Selection Of Beam Orientations Research Articles

PD-0732 Dosimetric feature-based beam orientation selection in intensity-modulated radiation therapy

PD-0732 Dosimetric feature-based beam orientation selection in intensity-modulated radiation therapy

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

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. 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. 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 1.5s to select a set of five beam orientations and 300s to calculate the dose influence matrices for 5 beams and finally 20s to solve the fluence map optimization (FMO). However, CG needed around 15h to calculate the dose influence matrices of all beams and at least 400s 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. 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.

Parallel beamlet dose calculation via beamlet contexts in a distributed multi-GPU framework.

Dose calculation is one of the most computationally intensive, yet essential tasks in the treatment planning process. With the recent interest in automatic beam orientation and arc trajectory optimization techniques, there is a great need for more efficient model-based dose calculation algorithms that can accommodate hundreds to thousands of beam candidates at once. Foundational work has shown the translation of dose calculation algorithms to graphical processing units (GPUs), lending to remarkable gains in processing efficiency. But these methods provide parallelization of dose for only a single beamlet, serializing the calculation of multiple beamlets and under-utilizing the potential of modern GPUs. In this paper, the authors propose a framework enabling parallel computation of many beamlet doses using a novel beamlet context transformation and further embed this approach in a scalable network of multi-GPU computational nodes. The proposed context-based transformation separates beamlet-local density and TERMA into distinct beamlet contexts that independently provide sufficient data for beamlet dose calculation. Beamlet contexts are arranged in a composite context array with dosimetric isolation, and the context array is subjected to a GPU collapsed-cone convolution superposition procedure, producing the set of beamlet-specific dose distributions in a single pass. Dose from each context is converted to a sparse representation for efficient storage and retrieval during treatment plan optimization. The context radius is a new parameter permitting flexibility between the speed and fidelity of the dose calculation process. A distributed manager-worker architecture is constructed around the context-based GPU dose calculation approach supporting an arbitrary number of worker nodes and resident GPUs. Phantom experiments were executed to verify the accuracy of the context-based approach compared to Monte Carlo and a reference CPU-CCCS implementation for single beamlets and broad beams composed by addition of beamlets. Dose for representative 4π beam sets was calculated in lung and prostate cases to compare its efficiency with that of an existing beamlet-sequential GPU-CCCS implementation. Code profiling was also performed to evaluate the scalability of the framework across many networked GPUs. The dosimetric accuracy of the context-based method displays <1.35% and 2.35% average error from the existing serialized CPU-CCCS algorithm and Monte Carlo simulation for beamlet-specific PDDs in water and slab phantoms, respectively. The context-based method demonstrates substantial speedup of up to two orders of magnitude over the beamlet-sequential GPU-CCCS method in the tested configurations. The context-based framework demonstrates near linear scaling in the number of distributed compute nodes and GPUs employed, indicating that it is flexible enough to meet the performance requirements of most users by simply increasing the hardware utilization. The context-based approach demonstrates a new expectation of performance for beamlet-based dose calculation methods. This approach has been successful in accelerating the dose calculation process for very large-scale treatment planning problems - such as automatic 4π IMRT beam orientation and VMAT arc trajectory selection, with hundreds of thousands of beamlets - in clinically feasible timeframes. The flexibility of this framework makes it as a strong candidate for use in a variety of other very large-scale treatment planning tasks and clinical workflows.

Integrated beam orientation and scanning-spot optimization in intensity-modulated proton therapy for brain and unilateral head and neck tumors.

