Meeting the detector challenges for pre-clinical proton and ion computed tomography

Six decades after its conception, proton computed tomography (pCT) and proton radiography have yet to be used in medical clinics. However, good progress has been made on relevant detector technologies in the past two decades, and a few prototype pCT systems now exist that approach the performance needed for a clinical device. The tracking and energy-measurement technologies in common use are described, as are the few pCT scanners that are in routine operation at this time. Most of these devices still look like detector R&D efforts as opposed to medical devices, are difficult to use, are at least a factor of five slower than desired for clinical use, and are too small to image many parts of the human body. Recommendations are made for what to consider when engineering a pre-clinical pCT scanner that is designed to meet clinical needs in terms of performance, cost, and ease of use.


Introduction
Medical imaging by means of proton computed tomography (pCT) has a history dating back as far as 1963, as one of three radiological applications that Cormack proposed for his image reconstruction ideas (Cormack 1963).More recent interest in this imaging modality has primarily revolved around the possibility of using it for treatment planning in proton therapy.Up to now, such planning is based on x-ray CT, sometimes aided by magnetic-resonance imaging (MRI). 1 Since x-rays interact with matter very differently from protons, it is plausible that measuring the proton stopping power using protons themselves can result in better treatment.Some other potential advantages of using proton beams for imaging, such as greater contrast (Koehler 1968), lack of artifacts from metallic inserts, and much lower radiological dose (Kramer et al 1980), have been emphasized since the early days of this imaging modality.
The two principal disadvantages of pCT are multiple-Coulomb scattering of protons, which degrades spatial resolution, and the relatively high cost of beam time at a proton therapy center.Experimentation has shown that sufficient spatial resolution can be achieved if protons are measured individually, 2 to estimate their individual paths through the patient, as well as their energy loss.To do that requires instrumentation with high data throughput, as well as large computational power for image reconstruction, in order to achieve a reasonable elapsed time for acquiring and processing the image.Technology circa 1980 was not up to the task, as noted in Hanson (1979), but today capable technology is available, perhaps even at reasonable cost.References Parodi (2014), Poludniowski et al (2015), and Johnson (2017) have thoroughly reviewed the history and the status of particle radiography and tomography.In this paper I provide updates on progress achieved since publication of those reviews and discuss some of the challenges that remain before the technology will find use in clinical settings.
The high cost of proton beam lines is problematic for proton therapy in general (Furlow 2013), and requiring each patient to access the beam line for treatment planning as well as the treatment itself adds to the cost, such that it may be warranted only in special cases.This problem could be ameliorated if pCT images could be acquired and processed rapidly into a treatment plan while the patient remains in the room.However, that is a concept that would be very challenging to execute and would leave little time for careful checking and analysis of the plan, although it is perhaps technically conceivable with modern scanned-pencil-beam technology. 3More realistically, the pCT apparatus might be used immediately before treatment only as a quality check on an existing treatment plan and verification of patient alignment (Krah et al 2019), in which case a few proton radiographs could suffice, in place of a full CT scan.
Still, how long it will be before pCT finds use within clinics is an open question, despite a number of dedicated research and development efforts made over the past two or three decades.This pragmatic objective should be the focus of ongoing research in pCT, and indeed recent years have seen some efforts to commercialize pCT technology (DeJongh et al 2021b) and to introduce development-level pCT instruments into clinics as cross-calibration tools to be used on phantoms, not patients (Fogazzi et al 2023).
Ions such as helium nuclei have also been used for imaging (Volz et al 2021).The detector challenges do not change much, relative to proton CT, but reduced multiple Coulomb scattering can result in improved image quality (Hansen et al 2014).Higher nuclear charge results in larger signals in detectors, making detection somewhat easier, but it also reduces the range for a given energy, which can be an important constraint for imaging, given that the ions must pass all the way through the patient.Heavier ions also result in more nuclear interactions and spallation, which can confuse the CT imaging.In the following, for brevity the term pCT (particle-CT) is used to refer to all ion imaging.

