Mevion S250i Hyperscan Proton Therapy
- The Mevion S250i Hyperscan System is a compact pencil beam scanning proton therapy platform featuring fixed 230 MeV beam extraction, 18 range shifter plates, and adaptive aperture collimation.
- It enables robust free-breathing intensity-modulated proton therapy for lung and esophageal cancers by integrating advanced 4D planning and interplay evaluation without rescanning.
- The platform supports cutting-edge research in proton dosimetry with β-Ga₂O₃ detectors and machine learning-based modeling of beam delivery times.
Searching arXiv for papers on the Mevion S250i Hyperscan system and closely related topics. The Mevion S250i Hyperscan System is a compact pencil beam scanning (PBS) proton therapy system with Hyperscan technology, described as utilizing a fixed 230 MeV beam and modulating range with eighteen range shifter plates inside the nozzle; in beam-delivery modeling it is further described as a compact, gantry-mounted superconducting synchrocyclotron proton therapy system with fixed-beam extraction at MeV, 18 Lexan plates inside the Range Modulator System (RMS), and 161 distinct energy layers (Cartechini et al., 22 Sep 2025, Cartechini et al., 26 Sep 2025). In the recent literature, the platform appears in several technically distinct roles: as a clinical intensity-modulated proton therapy (IMPT) delivery system for lung and esophageal treatments under free breathing, as a benchmark for detector development and proton dosimetry, and as the basis for machine-specific modeling of beam delivery time and temporal interplay. Comparative neutron work on the related Mevion S250 passive-scattering system is also used to contextualize the S250i within the broader Mevion product family (Ellis et al., 5 Aug 2025, Ellis et al., 21 Jan 2026, Tatari, 2022).
1. Beam generation, range modulation, and spot delivery
The system is characterized in the cited work as a PBS platform that delivers treatment spot-by-spot and layer-by-layer, with relatively large spot sizes compared to other PBS systems; this attribute is described as advantageous for mitigating the dose inhomogeneities due to respiratory motion interplay (Cartechini et al., 22 Sep 2025). Desired clinical energies, corresponding to different penetration depths, $0$–$32.2$ cm water, are achieved by inserting and retracting combinations of 18 Lexan plates inside the RMS, and each transition to a new energy layer introduces temporal delays (Cartechini et al., 26 Sep 2025). This combination of fixed high-energy extraction and in-nozzle range modulation is central to the delivery physics reported for the system.
A second defining hardware element is the Adaptive Aperture (AA), described as a nozzle-mounted, multi-leaf collimator-inspired dynamic field collimation system that trims the lateral penumbra of each spot, particularly important for the large spot sizes resulting from the in-nozzle range modulation (Cartechini et al., 26 Sep 2025). At every spot, the AA may need to move one or multiple leaves, which takes time and is often the limiting factor for rapid beam delivery. In the same literature, each spot is delivered as one or more pulses; each pulse can carry up to pC, and when the spot charge is higher it is split into multiple pulses. Plans exported from RayStation are reported to be reordered by a proprietary Spot Map Converter (SMC) algorithm to optimize delivery, affecting the delivery sequence, particularly in spot and energy order (Cartechini et al., 26 Sep 2025).
These reports collectively frame the S250i Hyperscan as a system whose operational behavior is not determined by scanning magnets alone. Range-modulator motion, AA motion, pulse splitting, and proprietary spot sequencing are all treated as first-order machine features. A plausible implication is that machine-specific temporal models are required for accurate 4D dose calculations, because generic PBS timing assumptions would not represent the reported interactions between spot scanning, energy switching, and frequent AA movements.
2. Clinical planning workflow and motion-robust delivery
In the reported thoracic workflow, the S250i was used to treat lung and esophageal cancer patients under free-breathing conditions, without additional motion mitigation techniques like rescanning (Cartechini et al., 22 Sep 2025). The planning pathway began with 4DCT acquisition yielding eight respiratory phases. The Internal Target Volume (ITV) was defined as the union of CTVs contoured in all phases, and an average-intensity 4DCT was used for robust plan optimization. Planning was performed in RayStation with 3D robust optimization considering range and $5$ mm setup uncertainties, with an additional ITV expansion of $1$–$3$ mm for delineation error. Treatments were delivered according to plan with in-room monitoring of respiratory signals using C-RAD/ANZAI, and machine log-files together with breathing traces were recorded per fraction.
The same study implemented two complementary interplay-evaluation pathways. The predictive pathway used 24 synthetic sinusoidal breathing traces with variable periods and phases, assigned each beam spot to a 4DCT phase on the basis of the synthetic trace and pre-treatment delivery logs from plan QA dry runs, recalculated dose per 4DCT phase, and mapped dose to a reference phase using deformable image registration (DIR). For each breathing scenario, a random worst-case combination of setup and range error was applied. The in vivo pathway used fraction-specific, patient-measured breathing traces and recorded delivery machine log-files to match each spot to the actual respiratory phase present during delivery, followed by the same per-phase dose recalculation, mapping, and fraction-wise accumulation. Both pathways were benchmarked against the FRED GPU-based Monte Carlo engine, and the full workflow was automated in RayStation via Python scripting with the RayStation API (Cartechini et al., 22 Sep 2025).
Agreement metrics reported for this framework were mean dose difference between RayStation and FRED and across DVH metrics between predictive and in vivo models. Cumulative dose distributions for the primary CTV converged after five fractions, and the paper states that preliminary results support clinical utility, especially for hypofractionation and targets with large motion $0$0 cm) (Cartechini et al., 22 Sep 2025). This directly qualifies a common assumption that thoracic PBS necessarily requires rescanning: in the reported institutional implementation, robust ITV-based planning combined with the S250i’s large spots yielded clinically acceptable convergence without rescanning, although the same paper limits this conclusion to preliminary results and notes that a larger cohort study is ongoing.
3. Dosimetry studies with $0$1-Ga$0$2O$0$3 detectors
Two recent studies use the S250i Hyperscan as an experimental platform for proton dosimetry with $0$4-Ga$0$5O$0$6-based metal-semiconductor-metal (MSM) detectors (Ellis et al., 5 Aug 2025, Ellis et al., 21 Jan 2026). In one configuration, the accelerator was operated in single-pulse mode with a fixed beam at $0$7 MeV and in raster-scan mode with a scanned $0$8 MeV proton beam across a $0$9 cm $32.2$0 $32.2$1 cm area covering 441 spots (Ellis et al., 5 Aug 2025). The detector was an epitaxially grown $32.2$2-Ga$32.2$3O$32.2$4 MSM device on sapphire with Ni/Au Schottky contacts in interdigitated geometry, using $32.2$5 width/spacing, 25 fingers, and $32.2$6 finger length, mounted on a PCB and placed on the carbon-fiber bed directly beneath the beam using the same geometry as patient QA. Readout was performed with a Keithley 2470 SourceMeter outside the treatment room, and a commercial PPC05 plane parallel ion chamber was positioned below for reference measurements (Ellis et al., 5 Aug 2025).
For that detector, the minimum detectable dose at $32.2$7 V bias was $32.2$8 MU, approximately $32.2$9 mGy; at lower bias the minimum detectable dose was 0 MU at 1 V and 2 MU at 3 V. The charge-to-dose relationship was reported as strictly linear over the full MEVION S250i dynamic range of 4 to 5 MU in both single-pulse and raster-scan modes. Representative fit equations included 6, ion chamber 7, and, for a raster-scan example at 8 V, 9 in nC versus MU. Reported sensitivities were 0 nC/Gy at 1 V, 2 nC/Gy at 3 V, 4 nC/Gy at 5 V, and 6 nC/Gy for the ion chamber. Variation in repeated deliveries was approximately 7 at 8 V and 9 V, comparable to the ion chamber, and $5$0 at $5$1 V, likely due to increased noise or leakage currents at higher bias. The study therefore recommends moderate bias, $5$2–$5$3 V, to balance sensitivity and stability (Ellis et al., 5 Aug 2025).
