration along with the Biograph mCT PET scanner are summarized in table 1.

Zoom In Zoom Out Reset image size

Table 1. Active-PET and Biograph mCT scanner design specifications as implemented in GATE simulations.

	Biograph mCT	Small gantry high-resolution	Small gantry high-sensitive	Large gantry high-resolution high-sensitive	Partial-ring
Axial FOV	218	218	218	218	218
Transaxial FOV	700	350	350	700	350
Detector ring diameter	842	421	421	842	421
Crystal pitch	4 × 4	2 × 2	4 × 4	4 × 4 and 2 × 2	4 × 4 $\to$ 2 × 2
Crystal thickness	20	10	20	20 and 10	20 $\to$10
Number of rings	4	4	4	4	4
Detector blocks per ring (Thick + Thin)	48 + 0	0 + 24	24 + 0	24 + 24	0 + 17 $\to$ 17 + 0
Crystals per block	13 × 13 = 169	26 × 26 = 676	13 × 13 = 169	169 and 676	13×13 =169

2.2. Active-PET detector modules

2.3. Monte Carlo simulations

2.4. Image reconstruction

2.5. Validation and performance evaluation

2.6. Mechanical engineering design aspects

2.7. Electronical engineering design aspects

3.1. Validation

3.2. Active-PET performance evaluation

Figure 2 shows the sinograms of three-point sources in the central plane located at 1 mm and 10 mm off center along the Y-axis and 10 mm off center along the X-axis reflecting the simulation of all PET geometrical configurations. The sinograms provide insight regarding the performance of the different configurations.

Zoom In Zoom Out Reset image size

The spatial distribution of the underlying signals, intensity, and uniformity of the sinograms are related to the spatial-resolution, sensitivity, and geometrical aspects. The widest and thinnest sinograms are associated with scanners with small bore and high-sensitivity and high-resolution detector modules, respectively. The partial-ring PET illustrates a gap/uniformity in the sinogram corresponding to the lack of detector modules on two sides of the scanner.

The estimated sensitivity for the different configurations at the center of the FOV and at 10 mm off-center are presented in figure 3. For all configurations, the sensitivity reaches its peak at the center of the FOV and starts to decrease towards the corners of the FOV. At the center of the FOV, the highest sensitivity (15.98 kcps MBq⁻¹) was achieved by a small gantry and high-sensitivity detector, while the configuration with a small gantry and high-resolution had the smallest sensitivity (6.12 kcps MBq⁻¹). The scanner equipped with high-sensitivity and high-resolution detectors has 16% sensitivity less than the Biograph mCT (7.31 versus 8.78 kcps MBq⁻¹).

Zoom In Zoom Out Reset image size

The tangential, radial, and axial spatial resolution for the full-ring cylindrical configurations and partial-ring configuration are reported in tables 2 and 3, respectively. The results support well the observations from the sinograms where the scanner with a small gantry and thin crystals (high resolution detectors) showed the smallest tangential, radial, and axial spatial resolution (2.2 mm, 2.3 mm, 2.5 mm FWHM, respectively). The large gantry (with high-sensitivity and high-resolution blocks) outperformed by 10%, 10%, and 8.5% the Biograph mCT scanner in terms of tangential, radial, and axial spatial resolution, respectively. The small gantry with high-sensitivity detectors showed a spatial resolution very close to the Biograph mCT at the center of the axial FOV (3.5 mm versus 3.7 mm, 3.6 mm versus 3.8 mm, and 3.5 mm versus 3.4 mm FWHM for tangential, radial, and axial directions, respectively). The partial-ring PET scanner outperformed the mCT scanner in terms of axial spatial resolution at the center of the FOV (2.9 versus 3.5 mm FWHM).

Table 2. Spatial-resolution results for three different configurations along with the Biograph mCT scanner. The FWHM and FWTM were reported for tangential (T), radial (R), and axial (A) directions (in mm) for a point source located at 1 cm and 10 cm distances from the CFOV.

