Simultaneous Monte Carlo dose computation and optimization for proton therapy: the beamlet-free approach

Details

Supervisors

Lee, John Aldo

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil biomédical, à finalité spécialisée
Master [120] : ingénieur civil biomédical, à finalité spécialisée

Abstract

enRadiotherapy is one of the most common treatments for cancer. Proton therapy, a more advanced technique that uses proton beams instead of X-rays, is growing rapidly due to its ability to target tumors more precisely. Today, proton therapy treatment planning is typically based on so-called "beamlet-based" methods, which involve dividing each beam into smaller units called "beamlets", whose intensities are individually optimized. These techniques are effective but require storing all the beamlets and involve high computational demands, which can be a limitation in situations where speed is essential, such as in adaptive proton therapy. This thesis introduces a new approach called beamlet-free, in which the optimization is performed without storing all individual beamlets. Instead, the method simulate randomly beamlet pairs to gradually estimate their optimal intensities. We present a proof of concept of this algorithm and analyze its convergence on both a simplified case and a realistic patient case, comparing it to the traditional beamlet-based method. Although our algorithm consistently converges toward a solution, our experiments on a simple case with few beamlets reveal residual noise stemming from its stochastic nature and the beamlet approximations. This noise slows convergence and prevents the algorithm from achieving a clinically acceptable treatment plan within a practical number of iterations. In contrast, for a more complex case with many beamlets, the beamlet-free approach appears to approximate the performance of the beamlet-based method more closely. Further work is needed to assess whether these limitations can be overcome, enabling the full theoretical advantages of this approach to be realized. It is also important to investigate whether, in practice, the beamlet-free algorithm can outperform the beamlet-based method in terms of speed and computational cost.