Y Graves

GPU-based fast Monte Carlo simulation for radiotherapy dose calculation

Monte Carlo (MC) simulation is commonly considered to be the most accurate dose calculation method in radiotherapy. However, its efficiency still requires improvement for many routine clinical applications. In this paper, we present our recent progress toward the development of a graphics processing unit (GPU)-based MC dose calculation package, gDPM v2.0. It utilizes the parallel computation ability of a GPU to achieve high efficiency, while maintaining the same particle transport physics as in the original dose planning method (DPM) code and hence the same level of simulation accuracy. In GPU computing, divergence of execution paths between threads can considerably reduce the efficiency. Since photons and electrons undergo different physics and hence attain different execution paths, we use a simulation scheme where photon transport and electron transport are separated to partially relieve the thread divergence issue. A high-performance random number generator and a hardware linear interpolation are also utilized. We have also developed various components to handle the fluence map and linac geometry, so that gDPM can be used to compute dose distributions for realistic IMRT or VMAT treatment plans. Our gDPM package is tested for its accuracy and efficiency in both phantoms and realistic patient cases. In all cases, the average relative uncertainties are less than 1%. A statistical t-test is performed and the dose difference between the CPU and the GPU results is not found to be statistically significant in over 96% of the high dose region and over 97% of the entire region. Speed-up factors of 69.1 ∼ 87.2 have been observed using an NVIDIA Tesla C2050 GPU card against a 2.27 GHz Intel Xeon CPU processor. For realistic IMRT and VMAT plans, MC dose calculation can be completed with less than 1% standard deviation in 36.1 ∼ 39.6 s using gDPM.

GPU-based Monte Carlo radiotherapy dose calculation using phase-space sources

A novel phase-space source implementation has been designed for graphics processing unit (GPU)-based Monte Carlo dose calculation engines. Short of full simulation of the linac head, using a phase-space source is the most accurate method to model a clinical radiation beam in dose calculations. However, in GPU-based Monte Carlo dose calculations where the computation efficiency is very high, the time required to read and process a large phase-space file becomes comparable to the particle transport time. Moreover, due to the parallelized nature of GPU hardware, it is essential to simultaneously transport particles of the same type and similar energies but separated spatially to yield a high efficiency. We present three methods for phase-space implementation that have been integrated into the most recent version of the GPU-based Monte Carlo radiotherapy dose calculation package gDPM v3.0. The first method is to sequentially read particles from a patient-dependent phase-space and sort them on-the-fly based on particle type and energy. The second method supplements this with a simple secondary collimator model and fluence map implementation so that patient-independent phase-space sources can be used. Finally, as the third method (called the phase-space-let, or PSL, method) we introduce a novel source implementation utilizing pre-processed patient-independent phase-spaces that are sorted by particle type, energy and position. Position bins located outside a rectangular region of interest enclosing the treatment field are ignored, substantially decreasing simulation time with little effect on the final dose distribution. The three methods were validated in absolute dose against BEAMnrc/DOSXYZnrc and compared using gamma-index tests (2%/2 mm above the 10% isodose). It was found that the PSL method has the optimal balance between accuracy and efficiency and thus is used as the default method in gDPM v3.0. Using the PSL method, open fields of 4 × 4, 10 × 10 and 30 × 30 cm(2) in water resulted in gamma passing rates of 99.96%, 99.92% and 98.66%, respectively. Relative output factors agreed within 1%. An intensity modulated radiation therapy patient plan using the PSL method resulted in a passing rate of 97%, and was calculated in 50 s (per GPU) compared to 8.4 h (per CPU) for BEAMnrc/DOSXYZnrc.

Comprehensive evaluations of cone-beam CT dose in image-guided radiation therapy via GPU-based Monte Carlo simulations.

Cone beam CT (CBCT) has been widely used for patient setup in image-guided radiation therapy (IGRT). Radiation dose from CBCT scans has become a clinical concern. The purposes of this study are (1) to commission a graphics processing unit (GPU)-based Monte Carlo (MC) dose calculation package gCTD for Varian On-Board Imaging (OBI) system and test the calculation accuracy, and (2) to quantitatively evaluate CBCT dose from the OBI system in typical IGRT scan protocols. We first conducted dose measurements in a water phantom. X-ray source model parameters used in gCTD are obtained through a commissioning process. gCTD accuracy is demonstrated by comparing calculations with measurements in water and in CTDI phantoms. Twenty-five brain cancer patients are used to study dose in a standard-dose head protocol, and 25 prostate cancer patients are used to study dose in pelvis protocol and pelvis spotlight protocol. Mean dose to each organ is calculated. Mean dose to 2% voxels that have the highest dose is also computed to quantify the maximum dose. It is found that the mean dose value to an organ varies largely among patients. Moreover, dose distribution is highly non-homogeneous inside an organ. The maximum dose is found to be 1-3 times higher than the mean dose depending on the organ, and is up to eight times higher for the entire body due to the very high dose region in bony structures. High computational efficiency has also been observed in our studies, such that MC dose calculation time is less than 5 min for a typical case.

Effect of statistical fluctuation in Monte Carlo based photon beam dose calculation on gamma index evaluation.

The γ-index test has been commonly adopted to quantify the degree of agreement between a reference dose distribution and an evaluation dose distribution. Monte Carlo (MC) simulation has been widely used for the radiotherapy dose calculation for both clinical and research purposes. The goal of this work is to investigate both theoretically and experimentally the impact of the MC statistical fluctuation on the γ-index test when the fluctuation exists in the reference, the evaluation, or both dose distributions. To the first order approximation, we theoretically demonstrated in a simplified model that the statistical fluctuation tends to overestimate γ-index values when existing in the reference dose distribution and underestimate γ-index values when existing in the evaluation dose distribution given the original γ-index is relatively large for the statistical fluctuation. Our numerical experiments using realistic clinical photon radiation therapy cases have shown that (1) when performing a γ-index test between an MC reference dose and a non-MC evaluation dose, the average γ-index is overestimated and the gamma passing rate decreases with the increase of the statistical noise level in the reference dose; (2) when performing a γ-index test between a non-MC reference dose and an MC evaluation dose, the average γ-index is underestimated when they are within the clinically relevant range and the gamma passing rate increases with the increase of the statistical noise level in the evaluation dose; (3) when performing a γ-index test between an MC reference dose and an MC evaluation dose, the gamma passing rate is overestimated due to the statistical noise in the evaluation dose and underestimated due to the statistical noise in the reference dose. We conclude that the γ-index test should be used with caution when comparing dose distributions computed with MC simulation.

Optimal surface marker locations for tumor motion estimation in lung cancer radiotherapy

Using fiducial markers on the patients body surface to predict the tumor location is a widely used approach in lung cancer radiotherapy. The purpose of this work is to propose an algorithm that automatically identifies a sparse set of locations on the patients surface with the optimal prediction power for the tumor motion. In our algorithm, it is assumed that there is a linear relationship between the surface marker motion and the tumor motion. The sparse selection of markers on the external surface and the linear relationship between the marker motion and the internal tumor motion are represented by a prediction matrix. Such a matrix is determined by solving an optimization problem, where the objective function contains a sparsity term that penalizes the number of markers chosen on the patients surface. Bregman iteration is used to solve the proposed optimization problem. The performance of our algorithm has been tested on realistic clinical data of four lung cancer patients. Thoracic 4DCT scans with ten phases are used for the study. On a reference phase, a grid of points are casted on the patients surfaces (except for the patients back) and propagated to other phases via deformable image registration of the corresponding CT images. Tumor locations at each phase are also manually delineated. We use nine out of ten phases of the 4DCT images to identify a small group of surface markers that are mostly correlated with the motion of the tumor and find the prediction matrix at the same time. The tenth phase is then used to test the accuracy of the prediction. It is found that on average six to seven surface markers are necessary to predict tumor locations with a 3D error of about 1 mm. It is also found that the selected marker locations lie closely in those areas where surface point motion has a large amplitude and a high correlation with the tumor motion. Our method can automatically select sparse locations on the patients external surface and estimate a correlation matrix based on 4DCT, so that the selected surface locations can be used to place fiducial markers to optimally predict internal tumor motions.

Automatic commissioning of a GPU-based Monte Carlo radiation dose calculation code for photon radiotherapy.

