Hybrid Artificial Intelligence–Monte Carlo Methods for Radiotherapy Dose Calculation

Monte Carlo (MC) computational techniques are widely used for applications in radiation medicine and have traditionally covered the areas of radiation dosimetry, shielding, radiotherapy treatment planning, and radiological imaging. Moreover, they have contributed to the improvement and understanding of the link between physical parameters of radiation delivery and therapy success. Recently, MC methods are being integrated with other technologies used in radiation therapy, such as inverse optimization, deformable image registration, and machine learning techniques for outcomes studies, etc.

Emerging magnetic resonance (MR) guided radiotherapy affords significantly improved anatomy visualization and, subsequently, more effective personalized treatment. The new therapy paradigm imposes significant demands on radiation dose calculation quality and speed, creating an unmet need for the acceleration of Monte Carlo (MC) dose calculation. Existing deep learning approaches to denoise the final plan MC dose fail to achieve the accuracy and speed requirements of large-scale beamlet dose calculation in the presence of a strong magnetic field for online adaptive radiotherapy planning. Our deep learning dose calculation method, DeepMC, addresses these needs by predicting low-noise dose from extremely noisy (but fast) MC-simulated dose and anatomical inputs, thus enabling significant acceleration. DeepMC simultaneously reduces MC sampling noise and predicts corrupted dose buildup at tissue-air material interfaces resulting from MR-field induced electron return effects. Here we demonstrate our model’s ability to accelerate dose calculation for daily treatment planning by a factor of 38 over traditional low-noise MC simulation with clinically meaningful accuracy in deliverable dose and treatment delivery parameters. As a post-processing approach, DeepMC provides compounded acceleration of large-scale dose calculation when used alongside established MC acceleration techniques in variance reduction and graphics processing unit-based MC simulation.

The low computation efficiency impeded the broad application of Monte Carlo (MC) simulation to particle therapy. The existing deep learning (DL) methods for fast dose calculation lacked physics-based interpretability, hence, may introduce additional risks, especially for the more complex carbon ion radiotherapy. To develop and validate a multi-modal diffusion model, Diff-MC, for noise reduction of particle number limited MC dose calculation, potentially supporting better optimization and faster online adaptation for carbon ionradiotherapy. By using multi-modal data such as CT images, dose maps using a low number of primary particles and beam parameters, and so forth, Diff-MC was developed to generate a dose map adaptively based on the beam state. To enable effective inter-modal interactions, a hybrid-fusion strategy was applied to integrate the data-level, feature-level, and decision-level fusion. The model was evaluated on a highly heterogeneous dataset, including 15000 paired beamlet data cropped from 20 CTs for training and validating, 500 paired beamlet data cropped from other 5 CTs for testing, and 500 paired beamlet data cropped from another 100 CTs for generalizability test. All datasets encompassed various geometry and beamlet physics parameters such as energy distribution and number of primary particles, and soforth. Using the MC simulation based on high number of primary particles as ground-truth, the Diff-MC achieved nearly linear acceleration and high accuracy of gamma passing rate up to 99.25% under the criteria of 3mm, 3%, 10% cutoff. The performance was significantly higher (all ) than the UNet-based models (96.17%) and transformer-based models (97.81%). The accuracy achieved by Diff-MC in the generalizability test was 99.22%. The lateral dose, integral depth dose (IDD), and percentage depth dose (PDD) of Diff-MC were also more consistent with the ground-truth than that of conventional AImodels. The proposed Diff-MC method displayed high efficiency and robustness in carbon ion dose calculation. By maintaining the physics features of MC, the results of Diff-MC were more interpretable and generalizable than the conventional AImodels.

Monte Carlo (MC) method is the most accurate computational method for dose calculation, but its wide application on clinical accurate radiotherapy is hindered due to its poor speed of converging and long computation time. In the MC dose calculation research, the main task is to speed up computation while high precision is maintained. The purpose of this paper is to enhance the calculation speed of MC method for electron-photon transport with high precision and ultimately to reduce the accurate radiotherapy dose calculation time based on normal computer to the level of several hours, which meets the requirement of clinical dose verification. Based on the existing Super Monte Carlo Simulation Program (SuperMC), developed by FDS Team, a fast MC method for electron-photon coupled transport was presented with focus on two aspects: firstly, through simplifying and optimizing the physical model of the electron-photon transport, the calculation speed was increased with slightly reduction of calculation accuracy; secondly, using a variety of MC calculation acceleration methods, for example, taking use of obtained information in previous calculations to avoid repeat simulation of particles with identical history; applying proper variance reduction techniques to accelerate MC method convergence rate, etc. The fast MC method was tested by a lot of simple physical models and clinical cases included nasopharyngeal carcinoma, peripheral lung tumor, cervical carcinoma, etc. The result shows that the fast MC method for electron-photon transport was fast enough to meet the requirement of clinical accurate radiotherapy dose verification. Later, the method will be applied to the Accurate/Advanced Radiation Therapy System ARTS as a MC dose verification module.

Artificial intelligence in reaction prediction and chemical synthesis

Clinical research in boron neutron capture therapy (BNCT) has been conducted worldwide. Currently, the Monte Carlo (MC) method is the only dose calculation algorithm implemented in the treatment planning system for the clinical treatment of BNCT. We previously developed the MC-RD calculation method, which combines the MC method and the removal-diffusion (RD) equation, for fast dose calculation in BNCT. This study aimed to verify the partial-MC-RD calculation method, which utilizes the MC-RD calculation method for a portion of the entire neutron energy range, in terms of calculation accuracy and time as the dose calculation method. We applied the partial-MC-RD calculation method to calculate the total dose for head phantom, comprising soft tissue, brain tissue, and bone. The calculation time and accuracy were evaluated based on the full-MC method. Our accuracy verifications indicated that the partial-MC-RD calculation was mostly comparable with full-MC calculation in the accuracy. However, the assumptions and approximation used in the RD calculation mainly occurred the discrepancy from the full-MC calculation result. Additionally, the partial-MC-RD calculation reduced the time required to approximately 45% for the irradiation to the top and cheek region of head phantom, compared to the full-MC calculation. In conclusion, the MC-RD calculation method can be the basis of a fast dose calculation method in BNCT.

Dose calculation in radiotherapy aims to accurately estimate and assess the dose distribution of a treatment plan. Monte Carlo (MC) dose calculation is considered the gold standard owing to its ability to accurately simulate particle transport in inhomogeneous media. However, uncertainties such as the patient's dynamically deforming anatomy can still lead to differences between the delivered and planned dose distribution. Development and validation of a deformable voxel geometry for MC dose calculations (DefVoxMC) to account for dynamic deformation in the dose calculation process of photon- and electron-based radiotherapy treatment plans for clinically motivated cases. DefVoxMC relies on the subdivision of a regular voxel geometry into dodecahedrons. It allows shifting the dodecahedrons' corner points according to the deformation in the patient's anatomy using deformation vector fields (DVF). DefVoxMC is integrated into the Swiss Monte Carlo Plan (SMCP) to allow the MC dose calculation of photon- and electron-based treatment plans on the deformable voxel geometry. DefVoxMC is validated in two steps. A compression test and a Fano test are performed in silico. Delta4 (for photon beams) and EBT4 film measurements in a cubic PMMA phantom (for electron beams) are performed on a TrueBeam in Developer Mode for clinically motivated treatment plans. During these measurements, table motion is used to mimic rigid dynamic patient motion. The measured and calculated dose distributions are compared using gamma passing rate (GPR) (3% / 2mm (global), 10% threshold). DefVoxMC is used to study the impact of patient-recorded breathing motion on the dose distribution for clinically motivated lung and breast cases, each prescribed 50Gy to 50% of the target volume. A volumetric modulated arc therapy (VMAT) and an arc mixed-beam radiotherapy (Arc-MBRT) plan are created for the lung and breast case, respectively. For the dose calculation, the dynamic deformation of the patient's anatomy is described by DVFs obtained from deformable image registration of the different phases of 4DCTs. The resulting dose distributions are compared to the ones of the static situation using dose-volume histograms and dose differences. DefVoxMC is successfully integrated into the SMCP to enable the MC dose calculation of photon- and electron-based treatments on a dynamically deforming patient anatomy. The compression and the Fano test agree within 1.0% and 0.1% with the expected result, respectively. Delta4 and EBT4 film measurements agree with the calculated dose by a GPR>95%. For the clinically motivated cases, breathing motion resulted in areas with a dose increase of up to 26.9Gy (lung) and up to 7.6Gy (breast) compared to the static situation. The largest dose differences are observed in high-dose-gradient regions perpendicular to the beam plane, consequently decreasing the planning target volume coverage (V95%) by 4.2% for the lung case and 2.0% for the breast case. A novel method for MC dose calculation for photon- and electron-based treatments on dynamically deforming anatomy is successfully developed and validated. Applying DefVoxMC to clinically motivated cases, we found that breathing motion has non-negligible impact on the dosimetric plan quality.