Intensity-Modulated Proton Therapy (IMPT) is the state-of-the-art method of delivering proton radiotherapy. Previous research has been mainly focused on optimization of scanning spots with manually selected beam angles. Due to the computational complexity, the potential benefit of simultaneously optimizing beam orientations and spot pattern could not be realized. In this study, we developed a novel integrated beam orientation optimization (BOO) and scanning-spot optimization algorithm for intensity-modulated proton therapy (IMPT). A brain chordoma and three unilateral head-and-neck patients with a maximal target size of 112.49cm3 were included in this study. A total number of 1162 noncoplanar candidate beams evenly distributed across 4π steradians were included in the optimization. For each candidate beam, the pencil-beam doses of all scanning spots covering the PTV and a margin were calculated. The beam angle selection and spot intensity optimization problem was formulated to include three terms: a dose fidelity term to penalize the deviation of PTV and OAR doses from ideal dose distribution; an L1-norm sparsity term to reduce the number of active spots and improve delivery efficiency; a group sparsity term to control the number of active beams between 2 and 4. For the group sparsity term, convex L2,1-norm and nonconvex L2,1/2-norm were tested. For the dose fidelity term, both quadratic function and linearized equivalent uniform dose (LEUD) cost function were implemented. The optimization problem was solved using the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA). The IMPT BOO method was tested on three head-and-neck patients and one skull base chordoma patient. The results were compared with IMPT plans created using column generation selected beams or manually selected beams. The L2,1-norm plan selected spatially aggregated beams, indicating potential degeneracy using this norm. L2,1/2-norm was able to select spatially separated beams and achieve smaller deviation from the ideal dose. In the L2,1/2-norm plans, the [mean dose, maximum dose] of OAR were reduced by an average of [2.38%, 4.24%] and[2.32%, 3.76%] of the prescription dose for the quadratic and LEUD cost function, respectively, compared with the IMPT plan using manual beam selection while maintaining the same PTV coverage. The L2,1/2 group sparsity plans were dosimetrically superior to the column generation plans as well. Besides beam orientation selection, spot sparsification was observed. Generally, with the quadratic cost function, 30%~60% spots in the selected beams remained active. With the LEUD cost function, the percentages of active spots were in the range of 35%~85%.The BOO-IMPT run time was approximately 20min. This work shows the first IMPT approach integrating noncoplanar BOO and scanning-spot optimization in a single mathematical framework. This method is computationally efficient, dosimetrically superior and produces delivery-friendly IMPT plans.

Pseudo-automatic beam orientations in multi-criteria intensity-modulated radiation therapy

Selecting a number of beams and their orientations in intensity-modulated radiation therapy (IMRT) is an important, but large-scale NP-hard optimization problem. Unfortunately, the quality of beam directions depends heavily on its corresponding beam intensity profiles. Usually, the number of beams is manually pre-determined and passed for a stochastic selector which optimizes beam orientations, and then a single objective inverse treatment planning algorithm is used for optimizing beam intensity profiles. The overall time needed to solve the inverse planning for every random selection of beam orientations becomes excessive. Therefore selecting an appropriate set of beam directions in IMRT is still a time-consuming manual trial-and-error search procedure that depends on intuition and empirical knowledge. During the last decade considerable progress has been made in optimizing beam intensity profiles by using multi-objective inverse treatment planning. Multi-criteria optimization results in a variety of beam intensity profiles for every selection of beam orientations. The planner then needs to search, manually, through this variety of solutions to decide upon an optimal plan. This work takes advantage of this human interaction to present a two-phase algorithm in which both processes of beam orientations and multi-criteria optimization with human guidelines cooperate together to determine a suitable number of beams and their orientations in a clinically practical time-slot. To illustrate the efficiency of the algorithm it has been applied to one artificial example and three real clinical cases. The results show that using the algorithm, both target coverage and sparing critical structures were significantly improved compared to the standard equally spaced beam plans. Moreover in some cases the algorithm suggested plans with one beam fewer than the equally spaced beam plans without compromising the quality of the treatment plan.

Optimal selection of beam orientations in intensity-modulated radiation therapy

Radiation therapy is a common and important treatment for some specific tumors in the treatment of cancer. In recent years, there has been a new development in radiation therapy, intensity-modulated radiation therapy (IMRT). IMRT modulates the intensity of the radiation beam to focus a higher radiation dose on the tumor while minimizing radiation exposure to surrounding normal tissues. This study aims to investigate the optimal selection of beam angles. Besides limiting the radiation dose to the tumor and the adjacent organs, we also take the dose distribution into consideration and optimize the beam angles and beam intensity maps simultaneously. Compared with the typical equi-spaced model, our proposed approach can provide optimal beam configuration to help physicians in decision-making.

Automatic Selection of Beam Orientations in Intensity-Modulated Radiation Therapy

Abstract Finding an optimal set of beam orientations in intensity-modulated radiation therapy (IMRT) planning is an important, but unfortunately, large-scale NP-hard optimization problem with impractical and excessive time complexity. Selecting appropriate beam directions in IMRT is still a time-consuming manual trial-and-error search procedure that depends on intuition and empirical knowledge. This work utilizes the methods of beam's eye view and observer's view, which are recognized for beam orientations in conventional conformal radiation therapy, to present an automatic algorithm that works in clinically practical time and aims at determining a suitable set of beam orientations for IMRT.

