Basis Vector Model Method for Proton Stopping Power Estimation using Dual-Energy Computed Tomography
Date of Award
Doctor of Philosophy (PhD)
Accurate estimation of the proton stopping power ratio (SPR) is important for treatment planning and dose prediction for proton beam therapy. The state-of-the-art clinical practice for estimating patient-specific SPR distributions is the stoichiometric calibration method using single-energy computed tomography (SECT) images, which in principle may introduce large intrinsic uncertainties into estimation results. One major factor that limits the performance of SECT-based methods is the Hounsfield unit (HU) degeneracy in the presence of tissue composition variations. Dual-energy computed tomography (DECT) has shown the potential of reducing uncertainties in proton SPR prediction via scanning the patient with two different source energy spectra. Numerous methods have been studied to estimate the SPR by dual-energy CT DECT techniques using either image-domain or sinogram-domain decomposition approaches.
In this work, we implement and evaluate a novel DECT approach for proton SPR mapping, which integrates image reconstruction and material characterization using a joint statistical image reconstruction (JSIR) method based on a linear basis vector model (BVM). This method reconstructs two images of material parameters simultaneously from the DECT measurement data and then uses them to predict the electron densities and the mean excitation energies, which are required by the Bethe equation for computing proton SPR.
The proposed JSIR-BVM method is first compared with image-domain and sinogram-domain decomposition approaches based on three available SPR models including the BVM in a well controlled simulation framework that is representative of major uncertainty sources existing in practice. The intrinsic SPR modeling accuracy of the three DECT-SPR models is validated via theoretical computed radiological quantities for various reference human tissues. The achievable performances of the investigated methods in the presence of image formation uncertainties are evaluated using synthetic DECT transmission sinograms of virtual cylindrical phantoms and virtual patients, which consist of reference human tissues with known densities and compositions. The JSIR-BVM method is then experimentally commissioned using the DECT measurement data acquired on a Philips Brilliance Big Bore CT scanner at 90 kVp and 140 kVp for two phantoms of different sizes, each of which contains 12 different soft and bony tissue surrogates. An image-domain decomposition method that utilizes the two HU images reconstructed via the scanner's software is implemented for comparison
The JSIR-BVM method outperforms the other investigated methods in both the simulation and experimental settings. Although all investigated DECT-SPR models support low intrinsic modeling errors (i.e., less than 0.2% RMS errors for reference human tissues), the achievable accuracy of the image- and sinogram-domain methods is limited by the image formation uncertainties introduced by the reconstruction and decomposition processes. In contrast, by taking advantage of an accurate polychromatic CT data model and a joint DECT statistical reconstruction algorithm, the JSIR-BVM method accounts for both systematic bias and random noise in the acquired DECT measurement data. Therefore, the JSIR-BVM method achieves much better accuracy and precision on proton SPR estimation compared to the image- and sinogram-domain methods for various materials and object sizes, with an overall RMS-of-mean error of 0.4% and a maximum absolute-mean error of 0.7% for test samples in the experimental setting. The JSIR-BVM method also reduces the pixel-wise random variation by 4-fold to 6-fold within homogeneous regions compared to the image- and sinogram-domain methods while exhibiting relatively higher spatial resolution. The results suggest that the JSIR-BVM method has the potential for better SPR prediction in clinical settings.
Joseph A. O'Sullivan
Mark A. Anastasio, R. M. Arthur, David G. Politte, Tianyu Zhao,
Permanent URL: https://doi.org/10.7936/1j70-h466