Purpose:Investigating the relative sensitivity of Monte Carlo (MC) and Pencil Beam (PB) dose calculation algorithms to low‐Z (titanium) metallic artifacts is important for accurate and consistent dose reporting in post¬operative spinal RS.Methods:Sensitivity analysis of MC and PB dose calculation algorithms on the Monaco v.3.3 treatment planning system (Elekta CMS, Maryland Heights, MO, USA) was performed using CT images reconstructed without (plain) and with Orthopedic Metal Artifact Reduction (OMAR; Philips Healthcare system, Cleveland, OH, USA). 6MV and 10MV volumetric‐modulated arc (VMAT) RS plans were obtained for MC and PB on the plain and OMAR images (MC‐plain/OMAR and PB‐plain/OMAR).Results:Maximum differences in dose to 0.2cc (D0.2cc) of spinal cord and cord +2mm for 6MV and 10MV VMAT plans were 0.1Gy between MC‐OMAR and MC‐plain, and between PB‐OMAR and PB‐plain. Planning target volume (PTV) dose coverage changed by 0.1±0.7% and 0.2±0.3% for 6MV and 10MV from MC‐OMAR to MC‐plain, and by 0.1±0.1% for both 6MV and 10 MV from PB‐OMAR to PB‐plain, respectively. In no case for both MC and PB the D0.2cc to spinal cord was found to exceed the planned tolerance changing from OMAR to plain CT in dose calculations.Conclusion:Dosimetric impacts of metallic artifacts caused by low‐Z metallic spinal hardware (mainly titanium alloy) are not clinically important in VMAT‐based spine RS, without significant dependence on dose calculation methods (MC and PB) and photon energy ≥ 6MV. There is no need to use one algorithm instead of the other to reduce uncertainty for dose reporting. The dose calculation method that should be used in spine RS shall be consistent with the usual clinical practice.

The safety and clinical efficacy of 125 I seed-loaded stent for the treatment of portal vein tumor thrombosis (PVTT) have been shown. Accurate and fast dose calculation of the 125 I seeds with the presence of the stent is necessary for the plan optimization and evaluation. However, the dosimetric characteristics of the seed-loaded stents remain unclear and there is no fast dose calculation technique available. This paper aims to explore a fast and accurate analytical dose calculation method based on Monte Carlo (MC) dose calculation, which takes into account the effect of stent and tissue inhomogeneity. A detailed model of the seed-loaded stent was developed using 3D modeling software and subsequently used in MC simulations to calculate the dose distribution around the stent. The dose perturbation caused by the presence of the stent was analyzed, and dose perturbation kernels (DPKs) were derived and stored for future use. Then, the dose calculation method from AAPM TG-43 was adapted by integrating the DPK and appropriate inhomogeneity correction factors (ICF) to calculate dose distributions analytically. To validate the proposed method, several comparisons were performed with other methods in water phantom and voxelized CT phantoms for three patients. The stent has a considerable dosimetric effect reducing the dose up to 47.2% for single-seed stent and 11.9%-16.1% for 16-seed stent. In a water phantom, dose distributions from MC simulations and TG-43-DP-ICF showed a good agreement with the relative error less than 3.3%. In voxelized CT phantoms, taking MC results as the reference, the relative errors of TG-43 method can be up to 33%, while those of TG-43-DP-ICF method were less than 5%. For a dose matrix with 256 × 256 × 46 grid (corresponding to a phantom of 17.2 × 17.2 × 11.5cm3 ) for 16-seed-loaded stent, it only takes 17s for TG-43-DP-ICF to compute, compared to 25h for the full MC calculation. The combination of DPK and inhomogeneity corrections is an effective approach to handle both the presence of stent and tissue heterogeneity. Exhibiting good agreement with MC calculation and computational efficiency, the proposed TG-43-DP-ICF method is adequate for dose evaluation and optimization in seed-loaded stent implantation treatment planning.

