Delivery-aware Monte Carlo second checks reveal unresolved verification gaps in particle therapy

A particle-therapy second-check platform modeled delivery and variable RBE, but its reported proton dose comparison achieved 84.9% gamma agreement and required review.

KEY POINTS

  • This technical study from Mayo Clinic Florida developed an independent second-check system for proton and carbon-ion pencil beam scanning (PBS), integrating treatment-plan data, predicted delivery sequences, Monte Carlo particle transport, and microdosimetric RBE-weighted dose calculation. The system was assessed using three de-identified proton plans involving cranial, head and neck, and breast/regional-nodal irradiation, while the carbon-ion branch was evaluated using a previously generated Monte Carlo dataset. The objective was to establish a common verification framework rather than demonstrate clinical equivalence with an established treatment planning system.
  • Unlike conventional independent dose calculations based solely on exported spot positions and weights, the proposed platform incorporated dose-driven continuous scanning (DDCS) delivery modeling. The algorithm converted discrete planned spots into a time-ordered irradiation sequence accounting for beam delivery during magnet motion, stationary irradiation, energy transitions, beam interruptions, and skip or break events. Proton source-generation and transport functionality were tested at 70.0, 170.2, and 229.1 MeV, although these checks did not quantitatively validate absolute output, proton range, or distal dose fall-off.
  • TOPAS/Geant4 Monte Carlo transport independently calculated physical dose, dose-averaged LET, and voxel-level microdosimetric quantities. Using the abridged microdosimetry distribution methodology (AMDM), each voxel retained 20 microdosimetric descriptors, comprising ten dose-mean lineal-energy values and ten corresponding fractional-dose contributions. These quantities were processed using the Mayo Clinic Florida microdosimetric kinetic model (MCF MKM) to generate variable-RBE-weighted dose. For protons, both fixed RBE 1.1 and variable-RBE distributions were reported separately; for carbon ions, only the model-derived variable-RBE approach was used.
  • The principal quantitative proton assessment did not satisfy all dose-comparison criteria. In a breast and regional-nodal plan with two treatment fields, the system converted 20,644 and 19,564 planned spot entries into 21,343 and 20,293 time-ordered source entries, respectively. Mean dose difference in the high-dose region was −1.67%, while the 3%/3 mm global gamma passing rate was 84.9%, using a 10% low-dose threshold. Although all assessed supraclavicular PTV dose-volume metrics met the configured criteria, the internal mammary and additional target structures triggered review, resulting in an overall REVIEW REQUIRED classification.
  • Computational performance was evaluated on a high-performance cluster using 256 CPU cores per treatment field. Across eight proton field calculations, runtime ranged from 10.3 to 22.0 minutes per field, with a mean of 15.8 minutes; because fields were processed in parallel, the effective Monte Carlo calculation time per plan was 12.8–22.0 minutes, excluding additional workflow overhead. Peak memory usage ranged from approximately 109 to 292 GB per field, illustrating the substantial computational resources required for this implementation.
  • Monte Carlo statistical uncertainty was quantified, but biological-model uncertainty was not. Across the three proton plans, mean relative statistical uncertainty ranged from 0.79% to 1.14% in regions receiving ≥90% of prescription and 1.00% to 1.62% in regions receiving 50–90%, with corresponding 95th-percentile values reaching 1.68% and 2.57%. These results represent sampling uncertainty only and exclude uncertainties in CT calibration, beam modeling, microdosimetric descriptors, and variable-RBE calculations. The quantitative accuracy of the proton MCF MKM biological-dose implementation was not independently evaluated.
  • Carbon-ion verification remained limited to component-level testing, using an archived dataset containing physical dose, microdosimetric descriptors, RBE factors, and RBE-weighted dose. No carbon-ion case completed the full delivery-conversion and independent Monte Carlo workflow in the present study. Furthermore, predicted DDCS delivery timing and stationary-versus-moving irradiation contributions were not compared with machine logs or physical measurements, and clinical PASS/REVIEW acceptance thresholds were not commissioned. Additional beam-model commissioning, end-to-end carbon-ion testing, quantitative variable-RBE validation, and broader patient-level benchmarking are required before clinical implementation.

CLINICAL TAKEAWAY

Integrating predicted delivery sequences with independent Monte Carlo transport and microdosimetric RBE calculations could strengthen patient-specific verification in proton and carbon-ion therapy by addressing uncertainties beyond conventional physical-dose recalculation. However, the reported proton comparison failed to satisfy all research acceptance criteria, carbon-ion testing was not performed end to end, and neither delivery timing nor variable-RBE accuracy was independently validated; the system therefore remains a technical feasibility platform rather than a clinically commissioned QA solution.

SOURCE

Frontiers in Oncology

Browse more research Suggest a correction