We used a pencil beam scanning gantry-mounted synchrocyclotron (S250i Hyperscan^TM, Mevion Medical Systems, Littleton, MA, USA) in clinical operation to perform scintillation imaging experiments at conventional and UHDRs. UHDR beam delivery using the Hyperscan^TM system has been demonstrated (Darafsheh et al 2020, 2021). The Mevion FLASH product includes three FLASH-enabled components: (1) a FLASH-capable synchrocyclotron that enables the delivery of UHDRs, (2) a FLASH Research Kit that acts as a portable dosimetry system capable of measuring and controlling FLASH deliveries, and (3) a FLASH Research Mode in the treatment delivery system to ensure integrity of the clinical treatment delivery parameters. The FLASH Research Kit can be installed on a compatible system in ∼15 min and is completely removed when FLASH research is not being performed such that the clinical synchrocyclotron is unaffected. When in FLASH mode, the Mevion system can deliver ∼100 Gy s⁻¹ average dose rate at the BP. Just like the clinical deliveries, the FLASH deliveries are pulsed with a nominal pulse frequency of 750 Hz, however, the pulse frequency can be modified between 600–780 Hz in FLASH Mode, and generally the duration of each pulse can be changed between ∼5–30 µs to change the charge-per-pulse, and hence the average dose rate. The Mevion FLASH dosimetry system can terminate the beam based upon a set number of pulses or an accumulated charge in the FLASH-enabled transmission ionization chamber, which can be calibrated to the desired dose in a target with 2% accuracy at the calibrated dose rate and 5% linearity across the entire dose rate spectrum (Lowe et al 2025). UHDR deliveries are capable of being scanned across a 5 × 5 cm² area with the FLASH kit.

Maximum energy of the protons prior to beam extraction is 227 MeV (32.2 cm range in water). For clinical use, proton range is modulated through a series of 18 polycarbonate range shifter plates within a motorized apparatus inside the nozzle. However, for FLASH RT delivery, a mechanical platform is mounted outside the nozzle and the beam’s range is changed by placing appropriate thicknesses of B₄C and poly(methyl methacrylate) blocks with pre-determined water-equivalent thickness in the beam line (the dose rate remains within UHDR regime) (Darafsheh and Bey 2025).

UHDR proton irradiation was done with the gantry at 90°. The geometry of our scintillation imaging setup is shown in figures 1(a)–(d). BC-408 (Saint-Gobain Crystals, Hiram, OH, USA) plastic scintillators (mass density ρ = 1.032 g cm⁻³, peak emission wavelength λ_peak ∼ 425 nm, rise time t_rise = 0.9 ns, and decay time t_decay = 2.1 ns) were used in this work. Scintillation imaging was performed in an optically isolated housing made of plywood with ∼3 mm water-equivalent thickness. To measure the proton range and relative intensity-per-pulse we used a 30 × 30 × 5 cm³ scintillator block (S₁ in figure 1(a)). All faces of the scintillator block have been polished. For spot size and width measurements, a 30 × 30 × 0.5 cm³ scintillator block (S₂ in figure 1(b)) was placed perpendicular to the beam direction, and a mirror placed at 45° reflected the scintillation images onto the camera’s sensor.

We used a high-speed complementary metal–oxide–semiconductor camera (VC-12MX2-M 330F, Vision Systems Technology, Vista, CA, USA) system with ‘Viewworks Imaging Solution 7’ capable of imaging at over 750 frames per second. The camera allows external pulse to trigger its shutter to open and close at programmed intervals. The camera was shielded against scattered neutrons by using dual-layer borated polyethylene blocks. The camera was located at ∼110 cm distance from the isocenter and perpendicular to the central axis of the beam. The isocenter is located at 205 cm distance from the exit of the cyclotron. The camera was placed on a thick copper block and cooled by frozen gel packs and miniaturized cooling fans in effort to reduce thermal noise.

