Performance of the HYPERSCINT scintillation dosimetry research platform for the 1.5 T MR-linac

Objective. Adaptive radiotherapy techniques available on the MR-linac, such as daily plan adaptation, gating, and dynamic tracking, require versatile dosimetric detectors to validate end-to-end workflows. Plastic scintillator detectors (PSDs) offer great potential with features including: water equivalency, MRI-compatibility, and time-resolved dose measurements. Here, we characterize the performance of the HYPERSCINT RP-200 PSD (MedScint, Quebec, CA) in a 1.5 T MR-linac, and we demonstrate its suitability for dosimetry, including in a moving target. Approach. Standard techniques of detector testing were performed using a Beamscan water tank (PTW, Freiburg, DE) and compared to microDiamond (PTW, Freiburg, DE) readings. Orientation dependency was tested using the same phantom. An RW3 solid water phantom was used to evaluate detector consistency, dose linearity, and dose rate dependence. To determine the sensitivity to motion and to MRI scanning, the Quasar MRI4D phantom (Modus, London, ON) was used statically or with sinusoidal motion (A = 10 mm, T = 4 s) to compare PSD and Semiflex ionization chamber (PTW, Freiburg, DE) readings. Conformal beams from gantry 0° and 90° were used as well as a 15-beam 8 × 7.5 Gy lung IMRT plan. Main results. Measured profiles, PDD curves and field-size dependence were consistent with the microDiamond readings with differences well within our clinical tolerances. The angular dependence gave variations up to 0.8% when not irradiating directly from behind the scintillation point. Experiments revealed excellent detector consistency between repeated measurements (SD = 0.06%), near-perfect dose linearity (R 2 = 1) and a dose rate dependence <0.3%. Dosimetric effects of MRI scanning (≤0.3%) and motion (≤1.3%) were minimal. Measurements were consistent with the Semiflex (differences ≤1%), and with the treatment planning system with differences of 0.8% and 0.4%, with and without motion. Significance. This study demonstrates the suitability of the HYPERSCINT PSD for accurate time-resolved dosimetry measurements in the 1.5 T MR-linac, including during MR scanning and target motion.


