Validation of a deep learning-based material estimation model for Monte Carlo dose calculation in proton therapy
Chih-Wei Chang1, Shuang Zhou2, Yuan Gao1
1Department of Radiation Oncology and Winship Cancer Institute, Emory University, Atlanta, GA 30308, United States of America.
Physics in Medicine and Biology
|September 29, 2022
Summary
This study developed a framework to validate proton range accuracy using CT-based material models. Physics-informed deep learning models significantly improved Monte Carlo dose calculations, reducing proton range uncertainties in treatment planning.
Area of Science:
- Medical Physics
- Radiotherapy
- Computational Imaging
Background:
- Proton range uncertainty in CT-based material conversion impacts treatment quality.
- Dual-energy CT (DECT) and deep learning (DL) show promise for predicting proton ranges.
- Physics-informed DL offers advanced material property inference.
Purpose of the Study:
- Develop a framework to validate Monte Carlo dose calculation (MCDC) using CT material characterization models.
- Assess the accuracy of proton range prediction and dose distribution.
- Improve the quality of proton treatment planning.
Main Methods:
- Utilized anthropomorphic and porcine phantoms for validation experiments.
- Employed physics-informed residual networks (PRN) for mass density inference from DECT.
- Compared PRN with conventional DECT and single-energy CT (SECT) models using gamma index for dose analysis.
Main Results:
- MCDC with PRN achieved high gamma passing rates (95.9%-97.8%) compared to empirical DECT models (79.7%-86.0%).
- PRN models demonstrated minimal mean Water Equivalent Thickness (WET) variations (-0.06 mm) versus measurement.
- Validation confirmed consistent dose and WET distributions with PRN.
Conclusions:
- The proposed framework effectively validates CT-based material models for MCDC.
- Physics-informed DL models significantly enhance proton range accuracy and treatment planning.
- The framework supports online adaptive radiotherapy planning.
