How Proton Therapy Works and Why Monte Carlo Matters
Proton therapy delivers dose control that no photon beam can match. The Bragg peak — that sharp energy concentration at the end of the proton path — deposits maximum dose in the tumor with virtually nothing beyond it. But harnessing this potential requires dose simulations capable of handling the complex physics of nuclear interactions, multiple Coulomb scattering, and the critical range dependence on tissue composition.
Series overview: for the full roadmap and related articles, return to the complete guide on Monte Carlo in radiotherapy.
Analytical pencil-beam algorithms have dominated clinical planning for years. They perform reasonably in homogeneous geometries like the liver but show real limitations in regions with abrupt density interfaces: head and neck, lung, breast. Monte Carlo resolves these situations by tracking each particle individually, computing scattering and nuclear interactions voxel by voxel. For a comprehensive overview, check our complete guide on Monte Carlo Techniques in Radiation Therapy.
Proton Beam Delivery: Passive Scattering vs. Scanning
Two main methods deliver dose to the patient. Passive scattering uses double scatterers, rotary modulators, and shaped compensators to conform the beam to the target. Magnetic beam scanning steers individual pencil beams in x and y, sweeping energy by energy — most newer facilities use this approach.
In practice, beam scanning eliminates field-specific hardware (aperture and compensator), enables IMPT (Intensity-Modulated Proton Therapy), and offers far superior Monte Carlo efficiency. Since beam parameter variations are smaller (energy and magnet positions), phase spaces and beam models can be reused across fields. Passive scattering, on the other hand, requires a unique configuration for each field — energy, scatterers, modulator, aperture, and compensator all vary — consuming massive computation time.
Proton Physics and Monte Carlo Modelling
Dose deposition in proton therapy involves ionizations, excitations, multiple Coulomb scattering, and nuclear interactions. Energy loss follows the Bethe-Bloch equation, and multiple scattering typically uses Molière theory in condensed history class II methods.
Nuclear interactions deserve particular attention. While they do not significantly change the Bragg peak shape, they cause two crucial effects: primary proton fluence reduction with depth (rule of thumb: about 1% per cm of range) and secondary proton production contributing to dose in the entrance region. Primary and secondary protons account for roughly 98% of the deposited dose. The maximum range of delta electrons in water is approximately 2.5 mm for a 250 MeV proton.
When summing multiple pencil beams in scanning, the “nuclear halo” around each beamlet can be significant and must be modelled correctly. Uncertainties in mean excitation energy values for materials (5–15%) can lead to several-millimetre uncertainties in predicted range. Angle-integrated emission spectra for proton-nucleus interactions are known only within 20–30% according to the ICRU. Yet dose calculation still achieves typical accuracy of ~1–2%.
Monte Carlo Codes for Proton Therapy
Several codes are available: FLUKA, Geant4, MCNPX, VMCpro, and Shield-Hit, among others. To make them accessible to non-experts, frameworks like GAMOS, GATE, PTsim (all Geant4-based), FICTION (FLUKA-based), and TOPAS were developed.
| Code/Framework | Base | Characteristics |
|---|---|---|
| FLUKA | Standalone | Multi-purpose code with detailed nuclear models |
| Geant4 | C++ Toolkit | Object-oriented toolkit; requires programming |
| MCNPX | Standalone | Multi-purpose code with coupled particle transport |
| VMCpro | Standalone | Speed-optimized for proton therapy |
| TOPAS | Geant4 | User-friendly; no programming required; NCI-supported |
| GATE | Geant4 | Medical simulation framework with scripting interface |
| GAMOS | Geant4 | User-friendly framework for radiology |
| FICTION | FLUKA | Radiation therapy-specific wrapper |
Source: Monte Carlo Techniques in Radiation Therapy (2nd ed., CRC Press, 2022)
TOPAS deserves special mention. Developed with NCI support under the “Informatics Technologies for Cancer Research” initiative, it lets non-physicists run complex simulations without programming. See our article on Monte Carlo fundamentals in radiation therapy for the foundational concepts.
Beam Parameterization and Treatment Head Modelling
An MC simulation typically starts at the treatment head entrance or exit. At the entrance, relevant parameters include beam energy, energy spread, spot size (typically 2–8 mm in sigma), and angular distribution (in the order of 2–5 mm·mrad for cyclotrons). Cyclotron energy spread is typically below 1%, while synchrotrons can be two orders of magnitude lower.
These parameters are correlated — for instance, position within the spot and angular momentum. For beam scanning, this correlation must be modelled. In passive scattering, the scattering material in the treatment head tends to blur these correlations. A well-calibrated MC code should reproduce measured SOBP dose distributions in water with ~1 mm accuracy in range and ~3 mm in modulation width.
CT Conversion and Patient Dose Calculation
One of the most critical aspects is converting CT numbers to material composition, mass density, and ionization potential for each tissue. Unlike photons (where electron density suffices), protons interact through ionizations, multiple Coulomb scattering, and nuclear reactions — each type has a different relationship with CT-derived properties.
Soft tissues with CT numbers between 0 and 100 are difficult to distinguish because different elemental compositions can produce the same CT number. Mean excitation energy uncertainties (5–15%) translate directly into beam range uncertainties. For head and neck treatments, CT conversion schemes can influence range by 1–2 mm. Dual-energy CT significantly improves conversion by providing relative electron density and effective atomic number maps.
Monte Carlo vs. Analytical Algorithms: Clinical Impact
The fundamental difference lies in how multiple Coulomb scattering is handled. Analytical algorithms are less sensitive to density interfaces and in-beam variations, especially at bone-soft tissue, bone-air, or air-soft tissue interfaces.
Range Prediction Differences
Clinical range margins of 3.5% + 1 mm are typically applied to cover analytical algorithm errors. With routine Monte Carlo, these margins could be uniformly reduced to 2.4% + 1.2 mm regardless of geometry. For geometries with lateral heterogeneities (head and neck, lung, breast), analytical margins may need to reach 6.3% + 1.2 mm.
Dose Prediction Differences
Monte Carlo predicts larger scattering components, resulting in target dose generally lower than analytical predictions. Mean target dose differences can reach 4% in head and neck and lung patients, or about 2% in breast and liver. Small fields are more sensitive. For high-Z metallic implants (dental implants, tantalum markers), dose perturbations are not accurately predicted by analytical algorithms. Learn more about Monte Carlo patient dose calculation.
Monte Carlo as a QA Tool
Monte Carlo goes beyond dose calculation, serving as a robust quality assurance tool in proton therapy.
A direct application is treatment plan recalculation using delivery system log files, independently verifying delivered dose. Monte Carlo also aids TPS commissioning, validating analytical algorithms in hard-to-measure scenarios. By simulating dose distributions while varying beam parameters, tolerances can be defined that reduce the need for extensive experimental studies.
MC simulations also enable prompt gamma detector design for in vivo range verification and image reconstruction optimization for proton computed tomography. For LET distributions and radiobiological effects, codes like TOPAS-nBio extend Geant4 physics to the nanometre scale, simulating DNA damage and energy deposition event clustering. See also how dynamic beam delivery fits into this context.
Dose-to-Water vs. Dose-to-Tissue
Analytical algorithms compute dose-to-water because all clinical experience and dosimetry are water-based. Monte Carlo naturally computes dose-to-tissue. The difference is clinically insignificant for soft tissues (~2%), but in bony anatomy dose-to-water can be 10–15% higher. In most cases, retroactive conversion using energy-independent relative stopping powers is sufficiently accurate (~1%).
Clinical Monte Carlo Outlook
Major TPS vendors are already working on Monte Carlo-based treatment planning. Direct integration into the IMPT optimization loop is the next step, requiring strategies such as voxelized geometry re-segmentation and dose distribution smoothing. Leading proton centres have already implemented in-house MC systems via DICOM-RTion interfaces. For applications in ion beams and brachytherapy source modelling, see our dedicated articles.
