Author: Wesley S. Culberson, Albert Du, Ryan T. Flynn, Ryan Gardner, Alonso N. Gutierrez, Patrick M Hill, Daniel E. Hyer, Eric Jensen, Kaustubh A. Patwardhan, Blake R. Smith, Nhan Vu, Karsten K. Wake 👨🔬
Affiliation: University of Wisconsin, Department of Medical Physics, School of Medicine and Public Health, University of Wisconsin - Madison, Miami Cancer Institute, Baptist Health South Florida, Department of Human Oncology, University of Wisconsin School of Medicine and Public Health, Ion Beam Applications (IBA), University of Iowa, Iowa Health Care 🌍
Purpose: To develop an accurate predictive temporal model for the movement of Dynamic Collimation System (DCS) trimmers during collimated proton pencil beam scanning (PBS) deliveries using log data. The DCS is a collimation device featuring two orthogonally paired trimmers forming unique apertures for each collimated proton spot in a PBS plan. For DCS plans, a slew time is programed between successive spot irradiations to allow trimmer motion, which is currently based on a simple kinematic model dependent on the trimmer travel distance and the motor’s derivative of acceleration (jerk) value.
Methods: A set of 22 logs was obtained from the delivery of DCS-collimated PBS treatment fields, including 13 unique beams delivered at jerk values varying from 50k to 400k mm/s3 using a clinical IBA (Louvain-la-Neuve, Belgium) proton therapy system. The actual trimmer movement was fit with a sigmoid function featuring a rate constant, c1 (s-1), and a temporal midpoint, c2 (s), at which 50% of the trimmer transition has occurred. These fit parameters were calculated as a function of trimmer travel distance and jerk to obtain a model encompassing all possible trimmer transitions and jerk with real-world dynamics. The sigmoid function was used to more accurately estimate slew duration, allowing for more efficient delivery of collimated plans.
Results: The sigmoid model closely aligns with the actual log data in qualitative comparisons. Using this model to assign slew durations reduces the expected delivery time of a 1096-spot, 26-layer field from 243.6, 230.1, 219.2, 214.0, and 210.7 seconds to 155.4 (-36.2%), 138.6 (-39.8%), 116.8 (-46.7%), 105.8 (-50.6%), and 102.1 (-51.5%) seconds for plans delivered with jerk values of 50k, 100k, 200k, 300k, and 400k mm/s³, respectively.
Conclusion: The DCS trimmer dynamics were characterized with jerk-specific sigmoid fits to accurately predict motion during delivery, enabling more efficient DCS treatments.