Date of Award
Doctor of Philosophy (PhD)
Proton radiotherapy has the potential to treat tumors with better conformal dose distribution than competing modalities when the rapid dose falloff at the end of the proton-beam range is correctly aligned to the edge of the clinical target volume (CTV). However, its clinical potential is dependent on the accurate localization of the Bragg-peak position from predicted stopping-power ratio maps. The method that is most commonly used in today’s clinical practice for predicting stopping-power ratio (SPR) consists of a stoichiometric calibrationtechnique based on single-energy CT (SECT) for direct estimation of patient-specific SPR distribution from vendor-reconstructed Hounsfield Unit (HU) images. Unfortunately, this approach fails to disambiguate the dependence of HU on differences in density and composition of the underlying tissue and has been shown to have an overall inherent uncertainty of 2-3.5% or 1-6 mm in the range of the proton beam. An alternative is to use reconstructed dual-energy CT images (DECT) from two different spectra to estimate additional information of the scanned material such as effective electron density and mean excitation energy to address the degeneracies in today’s SECT-based HU-SPR calibration curves. Our lab previously proposed an innovative DECT-based SPR mapping approach, the Joint Statistical Image-Reconstruction Technique built on a Basis Vector Model (JSIR-BVM), which effectively reconstructs basis materials by using the spectral information from two different energies. In previous studies, we successfully showed the potential of using the basis-material maps reconstructed with JSIR-BVM to estimate highly accurate SPR maps with better noise suppression and resolution than competing DECT based approaches; however, its potential to reduce the 3.5% uncertainty margin in today’s SECT-based SPR maps was not addressed. The purpose of this work is to evaluate the true clinical impact of the JSIR-BVM SPR maps by assessing its ability to further reduce uncertainty margins and to then be translated clinically. To do so, the efforts presented in this dissertation are split into two main parts. In the first part, a series of experiments were designed to compare the SPR maps derived through the JSIR-BVM method to SPR values measured directly from a superconducting synchrocyclotoron. An extensive uncertainty analysis of the JSIR-BVM was then performed in five different scenarios: comparison of JSIR-BVM SPR/SP to International Commission on Radiation Measurements and Units (ICRU) benchmarks; comparison of JSIR-BVM SPR to measured benchmarks; and uncertainties in JSIR-BVM SPR/SP maps for patients of unknown composition. This allowed us for the first time to determine the estimated proton-range uncertainty of an iterative reconstruction SPR mapping technique in proton radiotherapy treatment planning. In the second part of this dissertation, the integration of JSIR-BVM into analytical and Monte Carlo based proton dose calculations was assessed. A preliminary analysis was first performed in a simulated and clinical scenario to compare the accuracy of dose distributions estimated from SECT and JSIR-BVM mass-density maps in a commercial Monte Carlo based proton-therapy treatment planning system. Then, a method was proposed to use the highly accurate JSIR-BVM basis vector model weights to estimate the atomic composition and density needed for simulating elastic, inelastic, and nuclear scattering cross-sections in Monte Carlo dose calculations. In the first part of this dissertation, our findings demonstrated that JSIR-BVM estimates SP and SPR with high accuracy for custom-made bony- and soft-tissue surrogates of known composition (0.4%-0.8% RMS error) with a total estimated uncertainty of 0.5%-1.5%. In a clinical scenario, our analysis showed that JSIR-BVM supports a k=1 uncertainty in range localization of 0.6% to 0.9% for three proton-beam treatment which translated to an uncertainty of 1-2 mm in range, a substantial improvement to the overall uncertainty in today’s SECT and DECT SPR mapping methods. Finally, preliminary studies in simulated and clinical scenarios demonstrated the potential of directly using the highly accurate estimated mass-density maps from JSIR-BVM to generate small but detectable improvement in dose-calculation accuracy in a commercial proton-therapy treatment planning system. The proposed BVM material indexing method also effectively integrated the highly accurate JSIR-BVM information in Monte Carlo based proton dose calculations and reduced the error in estimated Bragg peak positioning of a conventional material indexing method from 1.4 mm to 0.2 mm. These findings, hence, demonstrate the potential of JSIR-BVM to generate SPR maps with subpercentage uncertainty levels that can be integrated into analytical and Monte Carlo based proton-therapy dose calculations, a significant consecutive step for its future clinical deployment.
Joseph A. O'Sullivan Jeffrey Williamson
David G. Politte, Tianyu Zhao, Ulugbek Kamilov, Yuan-Chuan Tai,