Detector requirements for pCT
In this report, I consider only instrumentation intended to work in 'list-mode,' for which particles are measured and recorded individually, as opposed to inclusion of instruments that record a signal integrated from multiple particles.The price in complexity is large, but the reward is spatial resolution that can approach that achieved by typical medical x-ray CT scanners (Fogazzi et al 2023) and should be at an acceptable level for clinical use. 4 As a bonus, list-mode instruments typically create images with the lowest possible radiation dose by virtue of the fact that they must be highly efficient for detection of individual particles.Figure 1 illustrates a detector concept for pCT working in list mode, with tracking detectors located both in front of and behind the object being imaged (referred to as the 'phantom' in the following).The position and direction of each particle are measured as it exits the phantom.Information on the trajectory prior to entering the phantom is also important for optimal angular resolution.If the incoming beam direction is precisely known, then it is sufficient to measure a single coordinate in the front tracking detector.Otherwise, two coordinates are needed in order give a measurement of the direction of the incoming particle's momentum.All of the tracking measurements are then used, along with the incident energy and ionization energy loss, to estimate for each particle its 'most likely path' (MLP) through the phantom (Williams 2004).
Following the tracking instrument is an energy or range detector that measures each particle's residual energy or range.Since the accelerator beam kinetic energy is generally known to an accuracy of 1% or better  Johnson (2017).Note that the aperture dimension perpendicular to the rotation axis must encompass the full phantom width in order to be usable for computed tomography (unless the rotation axis is shifted from the center of the field of view, as in Civinini et al (2020)).The dimension parallel to the axis only needs to be large enough to encompass the region of interest in the phantom.Adapted from Johnson 2017.© IOP Publishing Ltd.All rights reserved.(Liang et al 2018), subtraction of the residual measurement yields the energy lost along the path through the phantom (and tracking-detector material), typically expressed as a water-equivalent path length (WEPL).The WEPL resolution is limited by fluctuations in energy loss in detector materials and signal fluctuations in an energy detector or, similarly, range fluctuations in a range detector, as well as range fluctuations in the phantom and tracking detectors.
Magnetic spectrometers have also been considered in place of energy and range detectors (Takada et al 1988), but they have not found significant application in pCT because of size and expense.Also, the possibility to construct a pCT scanner based on time-of-flight to measure the particle velocity has been studied (Ulrich-Pur et al 2022).
Rapid acquisition of data from the detectors is crucial, in order to be able to complete a scan in a reasonable amount of time, a factor that will become even more important if and when pCT scans are done on patients in a clinic.Similarly, a large amount of computing power is needed for processing the data into images within a reasonable time frame.The nonlinear trajectories of ions in the phantom significantly complicate image reconstruction.CPU intensive algebraic reconstruction techniques have been used most often, although algorithms based on filtered back-projection have been successfully adapted to pCT and require less brute-force computation (Rit et al 2013).

Impact of accelerator type
Most existing treatment centers are based upon beams from synchrotrons or, especially, isochronous cyclotrons, and the centralized accelerator typically supplies beams to multiple treatment rooms.A synchrotron beam arrives in discrete 'spills' that are spaced sufficiently in time that the CT system (or patient) may be able to rotate from one view to the next between spills.The spills are stretched out in time such that the overall duty cycle is around 50%.The beam from an isochronous cyclotron is continuous, although as is also the case for synchrotrons, there is a radio frequency temporal structure that generally is so rapid that the particle arrival times are effectively continuously distributed from the point of view of the CT system.For a continuous cyclotron beam, maximum efficiency is achieved if the pCT system rotates constantly while acquiring data (Johnson et al 2016).
Recently we have seen the advent of single-room particle-therapy systems based upon compact synchrocyclotrons (Contreras et al 2017), for example the MEVION S250 proton therapy system.Such an accelerator delivers its pencil beam in short (e.g.20 microsecond) pulses at several hundred Hertz, corresponding to about a 1% duty cycle.That magnifies by a factor of 100 the potential difficulty of doing pCT in list mode.This problem can be ameliorated by blowing the beam up in size, e.g. by scattering in a metal foil, as long as the pCT system has sufficient spatial segmentation.

Tracking technologies
The technologies for tracking ions were typically developed for experimentation in elementary-particle physics, and the tracking instruments employed in pCT have often been built by physicists from that research field.Solidstate detectors, notably silicon-strip detectors, have found widespread use, but detectors based on ionization of gas and on scintillation in narrow glass fibers are also commonly used.

