Author: Katja M. Langen, William Andrew LePain, Robert Muiruri, Vivi Nguyen, Mosa Pasha, Roelf L. Slopsema, Alexander Stanforth, Yinan Wang, Mingyao Zhu 👨🔬
Affiliation: Emory Healthcare, Emory University, Department of Radiation Oncology and Winship Cancer Institute, Emory University 🌍
Purpose: Intensity modulated proton therapy (IMPT) treatment planning for craniospinal irradiation (CSI) is complex and requires extensive effort from the planner. This study aims to enhance planning efficiency by automating organ-at-risk (OAR) contouring and plan generation.
Methods: CT images from 11 patients who previously received IMPT CSI treatment were used. "Model-based segmentation" (MBS) and "Deep learning segmentation" (DLS) within the clinical treatment planning system (TPS) (RayStation 2023B) generated OAR contours for each patient. A medical physicist evaluated the auto-contours, ranking them as "good=3," "useful with modification=2," and "not useful=1."
After the contours were completed, a Python script within the same TPS automated the following steps: determining the number and locations of isocenters based on target length; adding two posterior-oblique fields to the brain isocenter and one posterior-anterior field for each spine isocenter; creating dose-gradient matching regions-of interest (ROIs) and beam-specific spot-placement ROIs; setting energy layer spacing, spot spacing, and plan optimization parameters; and adding optimization objectives for targets and OARs. The plan was then ready for optimization.
Results: Out of 38 DLS contours, 34 contours scored "2" or "3" in at least 80% of the patients, while 8 out of 14 MBS contours achieved the same. DLS scores were higher than MBS ones for all structures except the brain and the brainstem. For the 11 patients, the script completed the plan setup in 38±13 seconds (range: 20-55 seconds), compared to the 1-2 hours typically required manually. The optimized plans met clinical standards.
Conclusion: We developed an automated IMPT treatment planning process for CSI patients using built-in auto-segmentation tools and scripting functions in a commercial TPS, significantly improving clinical efficiency. This process has been implemented for routine clinical use. Future work will integrate auto-segmentation into the script to further streamline the workflow and enhance efficiency.