Gradience-Boosted Automatic Robust Dose-Mimicking for Proton Therapy
Abstract
Purpose
Dose-mimicking aims to generate treatment plans based on a reference dose distribution. The importance of dose-mimicking has grown substantially due to the increasing demand for rapid plan adaptation in adaptive radiotherapy—particularly in proton therapy, where dose uncertainty remains a significant concern. However, developing a unified, automated dose-mimicking module for proton therapy is challenging, especially for planning a dose equivalent to or superior to the reference. We proposed a generally applicable dose-mimicking framework for proton therapy planning, combined with a gradient-boosted (GB) approach.
Methods
A GB cost function was designed and implemented in robust dose-mimicking. Ten clinically implemented plans from RayStation were used as references for automated dose-mimicking validation. The gradients of the cost function in optimization were analyzed to prove the effectiveness of the GB approach. Percentage differences of dose-volume histogram (DVH) metrics between reference and auto-generated plans were calculated. Moreover, the homogeneity index (HI) of clinical target volume (CTV) was also calculated. All dose-mimicking was made on an in-house treatment planning system.
Results
The GB approach increased the gradients of the cost function during optimization by 85.3% for organs at risk (OARs) and 54.1% for targets. After GB optimization, gradients decreased by 63.2% in OARs and 9.57% in targets, indicating that the GB approach enables further improvement in plan quality. The HI of CTV in the reference was 0.20±0.14, and for the GB dose-mimicking plans, it was 0.16±0.10. The dose difference (Planin-house – Planreference, %) for CTV D95 was 0.80±2.00%, for CTV D2 was -2.69±3.22%, for OARs was -3.39±7.59%, demonstrating improved dose conformality.
Conclusion
This feasibility study demonstrates the potential of the GB approach in proton dose-mimicking. Owing to its plug-and-play nature, the proposed approach may have broader applicability in adaptive and general radiotherapy planning. Funding: NSFC No.12375359, CIFMS, 2024-I2M-C&T-B-076, Academic Excellence Foundation of BUAA for PhD Students.