Inverse planning for photon and proton beams

Deep reinforcement learning in radiation therapy planning optimization: A comprehensive review

PURPOSE

The formulation and optimization of radiation therapy plans are complex and time-consuming processes that heavily rely on the expertise of medical physicists. Consequently, there is an urgent need for automated optimization methods. Recent advancements in reinforcement learning, particularly deep reinforcement learning (DRL), show great promise for automating radiotherapy planning. This review summarizes the current state of DRL applications in this field, evaluates their effectiveness, and identifies challenges and future directions.

METHODS

A systematic search was conducted in Google Scholar, PubMed, IEEE Xplore, and Scopus using keywords such as "deep reinforcement learning", "radiation therapy", and "treatment planning". The extracted data were synthesized for an overview and critical analysis.

RESULTS

The application of deep reinforcement learning in radiation therapy plan optimization can generally be divided into three categories: optimizing treatment planning parameters, directly optimizing machine parameters, and adaptive radiotherapy. From the perspective of disease sites, DRL has been applied to cervical cancer, prostate cancer, vestibular schwannoma, and lung cancer. Regarding types of radiation therapy, it has been used in HDRBT, IMRT, SBRT, VMAT, GK, and Cyberknife.

CONCLUSIONS

Deep reinforcement learning technology has played a significant role in advancing the automated optimization of radiation therapy plans. However, there is still a considerable gap before it can be widely applied in clinical settings due to three main reasons: inefficiency, limited methods for quality assessment, and poor interpretability. To address these challenges, significant research opportunities exist in the future, such as constructing evaluators, parallelized training, and exploring continuous action spaces.

Combined optimization of spot positions and weights for better FLASH proton therapy

Objective.In Intensity Modulated Proton Therapy (IMPT), the weights of individual pencil-beams or spots are optimized to fulfil dosimetric constraints. Theses spots are usually located on a regular lattice and their positions are fixed during optimization. In many cases, the range of spot weights may however be limited, leading sometimes to sub-optimal plan quality. An emblematic use case is the delivery of a plan at ultra-high dose rate (FLASH-RT), for which the spot weights are typically constrained to high values.Approach. To improve further the quality of IMPT FLASH plans, we propose here a novel algorithm to optimize both the spot weights and positions directly based on the objectives defined by the treatment planner.Main results. For all cases considered, optimizing the spot positions lead to an enhanced dosimetric score, while maintaining a high dose rate.Significance. Overall, this approach resulted in a substantial plan quality improvement compared to optimizing only the spot weights, and in a similar execution time.

Treatment planning of scanned proton beams in RayStation

The use of scanned proton beams in external beam radiation therapy has seen a rapid development over the past decade. This technique places new demands on treatment planning, as compared to conventional photon-based radiation therapy. In this article, several proton specific functions as implemented in the treatment planning system RayStation are presented. We will cover algorithms for energy layer and spot selection, basic optimization including the handling of spot weight limits, optimization of the linear energy transfer (LET) distribution, robust optimization including the special case of 4D optimization, proton arc planning, and automatic planning using deep learning. We will further present the Monte Carlo (MC) proton dose engine in RayStation to some detail, from the material interpretation of the CT data, through the beam model parameterization, to the actual MC transport mechanism. Useful tools for plan evaluation, including robustness evaluation, and the versatile scripting interface are also described. The overall aim of the paper is to give an overview of some of the key proton planning functions in RayStation, with example usages, and at the same time provide the details about the underlying algorithms that previously have not been fully publicly available.

Implementation and evaluation of an intelligent automatic treatment planning robot for prostate cancer stereotactic body radiation therapy

PURPOSE

We previously developed a virtual treatment planner (VTP), an artificial intelligence robot, operating a treatment planning system (TPS). Using deep reinforcement learning guided by human knowledge, we trained the VTP to autonomously adjust relevant parameters in treatment plan optimization, similar to a human planner, to generate high-quality plans for prostate cancer stereotactic body radiation therapy (SBRT). This study describes the clinical implementation and evaluation of VTP.

MATERIALS AND METHODS

We integrate VTP with Eclipse TPS using scripting Application Programming Interface. VTP observes dose-volume histograms of relevant structures, decides how to adjust dosimetric constraints, including doses, volumes, and weighting factors, and applies the adjustments to the TPS interface to launch the optimization engine. This process continues until a high-quality plan is achieved. We evaluated VTP's performance using the prostate SBRT case from the 2016 American Association of Medical Dosimetrist/Radiosurgery Society plan study with its plan scoring system, and compared to human-generated plans submitted to the challenge. Using the same scoring system, we also compared the plan quality of 36 prostate SBRT cases (20 planned with IMRT and 16 planned with VMAT) treated at our institution for both VTP and human-generated plans.

RESULTS

In the plan study case, VTP achieved a score of 142.1/150.0, ranking the third in the competition (median 134.6). For the clinical cases, VTP achieved 110.6 ± 6.5 for 20 IMRT plans and 126.2 ± 4.7 for 16 VMAT plans, similar to scores of human-generated plans with 110.4 ± 7.0 for IMRT plans and 125.4 ± 4.4 for VMAT plans. The workflow, plan quality and planning time of VTP were reviewed to be satisfactory by experienced physicists.

CONCLUSION

We successfully implemented VTP to operate a TPS for autonomous human-like treatment planning for prostate SBRT.

Treatment planning consideration for very high-energy electron FLASH radiotherapy

PURPOSE

Very high-energy electron (VHEE) can make up the insufficient treatment depth of the low-energy electron while offering an intermediate dosimetric advantage between photon and proton. Combining FLASH with VHEE, a quantitative comparison between different energies was made, with regard to plan quality, dose rate distribution (both in PTV and OAR), and total duration of treatment (beam-on time).

METHODS

In two patient cases (head and lung), we created the treatment plans utilizing the scanning pencil beam via the Monte Carlo simulation and a PTV-based optimization algorithm. Geant4 was used to simulate VHEE pencil beams and sizes of 0.3-5 mm defined by the full width at half maximum (FWHM). Monoenergetic beams with Gaussian distribution in x and y directions (ISOURC = 19) were used as the source of electrons. A large-scale non-linear solver (IPOPT) was used to calculate the optimal spot weights. After optimization, a quantitative comparison between different energies was made regarding treatment plan quality, dose rate distribution (both in PTV and OAR), and total beam duration.

RESULTS

For head (80 MeV, 100 MeV, and 120 MeV) and lung cases (100 MeV, 120 MeV, and 140 MeV), the minimum beam intensity needs to be ∼2.5 × 10¹¹ electrons/s and ∼9.375 × 10¹¹ electrons/s to allow > 90 % volume of PTV reaching the average dose rate (DADR) higher than 40 Gy/s. At this beam intensity (fraction dose: 10 Gy), the overall irradiation time for the head case is 5258.75 ms (80 MeV), 5149.75 ms (100 MeV), and 4976.75 ms (120 MeV), including scanning time 872.75 ms. For lung cases, this number is 1034.25 ms (100 MeV), 981.55 ms (120 MeV), and 928.15 ms (140 MeV), including scanning time 298.75 ms. The plan of higher energy always performs with a higher dose rate (both in PTV and OAR) and thereby costs less delivery time (beam-on time).

CONCLUSION

The study systematically investigated the currently known FLASH parameters for VHEE radiotherapy and successfully established a benchmark reference for its FLASH dose rate performance.

Bi-criteria Pareto optimization to balance irradiation time and dosimetric objectives in proton arc therapy

Objective. Proton arc therapy (PAT) is a new delivery technique that exploits the continuous rotation of the gantry to distribute the therapeutic dose over many angular windows instead of using a few static fields, as in conventional (intensity-modulated) proton therapy. Although coming along with many potential clinical and dosimetric benefits, PAT has also raised a new optimization challenge. In addition to the dosimetric goals, the beam delivery time (BDT) needs to be considered in the objective function. Considering this bi-objective formulation, the task of finding a good compromise with appropriate weighting factors can turn out to be cumbersome.Approach. We have computed Pareto-optimal plans for three disease sites: a brain, a lung, and a liver, following a method of iteratively choosing weight vectors to approximate the Pareto front with few points. Mixed-integer programming (MIP) was selected to state the bi-criteria PAT problem and to find Pareto optimal points with a suited solver.Main results. The trade-offs between plan quality and beam irradiation time (staticBDT) are investigated by inspecting three plans from the Pareto front. The latter are carefully picked to demonstrate significant differences in dose distribution and delivery time depending on their location on the frontier. The results were benchmarked against IMPT and SPArc plans showing the strength of degrees of freedom coming along with MIP optimization.Significance. This paper presents for the first time the application of bi-criteria optimization to the PAT problem, which eventually permits the planners to select the best treatment strategy according to the patient conditions and clinical resources available.

SWFT-Net: a deep learning framework for efficient fine-tuning spot weights towards adaptive proton therapy

Objective. One critical task for adaptive proton therapy is how to perform spot weight re-tuning and reoptimize plan, both of which are time-consuming and labor intensive. We proposed a deep learning framework (SWFT-Net) to speed up such a task, a starting point for us to move towards online adaptive proton therapy.Approach. For a H&N patient case, a reference intensity modulated proton therapy plan was generated. For data augmentation, spot weights were modified to generate three datasets (DS10, DS30, DS50), corresponding to different levels of weight adjustment. For each dataset, the samples were split into the training and testing groups at a ratio of 8:2 (6400 for training, 1706 for testing). To ease the difficulty of machine learning, the residuals of dose maps and spot weights (i.e. difference relative to a reference) were used as inputs and outputs, respectively. Quantitative analyses were performed in terms of normalized root mean square error (NRMSE) of spot weights, Gamma passing rate and dose difference within the PTV.Main results. The SWFT-Net is able to generate an adapted plan in less than a second with a NVIDIA GeForce RTX 3090 GPU. For the 1706 samples in the testing dataset, the NRMSE is 0.41% (DS10), 1.05% (DS30) and 2.04% (DS50), respectively. Cold/hot spots in the dose maps after adaptation are observed. The mean relative dose difference is 0.64% (DS10), 0.92% (DS30) and 0.88% (DS50), respectively. For all three datasets, the mean Gamma passing rate is consistently over 95% for both 1 mm/1% and 3 mm/3% settings.Significance. The proposed SWFT-Net is a promising tool to help realize adaptive proton therapy. It can be used as an alternative tool to other spot fine-tuning optimization algorithms, likely demonstrating superior performance in terms of speed, accuracy, robustness and minimum human interaction. This study lays down a foundation for us to move further incorporating other factors such as daily anatomical changes and propagated PTVs, and develop a truly online adaptive workflow in proton therapy.

Design of static and dynamic ridge filters for FLASH-IMPT: a simulation study

PURPOSE

This paper focused on the design and optimization of ridge filter-based intensity-modulated proton therapy (IMPT), and its potential applications for FLASH. Differing from the standard pencil beam scanning (PBS) mode, no energy/layer switching is required and total treatment time can be shortened.

METHODS

Unique dose influence matrices were generated as a proton beam traverse through slabs of different thicknesses (i.e. modulation by different layers). To establish the references for comparison, conventional IMPT plans (single field) were created using a large-scale non-linear solver. The spot weights from the reference IMPT plans were used as inputs for optimizing the design of ridge filters. Two designs were evaluated: model A (static) and model B (dynamic). The ridge filters designs were first verified (by GEANT4 simulation) in a water phantom and then in a H&N case. Direct comparison was made between the GEANT4 simulation results of two models and their respective references, with regard to plan quality, dose-averaged dose rate (DADR), and total treatment time.

RESULTS

In both the water phantom and the H&N case, two models are able to modulate dose distributions with high conformity, showing no significant difference relative to the reference plans. Dose rate volume histograms (DRVHs) suggest that in order to achieve a dose rate of 40 Gy/s over 90% PTV, the beam intensity needs to be 2.5×10¹¹ protons/s for both models. For a fraction dose of 10 Gy, the total treatment time (including both irradiation time and dead time) can be shortened by a factor of 4.9 (model A) and 6.5 (model B), relative to the reference plans.

CONCLUSION

Two proposed designs (both static and dynamic) can be used for PBS-IMPT requiring no layer switching. They are promising candidates for FLASH-IMPT capable of reducing treatment time and achieving high dose rates, while maintaining dose conformity simultaneously. This article is protected by copyright. All rights reserved.

Site-agnostic 3D dose distribution prediction with deep learning neural networks

PURPOSE

Typically, the current dose prediction models are limited to small amounts of data and require retraining for a specific site, often leading to suboptimal performance. We propose a site-agnostic, three-dimensional dose distribution prediction model using deep learning that can leverage data from any treatment site, thus increasing the total data available to train the model. Applying our proposed model to a new target treatment site requires only a brief fine-tuning of the model to the new data and involves no modifications to the model input channels or its parameters. Thus, it can be efficiently adapted to a different treatment site, even with a small training dataset.

METHODS

This study uses two separate datasets/treatment sites: data from patients with prostate cancer treated with intensity-modulated radiation therapy (source data), and data from patients with head-and-neck cancer treated with volumetric-modulated arc therapy (target data). We first developed a source model with 3D UNet architecture, trained from random initial weights on the source data. We evaluated the performance of this model on the source data. We then studied the generalizability of the model to the new target dataset via transfer learning. To do this, we built three more models, all with the same 3D UNet architecture: target model, adapted model, and combined model. The source and target models were trained on the source and target data from random initial weights, respectively. The adapted model fine-tuned the source model to the target domain by using the target data. Finally, the combined model was trained from random initial weights on a combined data pool consisting of both target and source datasets. We tested all four models on the target dataset and evaluated quantitative dose-volume histogram metrics for the planning target volume (PTV) and organs at risk (OARs).

RESULTS

When tested on the source treatment site, the source model accurately predicted the dose distributions with average (mean, max) absolute dose errors of (0.32%±0.14, 2.37%±0.93) (PTV) relative to the prescription dose, and highest mean dose error of 1.68%±0.76, and highest max dose error of 5.47%± 3.31 for femoral head right. The error in PTV dose coverage prediction is 3.21%±1.51 for D₉₈ , 3.04%±1.69 for D₉₅ , and 1.83%±1.01 for D₀₂ . Averaging across all OARs, the source model predicted the OAR mean dose within 1.38% and the OAR max dose within 3.64%. For the target treatment site, the target model average (mean, max) absolute dose errors relative to the prescription dose for the PTV were (1.08%±0.95, 2.90%±1.35). Left cochlea had the highest mean and max dose errors of 5.37%±5.82 and 8.33%±8.88, respectively. The errors in PTV dose coverage prediction for D₉₈ and D₉₅ were 2.88%±1.59 and 2.55%±1.28, respectively. The target model can predict the OAR mean dose within 2.43% and the OAR max dose within 4.33% on average across all OARs.

CONCLUSION

We developed a site-agnostic model for three-dimensional dose prediction and tested its adaptability to a new target treatment site via transfer learning. Our proposed model can make accurate predictions with limited training data.

Report of AAPM Task Group 290: Respiratory motion management for particle therapy

Dose uncertainty induced by respiratory motion remains a major concern for treating thoracic and abdominal lesions using particle beams. This Task Group report reviews the impact of tumor motion and dosimetric considerations in particle radiotherapy, current motion‐management techniques, and limitations for different particle‐beam delivery modes (i.e., passive scattering, uniform scanning, and pencil‐beam scanning). Furthermore, the report provides guidance and risk analysis for quality assurance of the motion‐management procedures to ensure consistency and accuracy, and discusses future development and emerging motion‐management strategies. This report supplements previously published AAPM report TG76, and considers aspects of motion management that are crucial to the accurate and safe delivery of particle‐beam therapy. To that end, this report produces general recommendations for commissioning and facility‐specific dosimetric characterization, motion assessment, treatment planning, active and passive motion‐management techniques, image guidance and related decision‐making, monitoring throughout therapy, and recommendations for vendors. Key among these recommendations are that: (1) facilities should perform thorough planning studies (using retrospective data) and develop standard operating procedures that address all aspects of therapy for any treatment site involving respiratory motion; (2) a risk‐based methodology should be adopted for quality management and ongoing process improvement.