Silicon strip sensors
Silicon strips have some highly attractive features for use in pCT tracking.They achieve very high signal-to-noise (S/N) ratios, which allows them to detect particles passing through the active silicon with essentially 100% efficiency while maintaining a nearly zero noise background.They use no consumables other than electricity, and they are very stable in operation, requiring little or no calibration and minimal monitoring.Also, they are efficiently integrated into detector assemblies by automated manufacturing processes developed for the electronics industry.The systems described in Johnson et al (2016), Esposito et al (2018), Scaringella et al (2023), and Ulrich-Pur et al (2020) all use silicon-strip technology for tracking.
Silicon strips do have some disadvantages for use in the next generation of pCT scanners.Their cost is frequently questioned, but I do not believe that to be a major detractor.It is true that the raw sensors are relatively expensive, especially in small quantities, but their simplicity and suitability for automated assembly greatly ameliorate the ultimate system cost.A more significant problem is the relatively small size of individual detectors, typically not much larger than 10 × 10 cm 2 when cut from 15 cm diameter wafers. 5Therefore, it is impossible to integrate them into large apertures without gaps or tiling, both of which can lead to image artifacts (although deleterious effects from tiling can be mitigated as described in Scaringella et al (2023)).Another drawback is that their high S/N is typically achieved through large integration times, leading to problems with signal pileup and inefficiency when used with pencil beams of the desired intensity.Nevertheless, high speed with such detectors has been well demonstrated in elementary-particle physics (for example, see Chalupkova (2013)), albeit with smaller sensors, high power density, and high cost.

Scintillating fibers
Scintillating optical fibers can instrument a large detector plane without inactive gaps, as long as at least a double layer is used to provide overlap between fibers.Using photomultiplier (PMT) or silicon-photomultiplier (SiPM) readout, they can be significantly faster than large-format silicon-strip detectors.However, their signal-to-noise is poor compared with silicon strips, although nevertheless acceptable if relatively large (e.g. 1 mm diameter) fibers are used.Since pCT tracking errors are dominated by scattering in the thick phantom, spatial resolution of the order of a millimeter is sufficient.
This technology was effectively employed in an early system for proton radiography at the Paul Scherrer institute (PSI) (Pemler et al 1999).More recently, it has been successfully used for the tracking system of the proton-VDA instrument (DeJongh et al 2021b).In the proton-VDA system, twelve fiber pairs are connected to each SiPM, in order to reduce the electronics channel count and avoid using VLSI ASICs in the readout.That system is designed for use with scanned pencil beams, and knowledge of the instantaneous pencil-beam location is used to resolve the twelve-fold ambiguity.

Monolithic active pixel sensors
Monolithic active pixel sensors (MAPS) represent a newer technology that is already finding applications in upgrades of CERN LHC detectors, notably in the ALICE experiment inner tracking system (Contin 2020).The Bergen pCT collaboration is basing its design entirely upon this technology, with a 'high granularity tracking calorimeter' that serves as both the tracker and the device for WEPL measurement (Alme et al 2020).See figure 2. MAPS offer very fine segmentation, with 30 × 27 μm 2 pixels in the Bergen application.The fine, twodimensional segmentation provides spatial resolution far better than what is needed, but its importance is to allow the system to operate at high particle fluence without suffering from pileup.
The Bergen system is not triggered, so it does not initiate a readout for each detected particle.Instead, it reads a full 'frame' every 10 μs.With a goal of measuring 10 7 ions per second, that means that each frame will contain on average 100 particle tracks.With such fine granularity in the detector, reconstructing the individual tracks, to project them back into the phantom and to measure their range, should be feasible, even in a pencil-beam scan (in 10 μs the scanned beam will not move significantly, so the 100 or so tracks will cluster within a field the size of the beam).
A disadvantage of this scheme is that it will not be possible to associate individual tracks with measurements from a tracking detector placed in front of the phantom, because of the very high number of simultaneous signals from different particles and the large amount of random scattering that occurs in the phantom.This limits the information available for estimating the MLP of the ion passage through the phantom.The impact of that limitation on spatial resolution was studied in Sølie et al (2020), where the authors found the 'extended MLP' method of Krah et al (2018) to be the only known 'viable MLP estimation algorithm for use in single-sided proton imaging.'All pCT systems would be significantly simplified if the front tracking could be eliminated, but the analysis showed a significant loss of spatial resolution.The authors' simulations of single-sided pCT with a realistic rear tracking system resulted in 'spatial resolution just above what is suggested for accurate treatment planning.' The proposed MAPS tracking calorimeter has not yet been built, but a prototype successfully measured 106 particles per second with a 2 kHz readout rate, corresponding to 500 tracks per frame (Pettersen et al 2017). 6At the location of the Bragg peak, 0.42% of the pixels in the layer were above threshold, corresponding to clusters of about 13 pixels per proton.The cluster size correlates with deposited energy, providing information to fit accurately the location of the Bragg peak and thereby obtain the WEPL.
It is interesting to note that of all the technologies discussed here, the MAPS imaging calorimeter may be the only one capable of performing well in the beam of a synchrocyclotron.As described above, such an accelerator provides particles with a duty cycle around 1%. Therefore, to achieve good imaging speed, hundreds of particles would have to be detected simultaneously.The MAPS fine spatial segmentation enables such operation.Interestingly, for the MEVION S250 proton therapy system, with its 500 Hz pulse rate, the 2 kHz prototype may be better optimized than the planned system with 10 μs frames.

