Probabilistic optimization and planning

Principal Investigators: dr. Zoltan Perko (TU Delft), prof. dr.  Mischa Hoogeman (Erasmus MC, HollandPTC)

PhD students: Jelte Rinus de Jong (TU Delft) / Jenneke de Jong (Erasmus MC)

Funding: Raysearch Laboratories


The purpose of the cooperation is to jointly facilitate the scientific development of methods for probabilistic optimization. In WP1 TU Delft will develop techniques for computationally feasible probabilistic optimization. For example, this can be done using Polynomial Chaos Expansion (PCE) methods, which once trained allow extremely fast evaluation of a polynomial meta-model accurately approximating more complex models. This technique has already been validated for the comprehensive robustness evaluation of proton therapy plans, and holds the key to allowing probabilistic plan optimization by coupling it to the plan optimization to evaluate probabilistic objective functions and constraints.

In WP2, ErasmusMC will test and evaluate these methods in a clinical setting. In particular, we will investigate, with engaged end-user, how best to present and evaluate probabilistically optimized treatment plans to the radiation oncologist. The goal is to establish guidelines and codes of practice to use probabilistic treatment planning. We will also systematically research the potential benefit of probabilistic optimization by benchmarking it against current practice, i.e., scenario-based robustly optimized treatment plans. Analysis of robustness, sensitivity to uncertainties and computational efficiency will be used for the comparisons. It will also be researched how the probabilistic planning method has to differ to be applicable to different treatment sites where the effects of interplay and random errors may differ. And finally the comparative performance of the new probabilistic planning method for photon and proton therapy will be evaluated in this work package