Introduction

During the past 70 years or so, since the invention of the first positron-emitting imaging device in Massachusetts General Hospital, Boston [1], PET instrumentation evolved drastically in terms of performance characteristics, application, and availability. It is evident that improving the performance of PET in terms of spatial resolution and sensitivity will lead to wider adoption of this imaging modality in the clinic. Yet, an increase in the number of reimbursed clinical indications does not necessarily lead to higher global availability and accessibility to this technology. Availability/Accessibility depends mostly on fabrication cost. Based on a recent report by the International Atomic Energy Agency (IAEA), among 212 countries, only 109 have access to PET technology [2]. The number of PET scanners in high-income countries is 3.52 per million population, while it falls to 0.004 per million in low-income countries. Gallach et al. showed that at least 96 countries need to increase the number of PET/CT scanners and more than 200 additional PET/CT scanners are necessary to address the main common types of cancer, including lung, colorectal, lymphoma, head and neck, melanoma, and esophagus [3]. They estimated that approximately 229.3 M US$ are needed to equip these 96 countries with 16-slice PET/CT scanners. These statistics raise a few important questions; should the medical physics community (mainly instrumentation research groups) focus on improving PET scanners’ performance or their accessibility? Is it really necessary to compete toward developing fancy detector modules or complex PET configurations to improve the spatial resolution by a few percent or compete on developing methods for reducing fabrication cost? As a thought experiment, is it more beneficial for the society to have more PET scanners with low performance or fewer PET scanners with high performance?

A dedicated or organ-specific PET scanner may be the answer to the above-mentioned questions. Dedicated PET scanners optimized for scanning one specific organ offer both high performance and low manufacturing cost in comparison with general-purpose high-end whole-body PET scanners, which makes them more affordable and accessible. Their easier commissioning, maintenance, and training in addition to their smaller fingerprint or space consumption make them ideal for low-income and middle-income countries and small clinics in high-income countries. Although dedicated PET scanners inherently bear limitations in common clinical scenarios requiring whole-body scans (e.g., staging in clinical oncology), it must be emphasized that these specialized scanners, though more accessible, are not intended to replace the broader utility of whole-body PET scanners in clinical setting.

Apart from organ-specific dedicated PET scanners, designing irregular whole-body PET scanners is of great significance in the field of PET instrumentation. Improving the design of whole-body PET scanners by introducing novel detector concepts and geometrical configurations holds a significant level of enthusiasm in applied research. Novel ideas include extendable axial field-of-view [4], adjustable gantry diameter/shape [5], or scanners equipped with thick and thin detectors’ modules [6], or using plastic scintillators [7] and pseudo-pixelated crystals [6]. Besides the novelties in PET hardware, software advances, specifically involving the use of Artificial Intelligence (AI), might play a crucial role in reducing fabrication costs and improving PET performance. A number of preliminary studies have shown that AI have the potential to reduce cost and complexity, for instance, by removing the time-of-flight (TOF) hardware [8, 9], CT for attenuation correction [10], or even improving the resolution of monolithic and pixelated crystals [11]. In this review, we describe irregular and dedicated PET scanners and discuss the technical innovations that are likely to drive the future of conventional PET scanners. This introductory paper summarizes briefly recent advances in the field and provide insights on potential future developments.

Advances in non-conventional PET detector geometries

Research on PET detectors, at both the hardware and software levels, has mostly been focused on improving the key performance characteristics of the detectors, namely, the spatial resolution and the intrinsic sensitivity. Increasing the detector's sensitivity elevates the collected true coincidence events at a decreased level of injected activity and shortened scanning time, whereas achieving better spatial resolution improves image quality and quantitative accuracy. On one hand, using thicker crystals allows higher sensitivity at the cost of reduced depth-of-interaction (DOI) localization accuracy, which leads to parallax errors [inaccuracy in positioning the line of response (LOR)] and deteriorated spatial resolution. On the other hand, higher spatial resolution is achievable with smaller crystal cross-sectional size, but this will worsen detector sensitivity due to poor scintillation light collection and crystal identification. Accordingly, there is an intrinsic trade-off between key characteristics of the PET detectors making technological advancements challenging [12].

The solution to this trade-off is substantiated in two key technologies, namely, DOI determination and TOF capability [13]. The DOI information minimizes the parallax error and allows providing a more uniform spatial resolution. This is gaining importance in organ-specific imagers with small gantries and/or irregular geometries [14, 15]. Last but not least, DOI might help improving the energy resolution by generating specific photopeaks for different depths of annihilation photons’ detection when the crystals have a rough surface finish [16]. Furthermore, TOF information has the potential to significantly increase the signal-to-noise ratio (SNR) of PET images by limiting the location of the positron annihilation point along the LOR to a smaller segment [17]. High temporal resolution TOF technology is being increasingly highlighted in recent PET instrumentation research. Although currently, coincidence time resolutions (CTR) below 150 picoseconds (ps) full width-at-half-maximum (FWHM) are challenging to obtain, the ultimate goal is 10 ps CTR FWHM, since theoretically, it would directly give access to a reasonably accurate position of the positron annihilation [18]. This information enables to alleviate the difficulties associated with image reconstruction. This is particularly important in organ-specific PET scanners owing to the multiple challenges introduced by irregular geometries in the reconstruction process. A couple of scanner manufacturers unveiled at the last annual meeting of the Society of Nuclear Medicine (June 2023) novel PET scanners achieving temporal TOF resolutions of 178 ps and 194 ps for Siemens Healthineers and United Imaging, respectively.

However, TOF and DOI measurements are not independent of each other. The uncertainty in the DOI can induce errors in timing resolution due to the speed of optical photons in dense crystal medium. In Table 1, we summarized the most innovative detector designs by considering important aspects, such as energy, DOI, TOF resolution, and type of scintillator and readout technology. We also summarized the golden innovation aspects of these studies in the innovation column. Common techniques include Phoswich detectors [19, 20] equipped with pulse-shape discrimination schemes [21], multiple-layered detectors with independent readout for each layer [22], dual-ended readout techniques [23, 24], light-sharing detectors using a particular arrangement of crystals and reflectors [13, 25, 26], detectors with phosphor-coated crystals [27], sub-surface laser-induced optical barriers [28, 29], monolithic crystals coupled with a retroreflector layer [30], and finally machine learning approaches [11, 31, 32].

Table 1 Summary of the most important/novel developments of detector modules
Full size table

The Phoswich detectors’ approach commonly consists of multiple layers of different types of scintillators with different decay times, stacked on each other. Although Phoswich detectors can achieve good DOI resolution, their timing resolution is degraded. In fact, the boundaries between the layers reduce the number of optical photons arriving at the photosensors, and the variability of the arrival time of photons from different layers degrades the timing resolution [13]. Light-sharing detectors couple two crystals together by a particular arrangement of reflectors to imitate a dual-ended readout with a single-ended design. Pizzichemi et al. developed a TOF-DOI-PET module containing an array of crystals with 4-to-1 crystal-to-SiPM arrangement at one end, and a uniform glass light guide on the opposite side to redirect upgoing photons back into neighboring crystals [25]. By using particular prisms for crystals at edges and corners, and optimizing inter-crystal light sharing due to the prism reflection, LaBella et al. achieved better crystal identification, DOI resolution of 2.5 mm, and energy resolution of 9% [26].

Another series of light-sharing detectors, known as crosshair light-sharing detectors [13, 33], consist of crystal pairs partially coupled with optical windows, attached to two different Multi-Pixel Photon Counters (MPPCs). Parts of the crystals that are not coupled with optical windows are attached to reflectors. The DOI and crystal identification is calculated based on the output pattern of the paired MPPCs.

Detector modules based on monolithic crystals have a number of advantages, such as higher sensitivity, the ability to extract DOI, no zero detection regions, decent performance in spatial resolution, and less manufacturing cost. However, these detectors commonly require complex calibration procedures, and complicated algorithms for the location, energy, and timing assignation of photon interactions. Moreover, the spatial resolution deteriorates around the edges, although multiple studies attempted to confront this issue by calibrating the detector using analytical [34], simulation-based [35], and experimental [36] approaches.