Monte Carlo (MC) simulation is commonly considered as the most accurate method for radiation dose calculations. Commissioning of a beam model in the MC code against a clinical linear accelerator beam is of crucial importance for its clinical implementation. In this paper, we propose an automatic commissioning method for our GPU-based MC dose engine, gDPM. gDPM utilizes a beam model based on a concept of phase-space-let (PSL). A PSL contains a group of particles that are of the same type and close in space and energy. A set of generic PSLs was generated by splitting a reference phase-space file. Each PSL was associated with a weighting factor, and in dose calculations the particle carried a weight corresponding to the PSL where it was from. Dose for each PSL in water was pre-computed, and hence the dose in water for a whole beam under a given set of PSL weighting factors was the weighted sum of the PSL doses. At the commissioning stage, an optimization problem was solved to adjust the PSL weights in order to minimize the difference between the calculated dose and measured one. Symmetry and smoothness regularizations were utilized to uniquely determine the solution. An augmented Lagrangian method was employed to solve the optimization problem. To validate our method, a phase-space file of a Varian TrueBeam 6 MV beam was used to generate the PSLs for 6 MV beams. In a simulation study, we commissioned a Siemens 6 MV beam on which a set of field-dependent phase-space files was available. The dose data of this desired beam for different open fields and a small off-axis open field were obtained by calculating doses using these phase-space files. The 3D γ-index test passing rate within the regions with dose above 10% of dmax dose for those open fields tested was improved averagely from 70.56 to 99.36% for 2%/2 mm criteria and from 32.22 to 89.65% for 1%/1 mm criteria. We also tested our commissioning method on a six-field head-and-neck cancer IMRT plan. The passing rate of the γ-index test within the 10% isodose line of the prescription dose was improved from 92.73 to 99.70% and from 82.16 to 96.73% for 2%/2 mm and 1%/1 mm criteria, respectively. Real clinical data measured from Varian, Siemens, and Elekta linear accelerators were also used to validate our commissioning method and a similar level of accuracy was achieved.

TH‐C‐137‐10: Development of a GPU Research Platform for Automatic Treatment Planning and Adaptive Radiotherapy Re‐Planning

PURPOSE To develop a research platform called SCORE (Super Computing Online Re-planning Environment) for automatic treatment planning and adaptive radiotherapy re-planning based on GPU. METHODS Our software is a Graphical User Interface (GUI) based on the Qt framework that allows users to easily and quickly create a new treatment plan based on a reference plan. It consists of several modules, including loading plan and patient geometry from DICOM RT files, automatic and manual rigid registration, deformable registration for contour propagation, previous plan based automatic plan optimization, physician-driven plan tuning, final dose calculation, and plan exporting in DICOM RT format. For automatic planning, a reference plan is identified from a library of previously delivered plans and it is used to guide the optimization process. For adaptive radiotherapy re-planning, the original plan of the same patient is used as the reference plan to guide the optimization process to generate a new plan on the new patient geometry defined by either a new CT or Cone-Beam CT image. All the computation modules have been implemented in CUDA to achieve high efficiency. The results for each step of the workflow can be visualized for review, revision, and approval. RESULTS SCORE has been well tested and validated for prostate and head/neck cancer cases. The validation was done by comparing SCORE plans against the same plans with re-calculated dose distributions and DVHs using a commercial planning system. We found that SCORE can generate clinically optimal treatment plans that are realistic and deliverable. The plans can be automatically created in only a few minutes, followed by another few minutes of physician fine-tuning using an interactive GUI. CONCLUSION We have developed a very efficient and user-friendly GPU-based research platform that can be used for clinical research on automatic treatment planning and adaptive radiotherapy re-planning.

MO-G-BRE-01: A Real-Time Virtual Delivery System for Photon Radiotherapy Delivery Monitoring

PURPOSE Treatment delivery monitoring is important for radiotherapy, which enables catching dosimetric error at the earliest possible opportunity. This project develops a virtual delivery system to monitor the dose delivery process of photon radiotherapy in real-time using GPU-based Monte Carlo (MC) method. METHODS The simulation process consists of 3 parallel CPU threads. A thread T1 is responsible for communication with a linac, which acquires a set of linac status parameters, e.g. gantry angles, MLC configurations, and beam MUs every 20 ms. Since linac vendors currently do not offer interface to acquire data in real time, we mimic this process by fetching information from a linac dynalog file at the set frequency. Instantaneous beam fluence map (FM) is calculated. A FM buffer is also created in T1 and the instantaneous FM is accumulated to it. This process continues, until a ready signal is received from thread T2 on which an inhouse developed MC dose engine executes on GPU. At that moment, the accumulated FM is transferred to T2 for dose calculations, and the FM buffer in T1 is cleared. Once the calculation finishes, the resulting 3D dose distribution is directed to thread T3, which displays it in three orthogonal planes overlaid on the CT image for treatment monitoring. This process continues to monitor the 3D dose distribution in real-time. RESULTS An IMRT and a VMAT cases used in our patient-specific QA are studied. Maximum dose differences between our system and treatment planning system are 0.98% and 1.58% for the two cases, respectively. The average time per MC calculation is 0.1sec with <2% relative uncertainty. The update frequency of ∼10Hz is considered as real time. CONCLUSION By embedding a GPU-based MC code in a novel data/work flow, it is possible to achieve real-time MC dose calculations to monitor delivery process.