A subsequent study added a low-noise amplifier comprising a charge-sensitive preamplifier, pulse shaping, and additional amplification stages, with maximum gain $5$4 dB and cut-off frequency $5$5 kHz, and examined single-spot irradiation from $5$6 MU to $5$7 MU $5$8 Gy), raster-scan exposures from $5$9 MU to $1$0 MU per spot, and energies from $1$1 MeV down to $1$2 MeV (Ellis et al., 21 Jan 2026). The detector consisted of a $1$3 thick $1$4-Ga$1$5O$1$6 film with donor concentration approximately $1$7, fabricated on sapphire with the same 25-finger interdigitated Ni/Au contact geometry. Clear voltage pulses were observed down to the minimum repeatable system dose of $1$8 MU $1$9 mGy), integrated signal remained highly linear with delivered dose in both single-spot and raster-scan modes, and the signal-to-noise ratio followed $3$0. At a fixed dose of $3$1 MU, peak voltage and integrated output increased with increasing proton energy; both the detector and the ion chamber exhibited a non-linear decrease in response at lower energies, which the paper attributes not to detector inefficiency but to the Mevion machine’s use of energy-degrading spoiling blocks that cause beam broadening and lower local fluence. The detector response closely matched ion chamber measurements and simulated dose distributions from the RayStation TPS across the $3$2–$3$3 MeV range, while the dominant noise source was identified as flicker $3$4 noise, and long fall times on the order of hundreds of milliseconds were identified as a limitation (Ellis et al., 21 Jan 2026).
The key dosimetric relations reported in these studies are the collected charge,
$3$5
and a transient-current fit,
$3$6
Within the S250i literature, these experiments position the system as a testbed for QA devices intended for single-pulse, spot-scanning, multi-depth, and angle-resolved IMPT verification.
4. Machine-specific beam delivery time modeling
A separate line of work developed the first machine learning-based beam delivery time (BDT) model for the Mevion S250i Hyperscan system (Cartechini et al., 26 Sep 2025). The model was trained on institutional machine log files from 11 patients, comprising 1120 files and approximately $3$7 million pulses. The target variable was the inter-pulse time,
$3$8
and the core plan-available features included spot position shift,
$3$9
energy layer change, 0, Adaptive Aperture change,
1
spot charge per pulse, and Boolean indicators such as isFirstPulse and isTxPulse. Derived features included logarithmic transforms, interaction terms such as 2 and 3, categorical bins, and a composite movement metric,
4
The model used Random Forest regression in scikit-learn with hyperparameters optimized by randomized search and 5-fold cross-validation, and SHAP analysis was used to quantify feature contributions.
The reported mean absolute errors were 5 ms for short intervals 6 ms), 7 ms for 8–9 ms, 0 ms for 1–2 ms, and 3 ms for long delays 4 ms). SHAP analysis identified AA movements as the dominant global predictor for 5 ms, while spot positions and pulse charge influenced short intervals; energy changes had minor global impact but locally contributed to large 6 values, consistent with range modulator physics (Cartechini et al., 26 Sep 2025). The system-level interpretation offered by the paper is explicit: on this platform, beam delivery time is shaped primarily by the interaction of AA adjustments, spot movement, and range-modulator motion rather than by a single fixed spot-switching constant.
The model was evaluated in volumetric repainting and 4D dose recalculation for interplay assessment. Predicted cumulative delivery times deviated by only 7 from machine log data, and dosimetric metrics including D98, D95, and V95 remained within intrinsic delivery variability (Cartechini et al., 26 Sep 2025). This suggests that the S250i Hyperscan can be represented by a machine-specific temporal surrogate accurate enough for plan-specific interplay simulations even when treatment logs are not available before delivery.
5. Comparative neutron production and secondary radiation context
Published neutron analyses in the Mevion family have centered on the passive-scattering Mevion S250 rather than the S250i Hyperscan, but these studies are repeatedly used to contextualize the scanning system (Tatari, 2022). In the passive-scattering S250, Geant4 modeled the full scattering system including synchrocyclotron, nozzle components, and a water phantom, and benchmarked proton depth-dose, flux, and neutron spectra against measured data. Howell et al.’s extended-range Bonner sphere (ERBS) measurement at 50 cm transverse to the isocenter gave 8 n/cm9/Gy, while the simulation gave $0$00 n/cm$0$01/Gy, within $0$02, and neutron $0$03 agreed within $0$04. The ERBS dose-equivalent value was $0$05 mSv/Gy, and Geant4 gave approximately $0$06 mSv/Gy at isocenter with a water phantom. The same paper reports that measured or simulated $0$07 values span $0$08 to $0$09 mSv/Gy depending on position and proximity to the neutron source, and that internal neutrons within the phantom can contribute up to $0$10 of total neutron dose for low RMW step and high energy (Tatari, 2022).
The same comparative discussion notes that earlier SWENDI or WENDI-2 chamber studies reported substantially lower values, including $0$11 mSv/Gy, and that the summary table associates this lower value with the S250i Hyperscan scanning system (Tatari, 2022). The paper attributes discrepancies between Bonner-sphere and WENDI-type measurements to differences in beam, geometry, detector sensitivity, and cross-sectional uncertainties. Its stated interpretation is that lower neutron $0$12 values for scanning systems highlight an advantage of PBS over passive scattering, with implications for reduced risks of secondary malignancies and for shielding design.
For the S250i specifically, this is comparative rather than component-resolved evidence. A plausible implication is that the S250i Hyperscan’s relevance to secondary-radiation research lies partly in contrast to the passive-scattering S250: the latter supplies a validated neutron benchmark, while the former is invoked as the lower-neutron scanning counterpart within the same vendor lineage.
6. Reported research roles and technical significance
Across the cited literature, the Mevion S250i Hyperscan appears less as a single narrowly defined device than as a platform around which several methodological developments are organized. It is the clinical system used for free-breathing PBS delivery in thoracic and upper gastrointestinal motion studies, the proton source used for evaluating $0$13-Ga$0$14O$0$15 dosimeters against a PPC05 ion chamber, and the machine for which the first machine-specific BDT model was built (Cartechini et al., 22 Sep 2025, Ellis et al., 5 Aug 2025, Ellis et al., 21 Jan 2026, Cartechini et al., 26 Sep 2025). The unifying technical theme is that the system’s delivery architecture—fixed high-energy extraction, in-nozzle range modulation, large spot size, AA motion, and pulse-structured spot delivery—directly shapes the research questions asked of it.
| Research domain | Reported use of the S250i Hyperscan | Paper |
|---|---|---|
| Motion and interplay | Free-breathing lung and esophageal IMPT without rescanning; predictive and in vivo interplay evaluation with RayStation and FRED | (Cartechini et al., 22 Sep 2025) |
| Proton dosimetry and QA | $0$16-Ga$0$17O$0$18 MSM detector evaluation in single-pulse, single-spot, and raster-scan modes against a PPC05 ion chamber | (Ellis et al., 5 Aug 2025, Ellis et al., 21 Jan 2026) |
| Temporal delivery modeling | Random Forest prediction of inter-pulse time from machine log-derived features; tested in volumetric repainting and 4D dose recalculation | (Cartechini et al., 26 Sep 2025) |
Several recurring misconceptions are explicitly addressed by these studies. The first is that low-energy response degradation in the $0$19-Ga$0$20O$0$21 detector necessarily reflects detector inefficiency; the later dosimetry paper states that the non-linear decrease at lower energies is attributable not to detector inefficiency but to the Mevion machine’s use of energy-degrading spoiling blocks (Ellis et al., 21 Jan 2026). The second is that thoracic PBS on the S250i must rely on rescanning; the interplay paper instead reports robust delivery under free breathing without rescanning in the cases studied, with convergence after five fractions and close agreement between predictive and in vivo evaluations (Cartechini et al., 22 Sep 2025). The third is that a single neutron dose value can characterize all Mevion systems; the neutron comparison paper shows that values depend strongly on detector, geometry, and system configuration, with the passive-scattering S250 and the scanning S250i appearing in distinct dose-equivalent regimes (Tatari, 2022).
Taken together, the available literature presents the Mevion S250i Hyperscan System as a compact PBS/IMPT platform whose distinctive engineering features have made it a focal point for research on motion robustness, detector physics, and machine-specific temporal modeling. A plausible implication is that future technical work on the system will continue to couple delivery-physics detail with workflow automation, because the cited studies consistently treat the S250i not only as a treatment unit but also as an experimental object whose timing, beam shaping, and dosimetric behavior must be modeled explicitly.