Various research groups developed semi-monolithic detectors based on different designs. Sabet et al. [29] proposed a semi-monolithic detector using laser-induced optical barriers (LIOB), which creates small defects inside the LYSO crystal bulk that operate as an optical reflector, to combine the advantages of monolithic and pixelated crystals. Sanaat et al. suggested a novel concept for deflecting the trajectory of optical photons passing through a monolithic scintillator [37]. The proposed technique consists of a reflective belt created from millions of optical barrier points covering the surroundings of the crystal, created by the LIOB method. A monolithic crystal with a belt of reflectors created by laser engraving can lead to better spatial resolution and sensitivity.

Most recently, artificial intelligence was introduced as an effective tool for both accurate TOF estimation and positioning of photon interactions in PET detectors [38,39,40]. The best performance for both event positioning and time stamping resolution have been attained by complex algorithms, such as gradient tree boosting [41], maximum-likelihood [41], nearest neighbors [42], and neural networks [43] applied on monolithic crystals.

Conceptual designs of dedicated/irregular PET scanners

Multi-purpose or conventional PET scanners are intended for almost all clinical applications, including static whole-body, dynamic, brain, cardiac, prostate, and breast scans, as well as absorbed dose verification in heavy-ion radiation therapy [44, 45]. In Tables 2, 3, 4, 5, we listed and categorized dedicated/irregular PET scanners for brain, breast, prostate, and cardiac imaging, respectively. The design and performance parameters, such as spatial resolution, sensitivity, type of scintillator, and geometrical configuration, were listed to enable a quick comparison between the models. One column provides the technical details to provide the hidden aspects of the scanner design. Figure 1 depicts a short history of dedicated/irregular PET scanners development from the first dual-head PET scanner designed and developed in 1953 to the most technically complex and expensive total-body PET scanner. This figure covers a range of different geometrical designs from the dual-panel to dodecahedral geometry and moveable gantries with adjustable detectors.

Table 2 Summary of dedicated/irregular brain PET scanners
Full size table
Table 3 List of dedicated/irregular breast PET scanners
Full size table
Table 4 List of dedicated/irregular prostate PET scanners
Full size table
Table 5 Summary of dedicated/irregular cardiac PET scanners
Full size table
Fig. 1
figure 1

Overview of dedicated/irregular PET scanner geometries over the last 70 years. Courtesy of various sources, Refs. [5, 55, 58, 68, 69, 118,119,120,121,122,123,124,125]. ©IEEE. Reprinted, with permission, from Refs. [68, 121,122,123]. ©RSNA, Reprinted, with permission, from Ref. [119]. ©SNMMI Reprinted, with permission, from Ref. 120. ©IOP Publishing Ltd. with permission, from [5, 55]. ©John Wiley & Sons. Reprinted, with permission, from Ref. [58].

Full size image

In Fig. 2, we illustrated the improvement in image quality through depicting the 3D brain phantom from 1975 to 2022. This anecdotal illustration provides a sense of images generated by these systems as the data acquisition and reconstruction protocols were different. Yet, one can observe that scanners with small geometrical coverage (e.g., Helmet and PET-Hat) lead to quality degradation and provide less anatomical information. The insert PET scanners, like RF-penetrable, also generated blurred images, which can be caused by inaccurate attenuation correction.

Fig. 2
figure 2

Brain PET images of humans or 3D Hoffman brain phantom acquired by different simulated or physical dedicated, conventional, and irregular PET scanners. Courtesy of various sources, Refs. [5, 50, 55, 111, 125, 126, 134, 136, 138, 142, 143, 146, 148, 156]. The images related to reference [126] were taken from https://www.cerebromente.org.br/n01/pet/pet_hist.htm. ©IEEE. Reprinted, with permission, from Refs. [138, 148], ©SNMMI Reprinted, with permission, from Refs. [50, 111, 134], ©IOP Publishing Ltd. with permission, from Refs. [5, 55, 136, 146]

Full size image

The key factors in the conventional PET scanners are robustness, reproducibility, and accuracy of quantitative imaging to guarantee/ensure a dependable/reliable examination for the different applications (screening, diagnosis, response to treatment, and follow-up) considering the high throughput of patients in clinical setting. To fulfill the clinical requirements, conventional/multi-purpose PET scanners should provide relatively high sensitivity, moderate spatial resolution, at reasonable cost, and last but not least accurate/reproducible image quantification (since quantitative imaging is crucial in most clinical indications). The compromise/trade-off among these factors is considered in the design of conventional PET scanners, wherein the equipment used for simultaneous or sequential transmission or anatomical imaging is well considered, since it plays a significant role in quantitative PET and synergistic functional–structural imaging [46].

