An objective function for radiation treatment optimization based on local biological measures

The implementation of biological optimization of radiation treatment plans is impeded by both computational and modelling problems. We derive an objective function from basic model assumptions which includes the normal tissue constraints as interior penalty functions. For organs that are composed of parallel subunits, a mean response model is proposed which leads to constraints similar to dose-volume constraints. This objective function is convex in the case when no parallel organs lie in the treatment volume. Otherwise, an argument is given to show that a number of local minima may exist which are near degenerate to the global minimum. Thus, together with the measure quality of the objective function, highly efficient gradient algorithms can be used. The number of essential biological model parameters could be reduced to a minimum. However, if the optimization constraints are given as TCP/NTCP values, Lagrange multiplier updates have to be performed by invoking comprehensive biological models.     

Derivative-free generation and interpolation of convex Pareto optimal IMRT plans

In inverse treatment planning for intensity-modulated radiation therapy (IMRT), beamlet intensity levels in fluence maps of high-energy photon beams are optimized. Treatment plan evaluation criteria are used as objective functions to steer the optimization process. Fluence map optimization can be considered a multi-objective optimization problem, for which a set of Pareto optimal solutions exists: the Pareto efficient frontier (PEF). In this paper, a constrained optimization method is pursued to iteratively estimate the PEF up to some predefined error. We use the property that the PEF is convex for a convex optimization problem to construct piecewise-linear upper and lower bounds to approximate the PEF from a small initial set of Pareto optimal plans. A derivative-free Sandwich algorithm is presented in which these bounds are used with three strategies to determine the location of the next Pareto optimal solution such that the uncertainty in the estimated PEF is maximally reduced. We show that an intelligent initial solution for a new Pareto optimal plan can be obtained by interpolation of fluence maps from neighbouring Pareto optimal plans. The method has been applied to a simplified clinical test case using two convex objective functions to map the trade-off between tumour dose heterogeneity and critical organ sparing. All three strategies produce representative estimates of the PEF. The new algorithm is particularly suitable for dynamic generation of Pareto optimal plans in interactive treatment planning.     

A deterministic iterative least-squares algorithm for beam weight optimization in conformal radiotherapy

Currently, inverse treatment planning in conformal radiotherapy is, in part, a trial-and-error process due to the interplay of many competing criteria for obtaining a clinically acceptable dose distribution. A new method is developed for beam weight optimization that incorporates clinically relevant nonlinear and linear constraints. The process is driven by a nonlinear, quasi-quadratic objective function and the solution space is defined by a set of linear constraints. At each step of iteration, the optimization problem is linearized by a self-consistent approximation that is local to the existing dose distribution. The dose distribution is then improved by solving a series of constrained least-squares problems using an established method until all prescribed constraints are satisfied. This differs from the current approaches in that it does not rely on the search for the global minimum of a specific objective function. Essentially, our proposed objective function can be construed as a functional that comprises a class of dose-based quadratic objective functions. Empirical adjustment for appropriate model parameters in the construction of objective function is minimized, since these parameters are in effect adaptively adjusted during optimization. The method is robust in solving difficult clinical cases using either aperture or pencil beam based planning techniques for intensity-modulated radiation therapy.     

Inverse treatment planning by physically constrained minimization of a biological objective function

In the current state-of-the art of clinical inverse planning, the design of clinically acceptable IMRT plans is predominantly based on the optimization of physical rather than biological objective functions. A major impetus for this trend is the unproven predictive power of radiobiological models, which is largely due to the scarcity of data sets for an accurate evaluation of the model parameters. On the other hand, these models do capture the currently known dose-volume effects in tissue dose-response, which should be accounted for in the process of optimization. In order to incorporate radiobiological information in clinical treatment planning optimization, we propose a hybrid physico-biological approach to inverse treatment planning based on the application of a continuous penalty function method to the constrained minimization of a biological objective. The objective is defined as the weighted sum of normal tissue complication probabilities evaluated with the Lyman normal-tissue complication probability model. Physical constraints specify the admissible minimum and maximum target dose. The continuous penalty function method is then used to find an approximate solution of the resulting large-scale constrained minimization problem. Plans generated by our approach are compared to ones produced by a commercial planning system incorporating physical optimization. The comparisons show clinically negligible differences, with the advantage that the hybrid technique does not require specifications of any dose-volume constraints to the normal tissues. This indicates that the proposed hybrid physico-biological method can be used for the generation of clinically acceptable plans.     

