Proton Therapy Treatment Planning Research Articles

A deep-learning-based surrogate model for Monte-Carlo simulations of the linear energy transfer in primary brain tumor patients treated with proton-beam radiotherapy

Objective. This study explores the use of neural networks (NNs) as surrogate models for Monte-Carlo (MC) simulations in predicting the dose-averaged linear energy transfer (LET d ) of protons in proton-beam therapy based on the planned dose distribution and patient anatomy in the form of computed tomography (CT) images. As LET d is associated with variability in the relative biological effectiveness (RBE) of protons, we also evaluate the implications of using NN predictions for normal tissue complication probability (NTCP) models within a variable-RBE context. Approach. The predictive performance of three-dimensional NN architectures was evaluated using five-fold cross-validation on a cohort of brain tumor patients (n = 151). The best-performing model was identified and externally validated on patients from a different center (n = 107). LET d predictions were compared to MC-simulated results in clinically relevant regions of interest. We assessed the impact on NTCP models by leveraging LET d predictions to derive RBE-weighted doses, using the Wedenberg RBE model. Main results. We found NNs based solely on the planned dose distribution, i.e. without additional usage of CT images, can approximate MC-based LET d distributions. Root mean squared errors (RMSE) for the median LET d within the brain, brainstem, CTV, chiasm, lacrimal glands (ipsilateral/contralateral) and optic nerves (ipsilateral/contralateral) were 0.36, 0.87, 0.31, 0.73, 0.68, 1.04, 0.69 and 1.24 keV µm−1, respectively. Although model predictions showed statistically significant differences from MC outputs, these did not result in substantial changes in NTCP predictions, with RMSEs of at most 3.2 percentage points. Significance. The ability of NNs to predict LET d based solely on planned dose distributions suggests a viable alternative to compute-intensive MC simulations in a variable-RBE setting. This is particularly useful in scenarios where MC simulation data are unavailable, facilitating resource-constrained proton therapy treatment planning, retrospective patient data analysis and further investigations on the variability of proton RBE.

Patient-specific quality assurance of dynamically-collimated proton therapy treatment plans.

The dynamic collimation system (DCS) provides energy layer-specific collimation for pencil beam scanning (PBS) proton therapy using two pairs of orthogonal nickel trimmer blades. While excellent measurement-to-calculation agreement has been demonstrated for simple cube-shaped DCS-trimmed dose distributions, no comparison of measurement and dose calculation has been made for patient-specific treatment plans. To validate a patient-specific quality assurance (PSQA) process for DCS-trimmed PBS treatment plans and evaluate the agreement between measured and calculated dose distributions. Three intracranial patient cases were considered. Standard uncollimated PBS and DCS-collimated treatment plans were generated for each patient using the Astroid treatment planning system (TPS). Plans were recalculated in a water phantom and delivered at the Miami Cancer Institute (MCI) using an Ion Beam Applications (IBA) dedicated nozzle system and prototype DCS. Planar dose measurements were acquired at two depths within low-gradient regions of the target volume using an IBA MatriXX ion chamber array. Measured and calculated dose distributions were compared using 2D gamma analysis with 3%/3mm criteria and low dose threshold of 10% of the maximum dose. Median gamma pass rates across all plans and measurement depths were 99.0% (PBS) and 98.3% (DCS), with a minimum gamma pass rate of 88.5% (PBS) and 91.2% (DCS). The PSQA process has been validated and experimentally verified for DCS-collimated PBS. Dosimetric agreement between the measured and calculated doses was demonstrated to be similar for DCS-collimated PBS to that achievable with noncollimated PBS.