This trade-off would be highly skewed in dedicated PET scanners, since one of these key factors may have central importance in organ- and/or application-specific PET scanners [47]. For instance, in dynamic whole-body PET imaging, sensitivity is the key factor for low-noise estimation of time-activity curves (or parametric maps). Hence, the tendency would be toward extended FOV PET scanners (through adding more detector rings) at the cost of increased product price [48] or having axial gaps and covering a larger AFOV with the same number of detectors as demonstrated on the PennPET Explorer [49]. On the other hand, in brain PET imaging, a higher spatial resolution would be appreciated to register underlying signals from fine brain structures and neuro-connections. Hence, the tendency is toward exploiting/designing high-resolution PET detectors (finely pixelated, thin monolithic crystals, DOI capability, and advanced electronic read-outs) [50].

Organ-specific dedicated PET scanners are often designed to accommodate the target organ while maximizing the sensitivity and SNR. Nonetheless, the compact design of such scanners with small gantry aperture potentially increases parallax errors, thus degrading the spatial resolution uniformity. To alleviate this issue, detectors with discrete or continuous DOI capability are frequently considered (see [51, 52] for a review on various DOI techniques). Careful detector modules’ geometrical optimization has been reported in the literature to maximize performance from various standpoints. The use of several multi-layer LYSO crystal arrangements to improve spatial resolution uniformity and count-rate performance of a compact MR-compatible PET insert were reported [53].

In this review, we attempted to cover all PET scanners belonging to the following categories:

  • Organ-specific dedicated PET scanners (e.g., brain, breast, prostate, etc.).

  • PET scanners with non-cylindrical “asymmetrical” geometries (e.g., planar, partial-ring, oval shape, spherical/hat shape, etc.).

  • Cylindrical geometries with moveable detectors or gantry.

  • Any kind of cylindrical PET scanners bearing some novelty in detector modules’ conceptual design and acquisition techniques.

Some conceptual designs never materialized in real systems demonstrating their potential in clinical setting. Yet, they are briefly discussed in this review for the sake of completeness.

Challenges of PET image reconstruction algorithms for multifunctional PET scanners

Fulfilling the desired high performance of dedicated PET scanners requires the application of proper correction and calibration algorithms. Geometrical symmetries in PET scanners are often used in the calculation of the geometrical components of normalization factors [54]. Nonetheless, organ-specific dedicated PET scanners are designed to maximize the sensitivity when imaging the target organ, which often calls for a geometrically asymmetrical scanner. For instance, a peak sensitivity of more than 10% was achieved by the helmet PET scanner with an added row of detectors along the chin [55, 56]. The proposed helmet-chin scanner achieved 40% higher peak noise equivalent count rate (NECR) compared to a cylindrical PET scanner with the same number of detectors [57]. A similar dodecahedral design benefiting from an almost 4π coverage was suggested by [58]. While such designs boost detection efficiency of an organ-dedicated PET scanner, their asymmetrical geometry adds complexity to the normalization and correction of PET data.

Regarding PET data correction, some organ-specific dedicated PET scanners use concurrently acquired anatomical images from CT or MRI scanner. For instance, NeuroPET [59] includes a CT scanner. Likewise, MR-compatible PET inserts [60, 61] can benefit from the anatomical MR images for both anatomical localization and attenuation and scatter correction (though converting MR images to attenuation maps is another source of complexity). Anatomical MR images have also been used to estimate motion vectors to correct for patient head motion in PET/MR neuroimaging applications [62]. However, the absence of an anatomical imaging modality on most organ-dedicated scanners brings new challenges to PET attenuation and scatter correction which might consequently compromise PET’s quantitative accuracy. In the case of brain imaging, atlas-based u-maps generation was extensively studied [63, 64]. Nonetheless, such approaches are increasingly more challenging and less reliable when imaging other organs. Therefore, attenuation and scatter corrections are sometimes ignored on such scanners [65, 66] or alternative innovative approaches are sought. In addition, detector gain adjustment is a critical consideration that can affect peak location, scatter contribution, and consequently overall image quality in all PET scanners and more specifically on MR-compatible PET inserts [67].