SU-FF-T-591: Improvement of IMRT Plan Quality with Non-Isocentric Beam Arrangement for Head and Neck Cancers

Purpose: Selection of beam orientations are sometime limited by geometric confliction raised from isocentric setup. The recent introduction of robotic couch (e.g., Treago, Siemens) in linac room opens door for using non-isocentric beams. We study the dosimetric advantages of using non-isocentric beams over conventional isocentric beams for IMRT of head and neck cancers. Method and Materials: Because of complicated anatomic geometric relationship between planning target volumes (PTV) and surrounding organs at risk (OAR), IMRT plan quality for head and neck cancers is often affected by the selection of beam orientations. In this work, isocentric and non-isocentric IMRT plans were generated for 10 representative head and neck cases using a treatment planning system (XiO, CMS). Non-isocentric beams were determined based on the locations, sizes and shapes of the PTV and OARs with appropriate gantry/couch angle and couch position, to maximally avoid direct expose of OARs. Same dose volume constraints were used in the generation of isocentric and non-isocentric plans. The two plans were compared using various parameters, including conformity index (CI), uniformity index (UI), V30 (volume receiving 30 Gy) and V15, and an overall plan quality indicator, fEUD calculated based on the equivalent uniform doses (EUD) for both targets and OARs. Results: The dose coverage and uniformity of PTV are generally comparable between the two plans for the cases studied. Normal tissue sparing is however improved with non-isocentric plan. For example, the values of V30 of parotid gland, V30 and V15 of cord, for non-isocentric plans are reduced with p-values = 0.032, 0.040 and 0.049, respectively. Comparison of fEUD shows that the overall plan quality for a non-isocentric plan is improved. Conclusion: Non-isocentric beam arrangement, which can be efficiently delivered with a robotic couch, can improve IMRT plan quality for head and neck cancers as compared to the conventional isocentric beams.

SU-FF-T-612: Developing a Treatment Planning Framework for Using Non-Coplanar Non-Isocentric Beams to Improve Plan Quality

Purpose: Historically, coplanar isocentric techniques gained popularity due to the ease of patient setup and treatment delivery. A new class of remote‐controlled robotic couches (e.g., Treago, Siemens) that are being introduced for linac‐based radiotherapy will make possible rapid and accurate couch translations and rotations between treatment fields and will open door for efficient delivery of non‐coplanar beams. In addition, treatments at extended distance can be easily introduced to achieve some non‐coplanar beam angles not possible with the isocentric setup because of a collision. We propose a framework for treatment planning employing non‐coplanar non‐isocentric beams to be delivered in conjunction with a robotic couch. Method and Materials: In the proposed algorithm, the design of a treatment plan will start with the selection of optimal beam orientations by varying gantry angle and couch angle and position based on a geometrical relationship between target(s) and organs at risk. A rank order list of candidate beam angles will be created. The candidate beams will be checked for possible gantry‐couch and gantry‐patient collisions. In most cases, a wide range of robotic couch translations will allow treatments at extended distance, which will greatly reduce the number of infeasible beam orientations. Finally, inverse treatment planning will be performed based on a physical or biological cost function using the highest ranked deliverable beams. Results: Once the proposed framework is implemented, we will quantify dosimetric gains and identify tumor sites most likely to benefit. The proposed approach is expected to take advantage of Cyberknife‐like angular flexibility with the added benefit of intensity modulation at each beam angle. Conclusion: It is hypothesized that for certain tumor sites a considerable improvement in plan quality can be achieved through the use of non‐coplanar nonisocentric treatments. General strategy for implementing such a treatment approach using a robotic couch has been proposed.