MRI-only based material mass density and relative stopping power estimation via deep learning for proton therapy: a preliminary study

Magnetic Resonance Imaging (MRI) is increasingly being used in treatment planning due to its superior soft tissue contrast, which is useful for tumor and soft tissue delineation compared to computed tomography (CT). However, MRI cannot directly provide mass density or relative stopping power (RSP) maps, which are required for calculating proton radiotherapy doses. Therefore, the integration of artificial intelligence (AI) into MRI-based treatment planning to estimate mass density and RSP directly from MRI has generated significant interest. A deep learning (DL) based framework was developed to establish a voxel-wise correlation between MR images and mass density as well as RSP. To facilitate the study, five tissue substitute phantoms were created, representing different tissues such as skin, muscle, adipose tissue, 45% hydroxyapatite (HA), and spongiosa bone. The composition of these phantoms was based on information from ICRP reports. Additionally, two animal tissue phantoms, simulating pig brain and liver, were prepared for DL training purposes. The phantom study involved the development of two DL models. The first model utilized clinical T1 and T2 MRI scans as input, while the second model incorporated zero echo time (ZTE) MRI scans. In the patient application study, two more DL models were trained: one using T1 and T2 MRI scans as input, and another model incorporating synthetic dual-energy computed tomography (sDECT) images to provide accurate bone tissue information. The DECT empirical model was used as a reference to evaluate the proposed models in both phantom and patient application studies. The DECT empirical model was selected as the reference for evaluating the proposed models in both phantom and patient application studies. In the phantom study, the DL model based on T1, and T2 MRI scans demonstrated higher accuracy in estimating mass density and RSP for skin, muscle, adipose tissue, brain, and liver. The mean absolute percentage errors (MAPE) were 0.42%, 0.14%, 0.19%, 0.78%, and 0.26% for mass density, and 0.30%, 0.11%, 0.16%, 0.61%, and 0.23% for RSP, respectively. The DL model incorporating ZTE MRI further improved the accuracy of mass density and RSP estimation for 45% HA and spongiosa bone, with MAPE values of 0.23% and 0.09% for mass density, and 0.19% and 0.07% for RSP, respectively. These results demonstrate the feasibility of using an MRI-only approach combined with DL methods for mass density and RSP estimation in proton therapy treatment planning. By employing this approach, it is possible to obtain the necessary information for proton radiotherapy directly from MRI scans, eliminating the need for additional imaging modalities.

Dosimetric Comparison Between Proton and Photon Radiation Therapies for Pediatric Neuroblastoma

PurposeThe aim of this study is to determine the most beneficial radiation treatment technique for pediatric patients with thoracic and abdominal neuroblastoma (NBL), through a dosimetric comparison between photon Volumetric Modulated Arc Therapy and proton Intensity-Modulated Proton Therapy treatment plans. Materials and MethodsA retrospective analysis was conducted on a multicentre case series of 19 patients with thoracic and/or abdominal NBL who underwent radiation therapy, following the recommendations of the European protocol for high-risk NBL (HR-NBL2/SIOPEN). The prescribed dose was 21.6 Gy in 12 fractions (1.8 Gy/fraction) delivered over the preoperative disease volume. The dose volume histograms were analyzed for each patient, and a Wilcoxon signed-rank test with a significance level of 0.01 was employed to assess statistical differences between the dosimetric parameters investigated. Two homogeneity indices (HI and newHI) were compared to evaluate the uniformity in dose, delivered to the adjacent vertebrae (VBs_Adj). ResultsBoth radiation techniques conform to the protocol regarding CTV/PTV coverage for every location. Proton therapy resulted in statistically significant dose sparing for the heart and lungs in supradiaphragmatic locations and for the contralateral kidney, liver, spleen, and bowel in subdiaphragmatic locations. For both techniques, sparing the non-adjacent vertebrae (VBs_NAdj) results more challenging, although promising results were obtained. Furthermore, the dose delivered to the VBs_Adj was not statistically different, in terms of homogeneity, for the 2 radiation techniques that both met the protocol’s requirements. ConclusionThis dosimetric analysis highlights the potential of protons to reduce radiation dose to healthy tissue. These findings apply to all the investigated patients, regardless of primary tumor location, making proton therapy a valuable option for the treatment of neuroblastoma. However, a multidisciplinary assessment of each case is essential to ensure the selection of the most effective and suitable treatment modality.

Feasibility and constraints of Bragg peak FLASH proton therapy treatment planning.

