Author: Michael Joseph Dance, Shiva K. Das, John Dooley, David V. Fried, Spencer Lynch, Michael Repka, Shivani Sud, Neil Ari Wijetunga 👨🔬
Affiliation: University of North Carolina 🌍
Purpose: The growing complexity of radiation therapy treatment planning presents challenges in maintaining efficient clinical workflows while ensuring plan quality. This study evaluates the use of an adaptive objective function template (AOF-Template) for automated plan optimization (AutoOpt) via Ray Station’s Python scripting API for patients with prostate cancer.
Methods: The AOF-Template was comprised of a list of target and OAR objective functions. Target objective functions were adapted based on patient prescription while OAR objective functions were adapted based on SunNuclear’s FeasabilityDVH tool. Optimization was initiated via the AOF-Template and, when necessary, objectives were modified or added by a set algorithm to ensure target coverage and homogeneity. 11 patients, previously treated definitively to 70 Gy by manually optimized plans, were planned using AutoOpt and the results were compared to the clinically delivered plans.
Results: The algorithm successfully generated plans for 11 cases achieving coverage to all targets and comparable homogeneity, conformality, and OAR sparing to the manual plans. The AutoOpt plans took an average of 20 minutes per case after 5 minutes of user interaction. The AutoOpt plans had better homogeneity and conformality than the delivered plans. The AutoOpt plans delivered lower mean dose to the rectum (-1.0Gy) and a higher mean dose to the femoral heads (+2.7Gy) with bladder and small bowel having similar mean doses (<1Gy difference). AutoOpt had similar max doses to the bladder, femoral heads, rectum, and small bowel (<1Gy difference).
Conclusion: Plans generated by AutoOpt were of comparable quality to manually optimized plans and took a fraction of the time. AutoOpt could reduce planner workload, allowing more time to be dedicated to refining an AutoOpt plan rather than starting from scatch.