Multi-Objective Radiation Therapy Optimization Method

• US 20130197878A1
• Filed: 06/07/2011
• Published: 08/01/2013
• Est. Priority Date: 06/07/2010
• Status: Active Grant

• Current Assignee: Andrew Cull, Boyd Mccurdy, Heather Champion, Jason Fiege, Murphy Berzish, Peter Potrebko
• Original Assignee: Andrew Cull, Boyd Mccurdy, Heather Champion, Jason Fiege, Murphy Berzish, Peter Potrebko
• Inventors: Fiege, Jason, McCurdy, Boyd, Potrebko, Peter, Cull, Andrew, Champion, Heather, Berzish, Murphy
• Granted Patent: US 9,507,886 B2
• Time in Patent Office: Days
• US Class Current: 703/2
• CPC Class Codes: A61N 2005/1032 Genetic optimization methods
• CPC Class Codes: A61N 5/1031 using a specific method of dose optimization
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G16H 20/40 relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
• CPC Class Codes: G16H 70/20 relating to practices or guidelines

Abstract

A novel and powerful fluence and beam orientation optimization package for radiotherapy optimization, called PARETO (Pareto-Aware Radiotherapy Evolutionary Treatment Optimization), makes use 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. PARETO generates a database of Pareto non-dominated solutions and allows the graphical exploration of trade-offs between multiple planning objectives during IMRT treatment planning PARETO offers automated and truly multi-objective treatment plan optimization, which does not require any objective weights to be chosen, and therefore finds a large sample of optimized solutions defining a trade-off surface, which represents the range of compromises that are possible.

Claims

1. A radiation therapy optimization method for a multi-objective optimization problem comprising a planning-target-volume surrounded by organs-at-risk in a patient to be treated using radiation, the method comprising:

representing computed tomography data of the patient as a three-dimensional grid of coordinates divided into voxels which contain the planning-target-volume and the organs-at-risk;

using a multi-objective optimizer to search successive generations of trial solutions to generate a database of optimized solutions that form a Pareto non-dominated set of solutions, which sample an estimation of a Pareto front to the multi-objective optimization problem where all solutions on the Pareto front are regarded as equally optimal, each trial solution being specified by a set of parameters defining beam orientations and fluence patterns which together define the radiation;

evaluating each generation of trial solutions by;

estimating a radiation dose proxy to each voxel for each trial solution to the multi-objective optimization problem;

associating a fitness function with each of the organs-at risk and the planning-target-volume;

calculating a set of fitness values for each trial solution by evaluating the fitness functions associated with the trial solution using the respective radiation dose proxies of the respective voxels associated with the trial solution such that the fitness values are optimized with respect to Pareto-optimality of the trial solutions; and

determining the Pareto non-dominated set of solutions to be optimized solutions by comparing the fitness values of the trial solutions to one another and to the fitness values of the optimized solutions of previous generations of the trial solutions;

continuing to generate successive generations of the trial solutions according to a defined convergence criterion until the optimized solutions are determined according to prescribed criteria to be sufficiently approximate Pareto-optimal solutions to the multi-objective optimization problem; and

displaying the Pareto non-dominated set of solutions to a user by means of an interactive graphical interface.

2. The method according to claim 1 wherein the set of parameters defining the radiation of each trial solution includes beam orientations and fluence patterns for each one of a number of beams of radiation.

4. The method according to claim 2 wherein the number of beams of radiation is optimized by:

first beams closer than a certain minimum angle being merged together to form a single beam at an intermediate angle defined by fluence parameters constructed by averaging together the parameters of the first beams;

a penalty function being defined to prefer solutions requiring a fewer number of beams;

beams being sorted by gantry angle, thus rearranging the order of their defining parameters, and being re-inserted into the genetic algorithm population;

the size of the parameter space being reduced by sorting beams by gantry angle; and

beam averaging being applied recursively.

21. The method according to claim 2 including fluence for each beam over a variable PTV margin size being defined by a parameter that is optimized during the run according to the fitness functions such that the fluence is zero external to the margin.

22. The method according to claim 2 including rejecting beam orientations that cannot be realized by treatment hardware by setting all fitness values to a large number that increases with the degree to which hardware constraints are violated.