Monte Carlo (MC) radiation transport methods are used for dose calculation as ‘gold standard.’ However, the method is computationally time-consuming and thus impractical for normal tissue dose reconstructions for the large number of proton therapy patients required for epidemiologic investigations of late health effects. In the present study, we developed a new dose calculation method for the rapid reconstruction of out-of-field neutron dose to patients undergoing pencil beam scanning (PBS) proton therapy. The new dose calculation method is based on neutron dose voxel kernels (DVKs) generated by MC simulations of a proton pencil beam irradiating a water phantom (60 × 60 × 300 cm3), which was conducted using a MC proton therapy simulation code, TOPAS. The DVKs were generated for 19 beam energies (from 70 to 250 MeV with the 10 MeV interval) and three range shifter thicknesses (1, 3, and 5 cm). An in-house program was written in C++ to superimpose the DVKs onto a patient CT images according to proton beam characteristics (energy, position, and direction) available in treatment plans. The DVK dose calculation method was tested by calculating organ/tissue-specific neutron doses of 1- and 5-year-old whole-body computational phantoms where intracranial and craniospinal irradiations were simulated. The DVK-based doses generally showed reasonable agreement with those calculated by direct MC simulations with a detailed PBS model that were previously published, with differences mostly less than 30% and 10% for the intracranial and craniospinal irradiations, respectively. The computation time of the DVK method for one patient ranged from 1 to 30 min on a single CPU core of a personal computer, demonstrating significant improvement over the direct MC dose calculation requiring several days on high-performance computing servers. Our DVK-based dose calculation method will be useful when dosimetry is needed for the large number of patients such as for epidemiologic or clinical research.

Treatment fields consist of tissue other than water equivalent tissue (soft tissue, bones, lungs, etc.). The inhomogeneity effect can be investigated by Monte Carlo (MC) simulation. MC simulation of the radiation transport in an absorbing medium is the most accurate method for dose calculation in radiotherapy. The aim of this work is to evaluate the effect of inhomogeneity phantom on dose calculations in photon beam radiotherapy obtained by different MC codes. MC code system EGSnrc and Geant4 was used in this study. Inhomogeneity phantom dimension is 39.5 × 30.5 × 30 cm3 and made of 4 material slices (12.5 cm water, 10 cm aluminium, 5 cm lung and 12.5 cm water). Simulations were performed for field size 4 × 4 cm2 at SSD 100 cm. The spectrum distribution Varian Trilogy Clinac iX 10 MV was used. Percent depth dose (PDD) and dose profile was investigated in this research. The effects of inhomogeneities on radiation dose distributions depend on the amount, density and atomic number of the inhomogeneity, as well as on the quality of the photon beam. Good agreement between dose distribution from EGSnrc and Geant4 code system in inhomogeneity phantom was observed, with dose differences around 5% and 7% for depth doses and dose profiles.

A new pencil beam model for photon dose calculations in heterogeneous media

Purpose:Accurate and fast dose calculation is a prerequisite of precision radiation therapy in modern photon and particle therapy. While Monte Carlo (MC) dose calculation provides high dosimetric accuracy, the drastically increased computational time hinders its routine use. Deterministic dose calculation methods are fast, but problematic in the presence of tissue density inhomogeneity. We leverage the useful features of deterministic methods and MC to develop a hybrid dose calculation platform with autonomous utilization of MC and deterministic calculation depending on the local geometry, for optimal accuracy and speed.Methods:Our platform utilizes a Geant4 based “localized Monte Carlo” (LMC) method that isolates MC dose calculations only to volumes that have potential for dosimetric inaccuracy. In our approach, additional structures are created encompassing heterogeneous volumes. Deterministic methods calculate dose and energy fluence up to the volume surfaces, where the energy fluence distribution is sampled into discrete histories and transported using MC. Histories exiting the volume are converted back into energy fluence, and transported deterministically. By matching boundary conditions at both interfaces, deterministic dose calculation account for dose perturbations “downstream” of localized heterogeneities. Hybrid dose calculation was performed for water and anthropomorphic phantoms.Results:We achieved <1% agreement between deterministic and MC calculations in the water benchmark for photon and proton beams, and dose differences of 2%–15% could be observed in heterogeneous phantoms. The saving in computational time (a factor ∼4–7 compared to a full Monte Carlo dose calculation) was found to be approximately proportional to the volume of the heterogeneous region.Conclusion:Our hybrid dose calculation approach takes advantage of the computational efficiency of deterministic method and accuracy of MC, providing a practical tool for high performance dose calculation in modern RT. The approach is generalizable to all modalities where heterogeneities play a large role, notably particle therapy.

