The first PET glimpse of a proton FLASH beam

We demonstrate the first ever recorded positron-emission tomography (PET) imaging and dosimetry of a FLASH proton beam at the Proton Center of the MD Anderson Cancer Center. Two scintillating LYSO crystal arrays, read out by silicon photomultipliers, were configured with a partial field of view of a cylindrical poly-methyl methacrylate (PMMA) phantom irradiated by a FLASH proton beam. The proton beam had a kinetic energy of 75.8 MeV and an intensity of about 3.5 × 1010 protons that were extracted over 101.5 ms-long spills. The radiation environment was characterized by cadmium–zinc–telluride and plastic scintillator counters. Preliminary results indicate that the PET technology used in our tests can efficiently record FLASH beam events. The instrument yielded informative and quantitative imaging and dosimetry of beam-activated isotopes in a PMMA phantom, as supported by Monte Carlo simulations. These studies open a new PET modality that can lead to improved imaging and monitoring of FLASH proton therapy.


Motivation
It is broadly acknowledged that in the 76 years since the Wilson seminal publication (Wilson 1946), proton therapy has not yet delivered on all of its potential and promises despite that it is generally considered to be the future of radiation oncology (PTCOG). Medical accelerators and beam delivery have gone through several hardware and software transformations. Treatment planning is now a complex, sophisticated and well established multi-step process assisted by advances in the software modeling of beam interactions in phantoms and patients (Moreno et al 2019). However, the in vivo accuracy and related functional imaging efficacy of each irradiation (e.g. proton range verification) has not kept up at the same pace of these development. There is much room for improvement of the monitoring and the assessment of treatment's progress that could boost results of the overall outcome of proton radiation therapy. This is particularly pertinent in the era of significant advances in hardware technology, data processing and image analysis.
A modern conventional intensity-modulated proton therapy (IMPT) treatment plan involves 25-35 fractions of about 2 Gy dose -1 each delivered over a few minutes and each fraction performed about once a day. A newer plan type, not yet widely prescribed, uses a significantly reduced number of fractions ('hypofractionations'), each of 15-20 Gy. The latest, an emerging, clinically not-yet-practiced plan is FLASH therapy. This is a single-fraction modality of ultra-high dose of up to about 40 Gy delivered in milliseconds or even shorter beam extraction times. It achieves an instantaneous dose rate that is several orders of magnitude higher than what is currently used in conventional clinical radiotherapy. There are reports indicating that FLASH radiotherapy results in a surprising, not yet understood, and unexpectedly promising better sparing effect of the healthy tissues (Favaudon et Hughes andParsons 2020, Lin et al 2021).
Radiation delivered at high dose rates and high doses (FLASH mode), seems to block or reduce oxygen-based DNA damage, as demonstrated in extensive animal studies (Montay-Gruel et al 2019). Although we measure the same delivered dose at conventional and FLASH rates, the biological response within each cell nucleus is completely different depending if it's FLASH or conventional dose rates. In addition, FLASH delivered radiation differentially protects more healthy cells over cancer cells, which is known as the FLASH effect. Since it is practically impossible to study how FLASH alters individual cells in patients and animals,all studies have focused on evaluating if the recipient's organ is functional or not, so a binary 'on/off' response of the organ. This is a major weakness of many of the FLASH animal studies, which are unable to identify and assess sub-regions within an organ where FLASH led to less damage and other sub-regions where FLASH didn't protect. Real-time functional positron-emission tomography (PET) studies with short lived tracers, as proposed in this work, will allow us to study 'how FLASH affects different regions within an organ' and 'why'.
The FLASH treatment technique has the potential to revolutionize radiation oncology, thus has received tremendous amount of attention and is vigorously pursued by many clinical research groups. However, the method must rigorously demonstrate that such radiation delivery indeed reduces the normal tissue toxicities, commonly associated with conventional radiotherapy, and is effective in tumor eradication. The underlying biomedical mechanism responsible for the FLASH positive effects must be understood and fully elucidated if the therapy is to be fully exploited and adopted. But even before then, the impact on medical personnel and instrumentation must be fully characterized so that such fast extraction can be safely experimented with and can ultimately be routinely employed in therapy.