Micro pattern gas detectors
Detectors based on ionization in gas, with gas amplification of 10 4 or more in high electric field regions near the electronic sensors, provide fast signals with no need for long electronic integration times to achieve high signalto-noise.GEM (Sauli 1997) and MicroMegas (Giomataris et al 1996, Bortfeldt et al 2016) detectors are popular configurations of thin, planar gas-based tracking detectors.The CERN RD51 collaboration has been intensively developing micro pattern gas detector technologies for use in elementary-particle physics, including in the enormous LHC experiments.Such detectors can instrument very large areas (square meters) without gaps in sensitivity, and MicroMegas strips typically have a readout pitch of 0.25-0.5 mm (see figure 3).Thus they appear to be well suited to applications in pCT.In fact, the AQUA proton radiography project employed GEM detectors for tracking a decade ago (Bucciantonio et al 2013).

Calorimeters
A calorimeter is designed to stop the particle and produce a signal proportional to its energy.For pCT the device measures the energy remaining after passage through the phantom, whereas what is of interest is the energy lost in the phantom.That is obtained by subtracting the residual energy from the known beam energy (and correcting for the small amount of energy lost while passing through the tracking detectors and air).A problem with this approach is that where the phantom is thin, the residual energy is large, and since the calorimeter's energy resolution will scale with energy, one ends up with a large uncertainty on a small WEPL measurement.For that reason, the calorimeter needs to have excellent resolution, for example σ E /E around one percent for 200 MeV protons.That can be achieved, for example with thallium-doped CsI crystals, as used in the pCT scanner described in Hurley et al (2012).However, the signal development in those crystals is too slow, with microsecond-scale decay constants, to be useful for a scanner that will work at rates relevant to clinical needs.YAG:Ce crystals are much faster, with a 70 ns scintillation-light decay time, and have seen use in at least one prototype pCT system (Sipala et al 2015).That system was designed to achieve 1% energy resolution at 200 MeV while measuring 10 6 protons per second.Its readout amplifiers have a 1 μs peaking time, however, so that rate could only be practical in a cone beam spread out across multiple crystals.Civinini et al (2020) and Scaringella et al (2023) describe operation of a 14-crystal calorimeter in a 211 MeV beam, where an energy resolution of 0.9% was achieved, although while operating at a proton rate of only 80 kHz.
Plastic scintillators (Moser et al 1993) read out by PMTs or SiPMs readily achieve high speed, but they have energy resolution an order of magnitude worse than obtained from the inorganic scintillators described above.Two solutions have been employed in pCT systems to use plastic scintillators in a way that avoids measuring large residual energy.The Proton-VDA system (DeJongh et al 2021b) uses a calorimeter based on a monolithic block of plastic scintillator read out by 16 PMTs.That readout configuration gives excellent light collection for optimisation of the resolution (and also provides rough information on the particle location), but the technique used to avoid measuring large residual energy is to modulate the beam energy such that it is low in regions where the phantom is thin.The calorimeter can then be much thinner than what would be needed to stop ions with the maximum beam energy.This method works well with pencil beams but does have a disadvantage of requiring significant integration between the beam delivery and the pCT system if the beam energy is to be modulated during the scan.In practice, the system has been tested by scanning the full phantom at least twice, with beams of different energy for each scan.
Bashkirov et al (2016) designed a scintillator-based energy detector with five 'stages' in depth, each 5 cm thick.That is enough material to stop a 200 MeV proton beam in the last stage.In regions where the phantom slows the ions significantly, they will stop in earlier stages.The key is to base the energy measurement on the signal from the stage in which the ion stops.The previous stages contribute to the WEPL measurement according to their known material and thickness, although they do influence the WEPL resolution according to their contribution to range straggling.The disadvantage to this approach is that calibration is complex, and it is particularly difficult to handle accurately ions that stop near an interface between stages, resulting in some ring artifacts in the images (Johnson et al 2016, Volz et al 2021).

Range detectors
Several prototype pCT systems directly measure the ion's range in plastic scintillator in order to infer the WEPL in the phantom.To do so requires finely segmenting the scintillator in depth.The range then roughly corresponds to the location of the last scintillator with a signal.Each scintillator should be 3-4 mm in depth in order for the range uncertainty to be commensurate with unavoidable range straggling (Bashkirov et al 2016).About 64 such layers are needed in order to stop a 200 MeV proton.There would be no advantage to use thinner scintillators, but thicker ones may be used if the signal size from each is measured with sufficient accuracy to infer the stopping location in the last segment.In fact, the 5-stage energy detector described in the previous section (Bashkirov et al 2016) can be thought of as such an example in the extreme.
Nuclear interactions in the range detector, a common occurrence, will confuse the range information and result in insufficient WEPL resolution if not detected.Therefore, using a simple threshold on the light signal from each layer is most likely not optimal, even when tails outside of two or three standard deviations are removed, as is commonly done.Ideally, each layer should give some information on the energy deposit, as done in Bucciantonio et al (2013), which can then be used to detect nuclear interactions.That additional information can also be used to refine the knowledge of the position of the Bragg peak.There will not be much advantage to such refinement if the layers are as thin as 3 mm, so a well optimized system might use somewhat thicker layers, each instrumented with a coarse measurement of the signal size.
Various detectors have been used to read the signals from range-counter scintillators, and range detectors have been made using different detector technologies.The Alme et al (2020) tracking calorimeter can be thought of as a range detector very finely instrumented with MAPS detectors.In the PRaVDA prototype pCT system of Esposito et al (2018), silicon-strip detectors are used not only for the tracking but also in a range detector, sandwiched between 2 mm planes of PMMA plastic.The latter instrument, with 21 range-detector layers comprising a total water-equivalent thickness (WET) of 55.4 mm, was tested with protons up to only 81 MeV energy, close to its maximum reach.Similarly, the tracker lateral dimensions are only about 10 cm, so the prototype is limited to making CT images of phantoms much smaller than, for example, a human head.It does nevertheless illustrate the technological principle, and the collaboration is currently working on building an advanced, well funded, full-scale system named OPTIma (Winter et al 2023).
Several other range-detector examples are composed of thin plastic scintillator layers read out via wavelength shifting fibers.The PSI proton radiography instrument referenced already in section 3.2 used 64 3 mm thick layers of BC-404 scintillator read out by 16-channel PMTs, whose signals were analyzed by simple discriminators (no pulse-size information was recorded).The AQUA proton radiography instrument referenced in section 3.4 used 48 layers of 3 mm thick BC-408 scintillator read out by SiPMs, whose signals were digitized by 12-bit pipelined ADCs.Similarly, Uzunyan et al (2016) describe a pCT range detector with 96 layers, read out by SiPMs and 12-bit pipelined ADCs.Unfortunately, none of these examples has operated as a pCT scanner with published results.

Time of flight
Ion residual energy in pCT could be inferred from a direct measurement of the particle velocity following passage through the phantom.That possibility was evaluated in Krah et al (2022) and studied in Ulrich-Pur et al (2022) for a particular detector modality.Low-gain avalanche detectors (LGAD) were proposed to track the ions and measure their time-of-flight.Such detectors have achieved timing resolutions of 30 ps (Cartiglia et al 2017) to 50 ps (Pietraszko et al 2020) in beam tests of very small prototype systems.The Monte Carlo simulation study in Ulrich-Pur et al (2022) showed, for example, that with 30 ps timing resolution and a 1.5 m flight distance, an energy resolution of 1% or better could be achieved for proton kinetic energy up to about 250 MeV.Such a lengthy detector would result in a bulky system and would require very large arrays of LGAD sensors to make a pCT instrument capable of imaging the human body.
In Ulrich-Pur et al (2023), a more compact system concept is simulated and studied, for which the time-offlight is measured through the phantom, using LGAD detectors placed both before and after the phantom.The same detectors provide the precision tracking needed to find the MLP through the phantom.An empirical calibration procedure is used to convert the timing information into values of proton stopping power relative to water (relative stopping power, or RSP).However, for 30 ps timing, RSP accuracy better than 1% is achieved only for beam energy well below 100 MeV, less than half the energy needed to image, for example, a human head.Of course, better timing and/or longer flight distances will improve the results but may not be practical.

Operational prototype pCT systems
Only a few prototype pCT scanners fabricated to date have reached a sufficient size, speed, and performance such that they could conceivably be used in a medical clinic.Those few have seen repeated use over the past decade in experimentation with published results.They are discussed here with an eye toward what improvements are needed in order to create a true pre-clinical pCT scanner.