Purpose: The shortcomings of the pencil beam (PB) algorithm in regions of tissue inhomogeneity have been well documented, yet multiple commercially available treatment planning systems still employ versions of this algorithm (e.g. Varian® Eclipse®, Brainlab® BrainScan-iPlan, Best Nomos® Corvus®). The continuing value of the algorithm lies in its ability to quickly/efficiently calculate dose in the iterative inverse treatment planning environment. A novel heterogeneity inclusive PB dose calculation method capable of improving dose prediction in areas of lateral disequilibrium, such as lung, which does not compromise the time efficiency characteristics of the PB algorithm has been proposed. The new method is evaluated in anthropomorphic phantom in comparison to Monte Carlo (MC) calculation and measurement. Method and Materials: Alderson Rando and CIRS anthropomorphic phantoms are used with multiple embedded target sizes to measure dose via MOSFET and film. Measured values are compared to the Corvus '08 effective path length (EPL) correction method, to a prerelease version of the new heterogeneity inclusive dose calculation model which is capable of modeling lateral disequilibrium and to Peregrine® Monte Carlo. Results: In the anthropomorphic lung scenarios, Corvus '08 EPL calculation experienced errors ranging from 1.4% to 19%, depending on tumor size and beam configuration, while the new model agreed within 1.4% of measurement for all cases. The new model was seen to agree well with MC predicted dose, as well. Conclusion: The novel heterogeneity inclusive, pencil-beam-based dose calculation method demonstrated excellent agreement with both Monte Carlo prediction and anthropomorphic phantom measured dose. The algorithm appears capable of providing accurate calculation of absorbed dose in lung without compromising the advantage of constant time computational complexity for each pencil-beam to each point of interest. Conflict of Interest: Research support provided by Best NOMOS.

Purpose: To develop and test a hybrid method for optimizing IMRT plans using Monte Carlo(MC) dose calculation algorithms which retains the accuracy of MC while reducing the MC dose computational requirements. Method and Materials: A hybrid dose calculation strategy in which initial optimization is performed with a fast pencil beam (PB) algorithm using deliverable‐based IMRToptimization. Following convergence, doses are re‐computed with the VMC++ MC algorithm to determine correction factors for further PB‐based optimization. The correction/re‐optimization procedure is repeated until convergence. The hybrid method was benchmarked with respect to MC‐deliverable—based optimization for 5‐prostate IMRT plans. Figures of merit included number of MC dose computations required, final plan quality score, and optimization dose‐volume indices. Results: The hybrid method required a maximum of 3 MC dose calculations to converge to a result which provided equivalent dose coverage to the complete MC‐based optimization plan. The complete MC‐based optimization required between 6 and 9 MC dose computations to converge, depending on the specific patient. After 2 MC dose computations, the hybrid plan quality score was equal to or less than the MC‐based score for 4 of the 5 plans, the remaining plan required 3 iterations to achieve a score equal to that for the MC‐based optimization. Monitor units for the hybrid and complete MC‐optimization were within 5%. Conclusion:Hybrid PB‐MC‐IMRT dose calculation method is practical and results in plans equivalent to those achieved when MC‐dose calculation is used for all optimization iterations. The hybrid method reduced the number of MC calculations by a factor of ∼3, reducing overall optimization time by a factor of 2.8, and allowing for VMC++ MC‐based optimization to be completed in <30 min on a 20×2.4 Ghz CPU cluster. (Supported in part by NIH‐R01CA98524).