Integration of an Open-Source Voxel Sampling and Optimization Pipeline into Matrad for Research Accessibility
Abstract
Purpose
To overcome barriers in Intensity-Modulated Radiation Therapy (IMRT) research caused by reliance on proprietary optimization solvers (e.g., CPLEX) and fragmented software environments. We present the integration of a voxel down-sampling and fluence map optimization (FMO) pipeline directly into the open-source treatment planning system, matRad. This work aims to democratize access to advanced sampling algorithms and facilitate multi-institutional collaboration by establishing a fully open-source workflow.
Methods
An existing sampling pipeline (RT_Sampling_Pipeline), originally developed as a standalone MATLAB application relying on a C++ MEX interface and the commercial CPLEX solver, was refactored for native integration with matRad. The proprietary CPLEX dependency was replaced by generalizing the optimization backend to support open-source solvers already utilized by matRad. The pipeline’s input structures were harmonized with matRad’s dij and cst data classes, allowing users to seamlessly apply various down-sampling techniques (e.g., random, grid-based, or importance sampling) directly within the matRad GUI or scripting environment. The integrated workflow was validated using both clinical phantoms and the CORT dataset to enable diverse planning techniques on an exportable set of data structures.
Results
The integrated pipeline eliminates the requirement for external commercial licenses within the matRad architecture. Preliminary validation on phantoms, as well as prostate and head-and-neck cases, demonstrates that custom plans can be generated in a manner that is compatible with clinical equipment. Furthermore, the integration allows researchers to leverage matRad’s existing multi-modality evaluation tools immediately after optimization, streamlining the iterative research process.
Conclusion
By embedding the sampling and optimization pipeline into matRad and removing proprietary dependencies, we provide a robust, accessible tool for the medical physics community. This integration lowers the entry barrier for IMRT optimization research and fosters reproducibility by utilizing a standardized, open-source environment.