High-Fidelity Anthropomorphic Phantom Construction Via Elemental Composition Modeling for Proton Therapy
Abstract
Purpose
This study proposes a novel elemental composition modeling framework for constructing high-fidelity anthropomorphic phantoms, addressing the critical bottleneck of clinical data scarcity in medical imaging research.
Methods
A parameterized HU-to-tissue elemental mapping was established, enabling the creation of a dynamic density-element matrix for 24 tissue types via piecewise polynomial interpolation, rigorously constrained by the ICRU database. To validate the phantom's fidelity, the ground-truth Relative Stopping Power (RSP) was calculated from its known elemental composition using the Geant4 toolkit. A virtual CT simulator generated corresponding X-ray projections. These CT images were then used to train a deep learning model for RSP prediction and were also processed via conventional stoichiometric calibration to obtain a comparative RSP map. Finally, all three RSP distributions were imported into the matRad toolkit for proton dose calculation and analysis.
Results
The accuracy of the deep learning (DL)-generated and stoichiometric-calibrated proton CT (PCT) was evaluated using the Mean Absolute Percentage Error (MAPE). For the prostate region, the DL method achieved a MAPE of 1.94%, a 22.09% improvement over the stoichiometric result of 2.49%. In the bladder region, the corresponding values were 1.60% and 1.91%, representing a 16.23% improvement. In the dosimetric evaluation, gamma analysis further demonstrated that the dose distributions calculated based on the DL-generated RSP were significantly superior to those derived from the stoichiometric method across all criteria. Specifically, the average gamma passing rate under the 1%/1 mm criterion reached 91.03% (compared to 87.53% for the stoichiometric method), 96.92% (94.58%) under 2%/2 mm, and 98.66% (96.78%) under 3%/3 mm.
Conclusion
The PCT generated through deep learning reduces the uncertainty in RSP calculation during the photon CT conversion process. This can effectively improve the precision and efficacy of proton therapy while minimizing the side effects on normal tissues.