The radiation output structure of our machine is shown in figure 1(e). Protons are delivered with macro pulses with 5–30 μs temporal width separated apart by 1333 μs (i.e. 756 Hz). The ion source pulse precedes each proton pulse by 500 μs (i.e. it takes approximately 500 μs for the injected ions at the center of the cyclotron to fully accelerate and exit the cyclotron). To perform pulse-by-pulse imaging, a gating module was used to synchronize the camera with the synchrocyclotron. The ion source pulse was used to provide a trigger for the camera’s shutter. The camera’s shutter opened 350 µs after the ion source pulse is detected and remained open for 980 μs ensuring imaging of a whole pulse in a single frame. This process repeated for all pulses. The images were saved in 8-bit tagged image file format. To create an average background image, images were acquired prior to and after each measurement, when the beam was off. The created background image was subtracted from the captured scintillation images prior to analysis. We used in-house scripts in MATLAB® (MathWorks, Natick, MA, USA) for image processing.

Geometrical scaling factors that convert pixel numbers to distance in vertical and horizontal directions were obtained through imaging a mm-scale grid pattern at the location of the scintillator block. In order to account for geometrical image distortions, we established non-linear scaling factors obtained by imaging a grid pattern.

Proton beams with water-equivalent ranges of 5, 10, 15, 18, 20, and 25 cm (corresponding to 79, 117, 147, 163, 173, and 196 MeV energy) were delivered to perform pulse-by-pulse evaluation of the range and relative dose-per-pulse using setup S₁. Proton beams with charge-per-pulse of ∼10–35 pC were delivered to investigate the impact of average dose rate on the response of the scintillator. The corresponding dose rates at the BP are ∼30–100 Gy s⁻¹. Figures 1(f) and (g) show typical images obtained from setups S₁ and S₂, respectively. After preprocessing, in-house scripts generated integrated depth intensity profiles by adding signal from all pixels in each column (within the two horizontal dotted lines in figure 1(f)) and assigning the integrated value to the respective depth. Then the peak depth was identified by the location of the maximum intensity. This process was conducted on a pulse-by-pulse basis. BP widths at 90% and 80% of the peak height were extracted for each pulse through the integrated depth intensity profiles. In addition, beam width at mid-range (R/2) and at the BP were computed by drawing vertical profiles and computing their full width at the half-maximum. The red circle in figure 1(f) (with ∼2.5 mm diameter (drawn not to scale)) shows a region-of-interest (ROI) over which the pixel values were integrated as a surrogate for intensity to correlate with the beam current. Depth dependent empirical correction factors were applied to correct for ionization quenching; the correction factors were obtained by dividing an integral depth dose (IDD) profile measured by an ionization chamber to that measured by the scintillation imaging system under the BP (Goddu et al 2022). We verified using radiochromic film dosimetry that the shape of the IDD did not change with the dose rate which allowed us to measure the IDD using a parallel-plate ionization chamber at a conventional dose rate (Darafsheh and Bey 2025). Since the Mevion system does not employ an energy selection system, the distal shape of the IDD profiles remain the same regardless of the proton energy (Goddu et al 2022, Darafsheh and Bey 2025). As such, we used the same correction factors for all ranges. We determined the distal 90% depth that represents the range of the beam.

UHDR proton beams with full energy and ∼30 pC-per-pulse were scanned over a 2 × 2 cm² area with 1 cm spot spacing to measure the spot size and position on a pulse-by-pulse basis using scintillator S₂ (figure 1(b)). After image preprocessing, in-house scripts determined the location of each spot. First, the grayscale image was converted to a binary image, and regions were labeled. Properties of these labeled regions, including the weighted centroid, were calculated, and the centroid of the largest, roughly circular region was identified as the spot. At this coordinate, horizontal and vertical profiles were drawn to determine the spot position and size (sigma) as shown in figure 1(g).