Introduction
The introduction of the MR-linac device, a linear ccelerator with integrated magnetic resonance imaging (MRI) functionality, increases the availability of (online) adaptive radiation treatments , Mutic and Dempsey 2014, Raaymakers et al 2017. In the current clinical MR-linac workflow, a daily MRI is used for softtissue based position verification and pre-beam plan adaptation to compensate for inter-fraction anatomical variation (Winkel et al 2019). Furthermore, the availability of real-time MRI imaging facilitates online-adaptive treatments such as gating and multileaf collimator (MLC) tracking to compensate for intra-fraction motion (Klüter 2019, Akdag et al 2022, Uijtewaal et al 2022. To validate gating and MLC tracking functionality on the MR-linac, motion phantoms with integrated dosimeters are needed to measure dose in the moving target (Keall et al 2021). The increasing treatment complexity and motion-delivery interplay also increase the demand for Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. time-resolved dosimetry detectors (Ravkilde et al 2013). Importantly, these detectors must be MR compatible so as not to disturb the MRI, while their performance should not be disturbed by the magnetic field and its gradients, or by radiofrequency (RF) pulses, such that they can validate online adaptive end-to-end workflows. Furthermore, their performance should not be affected by the motion of the phantom.
Dose measurements in a strong magnetic field are challenging because the Lorentz force deflects the path of secondary electrons, resulting in a disrupted dose deposition, especially in the presence of density gradients (Raaymakers et al 2004, Raaijmakers et al 2005, Jelen and Begg 2019. Previous studies demonstrated the performance of different detectors in a 1.5 T MR-linac, including: ion chambers, a diamond detector, and diode detectors (Meijsing et al 2009, Reynolds et al 2014, O'Brien et al 2018, Woodings et al 2018a. Although it was shown that these detectors are suitable for use in an MR-linac, many detectors are not suitable for real-time MR-imaging. The metallic components used in many conventional detectors induce MRI artifacts (Meijsing et al 2009, Reynolds et al 2014, which limit the possibility to accurately follow the target positions with real-time imaging. Additionally, for many detectors the performance in a magnetic field strongly depends on their orientation , Woodings et al 2018a. These limitations introduce the need for an alternative type of detector. A promising alternative is a plastic scintillator dosimeter (PSD). In this detector, the scintillator emits optical photons proportional to the received energy. This photon flux is then guided to optical readout equipment through an optical fiber for dose tallying (Beddar et al 1992a, Beddar 1994, Clift et al 2000, Beddar 2006). PSDs have a relatively small volume (1-2 mm 3 ), a high degree of water equivalence in the MV photon range (Beddar et al 1992a, Beddart et al 1992b, Lambert et al 2008, 2010, a linear dose response, and their composition does not disturb the MRI (Klavsen et al 2022). A challenge with PSDs is the stem effect that impacts the accuracy of the dosimetry when not fully accounted for. This background signal is produced during irradiation of the optical fiber and is due to Cerenkov and fluorescence emissions (Beddar et al 1992a, Beddar 2006, Therriault-Proulx et al 2013. Among the different methods implemented in the literature (Parwaie et al 2018), the hyperspectral approach is the only one that accounts for both Cerenkov and fluorescence independently (Therriault-Proulx et al 2018). In the presence of strong static magnetic fields, both the intensity and composition of the stem effect are affected and vary with magnetic field strength and orientation. These magnetic field induced effects increase the need for an effective stem effect removal technique (Stefanowicz et al 2013, Parwaie et al 2018, Simiele et al 2018, Therriault-Proulx et al 2018. Another challenge of PSD is the potential impact of the magnetic field on the scintillation yield shown by other studies (Blomker and Holm 1992, Green et al 1995, Bertoldi et al 1997. A recent study on a 0.35 T MR-linac showed that magnetic field dependency can be minimized by calibrating the scintillator system directly in the MR linac (Klavsen et al 2022). However, the suitability and performance of PSDs have never been characterized on a 1.5 T MR-linac.
Regarding the validation of online-adaptive treatments, it is essential that the PSDs provide accurate timeresolved dose measurements in moving targets. Previous studies established both for C-arm linacs (Lambert et al 2006, Beierholm et al 2008, Sibolt et al 2017 and for a 0.35 T MR-linac (Klavsen et al 2022) that motion does not affect the accuracy of time-resolved PSD measurements. However, the use of PSD for time-resolved dosimetry in moving targets has not been examined in a 1.5 T MR-linac. Moreover, motion studies on the 0.35 T were restricted to conformal treatment fields from a single gantry angle as a proof of principle (Klavsen et al 2022).
The purpose of this study was to demonstrate the suitability of the HYPERSCINT RP-200 scintillation dosimetry research platform (MedScint, Quebec, CA) in a 1.5 T MR-linac, and to characterize the PSD performance. With a stem effect removal that accounts for scintillation, Cerenkov, fluorescence as well as optical fiber attenuation, we first characterize the performance of the Hyperscint regarding dose profiles, PDD curves, field-size dependency, repeatability, dose linearity, dose rate dependency, and orientation dependency. We then demonstrate the suitability of the Hyperscint for real-time, motion-related dosimetry applications in combination with a Quasar MRI 4D motion phantom (Modus Medical Devices Inc., London ON). To our knowledge, this is the first study detailing the performance of this commercial PSD detector on a 1.5 T MR-linac.

Methods
All experiments were performed on a Unity MR-linac (Elekta AB, Stockholm, Sweden), featuring a 1.5 T MR scanner and a 7 MV flattening filter free (FFF) linac with its beam axis perpendicular to the main magnetic field. We used several experimental setups containing a PTW Beamscan MR water tank (PTW, Freiburg, Germany), an RW3 solid water slab phantom, or a Quasar MRI 4D phantom (Modus Medical Devices Inc., London ON).