FLASH proton beams with few fractions have a great potential for positively impacting the radiation therapy economy by substantially increasing the patient throughput, while improving the treatment outcome and significantly decreasing short-term and long-term post-radiation side effects. PET scanners may play a major role in assisting in the transition to a routine image-guided FLASH therapy. However, there are technical challenges related to the availability of the FLASH extraction at many existing proton treatment centers and the necessary in-beam PET scanners are still being developed (Crespo et al 2006, 2007, Lang 2022. In FLASH extractions, the instantaneous intensity of positron yield due to proton beam activation may be as much as 1000 times higher than in conventionally fractionated proton beam spills used in IMPT. Radiation effects associated with a typical 2 Gy dose, typically delivered over a minute, are compressed to a fraction of a second. This poses constraints and challenges for instruments, including PET detectors surrounding the irradiated tissue and presents opportunities for using the strong and fast signal emitted by isotopes activated by protons (e.g. 15 O, 13 N, or 11 C) and their minimal biological washout. A PET scanner must function in the spatial and temporal proximity to the beam that creates an intense radiation zone, including penetrating (i.e. radiation damaging) low-energy neutrons. It is critical and necessary to systematically and thoroughly explore the FLASH beam and its effects on the surroundings.
In this work we are motivated by potential benefits of in-beam PET imaging of FLASH beam activation that results in an untapped source of short-lived isotopes. An important added value is that this signal is not washed out from a tissue during a sub-second spill duration thus represents a crisp snapshot of the delivered beam dose. The strength of the signal is also unique since it is due to a dose of tens of grays delivered in one shot rather than over thirty days. If most of these spill events can be registered and used for imaging and dosimetry this will constitute unprecedented probing of irradiated tissue and will provide unique feedback and monitoring of proton therapy. We report here results of some of the first steps in the direction of characterizing FLASH radiation environment. We first employed detectors known to operate well in high radiation areas to help characterize the FLASH beam radiation surroundings. After an initial online assessment, we then introduced and exposed PET modules for measuring the yield and the time evolution of activated positron emitters. Details of our work are presented below.

Experimental setup
Our tests were conducted at the ocular beam of the Proton Center of the MD Anderson Cancer Center. This is a FLASH beam designed and characterized by the Center's staff (Titt et al 2021, Yang et al 2022. Various logistical factors constrained the time available for the PET exposure resulting in a limited scope of FLASH beam irradiation measurements. Beam: A Hitachi medical synchrotron at MD Anderson delivers a beam that has a nominal energy of 87 MeV that drops to 75.8 MeV after passing through the beam shaping passive elements and has a 2σ width of 1.16 MeV of the Gaussian energy distribution. The FLASH extraction (spill) lasts about 101.5 ms. The beam has an intensity of about 3.5 × 10 10 protons per spill and delivers about 163.7 Gy s -1 (PTW North America Corporation). The main features of the beam are illustrated in figure 1. The figure also shows the time structure of the specific spills used in our tests.
Radiation counters: We conducted measurements to characterize the beam and its radiation environment using the cadmium-zinc-telluride (CZT) crystals and common plastic scintillators. These are counters mostly sensitive to ionizing electromagnetic radiation (i.e. gammas and electrons) and to charged particles. The CZT counters were provided by H3D, Inc. The plastic scintillation counters were made out of BC408 (BC-408) scintillator of dimensions 19.1 × 50.8 × 76.2 mm 3 glued to a 50.8 mm diameter Philips photomultiplier tube (PMT) model XP2230 (Philips Amperex).
PET setup: We configured two PET modules used in a larger PET scanner completed recently for preclinical tests in proton beam therapy (TOF-PET for Proton Therapy (TPPT), TPPT Consortium 2021, Klein et al 2021Klein et al , 2022. The two PET modules were assembled out of two 8 × 8 arrays, each made out of Lu 1.8 Y 0.2 SiO 5 : Ce (LYSO:Ce) scintillation crystals.
Each 64-element LYSO scintillation crystal array features 'pixel' crystals of dimensions 3.005 × 3.005 × 15 mm 3 (Tong, ) and each pixel crystal was coupled to a Hamamatsu S14161-3050HS-08 silicon photomultiplier (SiPM), also known as a Multi-Pixel Photon Counter (MPPC) (Hamamatsu silicon photo-multipliers (SiPM)). Figure 2 shows one such module. All pixel crystals in the LYSO arrays were polished and separated by three layers of an enhanced specular reflection (ESR) film (ESR) so that the pitch of crystals with wrappings closely matched the SiPM pixel pitch that features 8 × 8 pixels of 3.0 mm × 3.0 mm dimensions set 0.2 mm apart.   (Tong). The left figure shows a conceptual design drawing while the center picture depicts a realized module. The right picture shows a two-array LYSO module coupled to their SiPMs and without the front-face cover.
The two modules were placed 390 mm apart facing each other and equally distant on either side of the polymethyl methacrylate (PMMA) cylindrical phantom of 25.4 mm diameter and 101.6 mm length. The phantom was centered on the beam axis delivered through a 30 mm diameter opening in a brass shield. This PET 'miniscanner' is shown schematically in figure 3 and in a picture in figure 4. The phantom was outside of the field of view of the top 8 × 8 arrays of PET modules and the bottom arrays were elevated by 6.35 mm with respect to the bottom edge of the phantom cylinder resulting in the reduction of the field of view of the bottom LYSO arrays, as is illustrated in figure 3 and 4.
The readout electronics was provided by PETsys Electronics (PETSys Electronics) and is identical to the readout employed in the time-of-flight proton therapy (TPPT) PET scanner (TOF-PET for Proton Therapy (TPPT)). The individual channel readout electronics use the PETsys TOFPET2 ASIC in conjunction with the PETsys FEB/S SiPM readout board and the PETsys FEB/I ASIC interface board (PETSys Electronics). The SiPM was bumped-soldered to the FEB/S. A proprietary data acquisition system was also provided by PETsys Electronics.

Measurements using CZT counters
To advance our initial FLASH beam studies a 20 × 20 × 5 cm 3 PMMA block, shown in figures 5(a)-(b), was irradiated with a 75.8 MeV FLASH proton beam. The target was positioned so that the beam central axis was aligned, as shown by the gray dashed lines in figures 5(d)-(e). For these irradiations an M400 gamma imager manufactured by H3D, Inc. (H3D,Inc.) was used to measure the spectrum of secondary gamma rays emitted from the PMMA target during and immediately after each irradiation. The M400 imagers contain a 2 × 2 array of CdZnTe crystals (2 × 2 × 1 cm 3 ) capable of measuring 511 keV annihilation gamma rays and prompt gammas up to 7 MeV at count rates up to 140 000 counts s -1 (Zhang et al 2012. The M400 was positioned at a lateral distance of 28 cm from the edge of the PMMA target to the front face of the imager. For dosimetric comparisons, an EBT-XD Gafchromic film was placed at 2 cm depth in a PMMA block and irradiated with a dose of 16.3 Gy (from a single 101.5 ms proton beam spill), determined using a previously published calibration protocol (Titt et al 2021) from the FLASH proton beam to provide a lateral profile of the beam size (figure 5(c)). Additionally, images of the discoloration in the PMMA block caused by the FLASH beam dose (>500 Gy total over all experiments) were analyzed, as shown in figures 5(d)-(e). One-dimensional (1D) profiles of the discoloration were extracted from images of the PMMA target. Additionally, 1D profiles of optical density (OD) were extracted from EBT-XD film (using ImageJ software (ImageJ)) and compared to 1D discoloration profiles from the reconstructed images of the 511 keV annihilation gammas measured with the M400. Figure 6(a) shows the measured spectrum of secondary gamma emission measured with the M400. The spectra show peaks for the positron annihilation gammas (511 keV) from induced carbon and oxygen isotopes and from pair production interactions of prompt gammas, as well as prompt gammas from 12 C (718 keV) emitted from the target during and immediately after FLASH beam irradiation. Figure 6(b) shows the resulting 2D image reconstructed from positron annihilation gammas in a (509-512)keV energy window, registered and overlaid onto the optical image of the experimental setup taken from the location of the M400 during irradiations. This gamma image represents a projection (integral) of the full gamma emission along the target thickness (y-direction) and, as such, has been registered to the front face (visible in image) of the target. This allows the 2D size and range of the FLASH pencil beam in the phantom to be visualized during and immediately after irradiation.
