Report on the ESTRO Physics Workshop
23-24 October 2025 I Toulouse, France
 

Chairs: Edmond Sterpin (Belgium), Thomas Bortfeld (USA)

At this year’s ESTRO Physics Workshop 2025 in Toulouse, one of the sessions explored how probabilities could reshape radiotherapy treatment planning — from prescription to optimisation and evaluation. The topic group “Towards the Use of Probabilities for Treatment Prescription, Optimisation, and Evaluation of Radiation Treatment Plans” gathered clinicians, physicists, and researchers who were eager to rethink how uncertainty is handled in daily practice.

From margins to probabilities

Invited speakers set the stage for lively discussions. Marcel van Herk introduced probabilistic planning treatment volumes (PTVs) that could replace fixed planning margins with uncertainty-driven optimisation, and warned us about the intricacies of correlations. Edmond Sterpin showed how sound probabilistic evaluation was necessary to judge the robustness of any probability-based method in planning. Jan Unkelbach presented the probabilistic clinical target volume (CTV) concept and demonstrated various ways to use probabilistic CTVs, such as binary presence models to reduce dosage to secondary nodal target volumes. Finally, Thomas R Mackie and Robert Jeraj presented a talk on how probabilistic planning and evaluation concepts will probably be distilled into the coming International Commission on Radiation Units and Measurements (ICRU) report. Together, these talks outlined a future in which dose and probability would be considered hand-in-hand — bridging physics-based uncertainty with clinical relevance.

Structured discussion and key questions

Ahead of the workshop, participants prepared through an online pre-meeting and defined five subgroups to address probabilistic planning questions for distinct treatment sites. The discussions then iterated through five questions on probabilistic planning with the following conclusions:

• Q1 - Clinical use cases for probabilistic PTVs: Finding the “killer application” is not easy, but overlapping structures (e.g. optic nerve, brainstem, prostate-rectum) and pancreas cases seem promising. Participants agreed that the use of probabilistic evaluation could support faster and more consistent clinical decision-making than is possible now.
• Q2 - Uncertainties to include in probabilistic PTV planning: Systematic, random, and contouring variations should all be captured. The groups recommended starting simple—reproducing what current margin recipes achieve—and gradually introducing probabilities in optimisation.

 

• Q3 - Minimal features required in treatment planning systems (TPSs) to perform probabilistic PTV planning: Systems need a statistically sound way to combine uncertainty sources such as set-up, range, and contouring. A clear pass/fail summary for each organ and metric ‑‑ obtained from probabilistic optimisation ‑‑ was proposed as a user-friendly feature.

 

• Q4 - Clinical use cases for probabilistic CTVs: Head-and-neck cancers, particularly those with lymph node involvement, were highlighted as strong clinical cases. In general, any infiltrating CTV has been found of interest, but requires sound infiltration modelling. The focus should be on the probability of tumour presence rather than tumour control probabilities, which remain model-dependent.

 

• Q5 - Uncertainties that should be included and TPS features for probabilistic CTVs: Scenario-based methods remain central for evaluation and optimisation, but clear visualisation of probabilistic results for clinicians is crucial — especially when defining “worst-case” scenarios. Probability maps might be useful for optimisation, but their use is intricate due to correlations. Comparison tools with classical plans are necessary.

Looking ahead

The group identified several next steps: contributing to the ICRU initiative on probabilistic planning, preparing a position paper, and developing clinical case studies and research tools to enable hands-on exploration of probabilistic concepts. A future clinical trial was discussed as a long-term goal.

While participants agreed that probabilistic planning would not replace the PTV overnight, there was clear consensus that it represents the next evolution in treatment planning — one that could make uncertainty quantification a natural part of clinical workflows.

“We might not abandon margins tomorrow, but the future of radiotherapy treatment planning will speak the language of probabilities.”

Niklas Wahl
DKFZ Heidelberg
Germany