24. The method according to claim 2 including estimating the radiation dose proxy to each voxel by using rays in a ray-tracing algorithm to determine which voxels are intersected by each beam and using a compact approximation of the exponential function to estimate the exponential function to calculated radiation intensities.

29. The method according to claim 1 including,

specifying an organ type for each organ-at-risk and associating a fitness function with each organ-at-risk based on the specified organ type;

and

defining a simple volumetric fitness function for volumetric organs-at-risk by the following equation;

30. The method according to claim 1 including:

specifying an organ type for each organ-at-risk and associating a fitness function with each organ-at-risk based on the specified organ type; and

defining a simple serial fitness function for serial organs-at-risk by the following equation;

31. The method according to claim 1 including defining a DVH fitness function for volumetric or serial organs-at-risk by the following equation:

33. The method according to claim 1 including defining a conformity fitness function of the planning-target-volume by the following equation:

35. The method according to claim 1 further comprising:

defining a hot zone as the set of voxels near the PTV such that d_i<

=f_PTV*d_PTV where d_i is the distance from voxel i to the nearest voxel within the PTV;

f_PTV is a fractional value less than or equal to 0.5; and

d_PTV is the maximum distance between voxels within the PTV;

defining a cool zone as the set of voxels outside the hot zone and within the patient skin;

defining a penalty function to suppress radiation hotspots by the equation;

F_penalty=F₀*[1+max{d_i/d₀|d_i>

f_PTV*d_PTV and D_i>

D_OHT;

iε

{1 . . . N}}]

where d_i is the distance from voxel i within the treatment volume to the nearest voxel within the PTV structure;

d₀ is the maximum value of d_i over all voxels in the treatment volume;

f_PTV is a fractional value less than or equal to 0.5;

d_PTV is the maximum distance between voxels within the PTV, such that the product f_PTV*d_PTV defines the size of the allowed hot zone near the PTV;

D_i is the dose received by voxel i;

D_OHT is the maximum desired dose to other healthy tissue outside the hot zone; and

N is the number of voxels within the treatment volume; and

wherein distance d_i can be calculated exactly, or rapidly estimated according to the following procedure;

defining n to be an integer greater than approximately 25;

defining a matrix R of radii such that R=[(X−

x₀)²+(Y−

y₀)²+(Z−

y₀)²]^1/2 where X, Y, and Z are matrices of Cartesian coordinate values for the voxels constructed from coordinate vectors x, y, and z; and

x₀, y₀, and z₀ are the coordinates of a central voxel such that x₀=x(floor((Nx/2)+1)), y₀=x(floor((Ny/2)+1)), and z₀=z(floor((Nz/2)+1)) where Nx, Ny, and Ny are the lengths of coordinate vectors x, y, and z respectively;

redefining values of R equal to zero to be equal to the smallest non-zero value of R; and

calculating a matrix of distances d={d_i} by performing the convolution;

d=[conv(M, R<sup>−

n</sup>)]<sup>−

1/n</sup>.

42. The method according to claim 1 wherein the set of parameters defining the radiation of each trial solution includes dose rate, parameters defining the orientation of non-coplanar arcs, beam directions, and aperture shape for volumetric modulated arc therapy.

43. The method according to claim 1 wherein the set of parameters defining the radiation of each trial solution includes dose rate, non-coplanar arcs, beam directions, and aperture shape for stereotactic radiosurgery.

44. The method according to claim 2 wherein all parameters defining the radiation and the fitness functions together comprise a single monolithic optimization problem in which all parameters are treated identically and varied simultaneously by the optimizer to optimize the fitness functions.

50. The method according to claim 1 including using a clustering algorithm to discover distinct families of solutions in the parameter space.

57. The method according to claim 1 including ghosting of previously viewed solutions, for rapid and intuitive comparison with previously viewed solutions.

59. The method according to claim 1 wherein the prescribed criteria for determining the optimized solutions to be a set of converged Pareto non-dominated solutions comprises no fitness values being improved by a prescribed amount in a prescribed number of previous generations of trial solutions.

3. (canceled)

5-20. -20. (canceled)

23. (canceled)

25-28. -28. (canceled)

32. (canceled)

34. (canceled)

38-41. -41. (canceled)

45-49. -49. (canceled)

51-56. -56. (canceled)

58. (canceled)