A unified approach for inversion problems in intensity-modulated radiation therapy

We propose and study a unified model for handling dose constraints (physical dose, equivalent uniform dose (EUD), etc) and radiation source constraints in a single mathematical framework based on the split feasibility problem. The model does not impose on the constraints an exogenous objective (merit) function. The optimization algorithm minimizes a weighted proximity function that measures the sum of the squares of the distances to the constraint sets. This guarantees convergence to a feasible solution point if the split feasibility problem is consistent (i.e., has a solution), or, otherwise, convergence to a solution that minimally violates the physical dose constraints and EUD constraints. We present computational results that demonstrate the validity of the model and the power of the proposed algorithmic scheme.     

Multiobjective anatomy-based dose optimization for HDR-brachytherapy with constraint free deterministic algorithms

In high dose rate (HDR) brachytherapy, conventional dose optimization algorithms consider multiple objectives in the form of an aggregate function that transforms the multiobjective problem into a single-objective problem. As a result, there is a loss of information on the available alternative possible solutions. This method assumes that the treatment planner exactly understands the correlation between competing objectives and knows the physical constraints. This knowledge is provided by the Pareto trade-off set obtained by single-objective optimization algorithms with a repeated optimization with different importance vectors. A mapping technique avoids non-feasible solutions with negative dwell weights and allows the use of constraint free gradient-based deterministic algorithms. We compare various such algorithms and methods which could improve their performance. This finally allows us to generate a large number of solutions in a few minutes. We use objectives expressed in terms of dose variances obtained from a few hundred sampling points in the planning target volume (PTV) and in organs at risk (OAR). We compare two- to four-dimensional Pareto fronts obtained with the deterministic algorithms and with a fast-simulated annealing algorithm. For PTV-based objectives, due to the convex objective functions, the obtained solutions are global optimal. If OARs are included, then the solutions found are also global optimal, although local minima may be present as suggested.     

Feasibility of a fast inverse dose optimization algorithm for IMRT via matrix inversion without negative beamlet intensities

A fast optimization algorithm is very important for inverse planning of intensity modulated radiation therapy (IMRT), and for adaptive radiotherapy of the future. Conventional numerical search algorithms such as the conjugate gradient search, with positive beam weight constraints, generally require numerous iterations and may produce suboptimal dose results due to trapping in local minima. A direct solution of the inverse problem using conventional quadratic objective functions without positive beam constraints is more efficient but will result in unrealistic negative beam weights. We present here a direct solution of the inverse problem that does not yield unphysical negative beam weights. The objective function for the optimization of a large number of beamlets is reformulated such that the optimization problem is reduced to a linear set of equations. The optimal set of intensities is found through a matrix inversion, and negative beamlet intensities are avoided without the need for externally imposed ad-hoc constraints. The method has been demonstrated with a test phantom and a few clinical radiotherapy cases, using primary dose calculations. We achieve highly conformal primary dose distributions with very rapid optimization times. Typical optimization times for a single anatomical slice (two dimensional) (head and neck) using a LAPACK matrix inversion routine in a single processor desktop computer, are: 0.03 s for 500 beamlets; 0.28 s for 1000 beamlets; 3.1 s for 2000 beamlets; and 12 s for 3000 beamlets. Clinical implementation will require the additional time of a one-time precomputation of scattered radiation for all beamlets, but will not impact the optimization speed. In conclusion, the new method provides a fast and robust technique to find a global minimum that yields excellent results for the inverse planning of IMRT.     

Speed and convergence properties of gradient algorithms for optimization of IMRT

Gradient algorithms are the most commonly employed search methods in the routine optimization of IMRT plans. It is well known that local minima can exist for dose-volume-based and biology-based objective functions. The purpose of this paper is to compare the relative speed of different gradient algorithms, to investigate the strategies for accelerating the optimization process, to assess the validity of these strategies, and to study the convergence properties of these algorithms for dose-volume and biological objective functions. With these aims in mind, we implemented Newton's, conjugate gradient (CG), and the steepest decent (SD) algorithms for dose-volume- and EUD-based objective functions. Our implementation of Newton's algorithm approximates the second derivative matrix (Hessian) by its diagonal. The standard SD algorithm and the CG algorithm with "line minimization" were also implemented. In addition, we investigated the use of a variation of the CG algorithm, called the "scaled conjugate gradient" (SCG) algorithm. To accelerate the optimization process, we investigated the validity of the use of a "hybrid optimization" strategy, in which approximations to calculated dose distributions are used during most of the iterations. Published studies have indicated that getting trapped in local minima is not a significant problem. To investigate this issue further, we first obtained, by trial and error, and starting with uniform intensity distributions, the parameters of the dose-volume- or EUD-based objective functions which produced IMRT plans that satisfied the clinical requirements. Using the resulting optimized intensity distributions as the initial guess, we investigated the possibility of getting trapped in a local minimum. For most of the results presented, we used a lung cancer case. To illustrate the generality of our methods, the results for a prostate case are also presented. For both dose-volume and EUD based objective functions, Newton's method far outperforms other algorithms in terms of speed. The SCG algorithm, which avoids expensive "line minimization," can speed up the standard CG algorithm by at least a factor of 2. For the same initial conditions, all algorithms converge essentially to the same plan. However, we demonstrate that for any of the algorithms studied, starting with previously optimized intensity distributions as the initial guess but for different objective function parameters, the solution frequently gets trapped in local minima. We found that the initial intensity distribution obtained from IMRT optimization utilizing objective function parameters, which favor a specific anatomic structure, would lead to a local minimum corresponding to that structure. Our results indicate that from among the gradient algorithms tested, Newton's method appears to be the fastest by far. Different gradient algorithms have the same convergence properties for dose-volume- and EUD-based objective functions. The hybrid dose calculation strategy is valid and can significantly accelerate the optimization process. The degree of acceleration achieved depends on the type of optimization problem being addressed (e.g., IMRT optimization, intensity modulated beam configuration optimization, or objective function parameter optimization). Under special conditions, gradient algorithms will get trapped in local minima, and reoptimization, starting with the results of previous optimization, will lead to solutions that are generally not significantly different from the local minimum.     

A novel approach to multi-criteria inverse planning for IMRT

Treatment plan optimization is a multi-criteria process. Optimizing solely on one objective or on a sum of a priori weighted objectives may result in inferior treatment plans. Manually adjusting weights or constraints in a trial and error procedure is time consuming. In this paper we introduce a novel multi-criteria optimization approach to automatically optimize treatment constraints (dose-volume and maximum-dose). The algorithm tries to meet these constraints as well as possible, but in the case of conflicts it relaxes lower priority constraints so that higher priority constraints can be met. Afterwards, all constraints are tightened, starting with the highest priority constraints. Applied constraint priority lists can be used as class solutions for patients with similar tumour types. The presented algorithm does iteratively apply an underlying algorithm for beam profile optimization, based on a quadratic objective function with voxel-dependent importance factors. These voxel-dependent importance factors are automatically adjusted to reduce dose-volume and maximum-dose constraint violations.     

Are dose calculations during dose optimization in brachytherapy necessary?

We consider the problem of anatomy based dose optimization in brachytherapy. A calculation method for some objective functions and their derivatives is proposed which significantly reduces the number of required operations. The optimization in some cases, ignoring a preprocessing step, is independent of the number of sampling points. The idea is that some of the objectives and their derivatives used for dose optimization do not require the explicit calculation of dose values. Dose optimization with the new modified computation method for the objectives and derivatives is, depending on the number of sampling points, up to 100 times faster than the conventional method with dose calculation.     

A theoretical approach to the problem of dose-volume constraint estimation and their impact on the dose-volume histogram selection

This paper outlines a theoretical approach to the problem of estimating and choosing dose-volume constraints. Following this approach, a method of choosing dose-volume constraints based on biological criteria is proposed. This method is called "reverse normal tissue complication probability (NTCP) mapping into dose-volume space" and may be used as a general guidance to the problem of dose-volume constraint estimation. Dose-volume histograms (DVHs) are randomly simulated, and those resulting in clinically acceptable levels of complication, such as NTCP of 5 +/- 0.5%, are selected and averaged producing a mean DVH that is proven to result in the same level of NTCP. The points from the averaged DVH are proposed to serve as physical dose-volume constraints. The population-based critical volume and Lyman NTCP models with parameter sets taken from literature sources were used for the NTCP estimation. The impact of the prescribed value of the maximum dose to the organ, D(max), on the averaged DVH and the dose-volume constraint points is investigated. Constraint points for 16 organs are calculated. The impact of the number of constraints to be fulfilled based on the likelihood that a DVH satisfying them will result in an acceptable NTCP is also investigated. It is theoretically proven that the radiation treatment optimization based on physical objective functions can sufficiently well restrict the dose to the organs at risk, resulting in sufficiently low NTCP values through the employment of several appropriate dose-volume constraints. At the same time, the pure physical approach to optimization is self-restrictive due to the preassignment of acceptable NTCP levels thus excluding possible better solutions to the problem.     

A new column-generation-based algorithm for VMAT treatment plan optimization

We study the treatment plan optimization problem for volumetric modulated arc therapy (VMAT). We propose a new column-generation-based algorithm that takes into account bounds on the gantry speed and dose rate, as well as an upper bound on the rate of change of the gantry speed, in addition to MLC constraints. The algorithm iteratively adds one aperture at each control point along the treatment arc. In each iteration, a restricted problem optimizing intensities at previously selected apertures is solved, and its solution is used to formulate a pricing problem, which selects an aperture at another control point that is compatible with previously selected apertures and leads to the largest rate of improvement in the objective function value of the restricted problem. Once a complete set of apertures is obtained, their intensities are optimized and the gantry speeds and dose rates are adjusted to minimize treatment time while satisfying all machine restrictions. Comparisons of treatment plans obtained by our algorithm to idealized IMRT plans of 177 beams on five clinical prostate cancer cases demonstrate high quality with respect to clinical dose-volume criteria. For all cases, our algorithm yields treatment plans that can be delivered in around 2?min. Implementation on a graphic processing unit enables us to finish the optimization of a VMAT plan in 25-55?s.     

Treatment planning optimization using constrained simulated annealing.

A variation of simulated annealing optimization called 'constrained simulated annealing' is used with a simple annealing schedule to optimize beam weights and angles in radiation therapy treatment planning. Constrained simulated annealing is demonstrated using two contrasting objective functions which incorporate both biological response and dose-volume considerations. The first objective function maximizes the probability of a complication-free treatment (PCFT) by minimizing the normal tissue complications subject to the constraint that the entire target volume receives a prescribed minimum turmourcidal dose with a specified dose homogeneity. Probabilities of normal tissue complication are based on published normal tissue complication probability functions and computed from dose-volume histograms. The second objective function maximizes the isocentre dose subject to a set of customized normal tissue dose-volume and target volume dose homogeneity constraints (MVDL). Although the PCFT objective function gives consistently lower estimates of normal tissue complication probabilities, the ability to specify individualized dose-volume limits, and therefore the individualized probability of complication, for an individual organ makes the MDVL objective function more useful for treatment planning.     

Optimal drug regimens in cancer chemotherapy: A multi-objective approach

In this paper, based on two mathematical models, a set of optimal anticancer drug regimens is designed. Two important difficulties in treating cancers with chemotherapy are drug resistance and toxicity. They were considered in our treatment design; a compartmental model was extended to appropriately describe drug resistance and two constraints were imposed on the value of the anticancer drug dynamics to avoid toxicity. Indeed, we optimized two objective functions simultaneously in order to minimize the size of the tumor as well as the side effects of the anticancer drug on the patient's body. Operating the described task, a bi-objective optimization problem was defined and solved by an evolutionary method. This optimization led to a set of optimal drug regimens, which can be used in several chemotherapy treatments. Numerical simulations were given to illustrate the flexibility of the method.     

The generalized equivalent uniform dose function as a basis for intensity-modulated treatment planning

The efficiency of intensity-modulated radiation therapy (IMRT) treatment planning depends critically on the presence or absence of multiple local minima in the feasible search space. We analyse the convexity of the generalized equivalent uniform dose equation (Niemierko A 1999 Med. Phys. 26 1100) when used either in the objective function or in the constraints. The practical importance of this analysis is that convex objective functions minimized over convex feasibility spaces do not have multiple local minima, likewise for concave objective functions maximized over convex feasibility spaces. Both of these situations are referred to as 'convex problems' and computationally efficient local search methods can be used for their solution. We also show that the Poisson-based tumour control probability objective function is strictly concave (if one neglects inter-patient heterogeneity), and hence it implies a single local minimum if maximized over a convex feasibility space. Even when including inter-patient heterogeneity, multiple local minima, although theoretically possible, are expected to be of minimal concern. The generalized equivalent uniform dose function (EUDa) is proved to be convex or concave depending on its only parameter a: when a is equal to or greater than 1, minimizing EUDa, on a convex feasibility space leads to a single minimum; when a is less than 1, maximizing EUDa, on a convex feasibility space leads to a single minimum. We also study a recently proposed practical, yet difficult, IMRT treatment planning formulation: unconstrained optimization of the objective function proposed by Wu et al (2002 Int. J. Radiat. Oncol. Biol. Phys. 52 224-35), which is expressed in terms of the EUDa for the target and normal tissues. This formulation may theoretically lead to multiple local minima. We propose a procedure for improving resulting solutions based on the convexity properties of the underlying objective function terms.     

A novel linear programming approach to fluence map optimization for intensity modulated radiation therapy treatment planning

We present a novel linear programming (LP) based approach for efficiently solving the intensity modulated radiation therapy (IMRT) fluence-map optimization (FMO) problem to global optimality. Our model overcomes the apparent limitations of a linear-programming approach by approximating any convex objective function by a piecewise linear convex function. This approach allows us to retain the flexibility offered by general convex objective functions, while allowing us to formulate the FMO problem as a LP problem. In addition, a novel type of partial-volume constraint that bounds the tail averages of the differential dose-volume histograms of structures is imposed while retaining linearity as an alternative approach to improve dose homogeneity in the target volumes, and to attempt to spare as many critical structures as possible. The goal of this work is to develop a very rapid global optimization approach that finds high quality dose distributions. Implementation of this model has demonstrated excellent results. We found globally optimal solutions for eight 7-beam head-and-neck cases in less than 3 min of computational time on a single processor personal computer without the use of partial-volume constraints. Adding such constraints increased the running times by a factor of 2-3, but improved the sparing of critical structures. All cases demonstrated excellent target coverage (> 95%), target homogeneity (< 10% overdosing and < 7% underdosing) and organ sparing using at least one of the two models.     

Including robustness in multi-criteria optimization for intensity-modulated proton therapy

We present a method to include robustness in a multi-criteria optimization (MCO) framework for intensity-modulated proton therapy (IMPT). The approach allows one to simultaneously explore the trade-off between different objectives as well as the trade-off between robustness and nominal plan quality. In MCO, a database of plans each emphasizing different treatment planning objectives, is pre-computed to approximate the Pareto surface. An IMPT treatment plan that strikes the best balance between the different objectives can be selected by navigating on the Pareto surface. In our approach, robustness is integrated into MCO by adding robustified objectives and constraints to the MCO problem. Uncertainties (or errors) of the robust problem are modeled by pre-calculated dose-influence matrices for a nominal scenario and a number of pre-defined error scenarios (shifted patient positions, proton beam undershoot and overshoot). Objectives and constraints can be defined for the nominal scenario, thus characterizing nominal plan quality. A robustified objective represents the worst objective function value that can be realized for any of the error scenarios and thus provides a measure of plan robustness. The optimization method is based on a linear projection solver and is capable of handling large problem sizes resulting from a fine dose grid resolution, many scenarios, and a large number of proton pencil beams. A base-of-skull case is used to demonstrate the robust optimization method. It is demonstrated that the robust optimization method reduces the sensitivity of the treatment plan to setup and range errors to a degree that is not achieved by a safety margin approach. A chordoma case is analyzed in more detail to demonstrate the involved trade-offs between target underdose and brainstem sparing as well as robustness and nominal plan quality. The latter illustrates the advantage of MCO in the context of robust planning. For all cases examined, the robust optimization for each Pareto optimal plan takes less than 5 min on a standard computer, making a computationally friendly interface possible to the planner. In conclusion, the uncertainty pertinent to the IMPT procedure can be reduced during treatment planning by optimizing plans that emphasize different treatment objectives, including robustness, and then interactively seeking for a most-preferred one from the solution Pareto surface.     

Beam orientation optimization for IMRT by a hybrid method of the genetic algorithm and the simulated dynamics

We have developed a new method for beam orientation optimization in intensity-modulated radiation therapy (IMRT). The problem of beam orientation optimization in IMRT is solved by a decoupled two-step iterative process: (1) optimization of the intensity profiles for given beam configurations; (2) selection of optimal beam configurations based on the ranking by an objective function score for the results of the intensity profile optimization. The simulated dynamics algorithm is used for the intensity profile optimization. This algorithm enforces both the hard constraints and dose-volume constraints. A genetic algorithm is used to select beam orientation configurations. The method has been tested for both a simulated and clinical case, and the results show that beam orientation optimization significantly improved IMRT plans within a time period that is clinically acceptable. The results also show the dependence of the optimal orientation configurations on the prescribed constraints. In addition, beam orientation optimization by the method described here can provide multiple plans with similar dose distributions. This degeneracy characteristic can be exploited to our advantage in introducing additional planning objectives, e.g., the smoothness of intensity profiles, for the selection of the optimal plan among the degenerate configurations for treatment delivery.     

Clinically relevant optimization of 3-D conformal treatments.

In this paper a method of computer-aided optimization of 3-D conformal treatment plans is presented which incorporates models to predict the clinical consequences of resulting dose distributions. Even though these models are simplistic, it is submitted that their intelligent use leads to treatment plans which indicate lower normal tissue complications and higher tumor control. Dose distribution data, biological models, and observed normal tissue and tumor response data are used to compute tumor control and normal tissue complication probabilities for each of the critical normal structures encountered in a treatment plan. These quantities are combined into a single score using an objective function which incorporates the importance of each end point as assessed by the physician. Using the "simulated annealing" method of optimization, the beam weights are adjusted to maximize the score. Additional constraints are applied to ensure consistency of the results of optimization with the judgment of the physician. These optimization methods have been applied to conformal treatment plans consisting of multiple fixed fields with conformal field shaping. The results indicate that the methods presented have considerable potential.     

Tools for the analysis of dose optimization: II. Sensitivity analysis

Dose optimization requires that the treatment goals be specified in a meaningful manner, but also that alterations to the specification lead to predictable changes in the resulting dose distribution. Within the framework of constrained optimization, it is possible to devise a tool that quantifies the impact on the objective of target volume coverage of any change to a dosimetric constraint of normal tissue or target dose homogeneity. This sensitivity analysis relies on properties of the Lagrange function that is associated with the constrained optimization problem, but does not depend on the method used to solve this problem. It is useful particularly in cases with multiple target volumes and critical normal structures, where constraints and objectives can interact in a non-intuitive manner.