Ponente
Descripción
When it comes to treating cancer in critical regions such as the brain, precision is vital, as reducing margins of error can significantly decrease negative side effects of treatment. Due to their finite range and deposition profile, protons are an ideal candidate for this sort of treatment. However, this finite range of protons is both their biggest advantage and their biggest challenge, as a minor miscalculation of penetration depth can have severe consequences for the patient. As such, continuous monitoring of the proton energy deposition throughout treatment is of the upmost importance. A class of continuous monitoring techniques that has shown great promise in recent years is prompt-gamma ray detection. Prompt-gamma ray detection functions on the principle that, as protons penetrate the patient, gamma rays are emitted along their path. Currently, this method depends largely on Monte Carlo simulations, which is restricting as these simulations are very computationally expensive.
In practice, a treatment of ~1000 spots typically lasts a few minutes, whereas the simulation of this treatment typically takes several hours. In cases where the continuous monitoring results during treatment do not match the simulation, it becomes completely impractical to re-simulate scenarios in search of the discrepancy between simulated and true treatment parameters. Therefore, if the computation time of a single spot could be reduced to approximately one-tenth of a second, real-time re-simulation of results would be possible. To achieve such simulation speeds, standard Monte Carlo simulations have been shown to be insufficient, and as such simulation with a high degree of parallelisation is the theoretical solution.
The goal of this project is to provide an open source solution to this problem. The code is currently being written in SYCL (SYstem-wide Compute Language), a hardware-agnostic language which creates parallelised software that can be ran on either GPU or CPU with minimal overhead. The present simulation model has shown great promise in terms of computation times and accuracy when tested on a water phantom. The code has shown an increase in computation speed by several orders of magnitude for energy deposition and depth of penetration calculations when compared with TOPAS, whilst maintaining a maximal deviation of 0.5 mm in 90% drop-off position (R90) and 3 MeV for total energy deposition across the full treatment range. These results are thought to be improvable with a more complex model and the appropriate optimisations. The future aim of this project is to gradually increase simulation complexity by incorporating more processes such as gamma-ray transport and detection in an external scintillation detector, whilst maintaining code clarity and structure for possible public contributions, and continuously simplifying user experience.
