Bayesian Optimization of a Novel Intensity Modulated X-Ray Source
B Whelan1*, P Keall1, J Perl2, J Wang4, S Trovati6, S Tantawi2, R Fahrig4, P Maxim5, M Shumail2, B Loo6, (1) University of Sydney, Sydney, NSW, (2) Stanford Linear Accelerator Center, Menlo Park, CA, (3) Varian Medical Systems, Sunnyvale, CA, (4) Innovation, Advanced Therapies, Siemens Healthineers, Forchheim, Germany(5) University of California, Irvine, Orange, CA, (9) Slac, Menlo Park, CA, (6) Department of Radiation Oncology, Stanford University School of Medicine. Stanford, CA 94305 USA
Presentations
MO-G-202-2 (Monday, 7/11/2022) 2:45 PM - 3:45 PM [Eastern Time (GMT-4)]
Room 202
Purpose: Conventional intensity modulation is performed using an MLC, a complex mechatronic device comprising over 100 moving parts. This results in fundamental limitations in delivery speed and mechanical robustness. We propose an alternative intensity modulation concept with no moving parts, and demonstrate the use of Bayesian methods to optimize the design. This device is termed SPHINX: Scanning Pencil-beam High-speed Intensity-modulated X-ray source.
Methods: A parametric simulation environment was developed in Topas enabling simulation of a SPHINX structure from linac phase space to dose in water tank including transport through scanning magnets. The SPHINX geometry was parameterized with 8 variables, which were optimized using Bayesian optimization. The basic goal was to maximize dose rate for a user input beamlet width. Additionally, penalties on minimum wall thickness, inter-channel cross talk, and surface electron contamination were included. Designs for beamlet widths of 5, 7, and 10 mm were generated. For the seven-mm design, a simulation of dose in water for a 50x50 mm square field was carried out, incorporating transport through custom scanning magnets.
Results: The Dose-per-charge was 3574, 6351, and 10015 Gy/C and the beamlet width 5.1 mm, 7.2 mm and 10.1 mm for the five, seven, and ten mm models respectively. All designs were within defined bounds for minimum wall thickness, inter-channel cross talk, and surface electron contamination. For the simulation of all beamlets in water, the alignment of the beamlets and channels was .01 ± .03 m, demonstrating the accuracy of the scanning magnets. A slight negative trend in dose per charge was observed for the outermost beamlets, with a decrease of approximately -2% per degree of scanning angle.
Conclusion: SPHINX designs for user-input beamlet widths were generated using Bayesian Optimization of topas monte carlo simulations. SPHINX is a promising candidate for delivery of IMRT with no moving parts.
Funding Support, Disclosures, and Conflict of Interest: BWL acknowledges funding from NCI grant 2R44CA217607. BWL has received research support from Varian Medical Systems. BWL, RF, ST, and PM are co-founders and BWL and ST are board members of TibaRay, which has licensed intellectual property relating to the SPHINX concept. BW acknowledges funding from NHMRC grant 5284296.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email