A reinforcement learning application of a guided Monte Carlo Tree Search algorithm for beam orientation selection in radiation therapy

Current beam orientation optimization algorithms for radiotherapy, such as column generation (CG), are typically heuristic or greedy in nature because of the size of the combinatorial problem, which leads to suboptimal solutions. We propose a reinforcement learning strategy using a Monte Carlo Tree Search (MCTS) that can find a better beam orientation set in less time than CG. We utilize a reinforcement learning structure involving a supervised learning network to guide the MCTS and to explore the decision space of beam orientation selection problems. We previously trained a deep neural network (DNN) that takes in the patient anatomy, organ weights, and current beams, then approximates beam fitness values to indicate the next best beam to add. Here, we use this DNN to probabilistically guide the traversal of the branches of the Monte Carlo decision tree to add a new beam to the plan. To assess the feasibility of the algorithm, we used a test set of 13 prostate cancer patients, distinct from the 57 patients originally used to train and validate the DNN, to solve five-beam plans. To show the strength of the guided MCTS (GTS) compared to other search methods, we also provided the performances of Guided Search, Uniform Tree Search and Random Search algorithms. On average, GTS outperformed all the other methods. It found a better solution than CG in 237 s on average, compared to 360 s for CG, and outperformed all other methods in finding a solution with a lower objective function value in less than 1000 s. Using our GTS method, we could maintain planning target volume (PTV) coverage within 1% error similar to CG, while reducing the organ-at-risk mean dose for body, rectum, left and right femoral heads; the mean dose to bladder was 1% higher with GTS than with CG.

A hierarchical deep reinforcement learning framework for intelligent automatic treatment planning of prostate cancer intensity modulated radiation therapy

Purpose.We have previously proposed an intelligent automatic treatment planning (IATP) framework that builds a virtual treatment planner network (VTPN) to operate a treatment planning system (TPS) to generate high-quality radiation therapy (RT) treatment plans. While the potential of IATP in automating RT treatment planning has been demonstrated, its poor scalability caused by an almost linear growth of network size with the number of treatment planning parameters (TPPs) is a bottleneck, preventing its application in complicate, but clinically relevant treatment planning problems. The decision-making behavior of the trained network is hard to understand. Motivated by the decision-making process of a human planner, this study proposes a hierarchical IATP framework.Methods and materials.The hierarchical VTPN (HieVTPN) consists of three networks, i.e. Structure-Net, Parameter-Net, and Action-Net. When interacting with a TPS, the networks are employed in a sequential order in each step to decide the structure to adjust, the TPP to adjust for the selected structure, and the specific adjustment manner for the parameter, respectively. We developed an end-to-end hierarchical deep reinforcement learning scheme to simultaneously train the three networks. We then evaluated the effectiveness of the proposed framework in the treatment planning problems for prostate cancer intensity modulated RT (IMRT) and stereotactic body RT (SBRT). We benchmarked the performance of our approach by comparing plans made by VTPN of a parallel architecture, and the human plans submitted for competition in the 2016 American Association of Medical Dosimetrist (AAMD)/Radiosurgery Society (RSS) Plan Study. We analyzed scalability of the network size with respect to the number of TPPs. Numerical experiments were also performed to understand the rationale of the decision-making behaviors of the trained HieVTPN.Results.Both HieVTPNs for prostate IMRT and SBRT were trained successfully using 10 training patient cases and 5 validation cases. For IMRT, HieVTPN was able to generate high-quality plans for 59 testing patient cases that were not included in training process, achieving an average plan score of 8.62 (±0.83), with 9 being the maximal score. The score was comparable to that of the VTPN, 8.45 (±0.48). For SBRT planning, HieVTPN achieved an average plan score of 139.07 on five testing patient cases compared to the score of 132.21 averaged over the human plans summited for competition in AAMD/RSS plan study. Different from VTPN with network size linearly scaling with the number of TPPs, the network size of HieVTPN is almost independent of the number of TPPs. It was also observed that the decision-making behaviors of HieVTPN were understandable and generally agreed with the human experience.Conclusions.With the scalability and explainability, the hierarchical IATP framework is more favorable than the previous framework in terms of handling treatment planning problems involving a large number of TPPs.

Development of robustness evaluation strategies for enabling statistically consistent reporting

Robustness evaluation of proton therapy treatment plans is essential for ensuring safe treatment delivery. However, available evaluation procedures feature a limited exploration of the actual robustness of the plan and generally do not provide confidence levels. This study compared established and more sophisticated robustness evaluation procedures, with quantified confidence levels. We have evaluated several robustness evaluation methods for 5 bilateral head-and-neck patients optimized considering spot scanning delivery and with a conventional CTV-to-PTV margin of 4 mm. Method (1) good practice scenario selection (GPSS) (e.g. +/- 4 mm setup error 3% range uncertainty); (2) statistically sound scenario selection (SSSS) either only on or both on and inside isoprobability hypersurface encompassing 90% of the possible errors; (3) statistically sound dosimetric selection (SSDS). In the last method, the 90% best plans were selected according to either target coverage quantified by D 95 (SSDS_D 95) or to an approximation of the final objective function (OF) used during treatment optimization (SSDS_OF). For all methods, we have considered systematic setup and systematic range errors. A mix of systematic and random setup errors were also simulated for SSDS, but keeping the same conventional margin of 4 mm. All robustness evaluations have been performed using the fast Monte Carlo dose engine MCsquare. Both SSSS strategies yielded on average very similar results. SSSS and GPSS yield comparable values for target coverage (within 0.5 Gy). The most noticeable differences were found for the CTV between GPSS, on the one hand, and SSDS_D 95 and SSDS_OF, on the other hand (average worst-case D 98 were 2.8 and 2.0 Gy larger than for GPSS, respectively). Simulating explicitly random errors in SSDS improved almost all DVH metrics. We have observed that the width of DVH-bands and the confidence levels depend on the method chosen to sample the scenarios. Statistically sound estimation of the robustness of the plan in the dosimetric space may provide an improved insight on the actual robustness of the plan for a given confidence level.

A biological effect-guided optimization approach using beam distal-edge avoidance for intensity-modulated proton therapy

PURPOSE

Linear energy transfer (LET)-guided methods have been applied to intensity-modulated proton therapy (IMPT) to improve its biological effect. However, using LET as a surrogate for biological effect ignores the topological relationship of the scanning spot to different structures of interest. In this study, we developed an optimization method that takes advantage of the continuing increase in LET beyond the physical dose Bragg peak. This method avoids placing high biological effect values in critical structures and increases biological effect in the tumor area without compromising target coverage.

METHODS

We selected the cases of two patients with brain tumors and two patients with head and neck tumors who had been treated with proton therapy at our institution. Three plans were created for each case: a plan based on conventional dose-based optimization (DoseOpt), one based on LET-incorporating optimization (LETOpt), and one based on the proposed distal-edge avoidance-guided optimization method (DEAOpt). In DEAOpt, an L₁ -norm sparsity term, in which the penalty of each scanning spot was set according to the topological relationship between the organ positions and the location of the peak scaled LET-weighted dose (c LETxD) was added to a conventional dose-based optimization objective function. All plans were normalized to give the same target dose coverage. Dose (assuming a constant relative biological effectiveness value of 1.1, as in clinical practice), biological effect (c LETxD), and computing time consumption were evaluated and compared among the three optimization approaches for each patient case.

RESULTS

For all four cases, all three optimization methods generated comparable dose coverage in both target and critical structures. The LETOpt plans and DEAOpt plans reduced biological effect hot spots in critical structures and increased biological effect in the target volumes to a similar extent. For the target, the c LETxD_98% and c LETxD_2% in the DEAOpt plans were on average 7.2% and 11.74% higher than in the DoseOpt plans, respectively. For the brainstem, the c LETxD_mean in the DEAOpt plans was on average 33.38% lower than in the DoseOpt plans. In addition, the DEAOpt method saved 30.37% of the computation cost over the LETOpt method.

CONCLUSIONS

DEAOpt is an alternative IMPT optimization approach that correlates the location of scanning spots with biological effect distribution. IMPT could benefit from the use of DEAOpt because this method not only delivers comparable biological effects to LETOpt plans, but also is faster.

Operating a treatment planning system using a deep-reinforcement learning-based virtual treatment planner for prostate cancer intensity-modulated radiation therapy treatment planning

PURPOSE

In the treatment planning process of intensity-modulated radiation therapy (IMRT), a human planner operates the treatment planning system (TPS) to adjust treatment planning parameters, for example, dose volume histogram (DVH) constraints' locations and weights, to achieve a satisfactory plan for each patient. This process is usually time-consuming, and the plan quality depends on planer's experience and available planning time. In this study, we proposed to model the behaviors of human planners in treatment planning by a deep reinforcement learning (DRL)-based virtual treatment planner network (VTPN), such that it can operate the TPS in a human-like manner for treatment planning.

