ESTRO 2025 - Abstract Book

S3516

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTRO 2025

Purpose/Objective: Carbon ion radiotherapy (CIRT) facilities face unique scheduling challenges because multiple treatment rooms share a single synchrotron. Traditional scheduling approaches often result in extended patient wait times and reduced facility efficiency. This study develops and validates a genetic algorithm-based optimization tool for patient scheduling in multi-gantry CIRT facilities, aiming to minimize treatment delays while maximizing the utilization of resources. Material/Methods: A simulation-optimization model was developed to emulate patient flow in a CIRT facility with one fixed-beam room and two to four rotating gantry rooms. This model incorporated deterministic factors (treatment times, setup durations) and stochastic elements (patient arrival patterns, room-specific requirements). The genetic algorithm employed a novel fitness function that balanced multiple competing objectives: minimizing wait times, maximizing room utilization, and maintaining treatment priority sequences. The algorithm was implemented to iteratively optimize patient sequencing and room assignments, with performance evaluated against both baseline scheduling and Bayesian optimization approaches. Success metrics included the average in-room wait time, measured as the duration between setup completion and treatment initiation, and the speed of algorithm execution. The model's robustness was evaluated across various operational scenarios, such as equipment downtime and emergency patient schedule adjustments.

Results: The genetic algorithm reduced average patient wait times from 45.2 (40.3 – 50.1) minutes in baseline scheduling to 21.7 (17.5 – 25.8) minutes, representing a 52% improvement. This performance matched or exceeded that of Bayesian optimization (24.3 minutes) while achieving significantly faster execution times (4.1 seconds vs 55.5 seconds, p < 0.001). The algorithm effectively handled high-stress scenarios, maintaining wait times below 30 minutes even during peak facility utilization periods. Sensitivity analysis revealed robust performance across varying facility configurations, with consistent wait time reductions of 45-55%, regardless of gantry room count. The optimization proved effective across various patient load scenarios, highlighting its scalability and adaptability.