FLASH proton therapy (FLASH-PT) requires ultra-high dose rate (≥40 Gy/s) protons to be delivered in a short timescale whilst conforming to a patient-specific target. This study investigates the feasibility and constraints of Bragg peak FLASH-PT treatment planning, and compares the in silico results produced to plans for intensity modulated proton therapy (IMPT). Bragg peak FLASH-PT and IMPT treatment plans were generated for bone (n=3), brain (n=3), and lung (n=4) targets using the MIROpt research treatment planning system and the Conformal FLASH library developed by Applications SA from the open-source version of UCLouvain. FLASH-PT beams were simulated using monoenergetic spot-scanned protons traversing through a conformal energy modulator, a range shifter, and an aperture. A dose rate constraint of ≥ 40 Gy/s was included in each FLASH-PT plan optimisation. Space limitations in the FLASH-PT adapted beam nozzle imposed a maximum target width constraint, excluding 4 cases from the study. FLASH-PT plans did not satisfy the imposed target dose constraints (D95% ≥ 95% and D2%≤ 105%) but achieved clinically acceptable doses to organs at risk (OARs). IMPT plans adhered to all target and OAR dose constraints. FLASH-PT plans showed a reduction in both target homogeneity (p < 0.001) and dose conformity (non-significant) compared to IMPT. Without accounting for a sparing effect, IMPT plans were superior in target coverage, dose conformity, target homogeneity, and OAR sparing compared to FLASH-PT. Further research is warranted in treatment planning optimisation and beam delivery for clinical implementation of Bragg peak FLASH-PT.

OPTIma: simplifying calorimetry for proton computed tomography in high proton flux environments

Objective. Proton computed tomography (pCT) offers a potential route to reducing range uncertainties for proton therapy treatment planning, however the current trend towards high current spot scanning treatment systems leads to high proton fluxes which are challenging for existing systems. Here we demonstrate a novel approach to energy reconstruction, referred to as ‘de-averaging’, which allows individual proton energies to be recovered using only a measurement of their integrated energy without the need for spatial information from the calorimeter. Approach. The method is evaluated in the context of the Optimising Proton Therapy through Imaging (OPTIma) system which uses a simple, relatively inexpensive, scintillator-based calorimeter that reports only the integrated energy deposited by all protons within a cyclotron period, alongside a silicon strip based tracking system capable of reconstructing individual protons in a high flux environment. GEANT4 simulations have been performed to examine the performance of such a system at a modern commercial cyclotron facility using a σ ≈ 10 mm beam for currents in the range 10–50 pA at the nozzle. Main results. Apart from low-density lung tissue, a discrepancy of less than 1% on the Relative Stopping Power is found for all other considered tissues when embedded within a 150 mm spherical Perspex phantom in the 10–30 pA current range, and for some tissues even up to 50 pA. Significance. By removing the need for the calorimeter system to provide spatial information, it is hoped that the de-averaging approach can facilitate clinically relevant, cost effective and less complex calorimeter systems for performing high current pCTs.

Creating uniform cluster dose spread-out Bragg peaks for proton and carbon beams.

Preliminary data have shown a close association of the generalized ionization cluster size dose (in short, cluster dose) with cell survival, independent of particle type, and energy, when cluster dose is derived from an ionization detail parameter preferred for its association with cell survival. Such results suggest cluster dose has the potential to replace RBE-weighted dose in proton and ion beam radiotherapy treatment plan optimization, should a uniform cluster dose lead to comparable biological effects. However, further preclinical investigations are warranted to confirm this premise. To present an analytical approach to create uniform cluster dose spread-out Bragg peaks (SOBP) for evaluation of the potential of cluster dose to result in uniform biological effect. We modified the coefficients of the Bortfeld and Schlegel weight formula, an analytical method typically used for the creation of radiation dose SOBP in particle therapy, to produce uniform cluster dose SOBP of different widths (1-5cm) at relevant clinical proton and carbon beam energies. Optimum parameters were found by minimization of the ratio between the maximum and minimum cluster dose in the SOBP region using the Nelder-Mead method. The coefficients of the Bortfeld and Schlegel weight formula leading to uniform cluster dose SOBPs were determined for each combination of beam energy and SOBP width studied. The uniformity of the resulting cluster dose SOBPs, calculated as the relative difference between the maximum and minimum cluster dose within the SOBP, was within 0.3%-3.5% for the evaluated proton beams and 1.3%-3.4% for the evaluated carbon beams. The modifications to the analytical approach to create radiation dose SOBPs resulted in uniform cluster dose proton and carbon SOBPs over a wide range of beam energies and SOBP widths. Such SOBPs should prove valuable in preclinical investigations for the selection of nanodosimetric quantities to be used in proton and ion therapy treatment planning.