Non-coplanar automatic beam orientation selection in cranial IMRT: a practical methodology

This paper proposes a method for automatic selection of beam orientations in non-coplanar cranial IMRT. Methods of computer vision, beam's eye view techniques and neural networks are used to define a new geometry-based methodology that leads to treatment plans for cranial lesions that are comparable in quality to those generated by experienced radiation physicists. The automatic beam selection (ABS) process can be carried out in clinically useful computation times, in 1 min or less for most cases. In the process of describing the ABS process, it is shown that the cranial beam orientation optimization problem is mathematically ill posed, with the expectation that a large number of solutions will lead to similar results. Nevertheless, there are better and worse solutions and we show that the proposed ABS process, by its design, has to lead to one of the better ones. We have carried out extensive tests with 14 patients with beam selection tasks ranging from the rather simple to quite complex. The ABS process has always yielded optimizations with results that are considered good for clinic use. Seven-beam coplanar optimizations for some of the patients have also been investigated. Comparisons with non-coplanar optimizations indicate in which cases the simpler coplanar plans can be used to advantage. Parameters used in the comparisons are dose–volume histograms, minimum and maximum PTV doses, equivalent uniform doses for the PTV and OARs, and treatment volume, conformity and normal tissue indices. It is felt that the current ABS methodology is ready for extensive clinical tests.

Sci-Fri AM: YIS-07: A new paradigm for improving IMRT: Selection of beam orientations by optimizing beam intersection volume.

A beam orientation optimization (BOO) algorithm based on optimizing beam intersection volume (BIV) components within an Organ-at-Risk (OAR) is proposed to improve conventional intensity-modulated radiation therapy (IMRT). A simulated annealing algorithm was employed to search for the optimal set of five beam orientations (5-opt) which simultaneously minimize the BIV components within an OAR. The 5-opt plans were compared to standard 5, 7, and 9 equiangular-spaced beam plans (5-equi, 7-equi, 9-equi) for: (1) gastric (2) Radiation Therapy Oncology Group (RTOG) P-0126 prostate and (3) RTOG H-0022 oropharyngeal (Stage-III, IV) cancer patients. In the gastric case, the coplanar 5-opt plan reduced the right kidney V 20 Gy by 41.1%, 32.1%, and 29.5% compared to the 5-equi, 7-equi, and 9-equi plans. In the prostate case, the coplanar 5-opt plan improved rectal sparing over all standard plans with a reduction of the V 75 Gy, V 70 Gy, V 65 Gy, and V 60 Gy of 3.9%, 6.2%, 8.1%, and 10.6% compared to the 5-equi plan. In both oropharyngeal cases, the non-coplanar 5-opt plan substantially reduced the V 30 Gy and mean dose to the contralateral parotid compared to the 5-equi, 7-equi, and 9-equi plans: (Stage-III) 8.9%, 7.0%, 8.6% and 4.1 Gy, 2.5 Gy, 2.7 Gy (Stage-IV) 11.2%, 11.2%, 10.8% and 7.8 Gy, 7.9 Gy, 8.0 Gy. In conclusion, the method of optimizing BIV to produce substantial improvements in OAR sparing over conventional IMRT has been demonstrated to be robust for application to a variety of IMRT treatment sites.

A comparison of an algorithm for automated sequential beam orientation selection (Cycle) with simulated annealing

Some time ago we developed and published a new deterministic algorithm (called Cycle) for automatic selection of beam orientations in radiotherapy. This algorithm is a plan generation process aiming at the prescribed PTV dose within hard dose and dose–volume constraints. The algorithm allows a large number of input orientations to be used and selects only the most efficient orientations, surviving the selection process. Efficiency is determined by a score function and is more or less equal to the extent of uninhibited access to the PTV for a specific beam during the selection process. In this paper we compare the capabilities of fast-simulated annealing (FSA) and Cycle for cases where local optima are supposed to be present. Five pancreas and five oesophagus cases previously treated in our institute were selected for this comparison. Plans were generated for FSA and Cycle, using the same hard dose and dose–volume constraints, and the largest possible achieved PTV doses as obtained from these algorithms were compared. The largest achieved PTV dose values were generally very similar for the two algorithms. In some cases FSA resulted in a slightly higher PTV dose than Cycle, at the cost of switching on substantially more beam orientations than Cycle. In other cases, when Cycle generated the solution with the highest PTV dose using only a limited number of non-zero weight beams, FSA seemed to have some difficulty in switching off the unfavourable directions. Cycle was faster than FSA, especially for large-dimensional feasible spaces. In conclusion, for the cases studied in this paper, we have found that despite the inherent drawback of sequential search as used by Cycle (where Cycle could probably get trapped in a local optimum), Cycle is nevertheless able to find comparable or sometimes slightly better treatment plans in comparison with FSA (which in theory finds the global optimum) especially in large-dimensional beam weight spaces.

Direct-aperture optimization applied to selection of beam orientations in intensity-modulated radiation therapy

Direct-aperture optimization (DAO) was applied to iterative beam-orientation selection in intensity-modulated radiation therapy (IMRT), so as to ensure a realistic segmental treatment plan at each iteration. Nested optimization engines dealt separately with gantry angles, couch angles, collimator angles, segment shapes, segment weights and wedge angles. Each optimization engine performed a random search with successively narrowing step sizes. For optimization of segment shapes, the filtered backprojection (FBP) method was first used to determine desired fluence, the fluence map was segmented, and then constrained direct-aperture optimization was used thereafter. Segment shapes were fully optimized when a beam angle was perturbed, and minimally re-optimized otherwise. The algorithm was compared with a previously reported method using FBP alone at each orientation iteration. An example case consisting of a cylindrical phantom with a hemi-annular planning target volume (PTV) showed that for three-field plans, the method performed better than when using FBP alone, but for five or more fields, neither method provided much benefit over equally spaced beams. For a prostate case, improved bladder sparing was achieved through the use of the new algorithm. A plan for partial scalp treatment showed slightly improved PTV coverage and lower irradiated volume of brain with the new method compared to FBP alone. It is concluded that, although the method is computationally intensive and not suitable for searching large unconstrained regions of beam space, it can be used effectively in conjunction with prior class solutions to provide individually optimized IMRT treatment plans.

Computer optimization of noncoplanar beam setups improves stereotactic treatment of liver tumors

To investigate whether computer-optimized fully noncoplanar beam setups may improve treatment plans for the stereotactic treatment of liver tumors. An algorithm for automated beam orientation and weight selection (Cycle) was extended for noncoplanar stereotactic treatments. For 8 liver patients previously treated in our clinic using a prescription isodose of 65%, Cycle was used to generate noncoplanar and coplanar plans with the highest achievable minimum planning target volume (PTV) dose for the clinically delivered isocenter and mean liver doses, while not violating the clinically applied hard planning constraints. The clinical and the optimized coplanar and noncoplanar plans were compared, with respect to D(PTV,99%), the dose received by 99% of the PTV, the PTV generalized equivalent uniform dose (gEUD), and the compliance with the clinical constraints. For each patient, the ratio between D(PTV,99%) and D(isoc), and the gEUD(-5) and gEUD(-20) values of the optimized noncoplanar plan were higher than for the clinical plan with an average increase of respectively 18.8% (range, 7.8-24.0%), 6.4 Gy (range, 3.4-11.8 Gy), and 10.3 Gy (range, 6.7-12.5). D(PTV,99%)/D(isoc), gEUD(-5), and gEUD(-20) of the optimized noncoplanar plan was always higher than for the optimized coplanar plan with an average increase of, respectively, 4.5% (range, 0.2-9.7%), 2.7 Gy (range, 0.6-9.7 Gy), and 3.4 Gy (range, 0.6-9.9 Gy). All plans were within the imposed hard constraints. On average, the organs at risk were better spared with the optimized noncoplanar plan than with the optimized coplanar plan and the clinical plan. The use of automatically generated, fully noncoplanar beam setups results in plans that are favorable compared with coplanar techniques. Because of the automation, we found that the planning workload can be decreased from 1 to 2 days to 1 to 2 h.

SU-FF-T-208: Efficient Selection of Beam Number and Orientations for Intensity Modulated Radiotherapy (IMRT) by Emulating An Ideal Dose Distribution

Purpose: Since radiotherapytreatments only utilize a limited number of beams, it is important that these beams be carefully selected to extract maximum benefit. Here we propose a novel, fast, scheme that selects beam number and orientations for IMRT with the goal of emulating an ideal dose distribution resulting from a very large number of beam orientations. Method and Materials: The procedure starts by optimizing IMRT fluences for a large number of beam orientations (ideal dose distribution). A subset of these orientations is then sequentially selected to emulate this ideal distribution, using an orthogonal QR factorization method (dose‐volume constraints are respected). The use of QR factorization, rather than a full optimization, makes the process time‐efficient. During the procedure, previously selected beams are replaced by beams that are more capable of emulating the ideal distribution. The number of beams is chosen as that above which there are only marginal gains in a least‐squares objective function. The procedure was tested on a head‐and‐neck case (non‐coplanar orientations) and a prostate case (coplanar orientations). Results: For the head‐and‐neck case, the objective function (normalized to 1 for the ideal distribution) decreased with increasing orientations as: 14.43 (3 beams), 9.26 (4), 2.59 (5), 1.81 (6), 1.64 (7). Beyond 6 beams, objective function gains are marginal. The actual plan used to treat this patient yielded a 455% higher objective function, for the same number of beams (6). For the prostate case, the corresponding numbers were: 1.69 (3 beams), 1.30 (4), 1.15 (5), 1.11 (6). In this case 5 beams appear sufficient. The actual clinically used beams were 6% higher (5 beams). Computer time was on the order of 1 –3 hours on a 1 GHz laptop processor. Conclusion: The procedure presented here shows great promise in extracting greater benefit from IMRT by optimizing beam number and orientations.

Effectiveness of noncoplanar IMRT planning using a parallelized multiresolution beam angle optimization method for paranasal sinus carcinoma

To determine the effectiveness of noncoplanar beam configurations and the benefit of plans using fewer but optimally placed beams designed by a parallelized multiple-resolution beam angle optimization (PMBAO) approach. The PMBAO approach uses a combination of coplanar and noncoplanar beam configurations for intensity-modulated radiation therapy (IMRT) treatment planning of paranasal sinus cancers. A smaller number of beams (e.g. 3) are first used to explore the solution space to determine the best and worst beam directions. The results of this exploration are then used as a starting point for determining an optimum beam orientation configuration with more beams (e.g. 5). This process is parallelized using a message passing interface, which greatly reduces the overall computation time for routine clinical practice. To test this approach, treatment for 10 patients with paranasal sinus cancer was planned using a total of 5 beams from a pool of 46 possible beam angles. The PMBAO treatment plans were also compared with IMRT plans designed using 9 equally spaced coplanar beams, which is the standard approach in our clinic. Plans with these two different beam configurations were compared with respect to dose conformity, dose heterogeneity, dose-volume histograms, and doses to organs at risk (i.e., eyes, optic nerve, optic chiasm, and brain). The noncoplanar beam configuration was superior in most paranasal sinus carcinoma cases. The target dose homogeneity was better using a PMBAO 5-beam configuration. However, the dose conformity using PMBAO was not improved and was case dependent. Compared with the 9-beam configuration, the PMBAO configuration significantly reduced the mean dose to the eyes and optic nerves and the maximum dose to the contralateral optical path (e.g. the contralateral eye and optic nerve). The maximum dose to the ipsilateral eye and optic nerve was also lower using the PMBAO configuration than using the 9-beam configuration, although this difference was not significant. The mean doses to the optic chiasm and brain are marginally lower using the PMBAO configuration than using 9-beam configuration. The maximum doses to the optic chiasm and brain are the same with the PMBAO configuration and the 9-beam configuration. Parallelized multiple-resolution beam angle optimization with an optimized noncoplanar beam configuration is an effective and practical approach for IMRT treatment planning. Five-beam treatment plans optimized using the PMBAO are at least equivalent to, and overall better than, the plans using 9 equally spaced coplanar beams.

Automated selection of beam orientations and segmented intensity-modulated radiotherapy (IMRT) for treatment of oesophagus tumors

Background and purpose For some treatment sites, there is evidence in the literature that five to nine equiangular input beam directions are enough for generating IMRT plans. For oesophagus cancer, there is a report showing that going from four to nine beams may even result in lower quality plans. In this paper, our previously published algorithm for automated beam angle selection (Cycle) has been extended to include segmented IMRT. For oesophagus cancer patients, we have investigated whether automated orientation selection from a large number of equiangular input beam directions (up to thirty-six) for IMRT optimisation can result in improved lung sparing. Materials and methods CT-data from five oesophagus patients treated recently in our institute were used for this study. For a prescribed mean PTV dose of 55 Gy, Cycle was used in an iterative procedure to minimise the mean lung dose under the following hard constraints: standard deviation for PTV dose inhomogeneity 2% (1,1 Gy), maximum spinal cord dose 45 Gy. Conformal radiotherapy (CFRT) and IMRT plans for a standard four field oesophagus beam configuration were compared with IMRT plans generated by automated selection from nine or thirty-six equiangular input beam orientations. Comparisons were also made with dose distributions generated with our commercial treatment planning system (TPS), and with observations in the literature. Results Using Cycle, automated orientation selection from nine or thirty-six input beam directions resulted in improved lung sparing compared to the four field set-ups. Compared to selection from nine input orientations, selection from thirty-six directions did always result in lower mean lung doses, sometimes with even fewer non-zero weight beams. On average only seven beams with a non-zero weight were enough for obtaining the lowest mean lung dose, yielding clinically feasible plans even in case of thirty-six input directions for the optimisation process. With our commercial TPS we observed the same contra-intuitive, unfavourable results as reported in the literature; nine field equiangular IMRT plans had substantially higher mean lung doses than plans for the conventional four field set-ups. For all cases, the Cycle plans generated from nine equiangular input directions were superior compared to similar plans generated with our commercial TPS. Conclusions For the studied oesophagus cancer patients the best plans for IMRT were obtained with Cycle, using automated beam orientation selection from thirty-six input beam directions. The lowest mean lung doses could be obtained with, on average, a selection of only seven beams with non-zero weight.

WE‐D‐T‐6E‐05: Geometrically Assisted Selection of Optimum Beams in IMRT Planning by Techniques of Computer Vision

Purpose: To define optimum beam orientations in IMRT by geometric analysis of the PTV and OARs using techniques of computer vision. The method is particularly applicable to situations when one or more OARs are adjacent to the PTV and where non‐coplanar beam orientations are employed, but can be used in cases requiring coplanar beams. The selected beam orientations are those that can most effectively yield desirable dose distributions in the PTV, sparing to a maximum extent adjacent OARs. Method and Materials: For each voxel of a PTV‐OAR interface, Gaussian Invariant techniques are used to obtain the two orthogonal components of a vector that defines a tangent plane. A two‐dimensional histogram of the number of voxels having specific values of those two vector components is formed. Clusters or peaks in the histogram correspond to dominant plane orientations in the interface. Considering the anatomy of a patient whose PTV and OARs have been projected on the surface of a hemisphere, each histogram cluster gives rise to an arc on that surface. Any beam oriented from a point in that arc towards the isocenter is then a possible optimum beam. At the present stage of development, the planner selects one or more optimum orientations for each OAR by using beam's‐eye‐view (BEV) tools. Results: Initial application of the method to clinical cases yields optimizations that are consistent with current clinical practice, with the geometrically assisted definition of optimal beams taking a small fraction of the time usually needed for that task. Conclusion: The work presented can be the basis of a nearly full automatic method of optimal beam orientation selection in which dominant arcs on the hemispheric surface are found by a self‐organizing network and the subsequent BEV analyses are carried out automatically by a rule based system.

WE‐D‐T‐6E‐04: EUD Based Beam Orientation Selection

Purpose: To develop a clinically sensible beam orientation ranking model with incorporation of dose‐volume effects and to show its utility for IMRT beam placement. Materials and methods: Generally, a beamlet/beam is more preferable if it can deliver a higher dose to the target without exceeding the sensitive structure(s) tolerance. In previous geometry‐ or dose‐based approach, the beamlets are treated independently and, to compute the maximally deliverable target dose, each beamlet is pushed to the maximum intensity without considering other beamlets. When volumetric‐structures are involved, there are numerous dose distributions corresponding to the same dose‐volume tolerance and the beamlets are no longer independent. We model a volumetric organ by using EUD and find the beam profile that delivers the maximum target dose without violating the EUD constraints using an iterative algorithm. Four clinical cases are planned with and without the guidance of the angular ranking information and the qualities of the two types of IMRT plans are compared. Results: An angular ranking model with consideration of volumetric effect has been developed. It is shown that the previously reported dose‐based angular ranking represents a special case of the formalism proposed here. Application to four IMRT cases indicated that the proposed technique is capable of producing clinically sensible angular ranking. In all cases, we found that the IMRT plans obtained under the guidance of EUD‐based angular ranking are significantly improved in comparison with that obtained using the conventional uniformly spaced beams. Conclusion: The EUD‐based function is a general approach for angular ranking and allows us to identify the potentially good and bad angles for clinically complicated cases. The ranking can be used either as a guidance to facilitate the beam placement or as prior information to speed up the beam configuration optimization.