Detectors
The HYPERSCINT scintillation dosimetry research platform (MedScint, CA) was connected by a 20 m long optical fiber to a probe containing a scintillation detector ( figure 1(a)). The sensitive volume of the detector is 1 mm diameter × 1 mm length and the effective measurement point is 2.5 mm from the tip of the probe ( figure 1(b)). The PSD can perform time-resolved measurements. For the experiments we used sampling frequencies (F s ) of 1 or 4 Hz. During the experiments, the optical read-out device was placed outside the radiation bunker to avoid noise readings in the spectra measurements induced by scattered radiation.
To evaluate the performance of a HYPERSCINT PSD in a 1.5 T MR-linac, we compared its performance to the following detectors in different experimental setups: a PTW microDiamond (60019)

PSD calibration procedure
The PSD was calibrated following the vendor-specified guidelines to allow for a stem effect , Therriault-Proulx et al 2013 removal that independently accounts for scintillation, Cerenkov, and fluorescence light sources. First we used an Orthovoltage (Xstrahl 200 X-Ray Therapy System) with kV beam to separate scintillation and fluorescence light sources. Then we used an MV beam on the MR-linac to correct for Cerenkov light and to normalize the scintillation signal.
2.3. Comparing dose measurements with reference detectors 2.3.1. Dose profiles, PDD curves and field-size dependence We used the PTW Beamscan MR water tank to compare the dose profiles, percentage depth dose (PDD) curves and the field-size dependency of the PSD to measurements with the microDiamond. The water tank enables software controlled movements of detectors with a 0.1 mm accuracy in three directions. The microDiamond was fixated in the water tank by a Trufix holder (PTW, Freiburg, Germany) and the PSD was fixated by a 3D printed version of the holder. Measurements were obtained at the machine isocenter at a depth of 10 cm with a source-surface distance (SSD) of 133.5 cm and a 0°gantry angle.
Inline and crossline (IEC 61217) profiles were measured for a 3 × 3 cm 2 field, to compare the field shapes measured by the PSD and the microDiamond. A 3 × 3 cm 2 field was used because it is a small field (Das et al, 2021, AAPM 2017) with a clear in-field plateau and a distinguishable left and right penumbra, which is ideal to test the performance of small field detectors. To correct for misalignment between the sensitive volume of the detectors, the PSD profiles were corrected based on the determined shift between the profiles. For the analysis of the profiles, the area within the beam, defined as the area within the penumbra, was based on the NCS Linac QA report guidelines for flattening-filter-free (FFF) beams (Gobets et al 2020). PDD curves were measured along beam central axis for a 10 × 10 cm 2 field. Relative output factors (ROF) were measured for squared field sizes of 0.5 × 0.5 cm 2 to 22 × 22 cm 2 and for the rectangular field sizes 22 × 40 cm 2 and 22 × 57.4 cm 2 . The rectangular fields correspond to equivalent square field sizes of respectively 28.4 × 28.4 cm 2 and 31.8 × 31.8 cm 2 . Rectangular fields were used because the field size in inline direction is limited to 22 cm on the MR-linac, while in the crossline direction the maximum field reaches to 57.4 cm. Small field correction factors from the IAEA TRS-483 (AAPM 2017) were applied to the measurements.

PSD characteristics
In the following experiments, the performance consistency and the possible dependencies of the PSD in the 1.5 T MR-linac were evaluated. Unless otherwise noted, for all experiments we used a 30 × 30 × 24 cm 3 RW3 solid water phantom with tailored PSD drill hole according to the size of the detector. The phantom was positioned with the detector at the machine isocenter with a source axis distance (SAD) of 143.5 cm and a depth of 10 cm. To avoid air pockets around the detector, the detector opening in the phantom was filled with water before inserting the probe, and then sealed.

Repeatability, linearity and dose rate dependency
The repeatability or consistency of the PSD was verified by comparing the detector response of 10 consecutive measurements of 100 MU from a 10 × 10 cm 2 field at gantry 0°using the maximum dose rate (425 MU min −1 ). To test dose linearity, we repeated the measurements for 10-1000 MU. Finally, the dose rate dependency was verified by delivering 50 MU at dose rates varying from 50-400 MU min −1 in increments of 50 MU min −1 .

Sensitivity to air gaps
Air gaps around a detector could affect the dose deposition and dose measurement in a magnetic field (Hackett et al 2016, Malkov and Rogers 2017. The influence of air gaps around the PSD on the detector reading was determined by comparing measurements with and without water around the detector in the phantom's detector opening. Inserting the probe in the phantom without water creates air gaps around the detector that could affect the dose reading. To characterize the PSD's sensitivity to these air gaps, we compared measurements of 100 MU from a 10 × 10 cm 2 field at gantry 0°that were performed with and without water around the detector.