Optically stimulated luminescence dosimeters for simultaneous measurement of point dose and dose-weighted LET in an adaptive proton therapy workflow.

To demonstrate the suitability of optically stimulated luminescence detectors (OSLDs) for accurate simultaneous measurement of the absolute point dose and dose-weighted linear energy transfer (LETD) in an anthropomorphic phantom for experimental validation of daily adaptive proton therapy. A clinically realistic intensity-modulated proton therapy (IMPT) treatment plan was created based on a CT of an anthropomorphic head-and-neck phantom made of tissue-equivalent material. The IMPT plan was optimized with three fields to deliver a uniform dose to the target volume covering the OSLDs. Different scenarios representing inter-fractional anatomical changes were created by modifying the phantom. An online adaptive proton therapy workflow was used to recover the daily dose distribution and account for the applied geometry changes. To validate the adaptive workflow, measurements were performed by irradiating Al2O3:C OSLDs inside the phantom. In addition to the measurements, retrospective Monte Carlo simulations were performed to compare the absolute dose and dose-averaged LET (LETD) delivered to the OSLDs. The online adaptive proton therapy workflow was shown to recover significant degradation in dose conformity resulting from large anatomical and positioning deviations from the reference plan. The Monte Carlo simulations were in close agreement with the OSLD measurements, with an average relative error of 1.4% for doses and 3.2% for LETD. The use of OSLDs for LET determination allowed for a correction for the ionization quenched response. The OSLDs appear to be an excellent detector for simultaneously assessing dose and LET distributions in proton irradiation of an anthropomorphic phantom. The OSLDs can be cut to almost any size and shape, making them ideal for in-phantom measurements to probe the radiation quality and dose in a predefined region of interest. Although we have presented the results obtained in the experimental validation of an adaptive proton therapy workflow, the same approach can be generalized and used for a variety of clinical innovations and workflow developments that require accurate assessment of point dose and/or average LET.

LETd Optimization Verification With an SOI Microdosimeter

A first of its kind experimental verification of dose-averaged linear energy transfer (LETd) optimized treatment plans for proton therapy has been carried out using a silicon-on-insulator microdosimeter at the Massachusetts General Hospital (MGH), Boston, USA. Three clinical treatment plans of a typical ependymoma structure set were designed using the standard clinical approach, the proposed protocol approach, and a one-field approach. The plans were then reoptimized to reduce the LETd-weighted dose in the brain stem. All six plans were delivered in a solid water phantom and the experimental yD‾ measured. After LETd optimization, a reduction in yD‾ was found within the brain stem by an average of 12%, 19%, and 4% for the clinical, protocol, and one-field plans, respectively, while maintaining adequate coverage of the tumor structure. The experimental LETd-weighted doses were in agreement with the treatment planning system calculations and Monte Carlo simulations and reinforced the improvement of the optimization. This work demonstrates the first experimental verification of the clinical implementation of LETd optimization for patient treatment with proton therapy.

Dosimetric comparison of gantry and horizontal fixed-beam proton therapy treatment plans for base of skull chordoma.

Australia's first proton beam therapy (PBT) centre will house a fixed-beam room and two gantry rooms. As the only PBT facility in Australia for at least the short term, there is a need to efficiently allocate treatment appointments between the gantry and fixed-beam rooms. This planning study assesses the dosimetric differences between fixed-beam and gantry-based treatment plans for base of skull chordoma, one of the core indications likely to be referred for PBT in Australia. Retrospective gantry-based and fixed-beam treatment plans were generated for five patients with base of skull chordoma. Fixed-beam plans were generated with a conventional horizontal patient positioning system. Robust intensity modulated proton therapy (IMPT) optimisation and evaluation techniques were used for both delivery systems. Plans were designed to maximise target coverage while adhering to maximum dose constraints to neighbouring critical organs at risk. Robust target coverage and integral dose were found to be approximately equivalent for the gantry-based and fixed-beam plans. Doses to specific organs at risk could be reduced with the gantry-based geometry; however, the gantry-based plans did not exhibit a general decrease in doses to organs at risk. A fixed-beam treatment plan was found to be non-inferior to a gantry-based treatment plan for all base of skull patients included in the current study.