METHODS AND MATERIALS

Using prostate cancer IMRT as an example, we established the VTPN using a deep neural network developed. We considered an in-house optimization engine with a weighted quadratic objective function. Virtual treatment planner network was designed to observe an intermediate plan DVHs and decide the action to improve the plan by changing weights and threshold dose in the objective function. We trained the VTPN in an end-to-end DRL process in 10 patient cases. A plan score was used to measure plan quality. We demonstrated the feasibility and effectiveness of the trained VTPN in another 64 patient cases.

RESULTS

Virtual treatment planner network was trained to spontaneously learn how to adjust treatment planning parameters to generate high-quality treatment plans. In the 64 testing cases, with initialized parameters, quality score was 4.97 (±2.02), with 9.0 being the highest possible score. Using VTPN to perform treatment planning improved quality score to 8.44 (±0.48).

CONCLUSIONS

To our knowledge, this was the first time that intelligent treatment planning behaviors of human planner in external beam IMRT are autonomously encoded in an artificial intelligence system. The trained VTPN is capable of behaving in a human-like way to produce high-quality plans.

A fast robust optimizer for intensity modulated proton therapy using GPU

Robust optimization has been shown to be effective for stabilizing treatment planning in intensity modulated proton therapy (IMPT), but existing algorithms for the optimization process is time‐consuming. This paper describes a fast robust optimization tool that takes advantage of the GPU parallel computing technologies. The new robust optimization model is based on nine boundary dose distributions — two for ±range uncertainties, six for ±set‐up uncertainties along anteroposterior (A‐P), lateral (R‐L) and superior‐inferior (S‐I) directions, and one for nominal situation. The nine boundary influence matrices were calculated using an in‐house finite size pencil beam dose engine, while the conjugate gradient method was applied to minimize the objective function. The proton dose calculation algorithm and the conjugate gradient method were tuned for heterogeneous platforms involving the CPU host and GPU device. Three clinical cases — one head and neck cancer case, one lung cancer case, and one prostate cancer case — were investigated to demonstrate the clinical feasibility of the proposed robust optimizer. Compared with results from Varian Eclipse (version 13.3), the proposed method is found to be conducive to robust treatment planning that is less sensitive to range and setup uncertainties. The three tested cases show that targets can achieve high dose uniformity while organs at risks (OARs) are in better protection against setup and range errors. Based on the CPU + GPU heterogeneous platform, the execution times of the head and neck cancer case and the prostate cancer case are much less than half of Eclipse, while the run time of the lung cancer case is similar to that of Eclipse. The fast robust optimizer developed in this study can improve the reliability of traditional proton treatment planning in a much faster speed, thus making it possible for clinical utility.

A fast deep learning approach for beam orientation optimization for prostate cancer treated with intensity-modulated radiation therapy

PURPOSE

Beam orientation selection, whether manual or protocol-based, is the current clinical standard in radiation therapy treatment planning, but it is tedious and can yield suboptimal results. Many algorithms have been designed to optimize beam orientation selection because of its impact on treatment plan quality, but these algorithms suffer from slow calculation of the dose influence matrices of all candidate beams. We propose a fast beam orientation selection method, based on deep learning neural networks (DNN), capable of developing a plan comparable to those developed by the state-of-the-art column generation (CG) method. Our model's novelty lies in its supervised learning structure (using CG to teach the network), DNN architecture, and ability to learn from anatomical features to predict dosimetrically suitable beam orientations without using dosimetric information from the candidate beams. This may save hours of computation.

METHODS

A supervised DNN is trained to mimic the CG algorithm, which iteratively chooses beam orientations one-by-one by calculating beam fitness values based on Karush-Kush-Tucker optimality conditions at each iteration. The DNN learns to predict these values. The dataset contains 70 prostate cancer patients - 50 training, 7 validation, and 13 test patients - to develop and test the model. Each patient's data contains 6 contours: PTV, body, bladder, rectum, and left and right femoral heads. Column generation was implemented with a GPU-based Chambolle-Pock algorithm, a first-order primal-dual proximal-class algorithm, to create 6270 plans. The DNN trained over 400 epochs, each with 2500 steps and a batch size of 1, using the Adam optimizer at a learning rate of 1 × 10-5 and a sixfold cross-validation technique.

RESULTS

The average and standard deviation of training, validation, and testing loss functions among the six folds were 0.62 ± 0.09%, 1.04 ± 0.06%, and 1.44 ± 0.11%, respectively. Using CG and supervised DNN, we generated two sets of plans for each scenario in the test set. The proposed method took at most 1.5 s to select a set of five beam orientations and 300 s to calculate the dose influence matrices for 5 beams and finally 20 s to solve the fluence map optimization (FMO). However, CG needed around 15 h to calculate the dose influence matrices of all beams and at least 400 s to solve both the beam orientation selection and FMO problems. The differences in the dose coverage of PTV between plans generated by CG and by DNN were 0.2%. The average dose differences received by organs at risk were between 1 and 6 percent: Bladder had the smallest average difference in dose received (0.956 ± 1.184%), then Rectum (2.44 ± 2.11%), Left Femoral Head (6.03 ± 5.86%), and Right Femoral Head (5.885 ± 5.515%). The dose received by Body had an average difference of 0.10 ± 0.1% between the generated treatment plans.

CONCLUSIONS

We developed a fast beam orientation selection method based on a DNN that selects beam orientations in seconds and is therefore suitable for clinical routines. In the training phase of the proposed method, the model learns the suitable beam orientations based on patients' anatomical features and omits time intensive calculations of dose influence matrices for all possible candidate beams. Solving the FMO to get the final treatment plan requires calculating dose influence matrices only for the selected beams.

Intelligent inverse treatment planning via deep reinforcement learning, a proof-of-principle study in high dose-rate brachytherapy for cervical cancer

Inverse treatment planning in radiation therapy is formulated as solving optimization problems. The objective function and constraints consist of multiple terms designed for different clinical and practical considerations. Weighting factors of these terms are needed to define the optimization problem. While a treatment planning optimization engine can solve the optimization problem with given weights, adjusting the weights to yield a high-quality plan is typically performed by a human planner. Yet the weight-tuning task is labor intensive, time consuming, and it critically affects the final plan quality. An automatic weight-tuning approach is strongly desired. The procedure of weight adjustment to improve the plan quality is essentially a decision-making problem. Motivated by the tremendous success in deep learning for decision making with human-level intelligence, we propose a novel framework to adjust the weights in a human-like manner. This study used inverse treatment planning in high-dose-rate brachytherapy (HDRBT) for cervical cancer as an example. We developed a weight-tuning policy network (WTPN) that observes dose volume histograms of a plan and outputs an action to adjust organ weighting factors, similar to the behaviors of a human planner. We trained the WTPN via end-to-end deep reinforcement learning. Experience replay was performed with the epsilon greedy algorithm. After training was completed, we applied the trained WTPN to guide treatment planning of five testing patient cases. It was found that the trained WTPN successfully learnt the treatment planning goals and was able to guide the weight tuning process. On average, the quality score of plans generated under the WTPN's guidance was improved by ~8.5% compared to the initial plan with arbitrarily set weights, and by 10.7% compared to the plans generated by human planners. To our knowledge, this was the first time that a tool was developed to adjust organ weights for the treatment planning optimization problem in a human-like fashion based on intelligence learnt from a training process, which was different from existing strategies based on pre-defined rules. The study demonstrated potential feasibility to develop intelligent treatment planning approaches via deep reinforcement learning.

A convex optimization approach to radiation treatment planning with dose constraints

We present a method for handling dose constraints as part of a convex programming framework for inverse treatment planning. Our method uniformly handles mean dose, maximum dose, minimum dose, and dose-volume (i.e., percentile) constraints as part of a convex formulation. Since dose-volume constraints are non-convex, we replace them with a convex restriction. This restriction is, by definition, conservative; to mitigate its impact on the clinical objectives, we develop a two-pass planning algorithm that allows each dose-volume constraint to be met exactly on a second pass by the solver if its corresponding restriction is feasible on the first pass. In another variant, we add slack variables to each dose constraint to prevent the problem from becoming infeasible when the user specifies an incompatible set of constraints or when the constraints are made infeasible by our restriction. Finally, we introduce ConRad, a Python-embedded open-source software package for convex radiation treatment planning. ConRad implements the methods described above and allows users to construct and plan cases through a simple interface.

