ESTRO 2025 - Abstract Book

S2734

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2025

878

Proffered Paper Advancing Photon Radiotherapy Plan Robustness Evaluation with Probabilistic Surrogate Models Tong Zhang 1,2 , Jiacheng Liu 1 , Ruoxi Wang 1 , Hao Wu 1,2 1 Radiation Oncology, Peking University Cancer Hospital/Institute, Beijing, China. 2 Institue of Medical Technology, Peking University Health Science Center, Beijing, China Purpose/Objective: Robustness evaluation is an indispensable process in ion therapy. With the hypo-fractionated treatment trends, robustness evaluation also began to reveal its potential for photon radiotherapy. This study aimed to develop a robustness assessment tool for photon plans with variable uncertainty input dimensions.

Material/Methods:

Unlike the worst-case-evaluation approach [1], we adopted the probabilistic approach with surrogate models, i.e., Polynomial Chaos Expansion (PCE) [2] and Gaussian Process (GP) [3]. Additionally, a combination of the two, named Polynomial Chaos Kriging (PCK) was implemented [4]. As illustrated in Figure 1, the workflow of the robustness evaluation follows three steps: 1. Generate error scenarios given the uncertainty distribution, calculate doses, and collect DVH metrics of interest; 2. Train the surrogate models with collected datasets; 3. Based on the trained surrogate models, assess DVH metrics and corresponding probabilities in unknown scenarios. The evaluation tool was implemented through interaction with the Eclipse TPS via ESAPI. Three HyperArc SRS (HA-SRS) plans with varying PTV margins (0, 1, 2 mm) were analyzed utilizing the three surrogate models, according to an uncertainty input space (including translation and rotation set-up errors) characterized by a six-dimensional normal distribution. The predictive accuracies of the three surrogate models were assessed against the ground truths calculated by the TPS across 100 out-of-sample error scenarios, using R 2 as the statistical criterion.