A prostate PET scanner was designed [68] with the unique feature that it can be tilted to minimize photon attenuation effects [69]. Existing CT images from a separate scan can be co-registered to PET images to perform attenuation correction on dedicated prostate PET scanners. Another challenge in the reconstruction of organ-specific scanners is that often a large part of the data might be missing due to the inevitable detector gaps. These can be handled through interpolation, forward projection of an initial image estimate, or directly in the projection domain using deep learning (DL)-based approaches [70,71,72,73].

Table 6 summarizes the challenges and innovations in image reconstruction for a selection of dedicated/irregular PET scanners. In this table, we categorized image reconstruction methods for each geometrical configuration (Cylindrical, Cylindrical with removed/added detector modules, Flat-Panel, Spherical/Pseudo-spherical, and Irregular configuration) and target organ. The potential challenges and drawbacks for each configuration as well as the strategies for addressing them were listed. Furthermore, the calibration and correction method used in these scanners were briefly mentioned.

Table 6 Summary of the unique challenges and features of some dedicated/irregular PET scanners from a reconstruction point-of-view
Full size table

Summary and future trends

To reflect the perspectives/opinions of experts in the field of PET instrumentation, a survey was designed for this review and sent to 15 PET scientists. This survey included six questions about the design and development of dedicated PET scanners and the future of this field of research. We asked experts in the field to give a score between 1 and 10 for less important to more important or less costly to high costly. For other questions, we also asked them to sort out the options. The response to the questionnaire was averaged and the answers are reported in Fig. 3. Based on this survey, we concluded that the major challenges in dedicated/irregular PET scanner fabrication lie in the optimization of electronics and image reconstruction methods which take the utmost of human resources, while the scintillators and photodetectors take the utmost of financial resources. The majority of the experts also felt that dedicated brain and prostate PET scanners have the highest request, and if the price can be reduced, there will be a large demand. Definitely, none claimed that they will replace whole-body PET scanners. The results also supported the argument that there is a large space for AI to play role in data acquisition, event positioning, and quantitative image reconstruction and that future research should focus more on improving the coincidence time resolution, depth of interaction, and optimal geometrical configurations.

Fig. 3
figure 3

Outcome of a survey from 15 experts in the field of PET instrumentation who provided feedback on six main questions about dedicated PET scanners and future developments

Full size image

The major motivation behind the design and manufacturing of organ-specific PET scanners is to reduce the cost of end products compared to conventional cylindrical multi-ring PET scanners without scarifying key image quality factors relevant in clinical applications. Yet, the aim of dedicated scanners is not to replace existing clinical whole-body PET systems. In this regard, many efforts have been spent toward the design and building of compact PET scanners dedicated for specific organs with remarkably decreased manufacturing costs by reducing the complexity of PET systems’ design (e.g., using flat panel detectors), reducing the number of detectors, and using cost-effective PET detectors, such as monolithic scintillation crystals [74,75,76]. Owing to high demand for brain, breast, and prostate PET scans, the majority of dedicated PET scanners were designed for the purpose of reducing overall public health costs and increasing the accessibility of PET scanners to remote areas and/or developing/underdeveloped countries [75, 77]. A 2-m-long total-body PET scanner with plastic scintillators, referred to as J-PET [7], is one example of attempts to reduce the cost of a total-body PET scanner. Plastic scintillators used in J-PET have a density of about 1 g/cm3 (whereas LSO and BGO have a density of 7.1 to 7.4 g/cm3, respectively) but can provide decent time resolution (about 220 ps CTR) at the cost of reduced sensitivity [7, 78]. To reduce fabrication cost, a number of groups considered rearranging and reducing the number of detectors while relying on DOI and TOF to compensate for the missing sections (see for instance Ref. [79]).

Since conventional PET scanners are normally capable of providing moderate spatial resolution, a major incentive for dedicated PET scanners is to achieve high spatial resolution of the desired structures, such as in brain imaging. The majority of high-resolution dedicated PET scanners are designed for brain imaging, wherein quantitative and high-resolution imaging of brain-specific radiotracers in small structures and neuro-connections is highly desirable [80].To this end, high-resolution pixelated detectors improved DOI and TOF capability, whereas high-speed electronic readout technologies are employed on dedicated brain PET scanners [81, 82]. Furthermore, owing to the small FOV required in brain PET imaging, high-sensitivity imaging could be easily achieved by covering the whole head area, as used on the helmet PET scanner [56].

