ESTRO 2024 - Abstract Book

S2888

Interdiscplinary - Other

ESTRO 2024

654

Poster Discussion

Is full automation of radiotherapy treatment planning from scan to plan possible?

Mark J Gooding 1 , Shafak Al-Uwini 2 , Teresa Guerrero-Urbano 3 , Yasmin McQuinlan 4 , Deborah Om 5 , Floor H E Staal 2 , Stine Korreman 6 , Jean-Eammanuel Bibault 7 1 Inpictura Ltd, -, Oxford, United Kingdom. 2 University Medical Center Groningen, Dept. of Radiation Oncology, Groningen, Netherlands. 3 Guys and St Thomas’s Hospital, Dept. of Radiation Oncology, London, United Kingdom. 4 Mirada Medical Ltd, Product, Oxford, United Kingdom. 5 Hôpital Européen Georges Pompidou, Dept. Medical Physics, Paris, France. 6 Aarhus University, Dept. of Clinical Medicine, Aarhus, Denmark. 7 Université de Paris, Dept. of Radiation Oncology, paris, France

Purpose/Objective:

The ESTRO physics workshop 2017 (Glasgow) “Automate or perish” focused on workflow automation in radiotherapy, specifically focusing on contouring and auto-planning. Since then, advances in technology have meant that many individual tasks in the radiotherapy treatment chain can be automated to a large extent. In this study, in part conducted at the ESTRO Physics Workshop 2023, a challenge for full automation from simulation CT to treatment plan was conducted, with the objective to assess how plausible fully automated treatment planning is at this time.

Material/Methods:

16 CT cases of “routine” prostate patients were retrospectively collected and anonymised. These consisted of 6 prostate only, 6 prostate & nodes, and 4 prostate bed & nodes. The challenge was set up using the grand challenge.org framework, where participants were invited to create treatment plans that met contouring guidelines and planning objectives, in a fully automated way. Manual editing of contours or plans was prohibited. The treatment plans were evaluated against the planning objectives using a published scoring method, weighting target coverage and OAR constraints following a similar approach to Nelms et al. [1] to produce a score out of 100. To calculate the DVH parameters, consensus contours for the cases were created from contours provided by three clinical teams, provided with the same contouring guidelines as the challenge participants. Challenge participants were blinded to the contours by which their plans were to be assessed, meaning participants would have to rely on the accuracy of their contouring. Scores were also assessed against the participant’s own contours, to understand the impact of the contours used for assessment. The challenge was split into three phases; Three cases (one of each type) were made available to allow participants to prepare their workflows and ensure participants could successfully submit to the challenge website for scoring. Ten cases (4 prostate, 4 prostate & nodes, 2 prostate bed & nodes) were then used for the main challenge. Participants were limited to a maximum of three submissions to prevent optimisation against the test set. The final three cases (one of each type) were used to allow the participants to demonstrate their approaches in-person at the ESTRO Physics Workshop 2023. A questionnaire regarding the methods used and the level of automation was completed by participants submitting to the main challenge.

Results: