GPU-Accelerated and Monte Carlo Based Robust Optimization for Spot Scanning Proton Therapy

Chris Beltran, Mayo Clinic

Usage Details

The main disadvantage of cancer therapy with traditional photon-based radiation is the difficulty of excluding healthy tissue from the treatment. Irradiation with protons, on the other hand, provides a much more targeted therapy tool. In a typical therapeutic plan, a 3D model of the cancerous region is constructed from CT scans to schedule a sequence of irradiations with different angles and doses to destroy all of the identified cancer cells. A computational model is necessary to calculate the trajectories of extremely large number of protons with sufficient confidence and stopping range within the tissue.

In a clinical setting, however, it is only computationally manageable to use analytical transport models, which have intrinsic simplifications in the material composition and scattering properties, and therefore carry large uncertainties. A more accurate approach involves performing a stochastic Monte Carlo (MC) simulation, where interaction between protons and tissue can be more accurately modeled, and a more realistic inclusion of heterogeneous tissue regions incorporated. Unfortunately, traditional Monte Carlo techniques are very time- and resource-consuming, and are impractical for a clinical setting.

We have developed a new efficient, accurate and very fast proton transport MC algorithm that includes the proper physics in realistic tissue. This MC code will be incorporated into a robust IMPT optimization system that we are developing.

The robust IMPT optimization system incorporates the location and radiation dose constraints of the tumor and normal tissues, and optimizes the number of protons for each of ~100,000 beam spots. The number of protons in each spot can range from 10⁶ to 10⁹, and the energy can range from 70 to 230 MeV in 1 MeV increments. To have sufficiently low statistical uncertainty, ~100,000 protons per beam spot must be simulated in the MC algorithm. To be clinically viable, the calculation and optimization must be done in under 15 minutes.

In this project on Blue Waters we aim to create a fast, massively parallel implementation of the robust IMPT optimization system running on a GPU architecture, to provide accurate, safe and effective calculations of proton irradiation dosage for treatment.