WE‐E‐213CD‐07: Deformable Registration Between CT and Truncated CBCT for Adaptive Therapy Dose Calculation

PURPOSE Deformable image registration (DIR) is a crucial step in adaptive radiation therapy (ART) to deform the planning CT to the current CBCT for dose calculation and for contour propagation. CBCT images are sometimes truncated in the axial plane due to limited field of view (FOV). DIR between CT and truncated CBCT often leads to unphysical results, especially in and near missing regions, which may result in significant errors in dose calculation afterwards. The purpose of this work is to develop and evaluate a method to improve existing DIR algorithms to solve the truncation problem. METHODS The radius of FOV of CBCT is first estimated. A recently developed robust CT-CBCT DIR algorithm, called Deformation with Intensity Simultaneously Corrected (DISC), is used for deformation field calculation. At each iteration of DISC, the calculated deformation vector field outside the FOV is replaced by a smooth propagation of the deformation field inside the FOV. Six head-and-neck cancer cases and two prostate cancer cases are used for evaluation. Dose calculation is performed to test the impacts on the resulting dose distribution. RESULTS In terms of DIR accuracy, it is found that the average normalized mutual information (NMI), normalized cross correlation (NCC) and feature similarity index (FSIM) increase from 0.638, 0.948 and 0.917, to 0.641, 0.951 and 0.919, respectively, when compared with DISC without propagation. In terms of dose distribution, the relative L2 distance of dose (inside the FOV) between the ground truth and that calculated on deformed CT with propagation reduces from 9.25%to 1.41%, compared with those without propagation. CONCLUSIONS We have developed a deformation field propagation method for DIR to register the planning CT and the CBCT image with small FOV. Tests on head-and-neck and prostate cancer cases have demonstrated that our algorithm can generate more accurate registration results for dose calculation. This work is supported in part by the University of California Lab Fees Research Program, the Master Research Agreement from Varian Medical Systems, Inc., and the grants from the National Natural Science Foundation of China (No.30970866).

MO-FG-202-08: Real-Time Monte Carlo-Based Treatment Dose Reconstruction and Monitoring for Radiotherapy

PURPOSE This proof-of-concept study is to develop a real-time Monte Carlo (MC) based treatment-dose reconstruction and monitoring system for radiotherapy, especially for the treatments with complicated delivery, to catch treatment delivery errors at the earliest possible opportunity and interrupt the treatment only when an unacceptable dosimetric deviation from our expectation occurs. METHODS First an offline scheme is launched to pre-calculate the expected dose from the treatment plan, used as ground truth for real-time monitoring later. Then an online scheme with three concurrent threads is launched while treatment delivering, to reconstruct and monitor the patient dose in a temporally resolved fashion in real-time. Thread T1 acquires machine status every 20 ms to calculate and accumulate fluence map (FM). Once our accumulation threshold is reached, T1 transfers the FM to T2 for dose reconstruction ad starts to accumulate a new FM. A GPU-based MC dose calculation is performed on T2 when MC dose engine is ready and a new FM is available. The reconstructed instantaneous dose is directed to T3 for dose accumulation and real-time visualization. Multiple dose metrics (e.g. maximum and mean dose for targets and organs) are calculated from the current accumulated dose and compared with the pre-calculated expected values. Once the discrepancies go beyond our tolerance, an error message will be send to interrupt the treatment delivery. RESULTS A VMAT Head-and-neck patient case was used to test the performance of our system. Real-time machine status acquisition was simulated here. The differences between the actual dose metrics and the expected ones were 0.06%-0.36%, indicating an accurate delivery. ∼10Hz frequency of dose reconstruction and monitoring was achieved, with 287.94s online computation time compared to 287.84s treatment delivery time. CONCLUSION Our study has demonstrated the feasibility of computing a dose distribution in a temporally resolved fashion in real-time and quantitatively and dosimetrically monitoring the treatment delivery.