Single energy CT-based mass density and relative stopping power estimation for proton therapy using deep learning method

BackgroundThe number of patients undergoing proton therapy has increased in recent years. Current treatment planning systems (TPS) calculate dose maps using three-dimensional (3D) maps of relative stopping power (RSP) and mass density. The patient-specific maps of RSP and mass density were obtained by translating the CT number (HU) acquired using single-energy computed tomography (SECT) with appropriate conversions and coefficients. The proton dose calculation uncertainty of this approach is 2.5%-3.5% plus 1 mm margin. SECT is the major clinical modality for proton therapy treatment planning. It would be intriguing to enhance proton dose calculation accuracy using a deep learning (DL) approach centered on SECT.ObjectivesThe purpose of this work is to develop a deep learning method to generate mass density and relative stopping power (RSP) maps based on clinical single-energy CT (SECT) data for proton dose calculation in proton therapy treatment.MethodsArtificial neural networks (ANN), fully convolutional neural networks (FCNN), and residual neural networks (ResNet) were used to learn the correlation between voxel-specific mass density, RSP, and SECT CT number (HU). A stoichiometric calibration method based on SECT data and an empirical model based on dual-energy CT (DECT) images were chosen as reference models to evaluate the performance of deep learning neural networks. SECT images of a CIRS 062M electron density phantom were used as the training dataset for deep learning models. CIRS anthropomorphic M701 and M702 phantoms were used to test the performance of deep learning models.ResultsFor M701, the mean absolute percentage errors (MAPE) of the mass density map by FCNN are 0.39%, 0.92%, 0.68%, 0.92%, and 1.57% on the brain, spinal cord, soft tissue, bone, and lung, respectively, whereas with the SECT stoichiometric method, they are 0.99%, 2.34%, 1.87%, 2.90%, and 12.96%. For RSP maps, the MAPE of FCNN on M701 are 0.85%, 2.32%, 0.75%, 1.22%, and 1.25%, whereas with the SECT reference model, they are 0.95%, 2.61%, 2.08%, 7.74%, and 8.62%. ConclusionThe results show that deep learning neural networks have the potential to generate accurate voxel-specific material property information, which can be used to improve the accuracy of proton dose calculation.Advances in knowledgeDeep learning-based frameworks are proposed to estimate material mass density and RSP from SECT with improved accuracy compared with conventional methods.

Incorporating variable RBE in IMPT optimization for ependymoma.

To study the dosimetric impact of incorporating variable relative biological effectiveness (RBE) of protons in optimizing intensity-modulated proton therapy (IMPT) treatment plans and to compare it with conventional constant RBE optimization and linear energy transfer (LET)-based optimization. This study included 10 pediatric ependymoma patients with challenging anatomical features for treatment planning. Four plans were generated for each patient according to different optimization strategies: (1) constant RBE optimization (ConstRBEopt) considering standard-of-care dose requirements; (2) LET optimization (LETopt) using a composite cost function simultaneously optimizing dose-averaged LET (LETd ) and dose; (3) variable RBE optimization (VarRBEopt) using a recent phenomenological RBE model developed by McNamara etal.; and (4) hybrid RBE optimization (hRBEopt) assuming constant RBE for the target and variable RBE for organs at risk. By normalizing each plan to obtain the same target coverage in either constant or variable RBE, we compared dose, LETd , LET-weighted dose, and equivalent uniform dose between the different optimization approaches. We found that the LETopt plans consistently achieved increased LET in tumor targets and similar or decreased LET in critical organs compared to other plans. On average, the VarRBEopt plans achieved lower mean and maximum doses with both constant and variable RBE in the brainstem and spinal cord for all 10 patients. To compensate for the underdosing of targets with 1.1 RBE for the VarRBEopt plans, the hRBEopt plans achieved higher physical dose in targets and reduced mean and especially maximum variable RBE doses compared to the ConstRBEopt and LETopt plans. We demonstrated the feasibility of directly incorporating variable RBE models in IMPT optimization. A hybrid RBE optimization strategy showed potential for clinical implementation by maintaining all current dose limits and reducing the incidence of high RBE in critical normal tissues in ependymoma patients.

