Author: Ming Chao, Hao Guo, Tenzin Kunkyab, Yang Lei, Tian Liu, Kenneth Rosenzweig, Robert Samstein, Junyi Xia, Jiahan Zhang 👨🔬
Affiliation: Icahn School of Medicine at Mount Sinai 🌍
Purpose:
This study aims to develop and validate an Eclipse Scripting Application Programming Interface (ESAPI)-based planning toolbox that incorporates preset human expertise to improve planning efficiency for non-small cell lung cancer (NSCLC) patients.
Methods:
The ESAPI-based toolbox incorporates custom-designed ESAPI modular scripts with predefined functionalities, mimicking planners’ actions in treatment planning process. The scripts summarize the common effective strategies to finetune a plan to meet clinical constraints. Validation was performed using 33 anonymized NSCLC cases, where a human operator executed scripts iteratively to optimize dosimetric parameters based on clinical guidelines. The beam geometry settings are kept constant to standardize plan generation. Prescription information and region of interest (ROI) structures were extracted from reference clinically delivered plans for plan generation. Plans generated using the ESAPI-based toolbox were compared with those produced by using a clinical RapidPlan model to evaluate quality, consistency, and adaptability. Dosimetric evaluations included planning target volume (PTV) coverage and dose thresholds for organs at risk (OARs) and ROIs, including the heart, esophagus, lungs, spinal cord, and the body.
Results:
All 33 toolbox-generated plans fully met clinical dosimetric requirements, achieving adequate PTV coverage and satisfying OAR and ROI constraints through user-driven adjustments. In comparison, 9 out of 33 RapidPlan-generated plans failed to fully meet clinical criteria due to excessive doses to OARs or insufficient PTV coverage. Although both strategies made clinically-acceptable plans in most (24 out of 33) cases, the toolbox demonstrated superior adaptability.
Conclusion:
The ESAPI-based toolbox provides a feasible framework for automated treatment planning workflow, demonstrated by a semi-automated process combining modular scripts with human operation to generate clinically acceptable plans. The future work is to enable fully automated planning by integrating AI agents to execute the script toolbox independently. This work lays the groundwork for transitioning to fully automated NSCLC treatment planning.