	Biograph mCT						Large gantry high-resolution high-sensitive
	FWHM (mm)			FWTM (mm)			FWHM (mm)			FWTM (mm)
Distance (mm)	Tangential	Radial	Axial	Tangential	Radial	Axial	Tangential	Radial	Axial	Tangential	Radial	Axial
10	3.7	3.8	3.5	7.5	7.5	7.7	3.3	3.4	3.2	7.2	7.3	7.5
100	4.2	4.7	5.4	8.2	8.6	10.3	3.9	4.4	5.1	7.8	8.2	9.9
	Small gantry high-sensitive						Small gantry high-resolution
	FWHM (mm)			FWTM (mm)			FWHM (mm)			FWTM (mm)
	Tangential	Radial	Axial	Tangential	Radial	Axial	Tangential	Radial	Axial	Tangential	Radial	Axial
10	3.5	3.6	3.4	7.3	7.4	7.6	2.2	2.3	2.5	3.6	3.8	4.2
100	4	4.5	5.2	8	8.5	10.1	2.3	2.5	3.8	6.4	6.6	7.3

Table 3. Spatial resolution results for partial-ring configurations. The FWHM and FWTM were reported for tangential, radial, and axial directions (in mm) for a point source at 1 cm and 10 cm distance from the CFOV.

Distance (mm)	Partial-ring PET
	FWHM (mm)			FWTM (mm)
	Tangential	Radial	Axial	Tangential	Radial	Axial
10	3.7	3.7	2.9	7.2	7.3	7.3
100	4.1	4.6	4.8	8.1	8.5	9.5

Figure 4 depicts the plots of prompt, true, random, and scatter counts for the different configurations. Figure 5 shows the NECR carve for the different configurations, supporting the results obtained for the sensitivity.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The reconstructed images of the Hoffman 3D brain phantom for the different configurations are illustrated in figure 6. The configuration with a large gantry (high-sensitivity and high-resolution detector modules) showed better image quality providing the highest anatomical details preservation relative to the Biograph mCT. As an example, the Caudate and Putamen regions are more distinguishable from each other and other regions. The scanner with a small gantry and high-resolution detector modules shows better image quality compared to the one with high-sensitivity detector modules, while it may lead to underestimation of radiotracer uptake (the red arrow in the occipital lobe).

Zoom In Zoom Out Reset image size

High spatial resolution and high sensitivity can minimize the partial volume effect and noise characteristics, especially when scanning small objects and when the scanning time is short or the injected activity is small. In the field of nuclear medicine imaging, especially in PET, there is always a desire to have both high sensitivity and high spatial resolution, while the tradeoff between these two parameters makes the realization of this goal barely feasible. One of the major reasons for this limitation is the fixed configuration of PET scanners from their first invention time in the 1970s. Furthermore, in recent years, the success of dedicated organ-specific PET scanners in providing high sensitivity/high spatial resolution raised the desire of using these categories of PET scanners, although having several scanners in one center is costly and space-consuming (González et al 2018, Catana 2019). In this study, we attempted to address the above-mentioned tradeoff/limitation by designing a new PET scanner with variable gantry size and two types of detector modules that can adapt its geometry based on the users' needs. Our suggested multipurpose PET scanner (Active-PET) can switch between different geometrical modes to provide systems dedicated for brain, axilla, breast, prostate, whole body, preclinical, cell tracking, and pediatrics imaging, as well as image guidance for interventions, without increasing manufacturing cost. Active-PET is able to get the detectors close to objects to increase the solid angle coverage of the scanner. A small bore size leads to the reduction of the non-collinearity effect whereas using a thin scintillator decreases the parallax error.