Angular dependency
The angular dependency was determined by comparing the detector readings from different gantry angles and different detector orientations. We fixated the PSD at the iso-center in the PTW Beamscan MR water tank at 16 cm depth using a 3D printed Trifux holder. A depth of 16 cm was used to create identical water-equivalent depths for the gantry angles 0°, 90°, and 270°. During the experiments we adjusted both the detector orientation and the gantry angle. The detector orientation was adjusted from its reference orientation (0°) by rotating it around the anteroposterior axis by increments of 90°(figure 2). For every detector orientation, three measurements of 100 MU from a 10 × 10 cm 2 field were performed at gantry angles: 0°, 90°, and 270°. The detector readings from the different gantry angles were normalized to gantry 0°using transmission factors, to correct for the difference in beam transmission through the cryostat between the different gantry angles (Woodings et al 2018b, Snyder et al 2020, Woodings et al 2021. The reproducibility uncertainty was estimated by repositioning the detector afterwards in the reference orientation and comparing the measured dose to the initial measurement.

Sensitivity to motion and to MRI scanning
To establish the sensitivity of the PSD to motion and to MRI scanning, the Quasar MRI 4D phantom was used. The phantom contained an ion chamber holder insert with a 3 cm spherical target (GTV) that was positioned centrally in a water-filled body oval. The Semiflex insert has an opening to fit a Semiflex ionization chamber or a scintillation probe, such that they could be positioned exactly at the center of the body oval. First, we used MV images, acquired on the MR-linac's integrated electronic portal imaging device (EPID) panel, to position the phantom and the detector exactly at the iso-center in the inline and crossline direction. Before inserting a detector in the phantom, the opening was filled with water to avoid air gaps around the detector. The phantom was static or programmed with sinusoidal motion (A = 10 mm, T = 4 s).
To determine the sensitivity of the PSD to motion, we compared a static delivery to a delivery with motion. We used a 20 × 20 cm 2 field with 100 MU at gantry 0°and at gantry 90°, and additionally a typical 15-beam IMRT 8 × 7.5 Gy lung SBRT plan with a 3 cm diameter target (GTV) and a 3 mm GTV-to-PTV margin. The IMRT plan was created in Monaco 5.40 (Elekta AB, Stockholm, Sweden). The reading of the PSD was compared to the reading of a Semiflex. Additionally, we determined if MRI scanning would affect the PSD's performance. To this end, we scanned the phantom with a clinically-used 3D balanced turbo field echo (TFE) sequence during the above described scenarios.

Dose profiles, PDD curves and field-size dependence
The inline and crossline profiles at depth 10 cm are shown in figure 3. The alignment of the PSD and the microDiamond profiles was within ±0.2 mm in both directions. The measured PSD profiles are in good agreement with the microDiamond profiles. The maximum difference within the beam is 0.5% for the inline profile and 1.0% for the crossline profile. For both profiles, the tails of the PSD and the microDiamond are in excellent agreement. Main differences are found in the penumbra region, where the microDiamond penumbra is steeper than that of the PSD in both the inline and the crossline profile. Figure 4(a) shows the 10 × 10 cm 2 PDD curve of the PSD and the microDiamond. The PSD and the microDiamond curve are in excellent agreement with a maximum difference of only 0.4% outside the build-up region. The maximum PDD value is found at a depth of 1.3 cm for both detectors.
For both the PSD and the microDiamond the ROF are shown in figure 4(b). The reference reading of the 10 × 10 cm 2 field was repeated throughout the experiments, and the standard deviation of these reference readings was 0.12% for the PSD and 0.16% for the microDiamond. The output of both detectors was very consistent for field sizes down to 1 × 1 cm 2 , with differences <1%. For the field sizes 0.5 × 0.5 cm 2 and 1 × 1 cm 2 a difference of respectively 3% and 4% was found.
3.2. PSD characteristics 3.2.1. Repeatability, linearity and dose rate dependency The standard deviation of the reproducibility measurement was 0.06%. The PSD dose linearity is shown in figure 5(a). The measured dose as a function of prescribed MU gives excellent linearity (r 2 = 1.0). The dose per MU was very similar for the different prescriptions with a relative standard deviation of 0.1%. The dose rate dependence was <0.3%.

