Room: Exhibit Hall | Forum 7
Purpose: To explore considerations and requirements for consistent, scalable, site-non-specific automated field-in-field planning aimed at exceeding the capabilities of manual planning.
Methods: A threshold-driven beamlet-based auto-planning algorithm determines fluence maps to be used as the â€œidealâ€? case for a field-in-field plan. The fluence patterns are stratified and then iteratively sequenced using a combination of graph-based optimization algorithms and heuristics to maintain deliverability and form clinically â€œreasonableâ€? apertures. Intensities are optimized to match the original fluenceâ€™s dose distribution and adhere to minimum MU limits. Leaf closing positions and jaw shifts are optimized for delivery speed and minimal leakage.
Results: Significant gains were made in both planning time and plan quality. Automatically selecting the â€œbestâ€? of multiple sequencing runs yielded field-in-field plans dosimetrically similar to the input fluence-based plans. Plans can be generated with our methodology in minutes as implemented in EZFluence, an Eclipse plugin application.
Conclusion: This work highlights the benefits of modeling the field-in-field problem to be scalable beyond what is currently being done clinically. Extending this framework, IMRT plans can be automatically converted to field-in-field, boost volumes can be integrated, and non-isocentric field-in-field planning can be easily achieved.
Funding Support, Disclosures, and Conflict of Interest: Work supported by and done at Radformation by its employees for commercial product, EZFluence.