Zein et al reported a spatial resolution degradation of 0.49 mm FWHM in the axial direction at 1 mm off-center of the transaxial FOV and 1% sensitivity reduction relative to the Biograph Vision PET scanner (Zein et al 2021). Using detectors with thick (10 mm) and thin (2 mm) monolithic scintillators in a preclinical PET scanner enabled to improve the spatial resolution (0.7 mm FWHM) and absolute sensitivity (3.7%) in comparison with conventional PET scanners with an average crystal thickness (6 mm) (Sanaat et al 2020a). The oval shape scanner presented by Huber et al achieved a pick absolute sensitivity of 25.54 kcps MBq⁻¹ at the center of the FOV and transverse spatial resolution of 4 mm FWHM at 1 cm (Huber et al 2006, 2011). In this scanner, the elliptical shape of the gantry has a 45 cm and 70 cm minor and major axes, respectively, and a total of 80 detectors. Although this scanner showed better sensitivity than our partial-ring PET, its spatial resolution is around 8% less than Active-PET. The comparison between this model and Active-PET highlights the influence of the distance between the two scintillation detectors' array thickness, and void regions in the scanner.

The non-cylindrical configuration with some adjustments provides access to the subject (void areas without the detectors) from left and right sides. Such PET scanners are required for online ion (proton) therapy verification. Partial angular coverage PET systems would offer direct access/irradiation of the proton beam to the target subject, while the proton range could be instantly verified by the PET detectors. Evidently, the beam should not encounter obstacles (detectors) in its path when entering into the patient from the left or right side. This can be achieved by adding another degree of freedom in the Z-axis (by adding rails parallel to this axis) to extend the axial FOV virtually.

In Li et al study, the absolute sensitivity in whole-body mode and brain mode is 4.2% and 9.2% for a ²²Na point source at the center of the FOV. The spatial resolution for the same point source at the center of FOV was found to be 4 mm and 3.9 mm FWHM for whole-body and brain modes, respectively. In comparison with Active-PET, the abovementioned scanner does not have different detector modules and is not able to switch to various geometrical shapes, such including the oval geometry. Moreover, the transformation time takes 1.5 h, which is carried out manually (Li et al 2007).

We assume that Active-PET uses a similar configuration as the Biograph mCT in terms of number of crystals, photodetectors, electronics, ..etc for imaging, and other mechanical parts for bed motion and other components (ignoring the lower amount of crystals used on Active-PET compared to the Biograph mCT, since half of the detectors have half crystal thickness). Hence, the main components' cost would not change, the only additional cost would be for linear motors plus the electronics. According to our estimates, its cost would be approximately equivalent to two detector modules with a large crystal. We appreciate that the fabrication cost for Active-PET would be higher than a conventional scanner. Yet, considering the multipurpose application of this scanner and space requirements, the overall cost will be compensated.

This work suffers from three main limitations. The first one is the lack of a dedicated reconstruction method for the switchable partial-ring scanner, wherein we considered the system matrix of the partial high-sensitivity scanner. This would lead to suboptimal image reconstruction since the system matrix would not correctly reflect the highest resolution that can be achieved. An optimized reconstruction algorithm would further enhance the performance of this configuration. We simply applied conventional reconstruction methods for the cylindrical geometry with some approximations. The second limitation is the non-uniformity in reconstructed images for the scanner with mixed high-resolution and high-sensitivity detector modules owing to differences in detector sensitivity. Although in the Hoffman 3D brain phantom images, we did not observe severe non-uniformities, it could be more visible in the uniform phantom. The last limitation is the lack of experimental evaluation to confirm the simulation results.

We simulated and evaluated a new PET scanner design, referred to as Active-PET, which unlike conventional PET scanners contains two types of high-resolution and high-sensitivity detector modules whose position can change depending on the application. By using the linear motor behind each array of detector modules, the gantry size/diameter can be adapted. Active-PET can cover a wide range of applications, such as brain, axilla, breast, prostate, whole-body, preclinical and pediatric imaging, as well as cell tracking, and image guidance for interventions.

This work was supported by the Eurostars program of the European commission under grant E! 114021 ProVision and the Private Foundation of Geneva University Hospitals under Grant RC-06–01.