From the depth dose curves shown in figure 7(a) we see good agreement with the dose curve measured by the discoloration in the PMMA phantom (prior to the Bragg peak), and the Monte Carlo calculated curve. Discoloration in PMMA phantom beyond the Bragg peak is due to prior irradiations of the PMMA block during routine beam quality assurance measurements. The 1D depth profile from the reconstructed annihilation gamma image also shows a similar profile with a peak about 1 cm prior to the Bragg Peak depth as well as a falloff over a much longer range of depths. The shallower peak of the gamma image is not surprising because the cross section for positron emitting isotope production from carbon and oxygen atoms peak for proton energies from 25 to 50 MeV and falls off rapidly for lower energies ( This results in a peak in positron emitting isotope production at depths from about 0.5 to 2 cm proximal to the Bragg peak (Parodi et al 2002, Horst et al 2019. The elongated distal falloff of the gamma image is likely due to background artifacts introduced by the relatively simple back-projection method used for image reconstruction. Based on the results of previous studies Similarly, 1D lateral profiles of the measured FLASH beam dose profile at 2 cm depth are shown in figure 7(b). The lateral 1D profile extracted from the film shows the flattened dose profile extracted from the beam nozzle as described previously in (Titt et al 2021). However, the lateral dose profile from the PMMA phantom and from the gamma image show more of a Gaussian shape.

Measurements using plastic scintillation counters
The two plastic scintillator counters were configured on either side of a PMMA phantom, as pictured in figure 8 to count events during and after each spill of initial beam measurements. The outputs of the photomultipliers were connected to a discriminator whose outputs were connected to an Agilent 53230A 350 MHz universal Frequency Counter/Timer (Agilent) to count the number of pulses ('hits') as a function of time. The counter was set to sum counts in 100 ms time bins. It counted hits in individual counters ('singles') and doublecoincidence hits of both counters in the time proximity to the FLASH spills.
In figure 9, we display the number of 'singles' (i.e. counts in one of the plastic scintillation counters) and double-coincidence counts versus time in the counters during and after a single spill irradiated the phantom and empty target.   The nearly identical counts during the spill in all cases is the effect of prompt gammas hitting the counters and originating either in a phantom or a beam stop. The plastic scintillation counters registered about 2.7 × 10 5 counts in two 100 ms bins that included a FLASH spill. This is in contrast to our PET modules that saturated the data acquisition system for the 100 ms duration of a spill, as shown in figure 10 and discussed later. We used the number of hits in these counters during the first two seconds after a FLASH spill and compared it with the number of events recorded by the PET modules during the same time period. This ratio provides an estimate of the PET modules' response during the FLASH spill, as we describe it in section 4. Figure 9. Time spectra of 'singles' (left) and double-coincidence counts in plastic scintillators counter exposed to one FLASH spill. The left and middle spectra were obtained with the PMMA phantom while the right spectrum with an empty target (i.e. the air). Figure 10. Three FLASH extractions detected by PET modules. The top panel shows the overall time structure while bottom panels show magnified times scales around each spill (from 1 to 3 left to right). They clearly show dead-times of the PET data acquisition during the 101.5 ms time of each spill. The source of this dead-time is under investigation. The phantom activity prior to the first spill recorded by PET modules was induced by beam spills before PET modules were set in place.

PET measurements
We used the two-module PET system-a mini-scanner-for imaging and dosimetry of the FLASH beam irradiating a PMMA phantom. As described in section 2 and shown in figure 2, each module consists of two 64crystal LYSO arrays arranged vertically. However, the setup geometry was such that only the bottom arrays were sensitive to back-to-back gammas emitted from the activated phantom, as illustrated in figure 3. We are reporting here results from one short run lasting about 200 s. During this run our PET mini-scanner was exposed to register three FLASH spills that irradiated the phantom.
Each FLASH extraction had the intensity of about 3.5 × 10 10 protons. It lasted 101.5 ms and had the time structure shown in figure 1. The three FLASH spills were in close time proximity to one another, as shown in figure 10(a). We labeled the time axes such that the first FLASH spill occurred at 0 s ( figure 10(b)). The second spill (figure 10(c)) was 4.1 s after the first, and the third spill (figure 10(d)) was 6.4 s after the first one. The phantom was irradiated in prior extractions for studies with CZT detectors and plastic scintillators and its activation is clearly visible in figure 10(a). The data acquisition of the PET system started about 45 s before the first spill and continued past the third spill, as shown in figure 10(a).

Time distribution of PET events
The PET system readout dead-timed during each spill, as is clearly visible in figure 10. The reason for this deadtime is the high rate of hits in PET crystals which our readout and data acquisition system were unable to process. The exact nature of this limitation is related to a specific readout arrangement which will be improved during future PET exposure tests. During the 200 s of acquisition time, charge (proportional to energy) and time values were recorded and later processed into groups and coincidences. First, all detections were organized into groups. Each group consists of all charge depositions detected in the same crystal array within 100 ns of each other. Each event within a group was then ordered from highest to lowest charge. For the highest charge event within a group, a search was done among all the highest charge events in groups from the other crystal array to find another event detected within 10 ns of the first event. If such an event was found, these two events formed a coincidence and both of their charge and time values were recorded.
The phantom activity observed before the first PET spill was induced by a few beam spills before PET modules were set in place. A characteristic response for a typical crystal pair is shown in figure 11 for a representative example. A typical charge distribution reflects the underlying energy spectrum featuring a photopeak and a Compton shoulder (continuum). During the run, 163394 coincidences were detected in 3815 distinct pairs of readout channels. Each coincidence crystal pair allowed to construct a line of response (LOR) that was used for imaging. In the analysis described below we did not place any requirement on the energy value of the recorded events. The number of coincidences (equivalent to the number of LORs) as a function of time are plotted in figure 10 and 12. Of these 163394 coincidences, 119 390 were detected after the third spill.
As figure 10 shows, the data acquisition during spills was dead-timed. We used the plastic scintillation counters that showed no such problems to evaluate how many PET events were not recorded during each FLASH spill. By taking a ratio of the number of hit coincidences in the plastic counters (figure 9) to the number Figure 11. Characteristic energy spectrum expressed in data acquisition (DAQ) units detected in PET crystals shown for crystals of three lines of response (LOR). Each LOR is reconstructed using positions of two crystals with deposited energy read out as charge. Each panel shows such charge spectra in blue and red corresponding to two crystals. of PET events (figure 10) recorded over the period of two seconds past the first spill (394 counts and 3136 counts, respectively), and correcting for their relative distance to the phantom, we estimated that 534, 657 PET coincidences (of prompt gammas) may have been missed in each FLASH spill. In this estimation we assumed no pile-up events in neither the plastic scintillators nor in PET crystals.