Sensitivity to air gaps
The sensitivity to air gaps was tested by removing the water around the PSD. The diameter of the PSD housing and of the drill hole in the phantom were 5 mm and 6 mm, providing a layer of air around the detector of maximally 1 mm. The air around the detector reduced the measured dose by 1.4%. Figure 5(b) shows the dose measured from different detector orientations and from different gantry angles to determine the angular dependency of the PSD. All measurements were compared to a reference measurement without any rotations (detector orientation 0°and gantry 0°). Measurements with the detector anti-parallel (0°) and parallel (180°) to the magnetic field were in near-perfect agreement with the reference measurement with a maximum dose difference 0.2% for all three gantry angles. For the detector orientations perpendicular to the  magnetic field (90°and 270°), the measured dose was also very comparable to the reference dose with differences in the range of 0.3%-0.8%, when the detector was not positioned parallel to the beam (detector 90°and gantry 90°or detector 270°and gantry 270°). Positioning the detector parallel to the beam increased the dose by respectively 2.3% and 2.5%. Note that the parallel orientation of the detector and the beam is not recommended for any detector. We estimated a maximum setup uncertainty of 0.3%. Table 1 summarizes the planned dose and the PSD and Semiflex measurements in the Quasar phantom for different experimental setups. The dose measured by the PSD and the Semiflex were very comparable for all experiments, with differences 1%. The differences found when comparing the statically measured PSD dose to the planned dose were well within our clinical tolerances, with differences of 0.3% and 1.3% for the 20 × 20 cm 2 field measured at respectively gantry 0°and gantry 90°, and differences of 0.4% for the IMRT plan. These differences were similar for the Semiflex, except for the 20 × 20 cm 2 field measurement at gantry 90°w here we found a difference of 2.1%. Performing MR imaging during the measurements introduced negligible dosimetric differences of maximally 0.3% compared to a respective measurement without MR imaging. Applying motion introduced maximum differences of 0.2% for the PSD as well as for the Semiflex for the 20 × 20 cm 2 fields. However, for the IMRT plan we detected motion-induced dose reductions of up to 1.3% in both detectors. Here, the 20 mm peak-to-peak motion amplitude moves the detector through a shallow IMRT dose gradient, while for the 20 × 20 cm 2 field, the detector remains in the high dose area despite the introduced motion. Increasing F s from 1 Hz to 4 Hz during a measurement with moving phantom yielded dose differences <0.3%.