Robust optimization to reduce the impact of biological effect variation from physical uncertainties in intensity-modulated proton therapy

Robust optimization (RO) methods are applied to intensity-modulated proton therapy (IMPT) treatment plans to ensure their robustness in the face of treatment delivery uncertainties, such as proton range and patient setup errors. However, the impact of those uncertainties on the biological effect of protons has not been specifically considered. In this study, we added biological effect-based objectives into a conventional RO cost function for IMPT optimization to minimize the variation in biological effect. One brain tumor case, one prostate tumor case and one head & neck tumor case were selected for this study. Three plans were generated for each case using three different optimization approaches: planning target volume (PTV)-based optimization, conventional RO, and RO incorporating biological effect (BioRO). In BioRO, the variation in biological effect caused by IMPT delivery uncertainties was minimized for voxels in both target volumes and critical structures, in addition to a conventional voxel-based worst-case RO objective function. The biological effect was approximated by the product of dose-averaged linear energy transfer (LET) and physical dose. All plans were normalized to give the same target dose coverage, assuming a constant relative biological effectiveness (RBE) of 1.1. Dose, biological effect, and their uncertainties were evaluated and compared among the three optimization approaches for each patient case. Compared with PTV-based plans, RO plans achieved more robust target dose coverage and reduced biological effect hot spots in critical structures near the target. Moreover, with their sustained robust dose distributions, BioRO plans not only reduced variations in biological effect in target and normal tissues but also further reduced biological effect hot spots in critical structures compared with RO plans. Our findings indicate that IMPT could benefit from the use of conventional RO, which would reduce the biological effect in normal tissues and produce more robust dose distributions than those of PTV-based optimization. More importantly, this study provides a proof of concept that incorporating biological effect uncertainty gap into conventional RO would not only control the IMPT plan robustness in terms of physical dose and biological effect but also achieve further reduction of biological effect in normal tissues.

Robust Proton Treatment Planning: Physical and Biological Optimization

Accurate prediction of tumor control and toxicities in radiation therapy faces many uncertainties. Besides interpatient variability in the response to radiation, there are also dosimetric uncertainties, that is, differences between the dose displayed in a treatment planning system and the dose actually delivered to the patient. These uncertainties originate from several sources including imperfect knowledge of the patient geometry, approximation in the physics of radiation interaction with tissues, and uncertainties in the biological effectiveness of radiation. Generally, uncertainties are considered in the treatment planning process by applying margins. In intensity-modulated radiotherapy (IMRT), this leads to the planning target volume (PTV) concept. Intensity-modulated proton therapy (IMPT) is widely considered as the future of proton therapy. The treatment planning methods for IMPT and IMRT are similar and based on mathematical optimization techniques for both modalities. However, the PTV concept has fundamental limitations in IMPT. Therefore, researchers have developed robust optimization methods that directly incorporate uncertainties into the IMPT optimization problem. In recent years, vendors of commercial planning systems have started to implement these methods so that robust IMPT planning becomes available in clinical practice. This article summarizes uncertainties in proton therapy and the limitations of the PTV concept to deal with them. Subsequently, robust optimization techniques to overcome these limitations are reviewed.

The Scatter Search Based Algorithm for Beam Angle Optimization in Intensity-Modulated Radiation Therapy

This article introduces a new framework for beam angle optimization (BAO) in intensity-modulated radiation therapy (IMRT) using the Scatter Search Based Algorithm. The potential benefits of plans employing the coplanar optimized beam sets are also examined. In the proposed beam angle selection algorithm, the problem is solved in two steps. Initially, the gantry angles are selected using the Scatter Search Based Algorithm, which is a global optimization method. Then, for each beam configuration, the intensity profile is calculated by the conjugate gradient method to score each beam angle set chosen. A simulated phantom case with obvious optimal beam angles was used to benchmark the validity of the presented algorithm. Two clinical cases (TG-119 phantom and prostate cases) were examined to prepare a dose volume histogram (DVH) and determine the dose distribution to evaluate efficiency of the algorithm. A clinical plan with the optimized beam configuration was compared with an equiangular plan to determine the efficiency of the proposed algorithm. The BAO plans yielded significant improvements in the DVHs and dose distributions compared to the equispaced coplanar beams for each case. The proposed algorithm showed its potential to effectively select the beam direction for IMRT inverse planning at different tumor sites.

Optimization of treatment planning workflow and tumor coverage during daily adaptive magnetic resonance image guided radiation therapy (MR-IGRT) of pancreatic cancer

BACKGROUND

To simplify the adaptive treatment planning workflow while achieving the optimal tumor-dose coverage in pancreatic cancer patients undergoing daily adaptive magnetic resonance image guided radiation therapy (MR-IGRT).

METHODS

In daily adaptive MR-IGRT, the plan objective function constructed during simulation is used for plan re-optimization throughout the course of treatment. In this study, we have constructed the initial objective functions using two methods for 16 pancreatic cancer patients treated with the ViewRay™ MR-IGRT system: 1) the conventional method that handles the stomach, duodenum, small bowel, and large bowel as separate organs at risk (OARs) and 2) the OAR grouping method. Using OAR grouping, a combined OAR structure that encompasses the portions of these four primary OARs within 3 cm of the planning target volume (PTV) is created. OAR grouping simulation plans were optimized such that the target coverage was comparable to the clinical simulation plan constructed in the conventional manner. In both cases, the initial objective function was then applied to each successive treatment fraction and the plan was re-optimized based on the patient's daily anatomy. OAR grouping plans were compared to conventional plans at each fraction in terms of coverage of the PTV and the optimized PTV (PTV OPT), which is the result of the subtraction of overlapping OAR volumes with an additional margin from the PTV.

RESULTS

Plan performance was enhanced across a majority of fractions using OAR grouping. The percentage of the volume of the PTV covered by 95% of the prescribed dose (D₉₅) was improved by an average of 3.87 ± 4.29% while D₉₅ coverage of the PTV OPT increased by 3.98 ± 4.97%. Finally, D₁₀₀ coverage of the PTV demonstrated an average increase of 6.47 ± 7.16% and a maximum improvement of 20.19%.

CONCLUSIONS

In this study, our proposed OAR grouping plans generally outperformed conventional plans, especially when the conventional simulation plan favored or disregarded an OAR through the assignment of distinct weighting parameters relative to the other critical structures. OAR grouping simplifies the MR-IGRT adaptive treatment planning workflow at simulation while demonstrating improved coverage compared to delivered pancreatic cancer treatment plans in daily adaptive radiation therapy.

Performance of a hybrid Monte Carlo-Pencil Beam dose algorithm for proton therapy inverse planning

PURPOSE

Analytical algorithms have a limited accuracy when modeling very heterogeneous tumor sites. This work addresses the performance of a hybrid dose optimizer that combines both Monte Carlo (MC) and pencil beam (PB) dose engines to get the best trade-off between speed and accuracy for proton therapy plans.

METHODS

The hybrid algorithm calculates the optimal spot weights (w) by means of an iterative optimization process where the dose at each iteration is computed by using a precomputed dose influence matrix based on the conventional PB plus a correction term c obtained from a MC simulation. Updates of c can be triggered as often as necessary by calling the MC dose engine with the last corrected values of w as input. In order to analyze the performance of the hybrid algorithm against dose calculation errors, it was applied to a simplistic water phantom for which several test cases with different errors were simulated, including proton range uncertainties. Afterwards, the algorithm was used in three clinical cases (prostate, lung, and brain) and benchmarked against full MC-based optimization. The influence of different stopping criteria in the final results was also investigated.

RESULTS

The hybrid algorithm achieved excellent results provided that the estimated range in a homogeneous material is the same for the two dose engines involved, i.e., PB and MC. For the three patient cases, the hybrid plans were clinically equivalent to those obtained with full MC-based optimization. Only a single update of c was needed in the hybrid algorithm to fulfill the clinical dose constraints, which represents an extra computation time to obtain c that ranged from 1 (brain) to 4 min (lung) with respect to the conventional PB-based optimization, and an estimated average gain factor of 14 with respect to full MC-based optimization.

CONCLUSION

The hybrid algorithm provides an improved trade-off between accuracy and speed. This algorithm can be immediately considered as an option for improving dose calculation accuracy of commercial analytical treatment planning systems, without a significant increase in the computation time (≪5 min) with respect to current PB-based optimization.

Theoretical model for laser ablation outcome predictions in brain: calibration and validation on clinical MR thermometry images

PURPOSE

Neurosurgical laser ablation is experiencing a renaissance. Computational tools for ablation planning aim to further improve the intervention. Here, global optimisation and inverse problems are demonstrated to train a model that predicts maximum laser ablation extent.

METHODS

A closed-form steady state model is trained on and then subsequently compared to N = 20 retrospective clinical MR thermometry datasets. Dice similarity coefficient (DSC) is calculated to provide a measure of region overlap between the 57 °C isotherms of the thermometry data and the model-predicted ablation regions; 57 °C is a tissue death surrogate at thermal steady state. A global optimisation scheme samples the dominant model parameter sensitivities, blood perfusion (ω) and optical parameter (μ_eff) values, throughout a parameter space totalling 11 440 value-pairs. This represents a lookup table of μ_eff-ω pairs with the corresponding DSC value for each patient dataset. The μ_eff-ω pair with the maximum DSC calibrates the model parameters, maximising predictive value for each patient. Finally, leave-one-out cross-validation with global optimisation information trains the model on the entire clinical dataset, and compares against the model naïvely using literature values for ω and μ_eff.

RESULTS

When using naïve literature values, the model's mean DSC is 0.67 whereas the calibrated model produces 0.82 during cross-validation, an improvement of 0.15 in overlap with the patient data. The 95% confidence interval of the mean difference is 0.083-0.23 (p < 0.001).

CONCLUSIONS

During cross-validation, the calibrated model is superior to the naïve model as measured by DSC, with +22% mean prediction accuracy. Calibration empowers a relatively simple model to become more predictive.

Physically constrained voxel-based penalty adaptation for ultra-fast IMRT planning

Conventional treatment planning in intensity-modulated radiation therapy (IMRT) is a trial-and-error process that usually involves tedious tweaking of optimization parameters. Here, we present an algorithm that automates part of this process, in particular the adaptation of voxel-based penalties within normal tissue. Thereby, the proposed algorithm explicitly considers a priori known physical limitations of photon irradiation. The efficacy of the developed algorithm is assessed during treatment planning studies comprising 16 prostate and 5 head and neck cases. We study the eradication of hot spots in the normal tissue, effects on target coverage and target conformity, as well as selected dose volume points for organs at risk. The potential of the proposed method to generate class solutions for the two indications is investigated. Run-times of the algorithms are reported. Physically constrained voxel-based penalty adaptation is an adequate means to automatically detect and eradicate hot-spots during IMRT planning while maintaining target coverage and conformity. Negative effects on organs at risk are comparably small and restricted to lower doses. Using physically constrained voxel-based penalty adaptation, it was possible to improve the generation of class solutions for both indications. Considering the reported run-times of less than 20 s, physically constrained voxel-based penalty adaptation has the potential to reduce the clinical workload during planning and automated treatment plan generation in the long run, facilitating adaptive radiation treatments.

Projections onto the Pareto surface in multicriteria radiation therapy optimization

PURPOSE

To eliminate or reduce the error to Pareto optimality that arises in Pareto surface navigation when the Pareto surface is approximated by a small number of plans.

METHODS

The authors propose to project the navigated plan onto the Pareto surface as a postprocessing step to the navigation. The projection attempts to find a Pareto optimal plan that is at least as good as or better than the initial navigated plan with respect to all objective functions. An augmented form of projection is also suggested where dose-volume histogram constraints are used to prevent that the projection causes a violation of some clinical goal. The projections were evaluated with respect to planning for intensity modulated radiation therapy delivered by step-and-shoot and sliding window and spot-scanned intensity modulated proton therapy. Retrospective plans were generated for a prostate and a head and neck case.

RESULTS

The projections led to improved dose conformity and better sparing of organs at risk (OARs) for all three delivery techniques and both patient cases. The mean dose to OARs decreased by 3.1 Gy on average for the unconstrained form of the projection and by 2.0 Gy on average when dose-volume histogram constraints were used. No consistent improvements in target homogeneity were observed.

CONCLUSIONS

There are situations when Pareto navigation leaves room for improvement in OAR sparing and dose conformity, for example, if the approximation of the Pareto surface is coarse or the problem formulation has too permissive constraints. A projection onto the Pareto surface can identify an inaccurate Pareto surface representation and, if necessary, improve the quality of the navigated plan.

Heavy Ion Therapy: Status and Perspectives

Starting with the pioneering work at the University of California in Berkeley in 1977, heavy ion radiotherapy has been of increasing interest especially in Japan and Europe in the last decade. There are currently 3 facilities treating patients with carbon ions, two of them in Japan within a clinical setting. In Germany, a research therapy facility is in operation and the construction of a new hospital based facility at the Heidelberg university will be started soon. An outline of the current status of heavy ion radiotherapy is given with emphasis to the technical aspects of the respective facilities. This includes a description of passive and active beam shaping systems, as well as their implications for treatment planning and dosimetry. The clinical trials and routine treatments performed at the German heavy ion facility are summarized.

An overview over the upcoming new facilities and their technical possibilities is given. It is discussed what the necessary improvements are to fully exploit the potential of these facilities. Especially the new Heidelberg facility with the possibility of active beam scanning in combination with the first isocentric gantry for ions and offering beams of protons, helium, oxygen and carbon ions has implications on treatment planning, dosimetry and quality assurance. The necessary and ongoing developments in these areas are summarized. The new facilities also offer the possibilities to perform more extensive clinical studies and to explore future indications for radiotherapy with heavy ions. An overview over the indications and treatment schemes is also given.

Optimization of Physical Dose Distributions with Hadron Beams: Comparing Photon IMRT with IMPT

Intensity modulated radiotherapy with high enengy photons (IMRT) and with charged particles (IMPT) refer to the most advanced development in conformal radiation therapy. Their general aim is to increase local tumor control rates while keeping the radiation induced complications below desired thresholds. IMRT is currently widely introduced in clinical practice. However, the more complicated IMPT is still under development. Especially, spot-scanning techniques integrated in rotating gantries that can deliver proton or light ion-beams to a radiation target from any direction will be available in the near future.

We describe the basic concepts of intensity modulated particle therapy (IMPT). Starting from the potential advantages of hadron therapy inverse treatment planning strategies are discussed for various dose delivery techniques of IMPT. Of special interest are the techniques of distal edge tracking (DET) and 3D-scanning. After the introduction of these concepts a study of comparative inverse treatment planning is presented. The study aims to identify the potential advantages of achievable physical dose distributions with proton and carbon beams, if different dose delivery techniques are employed. Moreover, a comparison to standard photon IMRT is performed.

The results of the study are summarized as: i) IMRT with photon beams is a strong competitor to intensity modulated radiotherapy with charged particles. The most obvious benefit observed for charged particles is the reduction of medium and low doses in organs at risk. ii) The 3D-scanning technique could not improve the dosimetric results achieved with DET, although 10–15 times more beam spots were employed for 3D-scanning than for DET. However, concerns may arise about the application of DET, if positioning errors of the patient or organ movements have to be accounted for. iii) Replacing protons with carbon ions leads to further improvements of the physical dose distributions. However, the additional degree of improvement due to carbon ions is modest. The main clinical potential of heavy ion beams is probably related to their radiobiological properties.

Optimization of Beam Parameters and Treatment Planning for Intensity Modulated Proton Therapy

One of the objectives of the ongoing research and development work at the Northeast Proton Therapy Center (NPTC) in Boston is to perform optimized intensity modulated proton therapy (IMPT) treatments. Such treatments may be delivered by magnetically scanning a narrow proton pencil beam across the target volume, while both the scanning speed and the intensity of the beam are modulated. Localization of the proton dose in space allows one to yield dose distributions that are highly conformal to the target volume, thus minimizing the dose delivered to the surrounding healthy tissue. The aim of the current research is to determine technically optimal and clinically relevant specifications for the scanned beam delivery system, which is being developed in collaboration with Ion Beam Applications (IBA); and to create a link between the treatment planning and the beam delivery. IMPT treatment planning is performed for patient cases treated at the NPTC, with KonRad Pro software developed at the German Cancer Research Center (DKFZ). For the IMPT delivery, the proton intensity maps, optimized for discrete pencil beam spots, need to be translated into continuous scanning patterns. At the same time it is necessary to minimize the discrepancy between the planned and delivered doses which may result from such conversion, as well as from the technical limitations of the delivery system. Possibilities have been investigated for improving the proton dose conformity by optimizing the beam and scanning nozzle parameters, and by taking the specifics and limitations of the system into account in the treatment planning stage.

A critical evaluation of worst case optimization methods for robust intensity-modulated proton therapy planning

PURPOSE

To critically evaluate and compare three worst case optimization methods that have been previously employed to generate intensity-modulated proton therapy treatment plans that are robust against systematic errors. The goal of the evaluation is to identify circumstances when the methods behave differently and to describe the mechanism behind the differences when they occur.

METHODS

The worst case methods optimize plans to perform as well as possible under the worst case scenario that can physically occur (composite worst case), the combination of the worst case scenarios for each objective constituent considered independently (objectivewise worst case), and the combination of the worst case scenarios for each voxel considered independently (voxelwise worst case). These three methods were assessed with respect to treatment planning for prostate under systematic setup uncertainty. An equivalence with probabilistic optimization was used to identify the scenarios that determine the outcome of the optimization.

RESULTS

If the conflict between target coverage and normal tissue sparing is small and no dose-volume histogram (DVH) constraints are present, then all three methods yield robust plans. Otherwise, they all have their shortcomings: Composite worst case led to unnecessarily low plan quality in boundary scenarios that were less difficult than the worst case ones. Objectivewise worst case generally led to nonrobust plans. Voxelwise worst case led to overly conservative plans with respect to DVH constraints, which resulted in excessive dose to normal tissue, and less sharp dose fall-off than the other two methods.

CONCLUSIONS

The three worst case methods have clearly different behaviors. These behaviors can be understood from which scenarios that are active in the optimization. No particular method is superior to the others under all circumstances: composite worst case is suitable if the conflicts are not very severe or there are DVH constraints whereas voxelwise worst case is advantageous if there are severe conflicts but no DVH constraints. The advantages of composite and voxelwise worst case outweigh those of objectivewise worst case.

Noncoplanar VMAT for nasopharyngeal tumors: Plan quality versus treatment time

PURPOSE

The authors investigated the potential of optimized noncoplanar irradiation trajectories for volumetric modulated arc therapy (VMAT) treatments of nasopharyngeal patients and studied the trade-off between treatment plan quality and delivery time in radiation therapy.

METHODS

For three nasopharyngeal patients, the authors generated treatment plans for nine different delivery scenarios using dedicated optimization methods. They compared these scenarios according to dose characteristics, number of beam directions, and estimated delivery times. In particular, the authors generated the following treatment plans: (1) a 4π plan, which is a not sequenced, fluence optimized plan that uses beam directions from approximately 1400 noncoplanar directions and marks a theoretical upper limit of the treatment plan quality, (2) a coplanar 2π plan with 72 coplanar beam directions as pendant to the noncoplanar 4π plan, (3) a coplanar VMAT plan, (4) a coplanar step and shoot (SnS) plan, (5) a beam angle optimized (BAO) coplanar SnS IMRT plan, (6) a noncoplanar BAO SnS plan, (7) a VMAT plan with rotated treatment couch, (8) a noncoplanar VMAT plan with an optimized great circle around the patient, and (9) a noncoplanar BAO VMAT plan with an arbitrary trajectory around the patient.

RESULTS

VMAT using optimized noncoplanar irradiation trajectories reduced the mean and maximum doses in organs at risk compared to coplanar VMAT plans by 19% on average while the target coverage remains constant. A coplanar BAO SnS plan was superior to coplanar SnS or VMAT; however, noncoplanar plans like a noncoplanar BAO SnS plan or noncoplanar VMAT yielded a better plan quality than the best coplanar 2π plan. The treatment plan quality of VMAT plans depended on the length of the trajectory. The delivery times of noncoplanar VMAT plans were estimated to be 6.5 min in average; 1.6 min longer than a coplanar plan but on average 2.8 min faster than a noncoplanar SnS plan with comparable treatment plan quality.

CONCLUSIONS

The authors' study reconfirms the dosimetric benefits of noncoplanar irradiation of nasopharyngeal tumors. Both SnS using optimized noncoplanar beam ensembles and VMAT using an optimized, arbitrary, noncoplanar trajectory enabled dose reductions in organs at risk compared to coplanar SnS and VMAT. Using great circles or simple couch rotations to implement noncoplanar VMAT, however, was not sufficient to yield meaningful improvements in treatment plan quality. The authors estimate that noncoplanar VMAT using arbitrary optimized irradiation trajectories comes at an increased delivery time compared to coplanar VMAT yet at a decreased delivery time compared to noncoplanar SnS IMRT.

Shared data for intensity modulated radiation therapy (IMRT) optimization research: the CORT dataset

BACKGROUND

We provide common datasets (which we call the CORT dataset: common optimization for radiation therapy) that researchers can use when developing and contrasting radiation treatment planning optimization algorithms. The datasets allow researchers to make one-to-one comparisons of algorithms in order to solve various instances of the radiation therapy treatment planning problem in intensity modulated radiation therapy (IMRT), including beam angle optimization, volumetric modulated arc therapy and direct aperture optimization.

RESULTS

We provide datasets for a prostate case, a liver case, a head and neck case, and a standard IMRT phantom. We provide the dose-influence matrix from a variety of beam/couch angle pairs for each dataset. The dose-influence matrix is the main entity needed to perform optimizations: it contains the dose to each patient voxel from each pencil beam. In addition, the original Digital Imaging and Communications in Medicine (DICOM) computed tomography (CT) scan, as well as the DICOM structure file, are provided for each case.

CONCLUSIONS

Here we present an open dataset - the first of its kind - to the radiation oncology community, which will allow researchers to compare methods for optimizing radiation dose delivery.

Comparison of beam angle selection strategies for intracranial IMRT

PURPOSE

Various strategies to select beneficial beam ensembles for intensity-modulated radiation therapy (IMRT) have been suggested over the years. These beam angle selection (BAS) strategies are usually evaluated against reference configurations applying equispaced coplanar beams but they are not compared to one another. Here, the authors present a meta analysis of four BAS strategies that incorporates fluence optimization (FO) into BAS by combinatorial optimization (CO) and one BAS strategy that decouples FO from BAS, i.e., spherical cluster analysis (SCA). The underlying parameters of the BAS process are investigated and the dosimetric benefits of the BAS strategies are quantified.

METHODS

For three intracranial lesions in proximity to organs at risk (OARs) the authors compare treatment plans applying equispaced coplanar beam ensembles with treatment plans using five different BAS strategies, i.e., four CO techniques and SCA, to establish coplanar and noncoplanar beam ensembles. Treatment plans applying 5, 7, 9, and 11 beams are investigated. For the CO strategies the authors perform BAS runs with a 5°, 10°, 15°, and 20° angular resolution, which corresponds to a minimum of 18 coplanar and a maximum of 1400 noncoplanar candidate beams. In total 272 treatment plans with different BAS settings are generated for every patient. The quality of the treatment plans is compared based on the protection of OARs yet integral dose, target homogeneity, and target conformity are also considered.

RESULTS

It is possible to reduce the average mean and maximum doses in OARs by more than 4 Gy (1 Gy) with optimized noncoplanar (coplanar) beam ensembles found with BAS by CO or SCA. For BAS including FO by CO, the individual algorithm used and the angular resolution in the space of candidate beams does not have a crucial impact on the quality of the resulting treatment plans. All CO algorithms yield similar target conformity and slightly improved target homogeneity in comparison to equispaced coplanar setups. Furthermore, optimized coplanar (noncoplanar) beam ensembles enabled more than a 6% (5%) reduction of the integral dose. For SCA, however, integral dose was increased and target conformity was decreased in comparison to equispaced coplanar setups-especially for a small number of beams.

CONCLUSION

Both BAS strategies incorporating FO by CO and independent BAS strategies excluding FO provide dose savings in OARs for optimized coplanar and especially noncoplanar beam ensembles; they should not be neglected in the clinic.

The rationale for intensity-modulated proton therapy in geometrically challenging cases

Intensity-modulated proton therapy (IMPT) delivered with beam scanning is currently available at a limited number of proton centers. However, a simplified form of IMPT, the technique of field 'patching', has long been a standard practice in proton therapy centers. In field patching, different parts of the target volume are treated from different directions, i.e., a part of the tumor gets either full dose from a radiation field, or almost no dose. Thus, patching represents a form of binary intensity modulation. This study explores the limitations of the standard binary field patching technique, and evaluates possible dosimetric advantages of continuous dose modulations in IMPT. Specifics of the beam delivery technology, i.e., pencil beam scanning versus passive scattering and modulation, are not investigated. We have identified two geometries of target volumes and organs at risk (OAR) in which the use of field patching is severely challenged. We focused our investigations on two patient cases that exhibit these geometries: a paraspinal tumor case and a skull-base case. For those cases we performed treatment planning comparisons of three-dimensional conformal proton therapy (3DCPT) with field patching versus IMPT, using commercial and in-house software, respectively. We also analyzed the robustness of the resulting plans with respect to systematic setup errors of ±1 mm and range errors of ±2.5 mm. IMPT is able to better spare OAR while providing superior dose coverage for the challenging cases identified above. Both 3DCPT and IMPT are sensitive to setup errors and range uncertainties, with IMPT showing the largest effect. Nevertheless, when delivery uncertainties are taken into account IMPT plans remain superior regarding target coverage and OAR sparing. On the other hand, some clinical goals, such as the maximum dose to OAR, are more likely to be unmet with IMPT under large range errors. IMPT can potentially improve target coverage and OAR sparing in challenging cases, even when compared with the relatively complicated and time consuming field patching technique. While IMPT plans tend to be more sensitive to delivery uncertainties, their dosimetric advantage generally holds. Robust treatment planning techniques may further reduce the sensitivity of IMPT plans.

Characterizing the combinatorial beam angle selection problem

The beam angle selection (BAS) problem in intensity-modulated radiation therapy is often interpreted as a combinatorial optimization problem, i.e. finding the best combination of η beams in a discrete set of candidate beams. It is well established that the combinatorial BAS problem may be solved efficiently with metaheuristics such as simulated annealing or genetic algorithms. However, the underlying parameters of the optimization process, such as the inclusion of non-coplanar candidate beams, the angular resolution in the space of candidate beams, and the number of evaluated beam ensembles as well as the relative performance of different metaheuristics have not yet been systematically investigated. We study these open questions in a meta-analysis of four strategies for combinatorial optimization in order to provide a reference for future research related to the BAS problem in intensity-modulated radiation therapy treatment planning. We introduce a high-performance inverse planning engine for BAS. It performs a full fluence optimization for ≈3600 treatment plans per hour while handling up to 50 GB of dose influence data (≈1400 candidate beams). For three head and neck patients, we compare the relative performance of a genetic, a cross-entropy, a simulated annealing and a naive iterative algorithm. The selection of ensembles with 5, 7, 9 and 11 beams considering either only coplanar or all feasible candidate beams is studied for an angular resolution of 5°, 10°, 15° and 20° in the space of candidate beams. The impact of different convergence criteria is investigated in comparison to a fixed termination after the evaluation of 10 000 beam ensembles. In total, our simulations comprise a full fluence optimization for about 3000 000 treatment plans. All four combinatorial BAS strategies yield significant improvements of the objective function value and of the corresponding dose distributions compared to standard beam configurations with equi-spaced coplanar beams. The genetic and the cross-entropy algorithms showed faster convergence in the very beginning of the optimization but the simulated annealing algorithm eventually arrived at almost the same objective function values. These three strategies typically yield clinically equivalent treatment plans. The iterative algorithm showed the worst convergence properties. The choice of the termination criterion had a stronger influence on the performance of the simulated annealing algorithm than on the performance of the genetic and the cross-entropy algorithms. We advocate to terminate the optimization process after the evaluation of 1000 beam combinations without objective function decrease. For our simulations, this resulted in an average deviation of the objective function from the reference value after 10 000 evaluated beam ensembles of 0.5% for all metaheuristics. On average, there was only a minor improvement when increasing the angular resolution in the space of candidate beam angles from 20° to 5°. However, we observed significant improvements when considering non-coplanar candidate beams for challenging head and neck cases.

Influence of robust optimization in intensity-modulated proton therapy with different dose delivery techniques

PURPOSE

The distal edge tracking (DET) technique in intensity-modulated proton therapy (IMPT) allows for high energy efficiency, fast and simple delivery, and simple inverse treatment planning; however, it is highly sensitive to uncertainties. In this study, the authors explored the application of DET in IMPT (IMPT-DET) and conducted robust optimization of IMPT-DET to see if the planning technique's sensitivity to uncertainties was reduced. They also compared conventional and robust optimization of IMPT-DET with three-dimensional IMPT (IMPT-3D) to gain understanding about how plan robustness is achieved.

METHODS

They compared the robustness of IMPT-DET and IMPT-3D plans to uncertainties by analyzing plans created for a typical prostate cancer case and a base of skull (BOS) cancer case (using data for patients who had undergone proton therapy at our institution). Spots with the highest and second highest energy layers were chosen so that the Bragg peak would be at the distal edge of the targets in IMPT-DET using 36 equally spaced angle beams; in IMPT-3D, 3 beams with angles chosen by a beam angle optimization algorithm were planned. Dose contributions for a number of range and setup uncertainties were calculated, and a worst-case robust optimization was performed. A robust quantification technique was used to evaluate the plans' sensitivity to uncertainties.

RESULTS

With no uncertainties considered, the DET is less robust to uncertainties than is the 3D method but offers better normal tissue protection. With robust optimization to account for range and setup uncertainties, robust optimization can improve the robustness of IMPT plans to uncertainties; however, our findings show the extent of improvement varies.

CONCLUSIONS

IMPT's sensitivity to uncertainties can be improved by using robust optimization. They found two possible mechanisms that made improvements possible: (1) a localized single-field uniform dose distribution (LSFUD) mechanism, in which the optimization algorithm attempts to produce a single-field uniform dose distribution while minimizing the patching field as much as possible; and (2) perturbed dose distribution, which follows the change in anatomical geometry. Multiple-instance optimization has more knowledge of the influence matrices; this greater knowledge improves IMPT plans' ability to retain robustness despite the presence of uncertainties.

Robust optimization of intensity modulated proton therapy

PURPOSE

Intensity modulated proton therapy (IMPT) is highly sensitive to range uncertainties and uncertainties caused by setup variation. The conventional inverse treatment planning of IMPT optimized based on the planning target volume (PTV) is not often sufficient to ensure robustness of treatment plans. In this paper, a method that takes the uncertainties into account during plan optimization is used to mitigate the influence of uncertainties in IMPT.

METHODS

The authors use the so-called "worst-case robust optimization" to render IMPT plans robust in the face of uncertainties. For each iteration, nine different dose distributions are computed-one each for ± setup uncertainties along anteroposterior (A-P), lateral (R-L) and superior-inferior (S-I) directions, for ± range uncertainty, and the nominal dose distribution. The worst-case dose distribution is obtained by assigning the lowest dose among the nine doses to each voxel in the clinical target volume (CTV) and the highest dose to each voxel outside the CTV. Conceptually, the use of worst-case dose distribution is similar to the dose distribution achieved based on the use of PTV in traditional planning. The objective function value for a given iteration is computed using this worst-case dose distribution. The objective function used has been extended to further constrain the target dose inhomogeneity.

RESULTS

The worst-case robust optimization method is applied to a lung case, a skull base case, and a prostate case. Compared with IMPT plans optimized using conventional methods based on the PTV, our method yields plans that are considerably less sensitive to range and setup uncertainties. An interesting finding of the work presented here is that, in addition to reducing sensitivity to uncertainties, robust optimization also leads to improved optimality of treatment plans compared to the PTV-based optimization. This is reflected in reduction in plan scores and in the lower normal tissue doses for the same coverage of the target volume when subjected to uncertainties.

CONCLUSIONS

The authors find that the worst-case robust optimization provides robust target coverage without sacrificing, and possibly even improving, the sparing of normal tissues. Our results demonstrate the importance of robust optimization. The authors assert that all IMPT plans should be robustly optimized.