ESTRO 2025 - Abstract Book

S3444

Physics - Machine learning models and clinical applications

ESTRO 2025

4007

Digital Poster A constraint programming approach for the radiotherapy scheduling problem

Hugues RAUWEL 1 , Guillaume POVEDA 2 , Christophe LOUAT 3 , Florent TEICHTEL KOENIGSBUCH 2 , Laure VIEILLEVIGNE 1,4 1 Medical Physics Department, Oncopole Claudius Regaud, Institut Universitaire du Cancer de Toulouse (IUCT), Toulouse, France. 2 Central Research&Technology – Artificial Intelligence Research, Airbus, Toulouse, France. 3 Advanced Analytics Team and Artificial Intelligence, Airbus, Toulouse, France. 4 RADOPT, Centre de Recherches en Cancérologie de Toulouse, Toulouse, France Purpose/Objective: Radiotherapy scheduling is a complex task that requires careful consideration of medical and technical constraints, as well as resource availability. Human-made schedules often lead to suboptimal solutions, resulting in unnecessary patient waiting times. Adressing this issue could lead to significant improvement regarding both the treatment quality and the service performance. Material/Methods: A novel approach was developed based solely on a well-known operations research technique called constraint programming. The objective was to minimize the treatment start dates for all patients while enforcing the set of medical and technical constraints. Real-world data was provided by the IUCT-Oncopole comprehensive cancer center (Toulouse, France) with a given four-week existing schedule and a 65 patients flow to be scheduled over a week. The Oncopole has 7 linacs (2 Halcyon (Varian), 2 True Beam STx (Varian), 2 Tomo HD (Accuray) and 1 Radixact (Accuray) machines. Treating approximately 3000 patients per year, the center offers a range of advanced techniques : SRS/SRT, SBRT, SGRT, VMAT, IMRT, DIBH and Gating. Additionally, some patients require treatment on specialized linac. The scheduling problem was formulated as a multi-objective optimization problem, incorporating various critical variables. Different scheduling frequencies, were evaluated where a new schedule updated was computed with a batch of patients and the size of the batch depends on the frequency. The analysis was based on a frequency ranging from twice a day to once a week. An application was implemented to integrate this process in the IUCT-Oncopole's tools. Results: The results demonstrate that the constraint programming approach effectively solves the radiotherapy scheduling problem within a reasonable computation time. Preliminary results indicate that batch scheduling is the most promising strategy. Specifically, the twice-daily scheduling frequency emerged as the most efficient, reducing treatment start dates while maximizing other quality indicators such as treatment dates stability or machine preferences. For the 65-patient cohort, this approach reduced unnecessary waiting days up to 60 days compared to manual scheduling. Conclusion: Constraint programming is a well-adapted technique to address radiotherapy scheduling problem with reasonable scaling potential.

Keywords: constraint programming, radiotherapy scheduling