Optimization of Automated Radiotherapy Planning for Head and Neck Cancers and Brain Tumors Using Knowledge-Based...

Area of Science:

• Radiation Oncology
• Medical Physics
• Cancer Treatment Planning

Background:

• Knowledge-based planning (KBP) models, such as RapidPlan, aim to improve efficiency and consistency in radiation therapy planning.
• Evaluating the clinical utility of KBP models requires assessing their impact on dose coverage, organ-at-risk (OAR) sparing, and treatment delivery parameters.
• Standard statistical metrics may not fully capture the clinical relevance of KBP model performance.

Purpose of the Study:

• To develop and evaluate a locally trained KBP model using RapidPlan for head and neck (H&N), brain, and central nervous system (CNS) malignancies.
• To determine if statistical metrics like R² and outlier frequency predict the clinical utility of the KBP model.
• To assess the model's impact on OAR doses and monitor units (MUs).

Main Methods:

• A retrospective institutional dataset of 594 plans was used, with 497 plans for training and 370 paired plans for evaluation.
• Performance was assessed using identical beam geometry, quantifying training-validation overlap at plan and patient levels.
• A held-out sensitivity analysis, subgroup assessment, and OAR threshold achievement analysis were performed.

Main Results:

• The RapidPlan model maintained target coverage while significantly reducing OAR doses, including oral cavity Dmean (-7.62%) and larynx Dmean (-7.57%).
• Monitor units decreased significantly from 764.2 ± 275.5 to 695.8 ± 210.3 MU.
• A high R² was not necessary for clinically useful optimization, but patient-level overlap limited generalizability claims.

Conclusions:

• A locally trained KBP model can be a clinically useful tool for standardizing treatment planning and supporting decision-making in H&N, brain, and CNS cancers.
• Standard statistical metrics alone are insufficient to predict the clinical utility of KBP models.
• The model requires expert review and should be viewed as a decision-support tool, not a replacement for clinical judgment.