The reconstructed LORs were used in further analyzes that included their timing since spills, imaging, and dosimetry. The phantom material of PMMA, has a chemical formula of C O H n 5 2 8 ( ) and the density of 1.18 gm cm −3 . The main positron-emitting isotopes that a 75.8 MeV proton beam yields in such material are 10 C, 11 C, 13 N, 15 O whose half-lives range from about 20 s to about 20 min, as listed in table 1. The yield of these isotopes depends not only on the elemental composition of the phantom but also on various nuclear fragmentation processes, dominant examples of which are also shown in table 1. These nuclear reactions occur with energy-dependent cross-sections some of which are not well known, as we discuss it in section 5.
The number of observed LORs as a function of time is shown in figure 10. We marked the clearly visible jump in activation of the phantom by three FLASH spills as the start of time. We then fit the number of observed PET coincidences since the start time with the function that accounted for the main four isotopic species produced in the phantom. The specific fit function used for the phantom's activity t ( )  as a function of time was where the amplitudes a−d are the fit parameters and λ( x Y) denote the inverse of half-lives of isotopes x Y. The constant term e accounts for the activity due to earlier spills, when the phantom was irradiated for initial beam studies. This parameter is determined by fitting to the data before the first spill, as shown in figure 12. This function describes the time dependence very well and confirms the expected composition of isotopes which we also show in figure 12. The main discrepancy between this fit and data is visible in the early time after spill where several other short-lived isotopes are present but not accounted for in the fit. We discuss this in the simulation section 5.

PET imaging and dosimetry
The positions of PET modules were such that lines of response were effectively almost perpendicular to the direction of the beam striking the PMMA phantom, as illustrated in figure 3 and 4. We have used the CASToR imaging package (Merlin et al 2018) to create images using LORs recorded in the run. The three canonical views, coronal, transaxial, and sagittal, are shown figure 13, 14, and 15. The field of view of the PET mini-scanner is depicted by straight lines connecting PET modules. The image reconstruction used a voxel size of 1.5 × 1.5 × 1.5 mm 3 . Reconstruction is conducted iteratively many times but in our case the tomographic (i.e. angular) view is extremely limited so an overall good quality contrast image is usually obtained with a relatively low number of iterations. Upon examining up to 50 iterations we selected the 5th image iteration as qualitatively the sharpest and best reflecting the beam-phantom configuration. These images show some uncorrected for efficiencies of the PET modules that also demonstrate the dosimetric capability of this in-beam PET miniscanner by counting the numbers of LORs for (i.e. coincidences). Full analysis of these PET images and dosimetry is underway but here we offer a first glimpse of such unprecedented possibility for a FLASH beam. For the coronal and transaxial views (figures 13 and 14), we also show the dosimetric counts in PET crystals represented by the black bars which reflect the number of PET coincidences detected in columns (coronal view) and rows (transaxial view) of the crystal arrays. Red bars denote columns and rows that include either a dead or a noisy pixel. These figures constitute the synthesis of our imaging and dosimetry measurements.
Despite a limited field of view, the images clearly show the path of production of positron-emitting isotopes. For better visibility we show in figure 16 the magnified image portions of figures 13-15. As expected and supported by simulations discussed in the next section, the intensity of LORs reaches a maximum towards the end of range of the beam-about 10 mm shy of the maximum dose and the Bragg peak. The counting of events and LORs proves the ability of dosimetric capability of the system.

Geant4 simulations of beam interactions in the phantom
We simulated proton beam interactions with the PMMA phantom using the Geant4 toolkit (Geant4, 2016). The QGSP_BIC_ HP reference physics list was employed, enabling high precision neutron and electromagnetic physics modeling.
The proton beam was assumed to be an axially-symmetric pencil beam of intensity of 3.5 × 10 10 protons per 101.5 ms impinging onto a PMMA phantom cylinder of 25.4 mm diameter and 100 mm length along the cylinder's axis. The radial profile was modeled following the experimental data (figure 1) and the proton energy was sampled from a Gaussian distribution with the mean of 75.8 MeV and the 2σ width of 1.16 MeV. The chemical composition and density assumed in our simulations reflected those of our phantom, i.e. C 5 O 2 H 8 and 1.18 g cm −3 , respectively. There are reports pointing out some discrepancies of Geant4 modeling with lowenergy data and resulting variance in predicting spatial profiles of the positron emitting species (PES) (Bauer et al 2013, España et al 2014. For completeness and comparisons, we have adopted two approaches in our simulations. The first is based on the 'out-of-the-box' Geant4 and the second one is 'a modified Geant4' approach described in (Bauer et al 2013). The modified Geant4 refers to the introduction of 'custom' crosssections that are used to generate the most abundant PES.
