Enhancing accuracy in proton therapy: The impact of geometric uncertainty models in head and neck cancer treatment.

Abstract

Background: Anatomical changes present a major source of uncertainty in head and neck (H&N) cancer treatment. Accurate modeling of these changes is important for enhancing treatment precision and supporting better outcomes.

Purpose: The purpose of this study is to assess different anatomical uncertainty modeling methods in robust optimization for H&N cancer proton therapy.

Methods: This retrospective study involved five nasopharynx radiotherapy patients. We compared conventional robust optimization with anatomical robust optimization (aRO): (1) conventional robust optimization (cRO-3 mm), which used 3 mm setup shift and 3% range uncertainty. (2) aRO_AM which used three predicted images from an AM capturing systematic anatomical changes, with a 1 mm setup shift and 3% range uncertainty. (3) aRO_PM, which used three predicted images from a probability model (PM) capturing the most probable deformations, also with a 1 mm setup shift and 3% range uncertainty. We assessed weekly dose coverage of the clinical target volumes (CTVs). Normal tissue complication probability (NTCP) for grade $\ge$ 2 xerostomia and grade $\ge$ 2 dysphagia were calculated using the accumulated nominal dose (without errors).

Results: aRO_PM outperformed cRO-3 mm and aRO_AM, consistently achieving V94 ${}_{\text{voxmin}}$ $\ge$ 95% for all cases across treatment weeks. Additionally, aRO_PM reduced the NTCP for grade $\ge$ 2 xerostomia by an average of 4.88 %, with a maximum reduction of 8.03%, and reduced the NTCP for grade $\ge$ 2 dysphagia by an average of 1.80%, with a maximum reduction of 4.23 %.

Conclusion: The PM demonstrates potential for improving robust optimization by effectively managing anatomical uncertainties in H&N cancer proton therapy, thereby enhancing treatment effectiveness.