Discussion
In this study, we demonstrated the suitability of the HYPERSCINT PSD for time-resolved dosimetry measurements in a 1.5 T MR-linac. The results show that the PSD readings were consistent with the readings of previously validated detectors, and that the performance was excellent in terms of consistency, dose linearity, dose rate dependency, and angular dependency. Furthermore, the measurement accuracy was not affected by motion or by MRI scanning.
Measured PSD dose profiles, PDD curves, and field-size dependence were in excellent agreement with the microDiamond readings (differences 1%). Differences within the penumbra were larger (>5%) than within the beam area (1.0%) for both the inline and crossline profiles as a result of a steeper microDiamond penumbra. This was expected as both detectors have different densities (Francescon et al 2014, Brace et al 2020 and because the microDiamond has a smaller sensitive volume (0.004 mm 3 ) than the PSD (0.79 mm 3 ) with less volume averaging (Duggan and Coffey 1998). The PDD curves of both detectors agree very well beyond the build-up region, with differences 0.4%. Differences in the build-up region close to the surface can be due to density differences between the detectors (Francescon et al 2014, Brace et al 2020, different interactions with the detector holders that fixate the detectors, as well as to small positioning errors. Small shifts in the order of 0.2 mm can already lead to a change in response of the order of a few percent, because of the high dose gradient in the build-up region (Wegener et al 2020). For both detectors the depth of maximum dose (d max ) was found at a depth of approximately 1.3 cm inside water, a value in perfect agreement with previous studies on the 1.5 T MR- linac (Woodings et al 2018b, Chen et al 2019. This indicates the suitability of the PSD for measuring PDD curves. The PSD and the microDiamond gave very consistent ROF results for field sizes down to 2 × 2 cm 2 . The PSD measured a slightly lower value for the smallest field sizes, which is consistent with the greater volume averaging and lower density of the PSD. Other than comparing the PSD performance to another suitable MR-linac detector, we also performed standard detector characterization tests. The consistency of the PSD in terms of standard deviation of repetitive readings (0.06%) was well within the tolerance of 1.0% (Kutcher et al 1994). The results also showed perfect dose linearity (R 2 = 1), and very low dose rate dependence (<0.3%), which is in line with measurements without a magnetic field (Beddar et al 1992a, Beddar 2006, Lambert et al 2008, 2010. A minimal angular dependence (0.8%) was found, which is the first report of such performance. This excellent performance can be attributed to the use of an evolved hyperspectral approach that accounts for optical fiber Cerenkov and fluorescence emissions as well as its spectral attenuation. Only irradiating parallel to the optical fiber from behind the scintillation point leads to unacceptably large (2.3%-2.5%) variations. However, a parallel orientation of the detector to the beam is not recommended for any detector. In addition, it is a highly unlikely orientation for clinical use. Replacing the water around the detector with air reduced the measured dose (−1.4%), which is consistent with previous studies using ionization chambers (Hackett et al 2016, Malkov and Rogers 2017. This reduction is a result of deflections in the path length of secondary electrons due to the Lorentz force (Meijsing et al 2009). It is therefore important to perform PSD measurements in the presence of water or water-equivalent materials around the detector.
Validating online adaptive radiotherapy treatments requires versatile time-resolved detectors whose accuracy is not affected by target motion or by MRI scanning. Our results show that performing MRI scanning has only a minimal effect on the detector readings (0.3%). Previous studies established that PSDs do not disturb MR images (Klavsen et al 2022). As expected, target motion introduced dosimetric differences compared to a static delivery of 0.2% for a 20 × 20 cm 2 field and of maximally 1.3% for the IMRT plan. During the 20 × 20 cm 2 field delivery, the target did not move outside the high dose area, which is reflected by the measurements showing no changes compared to the static dose. For the IMRT plan, the applied motion kept the PSD well within the GTV. The PSD was thus only exposed to a shallow dose fall-off. Nonetheless, the measured dose reduction indicates the capability of the PSD to pick up subtle motion induced dose variations. The accuracy of the PSD readings was also confirmed by comparisons of a delivery with and without motion and MRI to Semiflex measurements (differences 1%) and a comparison of the static delivery to the planned dose (difference 0.5%). Increasing F s from 1 Hz to 4 Hz had only a minimal effect on the measured dose (<0.3%). An F s of 4 Hz would be suitable for time-resolved dosimetry for respiratory motion, as respiratory motion has a periodicity of typically 3-5 s (Seppenwoolde et al 2002, Krauss et al 2011. Together this shows the suitability of the PSD for use in MR-guided online adaptive radiotherapy workflows that requires continuous MR imaging to follow a moving target, such as gating and MLC-tracking (Akdag et al 2022, Uijtewaal et al 2022. Now that we have established the suitability of PSD in the 1.5 T environment, we can expand its use for new applications. An example would be increasing the number of scintillation points. More scintillation points would provide better volume coverage which could offer insight into the time-resolved dose delivered to the target in 2D or potentially even 3D. 2D arrays filled with PSDs have previously been prototyped (Petric et al 2006, Frelin et al 2008, Collomb-Patton et al 2009, Guillot et al 2011, Uijtewaal et al 2022. Future studies should further explore these detector arrays and establish their spatiotemporal dosimetric performance during motion experiments on the MR-linac. The flexibility of the individual positionable PSD also offers the potential for use in deformable targets (Cloutier et al 2021). Having individual detector points moving with the deformations of a target could provide great insight into the delivered dose and could potentially be used to validate the performance of deformable MLC tracking (Ge et al 2014).

Conclusion
This study demonstrates the suitability of the HYPERSCINT PSD for accurate time-resolved dosimetry measurements in the 1.5 T MR-linac. The excellent performance during continuous MR scanning and during dynamic movement indicates the great potential of the detector to validate end-to-end workflows of online adaptive radiotherapy.