In addition to achieving low-cost (for prevalent PET scans, such as prostate imaging) and high-resolution (for dedicated brain studies) PET imaging, the motivation behind designing dedicated PET scanners targets specific applications that cannot be accomplished with conventional PET scanners. Range verification in heavy-ion therapy (such as proton radiation therapy) plays a key role in accurate radiation treatment planning monitoring, wherein the identification of the Brag peak location is crucial to deliver the maximum radiation dose to the target volumes and spare healthy/normal tissues [83, 84]. Online (in-beam) PET imaging in heavy-ion radiation treatment requires an open gantry PET design for direct access of radiation beams to the patients [84]. A two-panel PET design is commonly considered for online PET imagers, where the patient could be accessed from two other sides. Due to the fact that the rate/probability of positron emission is not very high, these PET scanners should be equipped with high-sensitivity detectors to achieve acceptable SNR (sensitivity has higher priority than spatial resolution in this case) [85, 86]. Another interesting design, referred to as human-scale single-ring OpenPET system, providing an open space area by axially shifting the detectors to different sides in the axial direction, is suitable for online range verification in heavy-ion therapy [87].

In addition, simultaneous imaging of the target areas is crucial to achieve accurate whole-body dynamic and parametric PET imaging. This would also obviate the need for blood plasma sampling (input function) provided that the major body blood pools are covered in PET imaging [88]. In this regard, extended FOV or total-body PET systems gain attention for enabling fully parametric imaging as well as very low-dose or ultra-fast PET scans [49, 89]. The key factor in the design of such systems is the extensive coverage of the body at the cost of a dramatic increase in manufacturing expenses. To address this issue, extendable FOV PET scanners have been proposed/designed to reduce the number of required PET detectors (to reduce the overall manufacturing cost) while providing the required axial FOV. In these PET scanners, the detector rings are mounted on a mechanical mechanism allowing for an axial extension [90,91,92]. Furthermore, arterial blood sampling is crucial (regarded as gold standard) in dynamic PET studies. This has encouraged to design and build a dedicated small PET scanner for non-invasive image-guided input function estimation (SynchroPET ArterialPETTM scanner (Stony Brook, NY, USA) [92]. Such scanners require very small FOV (as large as a human arm diameter) with a good spatial resolution to reduce errors due to partial volume effect. Low-cost, ease of use, and high spatial resolution and sensitivity for input function estimation is the incentive behind building bracelet PET scanners.

Novel PET geometries, configurations, and detector designs are proposed in the context of conceptual design which could be promising for many applications [93]. However, in some conceptual designs, manufacturing cost is ignored, and sometimes, the improvements brought by complex designs are marginal [94]. A major challenge or drawback in most dedicated PET scanners is the lack of transmission scanning and/or structural imaging. Apart from the benefits of synergistic anatomical-functional imaging to realize the full potential of quantitative PET imaging, anatomical imaging is commonly required [95]. To address this challenge, maximum-likelihood activity and attenuation (MLAA) algorithms [96], attenuation map generation using background radiation [97], and template-based attenuation map estimation approaches were designed [98]. In this regard, a major tendency consists in designing PET inserts, which could be employed on commercial MR, PET/MR, and PET/CT scanners. This could address the challenge of attenuation map generation, since anatomical/transmission data are readily available on the host scanners [99, 100].

It should be noted that owing to the astonishing performance of artificial intelligence-based algorithms, in particular deep learning methods, novel approaches for performing attenuation and scatter correction (ASC) on PET data without using anatomical images have been developed [101,102,103]. These include ASC in the image domain [104], attenuation correction factor estimation in the sinogram domain [104], hybrid MLAA and deep learning methods [105], and attenuation map estimation from non-ASC PET images [106]. Moreover, deep learning algorithms are employed for accurate event positioning, calibration, post-reconstruction processing, and image quality enhancement. These techniques not only boost the overall quantitative accuracy and image quality of PET scans, but could also reduce manufacturing costs [9, 107]. A recently developed maximum-likelihood attenuation correction factor (MLACF) algorithm was adapted to a dedicated brain TOF-PET scanner and implemented in the commercialized system [108]. In this method, the authors combine an MLACF method that simultaneously synthesizes the emission data and attenuation sinogram from TOF-PET data, along with a scaling technique based on anatomical features.

