Click here to

Session: Therapy General ePoster Viewing [Return to Session]

Learning DVH Criteria in Radiation Therapy Treatment Planning

F Ahmadi*, T McNutt, K Ghobadi, Johns Hopkins University, Baltimore, MD

Presentations

PO-GePV-T-262 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Clinically acceptable radiation therapy plans might not be globally optimal, and improvements in terms of better target coverage or organs-at-risk (OARs) sparing are possible. The purpose of this work is to develop data-driven optimization models to improve OAR dose-volume objective tradeoffs based on available historical treatment plans.

Methods: We develop novel Inverse Learning methods to improve and personalize radiation therapy treatment plans. Inverse Learning is a mathematical method to learn optimal decisions from observed past decisions, i.e., historical clinical plans. Our models consider two explicit goals of (a) improving existing clinical plans and (b) providing insights on how such improvements can enhance planning time and effectiveness for future patients. Specifically, we consider improvements of dose-volume histogram (DVH) criteria. Our models can include additional constraints, e.g., minimizing perturbation from the historical plans to ensure homogeneity or other desired properties are maintained.

Results: We validate our methods on prostate cancer patients as a proof-of-concept. Our results for a representative patient case show a reduction of 3.3 Gy to 50% volume of rectum compared to the historical clinical plan, representing a 27.3% decrease over the average dose to Rectum. The planning target volume (PTV) coverage, while experiencing slightly reduced dose levels, remained comparable and within clinical tolerance in both plans at 95% volume with only 0.4% decrease in average dose. The rest of OARs received similar dose levels.

Conclusion: The results indicate that using inverse learning methods can improve the DVH objectives in radiation therapy treatment plans. The outcomes illustrate the potential for data-driven frameworks capable of identifying the best improvements for groups of patients with similar attributes. Inverse learning will be the core method to provide improved objective guidelines for future radiation therapy patients.

Funding Support, Disclosures, and Conflict of Interest: This work is supported by the Johns Hopkins Discovery Awards and the Johns Hopkins Malone Seed Grant Program.

Keywords

Inverse Planning, Optimization, Dose Volume Histograms

Taxonomy

TH- External Beam- Photons: IMRT/VMAT dose optimization algorithms

Contact Email

Share: