Search

Purpose: In radiation therapytreatment planning, the clinical objectives of uniform high dose to the planning target volume (PTV) and low dose to the organs-at-risk (OARs) are invariably in conflict, often requiring compromises to be made between them when selecting the best treatment plan for a particular patient. In this work, the authors introduce Pareto-Aware Radiotherapy Evolutionary TreatmentOptimization (pareto), a multiobjective optimization tool to solve for beam angles and fluence patterns in intensity-modulated radiation therapy(IMRT)treatment planning. Methods: pareto is built around a powerful multiobjective genetic algorithm (GA), which allows us to treat the problem of IMRTtreatment plan optimization as a combined monolithic problem, where all beam fluence and angle parameters are treated equally during the optimization. We have employed a simple parameterized beam fluence representation with a realistic dose calculation approach, incorporating patient scatter effects, to demonstrate feasibility of the proposed approach on two phantoms. The first phantom is a simple cylindrical phantom containing a target surrounded by three OARs, while the second phantom is more complex and represents a paraspinal patient. Results: pareto results in a large database of Pareto nondominated solutions that represent the necessary trade-offs between objectives. The solution quality was examined for several PTV and OAR fitness functions. The combination of a conformity-based PTV fitness function and a dose-volume histogram (DVH) or equivalent uniform dose (EUD) -based fitness function for the OAR produced relatively uniform and conformal PTV doses, with well-spaced beams. A penalty function added to the fitness functions eliminates hotspots. Comparison of resulting DVHs to those from treatment plans developed with a single-objective fluence optimizer (from a commercial treatment planning system) showed good correlation. Results also indicated that pareto shows promise in optimizing the number of beams. Conclusions: This initial evaluation of the evolutionary optimization software toolpareto for IMRTtreatment planning demonstrates feasibility and provides motivation for continued development. Advantages of this approach over current commercial methods for treatment planning are many, including: (1) fully automated optimization that avoids human controlled iterative optimization and potentially improves overall process efficiency, (2) formulation of the problem as a true multiobjective one, which provides an optimized set of Pareto nondominated solutions refined over hundreds of generations and compiled from thousands of parameter sets explored during the run, and (3) rapid exploration of the final nondominated set accomplished by a graphical interface used to select the best treatment option for the patient.

Purpose: To determine the best parameterization of fluence modulation for PARETO (Pareto‐Aware Radiotherapy Evolutionary Treatment Optimization) multi‐objective treatment planningsoftware, as judged by the complexity of each parameterization, the efficiency, and the quality of the plans produced. Methods: PARETO simultaneously optimizes the number of beams, beam orientations, and fluence modulation. For this investigation, we have tested four fluence parameterizations on a paraspinal tumour phantom and a spine patient data set using a pre‐determined number of beams. The first method applies linear gradients over the projection of Planning Target Volume (PTV) in the beam's‐eye‐view plane. The second method applies the two‐dimensional inverse cosine transform of a few low‐frequency parameters. Another method interpolates between the intensities of a coarse grid of square beamlets. Also, an isodose‐based contour method defines regions on the fluence maps that are assigned different intensities. Methods which produce plans of the best quality (as judged by dose‐volume histograms) will have a similar distribution of non‐dominated solutions. The complexity of each parameterization is determined by the number of parameters, and the efficiency is determined by the run time. Results: For both geometries, we found that the set of non‐dominated solutions of each parameterization overlapped in the projection of the PTV conformity fitness function and the quadrature‐averaged Organs‐At‐Risk (OAR) fitness function. The beam group and cosine transform methods produced some plans which did slightly better in simultaneously achieving good PTV conformity and OAR dose sparing. The cosine transform and linear gradient methods proved to be the most efficient. The cosine transform method also has the least number of parameters per beam. Conclusions: We have found that several different parameterizations of fluence modulation produce non‐dominated solutions of similar quality. The cosine transform method is the best choice for efficiently producing solutions that do well in PTV conformity and OAR sparing.

Purpose: We introduce a novel multi‐objective treatment planning system called PARETO (Pareto‐Aware Radiotherapy Evolutionary Treatment Optimization), which simultaneously optimizes beam angles and fluence patterns by treating the PTV conformity and dose to OARs as separate objectives that are optimized by a powerful multi‐objective parallel genetic algorithm (GA).Methods: PARETO treats radiotherapytreatment planning as a single monolithic optimization problem, in which beam angle configurations and fluence patterns are explored to simultaneously optimize the PTV dose conformity and a dose objective for each OAR. We use a Pareto‐ranking scheme to discover necessary compromises between objectives. A final non‐dominated database of solutions is compiled from all plans evaluated during the run such that no solution in the database is superior to any other solution in all objectives. A graphical tool allows rapid navigation of the pre‐optimized database to select a final treatment option for the patient.Results: PARETO is at the stage of a working prototype. Solutions are of high quality, as judged by DVH curves and dose distribution, and consistent results are obtained between runs. Only minor differences in trade‐off surfaces result from four different fluence parameterizations of varying complexity, but they differ by a factor of ∼4 in speed. A novel feature allows optimization of the number of beams and a realistic test case shows that no further improvements in conformity are found for more than ∼9 beams. A newly implemented GPU‐based ray tracer, including convolution with a patient dose kernel, results in a speedup factor of ∼3 compared to CPU‐based computation. Conclusions: This work demonstrates PARETO's feasibility as a treatment‐planning tool, which replaces manual iterative optimization of treatment plans with a rapid graphical exploration of pre‐optimized solutions. Clinically acceptable run times of less than an hour appear within reach by combining several of our GPU‐based systems in parallel. J. Fiege discloses authorship of the Ferret GA (used by PARETO) and ownership of nQube Technical Computing Corporation, which distributes this optimization software.

Intensity-modulated radiation therapy (IMRT) treatment planning requires tradeoffs to be made between delivering a prescribed dose to the planning target volume (PTV) and sparing the organ's-at-risk (OARs). Traditionally in clinical practice, treatment planners manually optimize beam orientations, objectives, and/or weights in a time-consuming, trial-and-error process to find some acceptable compromise, with no guarantee that this solution is actually optimal. We propose a novel and powerful fluence and beam orientation optimization package for radiotherapy optimization, called PARETO (Pareto-Aware Radiotherapy Evolutionary Treatment Optimization), which consists of a multi-objective genetic algorithm capable of optimizing several objective functions simultaneously and mapping the structure of their trade-off surface efficiently and in detail. Fitness functions, based on mean dose for the OARs and PTV, as well as fluence gradients are optimized. PARETO intelligently varies all beam orientations and beam fluence to simultaneously optimize all objectives. Over many generations, the entire family of Pareto-optimal treatment plans, spanning a multi-dimensional trade-off surface, is mapped out. Pareto-optimal solutions are stored in a database and trade-offs between the competing objectives can be visualized graphically and explored. The efficacy of the solutions provided by PARETO was evaluated using a commercial treatment planning system with five coplanar IMRT treatment plans for a homogenous phantom consisting of three OARs surrounding a central PTV. This work demonstrated that the fitness functions within PARETO have a strong correlation to the dose distribution. Thus, from many Pareto-optimal plans, the clinician may select the plan which they decide is the most appropriate multi-objective compromise for a patient.

Multi‐objective radiotherapy beam angle optimization is currently unavailable in commercial treatment planningsoftware. The PARETO (Pareto‐Aware Radiotherapy Evolutionary Treatment Optimization)software package presented here uses a sophisticated evolutionary algorithm that is capable of handling this difficult computational problem, while simultaneously incorporating IMRT fluence optimization. In this work, we focus on establishing the validity of the PARETO software in obtaining reasonable beam fluence maps for fixed gantry IMRT. PARETO IMRT solutions for a simple cylindrical phantom containing various structures (one planning target volume and one‐to‐three organs‐at‐risk) are compared to resulting fluence patterns at fixed beam angles generated by a commercial treatment planning system. We find that PARETO solutions agree well with those of the commercial system, with differences mainly attributable to the differences in the underlying dose calculation algorithm. Our approach to fluence modulation will allow PARETO to generate a database of optimal solutions for beam angles and IMRT plans that represent the best possible trade‐offs between competing treatment objectives.

Intensity modulated radiation therapy(IMRT) aims to deliver a uniform prescribed dose of radiation to a planning target volume (PTV), while also minimizing the dose to each nearby organ at risk (OAR). IMRTtreatment planning is therefore a multi‐objective optimization problem, which can be addressed by powerful multi‐objective optimization techniques from the field of evolutionary computing. We introduce a software package called PARETO (Pareto‐Aware Radiotherapy Evolutionary Optimization), under development at the University of Manitoba and CancerCare Manitoba, which uses a sophisticated multi‐objective genetic algorithm called Ferret to find Pareto‐optimal beam orientations and fluence maps for IMRTtreatment planning. We discuss our fitness functions and a novel geometric method for parameterizing fluence maps for use with the Ferret Genetic Algorithm. We show that our method replaces manual iterative optimization methods currently used with a faster navigation of a pre‐optimized database of solutions. We present an illustrative example that applies the code to a geometric phantom with three OARs, and show that the PARETO finds two distinct classes of solutions.

The quality of an IMRT treatment plan is ultimately limited by the trial‐and‐error planning process undertaken by the treatment planner using current treatment planning systems (TPS). We propose a novel multi‐objective genetic algorithm to optimize beam orientations and improve the quality and efficiency of IMRTtreatment planning. Our approach generates a database of Pareto‐optimal solutions for each patient, to be utilized by the human operator during IMRTtreatment planning for the graphical exploration of trade‐offs between multiple planning objectives. A 3D patient structure matrix is generated using the coordinates of the region‐of‐interest contours from patient plans stored in the TPS. A sphere of possible radiation beam orientations is pre‐computed and used to define a solution space. Multi‐objective optimization is achieved through a sophisticated genetic algorithm, called “Ferret”, which incorporates fitness functions based on clinical objectives for the planning target volume and the organ's‐at‐risk. Ferret intelligently varies all beam orientations to simultaneously optimize all objectives. Over many generations, the algorithm maps out the entire family of Pareto‐optimal solutions spanning a multi‐dimensional trade‐off surface. These optimal solutions are stored in a database where trade‐offs between the competing clinical objectives can be visualized graphically. The efficacy of the Pareto‐optimal solutions provided by Ferret was evaluated through a retrospective study comparing the quality of these treatment plans to standard plans for gastric, prostate, and oropharyngeal cancer patients. In each case, the utilization of the solution database and the graphical exploration of trade‐offs between multiple planning objectives produced superior treatment plans compared to conventional plans.