Room: Davidson Ballroom A
Purpose: Automated multi-objective treatment planning (AMTP) algorithms can consistently generate high quality treatment plans in clinical practice, but their configuration is an interactive and time-consuming process. Therefore, we propose an automated strategy for configuring our recently introduced fast AMTP algorithm, based on the Reference Point Method (RPM), using a set of training plans.
Methods: The starting point is a database of 287 prostate cancer input plans that were generated using our clinically applied (but relatively slow) AMTP algorithm. The automated strategy first determines an initial RPM configuration that aims for plan objective trade-offs similar to those in a set of training plans. Using this configuration, the output plans are generated for the training patients, and compared to the input plans. The differences in objective values are then used to iteratively improve the RPM configuration by guiding these differences towards ranges predefined by the user, e.g., emphasis can be put on improving the important objectives such as high-dose rectum sparing.
Results: We used 96 randomly selected training patients for automatically configuring the fast AMTP algorithm. In the configuration strategy, the aim was to improve the high-dose rectum sparing while maintaining similar high-dose bladder sparing and target coverage. After generating all 287 output plans, we observed that these reduced the average rectum Vâ‚‡â‚…(G)(y) by 1.12%-point (range 0 â€“ 2.78), and the average bladder Vâ‚†â‚…(G)(y) by 0.03%-point (range -3.35 â€“ 3.60). Clinically acceptable target coverage, Vâ‚‰â‚…(%) â‰¥ 99%, was observed for 286/287 (minimum Vâ‚‰â‚…(%): 98.97%) input plans and 284/287 output plans (minimum Vâ‚‰â‚…(%): 98.72%). The computation time for finding the final RPM configuration was 1.5 days.
Conclusion: The automated strategy successfully configures our fast AMTP algorithm for prostate cancer patients. This replaces the time-consuming manual configuration process, and therefore greatly improves the efficiency and effectiveness of an automated treatment planning workflow.
Funding Support, Disclosures, and Conflict of Interest: Erasmus MC Cancer Institute has research collaborations with Elekta AB, Stockholm, Sweden and Accuray Inc, Sunnyvale, USA.
Treatment Planning, Inverse Planning, Optimization
TH- External beam- photons: treatment planning/virtual clinical studies