The simulated dose profile is shown in figure 17 and features the Bragg peak that falls off at a distance of about 41 mm from the upstream entrance of the beam. The simulated combined depth profile for main PES (i.e. 11 C, 15 O, 13 N and 10 C) is also shown there. We note an approximately 7 mm difference in the distal edge position of the dose and the PES distributions. This difference is due to the energy threshold for production of these PES and the precipitating energy loss by protons towards the Bragg falloff.
The modified Geant4 approach of (Bauer et al 2013) was originally developed for an out-of-beam PET scanner where the effect of short-lived species (half-life of a minute or less) was ignored. A direct Geant4   figure 13, 14, and 15. Shown are contours of the phantom, the PET module and lines defining the field of view of the PET system. simulation suggests that 12 N (half-life of 11ms), 10 C (19.3 s) and 8 B (0.77 s) are also generated in the phantom and can be recorded by an in-beam scanner. While these species account for only a few percent of the long-lived ones, the short decay time results in a significant contribution to the activity at short times after the end of irradiation. Considering the data acquisition time in our measurements of about 200 s and ignoring an expected 'spike' right after the end of the spill, only the effect of 10 C is included in our considerations. The production cross-section used in our simulations was taken from (Horst et al 2019). In the energy range below the first experimental data point at 40.7 MeV, we have used linear interpolation, assuming the cross-section to be zero at 31.83 MeV, which is the Q-value of the C p, p2n C 12 10 ( ) reaction. In figure 18 we compare our PET data to the two models used in simulations. We have scaled the simulated activity to the PET mini-scanner data. This was done to compare the shapes of the timing distributions without conducting much more demanding full detector and setup simulations. For the scaling, we used a ratio factor between the data and the models in the time interval from 135 to 155 s. This 'normalization' clearly indicates a fair agreement among two models and the data past about 25 s after the spill. This comparison also suggests additional short-lived activity in the phantom that cannot be explained by the 10 C decays. A statistical measure of the comparison is expressed by a reduced χ 2 /ndf = 1.87 for the modified Geant4 model as compared to χ 2 /ndf = 4.08, if we limit the time range from 25 to 155 s. These values change to χ 2 /ndf = 3.48 and χ 2 /ndf = 4.89, respectively, if we use the time range from 25 to 75 s. Our measurements demonstrate that an  Figure 18. The distribution of times of recorded PET events overlayed with simulated scaled activity using Geant4 and a modified Geant4 approach discussed in the text. Note that for the modified Geant4 approach the simulation provides by-isotope activity after the spills, and the resulting time evolution is computed assuming the exponential decay with the known half-lives of isotopes. The statistical uncertainty of the results of the direct Geant4 simulation is about 9%.
in-beam PET scanner and judiciously chosen phantom materials can help constrain some cross-sections for PES production that could further improve simulations, imaging, and dosimetry.

Conclusions
In this report we demonstrate the unprecedented proof of principle of beneficial employment of an in-beam PET scanner for imaging and dosimetry of a FLASH proton beam. We have successfully realized tests which open a new PET modality with proton FLASH beams leading to improved monitoring of irradiations and imageguided FLASH proton therapy.
Further analysis exploiting our data and modeling will be forthcoming. We will also conduct additional studies of the in-spill PET data acquisition and of potential radiation damage of detector components. The main motivation for this report is to share with the public our recent results of studying FLASH beam data before more elaborate measurements can be conducted. Our results deliver encouraging news about PET imaging of FLASH beams.