Since the camera was synchronized with the synchrocyclotron, each image corresponds to a delivered pulse. A total of 11980 pules in 30 datasets, collected under Setup 1, that include three different ranges were analyzed. The scintillation intensity captured in each 2D image, which is proportional to the dose-per-pulse, is encoded in the pixel values. We integrated the pixel values over an ROI centered at the BP location as a surrogate to the relative dose delivered by the synchrocyclotron. The ROI was a circle with 30 pixels in diameter (∼2.5 mm) along central axis and centered at the BP location as shown in figure 1(f). Integrated scintillation intensity of each frame was compared to the corresponding delivered charge-per-pulse recorded in the machine log files, as exemplified in figures 2(a) and (b) for 400 pulses. In figure 2(a), the integrated intensity and charge-per-pulse for each data point was normalized by their respective average value within the 400 pulses. Figure 2(a) demonstrates on average ±10% variation in charge-per-pulse across the 400 delivered pulses that is well-correlated with the transmission ion-chamber measurements. In some instances, the variation in dose-per-pulse reaches ∼15%. It can be seen in figure 2(b) that the majority of points are within 0.5% when the scintillation signal in each pulse was divided by the respective delivered charge, indicating an excellent correlation between the scintillation intensities and the charge-per-pulse. Figure 2(c) shows the integrated scintillation intensity as a function of charge-per-pulse (i.e. dose rate). The linear trend indicates the linearity of the scintillator signal with dose-per-pulse (instantaneous dose rate) from low dose rates up to the UHDRs studied in our work. Each circle in figure 2(c) represents the average image intensity in a data set containing ∼400 images (∼400 pulses).

A typical scintillation image obtained through the S₁ setup is shown in figure 1(f). We extracted various beam parameters on a pulse-by-pulse basis through images obtained with scintillator S₁. These parameters include proton range (distal R₉₀), Bragg-peak width at 90% and 80% of the peak height, the beam’s width measured at mid-range and at the BP, as well as the BP’s vertical position. The variations presented in figure 3 are not absolute; they are computed from their respective average values in each set. The analysis of pulse-by-pulse range variation presented in figure 3(a) for three sets of proton beams each set with 6 different ranges with ∼300 pulses shows that the range variation is less than 0.1 mm in most cases. Figure 3(b) shows that the deviation in spot vertical position relative to the average value is mostly within ±0.5 mm. Figures 3(c) and (d) show the deviation with respect to the average value is mostly within 0.3 mm for the BP width at 90% and 80%, respectively. Figures 3(e) and (f) show the deviation with respect to the average value for the spots’ width is mostly within 0.5 mm at mid-range and BP, respectively.

Figure 4(a) shows the average measured range for all delivered beams. A linear trend was observed between the measured range using the scintillator and the range measured by the multi-layer ionization chamber (MLIC) (Zebra, IBA). In most cases the measured range is within ±1 mm of the programmed range. Figure 4(b) shows that scintillator underestimates the range at shorter ranges (R_MLIC − R_Scint. > 0) and overestimates the range at longer ranges (R_MLIC − R_Scint. < 0). This systematic behavior requires further investigation.

Figure 5 shows pulse-by-pulse analysis of proton spot size and position measurement along the x- and y-direction using setup S₂ for 1350 pulses delivered with full energy. Nine spots (each with 50 pulses), spatially separated by 10 mm distance, were delivered three times across a 2 × 2 cm² area as shown in figure 5(a). A typical scintillation image of a single pulse of a single spot is shown in figure 1(g). The difference between the spot size (σ) measurement using our scintillation imaging system and the expected value measured using a commercial scintillation screen (Lynx, IBA) along the x- and y-axis for all 9 spots is shown in figures 5(b) and (c). The nominal corresponding values are 4.5 and 5.3 mm, respectively (Darafsheh and Bey 2025). The majority of pulses are within ±0.3 mm of the expected value. The systematic differences between the expected and measured values may be due to a slight misalignment of the mirror with respect to the camera.

The difference between the nominal and measured position along the x- and y-axis is shown in figures 5(d) and (e) with majority of the spots within ±0.2 mm of the corresponding nominal spot position.

Ionization quenching is manifested as an under-response of the scintillator around the BP, where the LET is high, in proton depth dose curves (Birks 1964, Wang et al 2012, Robertson et al 2013). To further investigate any dose-rate-dependency of ionization quenching of our scintillator at UHDRs, we compared integrated depth intensity profiles obtained at three average dose rates of ∼22, 43, and 86 Gy s⁻¹ corresponding to 7, 14, and 28 pC-per-pulse in figure 6(a). In figure 6(b), we present the intensity of the images at the BP divided by that at 5 cm depth. These results as well as the linear trend in figure 2(c) do not point toward a noticeable dose rate dependency in ionization quenching, indicating that the same correction method can be applied in both low and UHDRs produced by our machine.