More aggressive efforts to achieve a coincidence time resolution of only a few tens of picoseconds and initial promising results suggest that future PET scanners can indeed rely even more on TOF to improve image quality [109]. There are ongoing debates on the technological limitations of achieving very small CTR values to reach the reconstruction-less capability. Nonetheless, more precision in TOF data leads to higher image SNR and better mitigation of limited-angle tomography.

Fortunately, in the presence of TOF, heavily compressed sinograms with axial rebinning and significant azimuthal mashing can be used without resolution loss [110]. Nevertheless, list-mode reconstruction remains a top choice for many researchers and even on commercial total-body [111, 112] and non-cylindrical PET scanners (e.g., Ref. [47]). With list-mode iterative reconstruction, an accurate physics model of the scanner, including the exact positioning of each LOR, DOI, shift-varying PSF, and TOF can be incorporated in the system matrix. Image artifacts that can be caused by asymmetrical geometries of some organ-specific dedicated PET scanners were elegantly discussed in Ref. [113], also highlighting how incorporation of a TOF can minimize such artifacts.

A new trend in high-resolution PET instrumentation includes dedicated specimen systems for intra-operative assessment of surgical samples for the assessment of lesions heterogeneity and surgical margins in three dimensions [114, 115] and organs-on-chips (OOCs) microdevices mimicking in vivo organs [116], which are finding promising applications in disease modeling and drug discovery. These developments are expected to grow in the future as there appears to be a market for these devices.

It is gratifying to see in perspective all innovative developments in PET instrumentation, from fully 3D imaging without septa to TOF, resolution recovery reconstruction, digital SiPM-based photodetectors, and more recently long axial field-of-view designs. Advances in PET instrumentation and associated image reconstruction and quantification techniques have been very swift and stimulating, and there is every reason to believe that the field will move forward more swiftly in the future with the advent of novel scintillators and photodetector technologies and the unlimited imagination of PET scientists. There is no shortage of challenges and opportunities for PET instrumentation and innovative conceptual designs.

While PET scanner technology witnessed spectacular advancements over the years, many innovative design concepts have not progressed to commercial products for various reasons. These motives can be summarized in five main aspects: fabrication cost and market readiness, service/maintenance cost, patient discomfort, suboptimal real-world performance, and difficulties associated with translating developments from academic to corporate settings. Total-body PET scanners or scanners with high temporal TOF resolution are usually expensive at the present time, which makes them less affordable in the clinic, particularly in low GDP countries [49, 107]. The extendable FOV PET design concept or compensation of the low TOF resolution through deep learning-based image quality enhancement might address this limitation [4, 8, 9]. Maintenance cost is another significant hurdle, especially for PET scanners with moveable detector configurations. Such scanners are more prone to mechanical damage, sensitive to calibration issues, and can contribute to patient discomfort, thereby increasing the maintenance cost and patient anxiety [5]. Hence, it is imperative to establish meticulous protocols for calibration and quality assurance.

Patient comfort is a fundamental consideration in PET scanner design and manufacturing. A few geometrical designs, such as spherical or dodecahedral shapes, may induce feelings of discomfort and claustrophobia [58]. Likewise, scanners with dynamic gantries could potentially cause anxiety when the detectors approach the patient [5, 117]. While these issues can be mitigated through the use of virtual reality headsets or anxiety-reducing medications, it is crucial that these factors are taken into account during the design process to ensure patient's comfort and cooperation. Another important aspect is the performance of the suggested designs in real-world scenario. Many of the suggested configurations were evaluated based on Monte Carlo simulation studies that have ignored several physical factors, which can downgrade the performance. Finally, some concepts like portable, handheld PET scanners also face significant hurdles. Despite the potential for point-of-care application, the need for radiation shielding, stringent regulatory requirements, and the difficulty of miniaturizing the necessary technology all contribute to the non-viability of these designs.