Modeling and Mitigating Anatomical Uncertainties and Motion in Proton Therapy Via Gaussian Process Prediction and Water Equivalent Thickness Mapping
Author(s)
Bohannon, Duncan Henry
Advisor(s)
Zhou, Jun
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Currently, proton radiotherapy accounts for uncertainties from setup and inter- and intra-fraction anatomical changes using population-based margins. However, by their nature, population-based margins tend to overestimate the actual setup error for most patients, while underestimating it for a minority. To address this, this work uses Gaussian Process (GP) models to predict patient-specific inter-fraction motion. For sites with high anatomical uncertainty such as thoracic cancers, plans are robustly optimized on the on the treatment planning CT (TPCT) and additional images that represent anatomical uncertainty. This optimization technique, termed 4D robust optimization, is inefficient because it optimizes over multiple images. To address this inefficiency, this work will develop water equivalent thickness mapping (WET-MAP), which converts beam path changes from anatomical uncertainty to contour changes on the TPCT.
This work applied Gaussian Process Prediction Water Equivalent Thickness Map (GPP-WET-MAP, or GWM), a planning strategy that uses GP regression models to predict patient-specific prostate inter-fraction motion and accounts for the predicted inter-fraction motion with WET mapping, to conventionally fractionated and stereotactic body radiotherapy (SBRT) prostate patients. GWM plans were optimized on just the TPCT, increasing efficiency, and use a smaller setup uncertainty. This work found that, compared to current clinical practice, GWM plans maintain coverage, have improved robustness to inter-fraction motion, and increased organ sparing for both groups of prostate patients, demonstrating a high potential for clinical use.
Sponsor
Date
2025-07-11
Extent
Resource Type
Text
Resource Subtype
Dissertation