Proton VDA
Proton VDA is the first commercial venture to market pCT and particle radiography (DeJongh et al 2021b). 7igure 4 shows a photograph of the apparatus set up to image an anthropomorphic head phantom.The Proton VDA design emphasizes low cost by minimizing the number of electronics channels and by reducing the tracking to measuring just a point before and after the phantom, instead of two points that would be needed for a direction vector.The tracking and calorimetry are described above, in sections 3.2 and 4.1, but it is worth reiterating that the system is meant to operate specifically with pencil-beam scanning systems.
The proton-VDA system has a large aperture and can, for example, view an entire human head in a single scan.It also has fast data acquisition, designed to reach speeds of 10 7 ions per second.Published results have been based on data acquired at a few megahertz, but even that is faster than any other operational pCT scanner.A particularly interesting publication by proton VDA is a comparison between pCT and x-Ray CT in measuring proton stopping power, which nicely demonstrates some of the potential for pCT (DeJongh et al 2021a).
Proton VDA has from the beginning emphasized radiography alongside pCT.In Miller et al (2019), they study the use of proton radiographs for image guidance in proton therapy, and they emphasize fast in situ reconstruction to support that modality (Ordoñez et al 2019).On the company's web page they have announced a partnership with Leo Cancer Care to offer a proton therapy treatment room that includes upright patient positioning (eliminating the need for a gantry in the proton beamline), dual-energy x-ray CT, and proton radiography.This may become the first clinical particle radiography system using list-mode data acquisition.

The LLU/UCSC/Baylor Phase-2 pCT scanner
An academic collaboration of medical physicists, elementary-particle physicists, and data scientists completed the 'Phase-2' pCT scanner in 2014 (Johnson et al 2016).Its tracking system is based on eight layers of siliconstrip sensors left over from the NASA Fermi LAT gamma-ray telescope fabrication (Atwood et al 2007) and read out by an ASIC developed specifically for this pCT project (Johnson et al 2013).Its five-stage plastic-scintillator energy detector is described above in section 4.1.Its custom data acquisition system is based on a hierarchy of FPGAs that culminates in a 100 megabyte-per-second TCP-IP link to a computer.It has been demonstrated repeatedly to acquire data at a steady rate exceeding a million measured particles per second and requires 5 or 6 minutes to complete a full CT scan.The aperture is wide enough to scan a human head, but its height is one fourth the width, so that two or three scans have to be assembled together to encompass an entire head.Figure 5 is a photograph of the scanner in operation to make a CT scan of an anthropomorphic head phantom.Several external power supplies, an external fan, and many cables are required to set it up, making it look a bit like a particle-physics experiment.Nevertheless, multiple groups have made use of the instrument to do published  research.It was originally designed to operate at the Loma Linda University Medical Center in the beam of a synchrotron, so the timing was designed to accommodate synchrotron spills, and the beam was expected to be a cone beam produced by scattering in a metal foil.However, most of its operation has been in the practically continuous cyclotron beam at the Northwestern Medicine Proton Center, with beam diameter of 4 to 7 centimeters 'wobbled' across the instrument aperture.It has also been shipped twice to Germany for experiments in an alpha-particle beam from the synchrotron at the Heidelberg Ion-beam Therapy Center.
In the following I highlight some of the publications based on data from this instrument.In Dedes et al (2022) we find a comparison of the performance of the Phase-2 scanner versus the Proton-VDA scanner.Both were found to 'fulfill the requirement of an RSP accuracy of about 1%,' and as expected, the design of the Phase-2 scanner, with ion direction as well as position measured before and after the phantom, yields noticeably better spatial resolution.In Dedes et al (2019) there is a comparison between pCT and dual-energy x-ray CT (DECT), and in Volz et al (2021) we find similar comparisons that also include helium-ion CT.Some advantage of pCT was seen for RSP measurement, and with helium ions the spatial resolution closely approached the capability of a state-of-the art DECT scanner.
Dedes et al (2018) explores a more sophisticated operational mode in which the fluence is modulated across the phantom, in a pencil-beam scan, to optimize the image in a region of interest.The Phase-2 scanner was not intended to operate with such a pencil beam (1.4 cm FWHM), so for these studies the data acquisition rate was lowered to 400 KHz.However, even at a rate of 900 kHz only a 5% degradation was observed in the tracking efficiency, due to pileup of signals in the silicon-strip tracker.

INFN Prima-RDH system
The Prima-RDH pCT system has a tracking system based on silicon-strip sensors in a configuration functionally equivalent to that used in the Phase-2 tracker, although with a somewhat smaller aperture.Its YAG:Ce crystal calorimeter is discussed above in section 4.1.The data acquisition was limited to 80 kHz, but given longer duration test runs it could complete excellent pCT scans.In Civinini et al (2020) we find that sub 1% RSP errors were obtained, similar to the performance of the Proton-VDA and Phase-2 scanners.The same publication also focuses on another advantage of using charged particles for imaging-the images are free from artifacts that in x-ray CT are caused by heavy metal, such as in false teeth, due to beam-hardening.See figure 6.
A novel approach to getting pCT used in the clinic in the near term is being taken by the Prima-RDH group (Fogazzi et al 2023).Their idea is to use pCT as a 'clinical tool for verification and correction of x-ray CT calibration in proton treatment planning.'In that application there is no need to obtain approval to use the system with human patients, and it is also not critical that the scans be completed in short time intervals.Furthermore, research by Fogazzi et al (2023) shows that such existing pCT instruments can be valuable as tools to measure reference RSP values in a laboratory and thereby develop a calibration procedure to be applied to x-ray CT images.

Engineering a true pre-clinical pCT scanner
With the exception of Proton-VDA, pCT scanners to date have been constructed as research instruments that by design could never be used in a clinic with human patients.I believe that what are needed now are pre-clinical scanners that meet the requirements of a clinical device and look like clinical devices, even if not intended for use in a clinic.Hopefully they will demonstrate that a true clinical device will one day be feasible and useful.
A set of design goals should include the following: • High speed data acquisition, to complete a CT scan in a minute or less.That roughly translates into the range 5 × 10 6 to 15 × 10 6 particles per second, depending on the size of the aperture.8 • Ability to handle pencil beams with currents that are not below what facilities are able to provide in normal operation.Compared with diffuse beams that illuminate a large part of, or all of, the aperture, pencil beams greatly increase the particle flux, potentially causing signal pileup problems in the detectors.Relatively high currents likely make it necessary to measure multiple particles simultaneously.
• Integrated fast image reconstruction.The physician should not have to wait significantly longer than the acquisition time to view images, so proposals and designs must include a sufficient emphasis on computation.
• RSP accuracy better than 1%, no visible artifacts, and spatial resolution at least as good as currently achieved by, for example, the Phase-2 scanner.
• Stable calibration.Calibration runs for checking or updating the system should not be needed more often than once per day.
• Turn-key operation.Besides positioning the patient, the operator should not need to interact with more than a power switch and a straightforward GUI on a computer screen.
• Clean appearance.Appearances matter, so the apparatus should look like a medical instrument, not an elementary-particle physics experiment.All the complexity should be hidden from the operator and the patient.
• Close integration with the beam delivery system.
The last item may be the most difficult to develop, because it will require significant prior interest by companies that provide the treatment facilities and, in particular, the particle accelerators.The integration will include patient positioning and rotation, with the rotation angle acquired along with the detector data.Other information such as beam energy, current, and position should also be acquired in real time.In some systems, such as Proton-VDA, it may be desirable for the beam energy to be controlled by the pCT scanner during a scan.Safety interlocks will be necessary, and it will be crucial for the accelerator operators to see what is happening in real time.In existing facilities, the operators have no detectors that can sense the presence and position of such low-intensity beams as have been used thus far for pCT.
All of this could present a chicken-and-egg problem, since the facility vendors are unlikely to become interested until they see a good pre-clinical scanner.To start with, it may be necessary on the side of the preclinical scanner to provide an electronic interface that presents the appropriate data and control paths and could conceivably be integrated into a future facility.It could include, for example, a video stream that shows the beam position, intensity, and even energy based on data from the scanner itself.
A crucial factor so far unmentioned is cost.A practical clinical device will have to be affordable.It is difficult, however, to estimate the ultimate cost of any of these technologies, as that will depend not only on the cost of specialized parts such as silicon-strip sensors, MAPS sensors, and readout ASICs, but also on to what degree the manufacturing can be readily automated, in a manner that is consistent with the market size.Nevertheless, cost and manufacturability should be kept closely in mind when designing a pre-clinical pCT system.I believe that it should be possible to manufacture a clinical pCT scanner with a cost no higher than the $100 000-$200 000 price of a moderately capable x-ray CT scanner.

Outlook and conclusions
It is frustrating that more than four decades after the first serious experimentation with pCT, the technology still has not made it into the clinic.On the positive side, large advances in electronics and computing over that time span suggest that it is now technically possible to fulfill the promise of pCT to improve cancer treatment planning and delivery.Nevertheless, numerous promising development programs that realized small, slow prototypes but promised a next generation of full-size (≈40 cm aperture) and high speed (>10 6 Hz) instrumentation appear to have been discontinued or are dormant.
Ion radiography is much easier than pCT because of relaxed requirements on speed and data processing.Even there, however, promising starts have not yet resulted in clinical use.The system installed in the PSI radiotherapy beamline (Pemler et al 1999) got as far as making a radiograph of a cancerous live dog (Schneider et al 2004) but did not see clinical use with humans.Nevertheless, a 2016 PSI publication based on simulations concluded that 'proton-based radiographic images can accurately monitor patient positioning and in vivo range verification, while providing equivalent WEPL information to a CT scan, with the advantage of a much lower imaging dose' (Hammi et al 2016).Similarly, the TERA AQUA PRR-30 system (Bucciantonio et al 2013) does not appear to have found clinical use.In fact, its range telescope, paired with a new silicon-strip based tracker, is currently being used in a prototype pCT instrument at the MedAustron center for ion therapy and research in Wiener Neustadt, Austria (Ulrich-Pur et al 2020).Nevertheless, it is heartening to see the recently initiated collaboration between Proton-VDA and Leo Cancer Care to bring proton radiography into the clinic.A number of other efforts in online treatment verification do not employ hardware intended for pCT (for example, see Volz et al (2020) and Schilling et al (2023)), but increasing interest in this area could be important to spur further development of pCT.
As presented in section 5, a few R&D programs have already built pCT scanners that approach the requirements needed in a clinic.The Proton-VDA system is certainly the one example we have of a pCT or particle-radiography scanner nearly ready to go into clinical use, and it may soon be used for clinical proton radiography.The Bergen collaboration is working on an instrument that could be the ultimate pCT scanner in terms of speed and RSP resolution and is inherently capable of measuring many simultaneous particles.The PRaVDA/OPTIma project is funded and well underway, with tracking based on silicon strip sensors and a segmented energy detector that is not yet well defined but may be based on scintillating bars (Winter et al 2023).It is aiming for advanced performance that would be useful in a clinical setting, in particular a capability to handle high rates in a scanned pencil beam, with up to seven charged-particle tracks measured simultaneously.The goal is to make pCT scans with as little as 11 s of beam time.In addition, the groups that built and used the Phase-2 scanner are collaborating on a proposal to go the next step up in size and speed to a new scanner better suited to a clinical setting.Some other groups mentioned above, including Prima-RDH and the MedAustron effort, are also still active in the field.
The community has not settled on what are the best detector technologies for pCT, and we find that all of those summarized in sections 3 and 4 are being employed in these various instrument concepts.It is interesting that the efforts in energy-detector design tend to be the most difficult in terms of handling high rates and beam current, because tracking detectors are inherently finely segmented (Winter et al 2023), whereas energy detectors typically are not.It is my hope that along with this ongoing technology development, this next generation of instrumentation will address the more general design and engineering requirements that I have outlined in section 6, which I believe will then position us to start introducing pCT into clinical use.

Figure 1 .
Figure1.Conceptual image of a pCT scanner designed to work in list mode, detecting and measuring charged particles one at a time, fromJohnson (2017).Note that the aperture dimension perpendicular to the rotation axis must encompass the full phantom width in order to be usable for computed tomography (unless the rotation axis is shifted from the center of the field of view, as inCivinini et al (2020)).The dimension parallel to the axis only needs to be large enough to encompass the region of interest in the phantom.Adapted from Johnson 2017.© IOP Publishing Ltd.All rights reserved.

Figure 2 .
Figure 2. The general structure of the Bergen pCT system.Figure published in Alme et al (2020), Reproduced from Alme et al 2020.CC BY 4.0.

Figure 4 .
Figure 4. Photograph of the proton-VDA pCT scanner in front of a beam nozzle at the Northwestern Medicine Proton Center in Warrenville Illinois.From https://physicsworld.com/a/proton-imaging-moves-a-step-closer-to-the-clinic/.Photo reproduced with permission from James Welsh.

Figure 5 .
Figure 5. Photograph of the Phase-2 pCT scanner in front of a beam nozzle at the Northwestern Medicine Proton Center in Warrenville Illinois.

Figure 6 .
Figure 6.Comparison of x-ray CT (a) and pCT (b) images of an anthropomorphic phantom head, showing large artifacts in the x-ray image caused by a tungsten dental filling.Reproduced from Civinini et al 2020.© 2020 Institute of Physics and Engineering in Medicine.All rights reserved.