[Paper Review] Optimization of patient-specific range modulators for conformal FLASH proton therapy

Sylvain Deffet, Kevin Souris+1 more|arXiv (Cornell University)|Mar 15, 2023

Radiation Therapy and DosimetryMedicine3 citations

TL;DR

This paper proposes a novel inverse planning method that jointly optimizes patient-specific voxelized range modulators and pencil beam scanning spot weights for conformal FLASH proton therapy. By directly optimizing modulator geometry and spot weights under dose constraints using a fast Monte Carlo gradient estimator, the method achieves dose distributions close to IMPT while enabling 3D-printable, CT-compatible designs with simplified simulation workflows.

ABSTRACT

Purpose: A promising approach to enable FLASH conformal proton therapy is to passively degrade a single energy layer using a patient-specific range modulator. We propose an innovative method to directly optimize the geometrical characteristics of the range modulator and the treatment plan with respect to user defined constraints, similarly to state-of-the-art IMPT inverse planning. Methods: The kind of range modulators proposed in this study is a voxelized object which can be placed in the CT for dose computation, which simplifies the simulation pipeline. Both the geometrical characteristics of the range modulator and the weights of the PBS spots were directly optimized with respect to constraints on the dose using a first-order method. A modified Monte Carlo dose engine was used to provide an estimate of the gradient of the relaxed constraints with respect to the elevation values of the range modulator. Results: Assessed on a head and neck case, dose conformity logically appeared to be significantly degraded compared to IMPT. We then demonstrated that this degradation came mainly from the use of a large range shifter and therefore from physical limitations inherent in the passive degradation of beam energy. The geometry of the range modulator, on the other hand, was shown to be very close to being optimal. PBS dose rates were computed and discussed with respect to FLASH objectives. Conclusions: The voxelized range modulators optimized with the proposed method were proven to be optimal on a head and neck case characterized by two rather large volumes, with irregular contours and variable depths. The optimized geometry differed from conventional ridge filters as it was arbitrarily set by the optimizer. This kind of range modulators can be directly added in the CT for dose computation and is well suited for 3D printing.

Motivation & Objective

• Address the lack of direct optimization of range modulators in existing FLASH proton therapy approaches.
• Overcome the limitations of dose-mimicking optimization that relies on precomputed IMPT plans and external Monte Carlo engines.
• Develop a unified optimization framework for both range modulator geometry and PBS spot weights, similar to IMPT inverse planning.
• Enable practical implementation via 3D-printable, voxelized range modulators that can be directly embedded in CT for dose calculation.
• Assess the feasibility and dose conformity of the method on a complex head and neck cancer case under FLASH dose rate constraints.

Proposed method

• Voxelized range modulators are defined as 3D elevation maps (1×1×1 mm³ resolution) placed directly in the CT for dose computation.
• A first-order optimization method simultaneously updates the elevation values of the modulator and the PBS spot weights.
• A modified Monte Carlo dose engine computes the gradient of relaxed dose constraints with respect to each voxel’s height.
• The optimization is performed under user-defined dose constraints for target and organs at risk, similar to IMPT inverse planning.
• The method avoids external parameterized geometry engines by using the CT-based modulator, simplifying the simulation pipeline.
• The approach enables direct, joint optimization of modulator shape and spot weights, avoiding sequential or decoupled procedures.

Figure 1: A 3D range modulator can be represented as a 2D elevation map. It is split into ’towers’ and a range shifter. A minimum thickness of plain material is left attached to the ’towers’ accordingly to 3D printing specification.

Experimental results

Research questions

• RQ1 Can joint optimization of range modulator geometry and PBS spot weights improve dose conformity in FLASH proton therapy compared to sequential or dose-mimicking methods?
• RQ2 To what extent does the physical limitation of passive beam energy degradation affect FLASH dose conformity when using a single energy layer?
• RQ3 How does the voxelized modulator design compare to conventional ridge filters in terms of geometric flexibility and 3D printability?
• RQ4 Can a simplified simulation pipeline using CT-based modulators replace complex external Monte Carlo engines with parameterized geometries?
• RQ5 What is the achievable dose rate in the optimized FLASH plan, and does it meet the 40 Gy/s threshold associated with the FLASH effect?

Key findings

• The optimized range modulator geometry was found to be very close to optimal, with asymmetric pyramidal structures emerging naturally from the optimization process.
• Dose conformity in the FLASH plan was significantly degraded compared to full IMPT, primarily due to the physical limitations of passive beam energy degradation.
• The PBS dose rate did not reach the 40 Gy/s threshold in the evaluated head and neck case, with scanning length being a key limiting factor.
• The voxelized modulator design enabled direct integration into the CT for dose calculation, eliminating the need for complex external Monte Carlo simulations with parameterized geometries.
• The method achieved a dose distribution very close to that of a single-field IMPT plan using the same range shifter, indicating high fidelity in dose delivery.
• The approach is well-suited for 3D printing due to its coarse 1×1×1 mm³ resolution, which enhances mechanical stability and manufacturability.

Figure 3: Experimental setup for the right part of the tumor. From nozzle to patient: range modulator (PMMA), range shifter (aluminium), and aperture (tungsten).