The dirty and clean dose concept: Towards creating proton therapy treatment plans with a photon-like dose response.

Applying tolerance doses for organs at risk (OAR) from photon therapy introduces uncertainties in proton therapy when assuming a constant relative biological effectiveness (RBE) of 1.1. This work introduces the novel dirty and clean dose concept, which allows for creating treatment plans with a more photon-like dose response for OAR and, thus, less uncertainties when applying photon-based tolerance doses. The concept divides the 1.1-weighted dose distribution into two parts: the clean and the dirty dose. The clean and dirty dose are deposited by protons with a linear energy transfer (LET) below and above a set LET threshold, respectively. For the former, a photon-like dose response is assumed, while for the latter, the RBE might exceed 1.1. To reduce the dirty dose in OAR, a MaxDirtyDose objective was added in treatment plan optimization. It requires setting two parameters: LET threshold and max dirty dose level. A simple geometry consisting of one target volume and one OAR in water was used to study the reduction in dirty dose in the OAR depending on the choice of the two MaxDirtyDose objective parameters during plan optimization. The best performing parameter combinations were used to create multiple dirty dose optimized (DDopt) treatment plans for two cranial patient cases. For each DDopt plan, 1.1-weighted dose, variable RBE-weighted dose using the Wedenberg RBE model and dose-average LETd distributions as well as resulting normal tissue complication probability (NTCP) values were calculated and compared to the reference plan (RefPlan) without MaxDirtyDose objectives. In the water phantom studies, LET thresholds between 1.5 and 2.5keV/µm yielded the best plans and were subsequently used. For the patient cases, nearly all DDopt plans led to a reduced Wedenberg dose in critical OAR. This reduction resulted from an LET reduction and translated into an NTCP reduction of up to 19 percentage points compared to the RefPlan. The 1.1-weighted dose in the OARs was slightly increased (patient 1: 0.45Gy(RBE), patient 2: 0.08Gy(RBE)), but never exceeded clinical tolerance doses. Additionally, slightly increased 1.1-weighted dose in healthy brain tissue was observed (patient 1: 0.81Gy(RBE), patient 2: 0.53Gy(RBE)). The variation of NTCP values due to variation of α/β from 2 to 3Gy was much smaller for DDopt (2 percentage points (pp)) than for RefPlans (5 pp). The novel dirty and clean dose concept allows for creating biologically more robust proton treatment plans with a more photon-like dose response. The reduced uncertainties in RBE can, therefore, mitigate uncertainties introduced by using photon-based tolerance doses for OAR.

Validation of an MR-based multimodal method for molecular composition and proton stopping power ratio determination using ex vivo animal tissues and tissue-mimicking phantoms

Objective. Range uncertainty in proton therapy is an important factor limiting clinical effectiveness. Magnetic resonance imaging (MRI) can measure voxel-wise molecular composition and, when combined with kilovoltage CT (kVCT), accurately determine mean ionization potential (I m ), electron density, and stopping power ratio (SPR). We aimed to develop a novel MR-based multimodal method to accurately determine SPR and molecular compositions. This method was evaluated in tissue-mimicking and ex vivo porcine phantoms, and in a brain radiotherapy patient. Approach. Four tissue-mimicking phantoms with known compositions, two porcine tissue phantoms, and a brain cancer patient were imaged with kVCT and MRI. Three imaging-based values were determined: SPRCM (CT-based Multimodal), SPRMM (MR-based Multimodal), and SPRstoich (stoichiometric calibration). MRI was used to determine two tissue-specific quantities of the Bethe Bloch equation (I m , electron density) to compute SPRCM and SPRMM. Imaging-based SPRs were compared to measurements for phantoms in a proton beam using a multilayer ionization chamber (SPRMLIC). Main results. Root mean square errors relative to SPRMLIC were 0.0104(0.86%), 0.0046(0.45%), and 0.0142(1.31%) for SPRCM, SPRMM, and SPRstoich, respectively. The largest errors were in bony phantoms, while soft tissue and porcine tissue phantoms had <1% errors across all SPR values. Relative to known physical molecular compositions, imaging-determined compositions differed by approximately ≤10%. In the brain case, the largest differences between SPRstoich and SPRMM were in bone and high lipids/fat tissue. The magnitudes and trends of these differences matched phantom results. Significance. Our MR-based multimodal method determined molecular compositions and SPR in various tissue-mimicking phantoms with high accuracy, as confirmed with proton beam measurements. This method also revealed significant SPR differences compared to stoichiometric kVCT-only calculation in a clinical case, with the largest differences in bone. These findings support that including MRI in proton therapy treatment planning can improve the accuracy of calculated SPR values and reduce range uncertainties.

A stopping criterion for iterative proton CT image reconstruction based on correlated noise properties.

Whereas filtered back projection algorithms for voxel-based CT image reconstruction have noise properties defined by the filter, iterative algorithms must stop at some point in their convergence and do not necessarily produce consistent noise properties for images with different degrees of heterogeneity. A least-squares iterative algorithm for proton CT (pCT) image reconstruction converges toward a unique solution for relative stopping power (RSP) that optimally fits the protons. We present a stopping criterion that delivers solutions with the property that correlations of RSP noise between voxels are relatively low. This provides a method to produce pCT images with consistent noise properties useful for proton therapy treatment planning, which relies on summing RSP along lines of voxels. Consistent noise properties will also be useful for future studies of image quality using metrics such as contrast to noise ratio, and to compare RSP noise and dose of pCT with other modalities such as dual-energyCT. With simulated and real images with varying heterogeneity from a prototype clinical proton imaging system, we calculate average RSP correlations between voxel pairs in uniform regions-of-interest versus distance between voxels. We define a parameter r, the remaining distance to the unique solution relative to estimated RSP noise, and our stopping criterion is based on r falling below a chosenvalue. We find large correlations between voxels for larger values of r, and anticorrelations for smaller values. For r in the range of 0.5-1, voxels are relatively uncorrelated, and compared to smaller values of r have lower noise with only slight loss of spatialresolution. Iterative algorithms not using a specific metric or rationale for stopping iterations may produce images with an unknown and arbitrary level of convergence or smoothing. We resolve this issue by stopping iterations of a least-squares iterative algorithm when r reaches the range of 0.5-1. This defines a pCT image reconstruction method with consistent statistical properties optimal for clinical use, including for treatment planning with pCTimages.

ScatterNet for projection-based 4D cone-beam computed tomography intensity correction of lung cancer patients.

Background and purpose: In radiotherapy, dose calculations based on 4D cone beam CTs (4DCBCTs) require image intensity corrections. This retrospective study compared the dose calculation accuracy of a deep learning, projection-based scatter correction workflow (ScatterNet), to slower workflows: conventional 4D projection-based scatter correction (CBCTcor) and a deformable image registration (DIR)-based method (4DvCT). Materials and methods: For 26 lung cancer patients, planning CTs (pCTs), 4DCTs and CBCT projections were available. ScatterNet was trained with pairs of raw and corrected CBCT projections. Corrected projections from ScatterNet and the conventional workflow were reconstructed using MA-ROOSTER, yielding 4DCBCTSN and 4DCBCTcor. The 4DvCT was generated by 4DCT to 4DCBCT DIR, as part of the 4DCBCTcor workflow. Robust intensity modulated proton therapy treatment plans were created on free-breathing pCTs. 4DCBCTSN was compared to 4DCBCTcor and the 4DvCT in terms of image quality and dose calculation accuracy (dose-volume-histogram parameters and / gamma analysis). Results: 4DCBCTSN resulted in an average mean absolute error of and when compared to 4DCBCTcor and 4DvCT respectively. High agreement was observed in targets with median dose differences of (4DCBCTSN-4DCBCTcor) and (4DCBCTSN-4DvCT). The gamma analysis showed high average / pass rates of for both 4DCBCTSN vs. 4DCBCTcor and 4DCBCTSN vs. 4DvCT. Conclusions: Accurate 4D dose calculations are feasible for lung cancer patients using ScatterNet for 4DCBCT correction. Average scatter correction times could be reduced from (4DCBCTcor) to , showing the clinical suitability of the proposed deep learning-based method.

