Impact of beam-hardening corrections on proton relative stopping power estimates from single- and dual-energy CT.

A major contributing factor to proton range uncertainty is the conversion of computed tomography (CT) Hounsfield units (HU) to proton relative stopping power (RSP). This uncertainty is heightened in the presence of X-ray beam-hardening artifact (BHA), which has two manifestations: cupping and streaking, especially in and near bone tissue. This uncertainty can affect the accuracy of proton RSP calculation for treatment planning in proton radiotherapy. Dual-energy CT (DECT) and iterative beam-hardening correction (iBHC) both show promise in mitigating CT BHA. This present work attempts to analyze the relative robustness of iBHC and DECT techniques on both manifestations of BHA. The stoichiometric method for HU to RSP conversion was used for single-energy CT (SECT) and DECT-based monochromatic techniques using a tissue substitute phantom. Cupping BHA was simulated by measuring the HU of a bone substitute plug in wax/3D-printed phantoms of increasing size. Streaking BHA was simulated by placing a solid water plug between two bone plugs in a wax phantom. Finally, the effect of varying calibration phantom size on RSP was calculated in an anthropomorphic head phantom. The RSP decreased -0.002 cm-1 as phantom size increased for SECT but remained largely constant when iBHC applied or with DECT techniques. The RSP varied a maximum of 2.60% in the presence of streaking BHA in SECT but was reduced to 1.40% with iBHC. For DECT techniques, the maximum difference was 2.40%, reduced to 0.6% with iBHC. Comparing calibration phantoms of 20- and 33-cm diameter, maximum voxel differences of 5 mm in the water-equivalent thickness were observed in the skull but reduced to 1.3 mm with iBHC. The DECT techniques excelled in mitigating cupping BHA, but streaking BHA still could be observed. The use of iBHC reduced RSP variation with BHA in both SECT and DECT techniques.

Accuracy of proton stopping power estimation of silicone breast implants with single and dual-energy CT calibration techniques.

A major contributing factor to proton range uncertainty is the conversion of computed tomography (CT) Hounsfield Units (HU) to proton relative stopping power (RSP). This uncertainty is elevated with implanted devices, such as silicone breast implants when computed with single energy CT (SECT). In recent years, manufacturers have introduced implants with variations in gel cohesivity. Deriving the RSP for these implants from dual-energy CT (DECT) can result in a marked reduction of the error associated with SECT. In this study, we investigate the validity of DECT calibration of HU to RSP on silicone breast implants of varying cohesivity levels. A DECT capable scanner was calibrated using the stoichiometric method of Bourque et al for SECT and DECT using a tissue substitute phantom. Three silicone breast implants of increasing gel cohesivity were measured in a proton beam of clinical energy to determine ground-truth RSP and water equivalent thickness (WET). These were compared to SECT-derived RSP at three CT spectrum energies and DECT with two energy pairs (80/140 kVp and 100/140 kVp) as obtained from scans with and without an anthropomorphic phantom. The RSP derived from parameters estimates from CT vendor-specific software (syngo.via) was compared. The WET estimates from SECT deviated from MLIC ground truth approximately +11%-19%, which would result in overpenetration if used clinically. Both the Bourque calibration and syngo.via WET estimates from DECT yielded error ≤0.5% from ground truth; no significant difference was found between models of varying gel cohesivity levels. WET estimates without the anthropomorphic phantom were significantly different than ground truth for the Bourque calibration. From these results, gel cohesivity had no effect on proton RSP. User-generated DECT calibration can yield comparably accurate RSP estimates for silicone breast implants to vendor software methods. However, care must be taken to account for beam hardening effects.

Prediction of the output factor using machine and deep learning approach in uniform scanning proton therapy.

PURPOSE: The purpose of this work is to develop machine and deep learning-based models to predict output and MU based on measured patient quality assurance (QA) data in uniform scanning proton therapy (USPT). METHODS: This study involves 4,231 patient QA measurements conducted over the last 6 years. In the current approach, output and MU are predicted by an empirical model (EM) based on patient treatment plan parameters. In this study, two MATLAB-based machine and deep learning algorithms - Gaussian process regression (GPR) and shallow neural network (SNN) - were developed. The four parameters from patient QA (range, modulation, field size, and measured output factor) were used to train these algorithms. The data were randomized with a training set containing 90% and a testing set containing remaining 10% of the data. The model performance during training was accessed using root mean square error (RMSE) and R-squared values. The trained model was used to predict output based on the three input parameters: range, modulation, and field size. The percent difference was calculated between the predicted and measured output factors. The number of data sets required to make prediction accuracy of GPR and SNN models' invariable was also evaluated. RESULTS: The prediction accuracy of machine and deep learning algorithms is higher than the EM. The output predictions with [GPR, SNN, and EM] within ± 2% and ± 3% difference were [97.16%, 97.64%, and 92.95%] and [99.76%, 99.29%, and 97.18%], respectively. The GPR model outperformed the SNN with a smaller number of training data sets. CONCLUSION: The GPR and SNN models outperformed the EM in terms of prediction accuracy. Machine and deep learning algorithms predicted the output factor and MU for USPT with higher predictive accuracy than EM. In our clinic, these models have been adopted as a secondary check of MU or output factors.