PO-1969 Microdosimetric approach for treatment planning in proton therapy

PO-1969 Microdosimetric approach for treatment planning in proton therapy

A deep learning-based approach for statistical robustness evaluation in proton therapy treatment planning: a feasibility study

Objective. Robustness evaluation is critical in particle radiotherapy due to its susceptibility to uncertainties. However, the customary method for robustness evaluation only considers a few uncertainty scenarios, which are insufficient to provide a consistent statistical interpretation. We propose an artificial intelligence–based approach that overcomes this limitation by predicting a set of percentile dose values at every voxel and allows for the evaluation of planning objectives at specific confidence levels. Approach. We built and trained a deep learning (DL) model to predict the 5th and 95th percentile dose distributions, which corresponds to the lower and upper bounds of a two-tailed 90% confidence interval (CI), respectively. Predictions were made directly from the nominal dose distribution and planning computed tomography scan. The data used to train and test the model consisted of proton plans from 543 prostate cancer patients. The ground truth percentile values were estimated for each patient using 600 dose recalculations representing randomly sampled uncertainty scenarios. For comparison, we also tested whether a common worst-case scenario (WCS) robustness evaluation (voxel-wise minimum and maximum) corresponding to a 90% CI could reproduce the ground truth 5th and 95th percentile doses. Main results. The percentile dose distributions predicted by DL yielded excellent agreements with the ground truth dose distributions, with mean dose errors below 0.15 Gy and average gamma passing rates (GPR) at 1 mm/1% above 93.9, which were substantially better than the WCS dose distributions (mean dose error above 2.2 Gy and GPR at 1 mm1% below 54). We observed similar outcomes in a dose-volume histogram error analysis, where the DL predictions generally yielded smaller mean errors and standard deviations than the WCS evaluation doses. Significance. The proposed method produces accurate and fast predictions (∼2.5 s for one percentile dose distribution) for a given confidence level. Thus, the method has the potential to improve robustness evaluation.

Stopping-power ratio of body tissues with updated effective energies and elemental I values for treatment planning of proton therapyand ion beam therapy with helium, carbon, oxygen, and neon ions.

The stopping-power ratio (SPR) of body tissues relative to water depends on the particle energy and mean excitation energy (I value) of the tissues. Effective energies to minimize the range error in proton therapyand ion beam therapy with helium, carbon, oxygen, and neon ions and elemental I values have been updated in recent studies. We investigated the effects of these updates on SPR estimation for computed tomography-based treatment planning. The updates led to an increase of up to 0.5% in the SPRs of soft tissues, whereas they led to a decrease of up to 1.9% in the SPRs of bone tissues compared with the current clinical settings. For 44proton beams planned for 15 randomly sampled patients, the mean water-equivalent target depth change was -0.2mm with a standard deviation of 0.2mm. The maximum change was -0.6mm, which we consider to be insignificant in clinical practice.

Rotating Gantries Provide Individualized Beam Arrangements for Charged Particle Therapy.

This study evaluates beam angles used to generate highly individualized proton therapy treatment plans for patients eligible for carbon ion radiotherapy (CIRT). We retrospectively evaluated patients treated with pencil beam scanning intensity modulated proton therapy from 2015 to 2020 who had indications for CIRT. Patients were treated with a 190° rotating gantry with a robotic patient positioning system. Treatment plans were individualized to provide maximal prescription dose delivery to the tumor target volume while sparing organs at risk. The utilized beam angles were grouped, and anatomic sites with at least 10 different beam angles were sorted into histograms. A total of 467 patients with 484 plans and 1196 unique beam angles were evaluated and characterized by anatomic treatment site and the number of beam angles utilized. The most common beam angles used were 0° and 180°. A wide range of beam angles were used in treating almost all anatomic sites. Only esophageal cancers had a predominantly unimodal grouping of beam angles. Pancreas cancers showed a modest grouping of beam angles. The wide distribution of beam angles used to treat CIRT-eligible patients suggests that a rotating gantry is optimal to provide highly individualized beam arrangements.