Multi-objective optimization method for radiotherapy plan

Abstract

The invention discloses a multi-objective optimization method for a radiotherapy plan, and the method comprises the steps: S1, building a multi-objective optimization model, and enabling the multi-objective optimization model to be used for the multi-objective intensity optimization of a radiotherapy reverse intensity modulation plan, and generating a Pareto solution; step S2, inputting a Pareto solution currently generated by the multi-objective optimization model into a weight search network model, wherein the weight search network model predicts a sub-objective weight of a next round of multi-objective intensity optimization of the multi-objective optimization model according to the current Pareto solution; s3, optimizing a single-target strength optimization problem with sub-target weights through a multi-target optimization model to update a current Pareto solution; and S4, circularly executing the steps S2 to S3 until a Pareto solution generated by the multi-objective optimization model forms a Hausdorff distance dHaus (Zin, Zout) between an upper bound and a lower bound of the Pareto in space, the distance dHaus (Zin, Zout) is less than threshold, and the threshold is a predefined threshold parameter which is greater than 0. According to the invention, a deep learning method is creatively used to solve the problem of constraint weight search in a radiation therapy reverse plan multi-objective optimization method.

Description

Technical Field

[0001] The invention relates to the technical field of radiotherapy equipment, and in particular to a multi-objective optimization method for radiotherapy planning. Background Art

[0002] In the intensity modulated radiation therapy industry, in order to solve the intensity modulation optimization problem, it is necessary to optimize a vector consisting of a set of objective functions. The traditional multi-objective method solves the dominant solution, the solution set space is large and the optimization speed is slow, and it fails to fully explore the internal laws of the Pareto solution.

[0003] The existing Chinese invention patent with application number 200810243746.X and titled Multi-objective optimization method and system for optimizing radiation therapy beam intensity distribution discloses a multi-objective optimization method for optimizing radiation therapy beam intensity distribution, including: establishing an optimization model for converting the requirements of the plan designer and clinical restrictions into an optimized mathematical model; and performing multi-objective optimization using an improved fast non-dominated multi-objective genetic algorithm to optimize the intensity distribution diagram of each beam. The system of the present invention includes: a data input module, an optimization model building module, a multi-objective optimization module, and a data output module. The present invention not only accurately establishes a mathematical model of the optimization problem and has a fast solution speed, but also can obtain multiple non-inferior solutions in one multi-objective optimization, from which the plan designer can select the optimal solution that meets the requirements.

[0004] Although the above-mentioned invention patent adopts a multi-objective optimization method, the multi-objective optimization method based on genetic algorithm has the characteristics of slow iteration times and high computational complexity, and the generated plan solution is not a dominant solution.

[0005] In view of this, the present invention patent is proposed. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a multi-objective optimization method for radiotherapy planning. Specifically, the following technical solutions are adopted:

[0007] A multi-objective optimization method for radiotherapy planning, comprising:

[0008] Step S1, establishing a multi-objective optimization model for multi-objective intensity optimization of inverse intensity modulation plan for radiotherapy and generating a Pareto solution;

[0009] Step S2, inputting the Pareto solution currently generated by the multi-objective optimization model into the weight search network model, and the weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution;

[0010] Step S3, optimizing the single-objective strength optimization problem with sub-objective weights through a multi-objective optimization model to update the current Pareto solution;

[0011] Step S4, looping through steps S2 to S3 until the multi-objective optimization model generates a Pareto solution that forms a Hausdorff distance d between the upper and lower bounds of the Pareto surface in space. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0012] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, the optimization form of establishing the multi-objective optimization model in step S1 is:

[0013]

[0014] Where Ψ is the radiation intensity value, d is the radiation dose, and f i and c i They are all convex functions, n ≥ 2.

[0015] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, the optimization form of establishing the multi-objective optimization model in step S1 is:

[0016]

[0017] Among them, w i is the search parameter, which remains unchanged in each round of optimization. TV(Ψ) is the total variation function, which is used as a regularization term to balance the intensity and characteristic structure, and has the form

[0018]

[0019] Where B represents the intensity surface index set, IJ b represents the set of indices of the intensity plane b, Represents the radiation intensity at point (i, j) on plane b.

[0020] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, the weight search network model in step S2 predicts the sub-objective weights of the next round of multi-objective intensity optimization of the multi-objective optimization model according to the current Pareto solution, including:

[0021] The Pareto solution currently generated by the multi-objective optimization model forms a Pareto surface in space;

[0022] The upper and lower bounds of the Pareto surface are estimated using discrete points on the Pareto surface. The upper bound of the Pareto surface is estimated using the inner approximate surface Z. in = {P T λ+Q T μ:λ,μ≥0,e T λ=1}, and the lower bound of the Pareto surface is represented by the outer approximate surface Z out ={z:Wz≥r}, where P=f(D) is the matrix of the current Pareto solution, Q is the matrix of the polar directions of the current Pareto solution space, each row of W is the optimization target weight w, and each element of r is the dot product of the function weight and the target function value;

[0023] The weight search network model searches for the most appropriate target weight for shortening Z in and Z out Hausdorff distance V is for Z out Vertex collection;

[0024] Said The solution formula is

[0025]

[0026] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, according to the The solution formula is, if Then the corresponding hyperplane H(π, ρ) = {z: πz = ρ, π ≥ 0} is Z in π is the sub-objective weight of the next round of multi-objective strength optimization.

[0027] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, the weight search network model adopts a recurrent neural network, and the network input of the weight search network model has the form In={(w, z): w=(w1, ..., w n ), z=(z1,...,z n )}, where w is the target weight of the previous round of optimization that generated the current Pareto solution, z is the objective function value of the current Pareto solution, and the network output of the weight search network model has the form Out = Netθ(In) = π, which represents the target weight of the next round of multi-objective intensity optimization, and Net represents the weight search network.

[0028] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, the weight search network model is trained using a supervised learning method, and the training loss is Where p is the function value of the current Pareto solution, is the estimated target weight for the next round of multi-objective intensity optimization, Represents estimated Z in Support surface The support surface H(π,ρ) constructed by π,ρ corresponds to Z out The pole v is the distance Z in The biggest point.

[0029] As an optional embodiment of the present invention, in a multi-objective optimization method for a radiotherapy plan of the present invention, the training process of the weight search network model includes:

[0030] Calculate the inner approximate surface Z of the corresponding Pareto surface based on the optimization results of the previous rounds of the multi-objective optimization model in = {P T λ+Q T μ:λ,μ≥0,e T λ=1} and the outer approximate surface Z out ={z:Wz≥r};

[0031] Calculate V = {v: v is Z out vertices};

[0032] Calculate v′=argmax v∈V d(v,Z in );

[0033] Get the support surface H(π, ρ) corresponding to v′;

[0034] Construct the network input of the weighted search network model In={(w, z): w=(w1, ..., w n ), z=(z1,...,z n )}, where w is the weight value of the planned sub-objective and z is the sub-objective function value;

[0035] The weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization

[0036] Calculating prediction bias

[0037] Calculating training loss

[0038] Update model parameters η is the learning rate;

[0039] Repeat the above steps until the weight search network model parameters converge.

[0040] As an optional embodiment of the present invention, a multi-objective optimization method for a radiotherapy plan of the present invention includes:

[0041] Step S101, generate N groups of anchor plans through a multi-objective optimization model, and the optimization goal of the i-th group of anchor plans is to minimize Ψ f i (d(Ψ)+ηTV(Ψ);

[0042] Step S201, generate a set of balancing plans, the optimization goal of the balancing plan is

[0043]

[0044] Step S301, construct a weighted search network model according to the original plan, input In = {(w, z): w = (w1, ..., w n ), z=(z1,...,z n )}, obtain the sub-objective weight value Out=Net for the next round of multi-objective strength optimization through the network search model θ (In) = π;

[0045] Step S401, generating an auxiliary plan according to the sub-objective weight values ​​of the next round of multi-objective strength optimization, the optimization goal of the auxiliary plan is to minimize Ψ w1f1(d(Ψ))+…+w n f n (d(Ψ))+ηTV(Ψ), where π=(w1,...,w n ) T ;

[0046] Step S501, repeat the above steps S301 to S401 until the multi-objective optimization model generates a Pareto solution corresponding to the Hausdorff distance d between the inner approximate surface and the outer approximate surface of the Pareto surface. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0047] The present invention also provides a multi-objective optimization device for radiotherapy inverse intensity modulation plan for realizing the multi-objective optimization method, comprising:

[0048] Multi-objective optimization model module, including multi-objective optimization model, used for multi-objective intensity optimization of inverse intensity modulation plan for radiotherapy, and generating Pareto solution;

[0049] A weight search network model module, which inputs the Pareto solution currently generated by the multi-objective optimization model into the weight search network model, and the weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution;

[0050] Optimize the single-objective strength optimization problem with sub-objective weights through a multi-objective optimization model to update the current Pareto solution;

[0051] The multi-objective optimization model module and the weight search network model module are executed cyclically until the multi-objective optimization model generates a Pareto solution that forms the Hausdorff distance d on the upper and lower bounds of the Pareto surface in space. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present invention provides a multi-objective optimization method for radiotherapy planning. The method predicts the sub-objective weights of the next round of multi-objective intensity optimization of the multi-objective optimization model through a weight search network model. Even if the number of optimization targets is large, the accuracy of weight screening can be guaranteed. At the same time, the search time is greatly shortened and the occupation of computing resources is reduced. Therefore, the method has practical application value.

[0054] Therefore, the multi-objective optimization method for radiotherapy planning provided by the present invention overcomes the shortcomings of the original single-objective optimization method in generating plan schemes with different objective balance numbers, and creatively uses a weight search network model based on a deep learning method to solve the constraint weight search problem in the multi-objective optimization method of radiotherapy inverse planning. The generated plan solution is dominant, which can save the time of planners while generating more plan combinations. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A flowchart of a multi-objective optimization method for radiotherapy planning according to an embodiment of the present invention;

[0056] Figure 2 A training flow chart of a weighted search network model according to an embodiment of the present invention;

[0057] Figure 3 A specific example flow chart of a multi-objective optimization method for a radiotherapy plan according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0059] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0061] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0062] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the invention product is usually placed when in use, or the orientation or positional relationship commonly understood by those skilled in the art. Such terms are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0063] See also Figure 1 As shown, a multi-objective optimization method for a radiotherapy plan in this embodiment includes:

[0064] Step S1, establishing a multi-objective optimization model for multi-objective intensity optimization of inverse intensity modulation plan for radiotherapy and generating a Pareto solution;

[0065] Step S2, inputting the Pareto solution currently generated by the multi-objective optimization model into the weight search network model, and the weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution;

[0066] Step S3, optimizing the single-objective strength optimization problem with sub-objective weights through a multi-objective optimization model to update the current Pareto solution;

[0067] Step S4, looping through steps S2 to S3 until the multi-objective optimization model generates a Pareto solution that forms a Hausdorff distance d between the upper and lower bounds of the Pareto surface in space. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0068] This embodiment provides a multi-objective optimization method for a radiotherapy plan, which predicts the sub-objective weights of the next round of multi-objective intensity optimization of the multi-objective optimization model through a weight search network model. Even if the number of optimization targets is large, the accuracy of weight screening can be guaranteed, while the search time is greatly shortened and the occupation of computing resources is reduced. Therefore, it has practical application value.

[0069] Therefore, the multi-objective optimization method for radiotherapy planning provided in this embodiment overcomes the shortcomings of the original single-objective optimization method in generating plan schemes with different objective balance numbers, and creatively uses a weight search network model based on a deep learning method to solve the constraint weight search problem in the multi-objective optimization method of radiotherapy inverse planning. The generated plan solution is dominant, which saves the time of planners while generating more plan combinations.

[0070] As an optional implementation of this embodiment, in a multi-objective optimization method for a radiotherapy plan of this embodiment, the optimization form of establishing the multi-objective optimization model in step S1 is:

[0071]

[0072] Where Ψ is the radiation intensity value, d is the radiation dose, and f i and c i They are all convex functions, n ≥ 2.

[0073] Specifically, f of this embodiment i It can be a DVH function, an EUD function, or a maximum (minimum) dose function. The multi-objective optimization model of this embodiment needs to optimize at least one target area and one organ at risk.

[0074] Specifically, the optimization form of establishing the multi-objective optimization model in step S1 of this embodiment is:

[0075]

[0076] Among them, w i is the search parameter, which remains unchanged in each round of optimization. TV(Ψ) is the total variation function, which is used as a regularization term to balance the intensity and characteristic structure, and has the form

[0077]

[0078] Where B represents the intensity surface index set, IJ b represents the set of indices of the intensity plane b, Represents the radiation intensity at point (i, j) on plane b.

[0079] As an optional implementation of this embodiment, in a multi-objective optimization method for a radiotherapy plan of this embodiment, the weight search network model in step S2 of the multi-objective optimization model predicts the sub-objective weights of the next round of multi-objective intensity optimization of the multi-objective optimization model according to the current Pareto solution, including:

[0080] The Pareto solution currently generated by the multi-objective optimization model forms a Pareto surface in space;

[0081] The upper and lower bounds of the Pareto surface are estimated using discrete points on the Pareto surface. The upper bound of the Pareto surface is estimated using the inner approximate surface Z. in = {P T λ+Q T μ:λ,μ≥0,e T λ=1}, and the lower bound of the Pareto surface is represented by the outer approximate surface Z out ={z:Wz≥r}, where P=f(D) is the matrix of the current Pareto solution, Q is the matrix of the polar directions of the current Pareto solution space, each row of W is the optimization target weight w, and each element of r is the dot product of the function weight and the target function value;

[0082] The weight search network model searches for the most appropriate target weight for shortening Z in and Z out Hausdorff distance V is for Z out Vertex collection;

[0083] Said The solution formula is

[0084]

[0085] Specifically, according to the The solution formula is, if Then the corresponding hyperplane H(π, ρ) = {z: πz = ρ, π ≥ 0} is Z in π is the sub-objective weight of the next round of multi-objective strength optimization.

[0086] As an optional implementation of this embodiment, the weight search network model of this embodiment predicts the target weight of the next round of optimization based on the current Pareto solution and the weight value of the optimization problem (2) that generates the Pareto solution. Therefore, the weight search network model adopts a recurrent neural network, and the network input of the weight search network model has the form In={(w, z): w=(w1, ..., w n ), z=(z1,...,z n )}, where w is the target weight of the previous round of optimization that generated the current Pareto solution, z is the objective function value of the current Pareto solution, and the network output of the weighted search network model has the form Out = Net θ (In) = π, which represents the target weight of the next round of multi-objective strength optimization, and Net represents the weight search network.

[0087] Furthermore, the weight search network model described in this embodiment is trained using a supervised learning method, and the training loss is Where p is the function value of the current Pareto solution, is the estimated target weight for the next round of multi-objective intensity optimization, Represents estimated Z in Support surface The support surface H(π,ρ) constructed by π and ρ corresponds to Z out The pole v is the distance Z in The Adam optimizer is used to optimize the above loss, and the learning rate is set to 0.001.

[0088] As an optional implementation of this embodiment, see Figure 2 As shown, the training process of the weight search network model described in this embodiment includes:

[0089] Calculate the inner approximate surface Z of the corresponding Pareto surface based on the optimization results of the previous rounds of the multi-objective optimization model in = {P T λ+Q T μ:μ≥0,e T λ=1} and the outer approximate surface Z out ={z:Wz≥r};

[0090] Calculate V = {v: v is Z out vertices};

[0091] Calculate v′=argmax v∈V d(v,Z in );

[0092] Get the support surface H(π,ρ) corresponding to v′;

[0093] The network input In={(w,z):w=(w1,…,w n ),z=(z1,…,z n )}, where w is the weight value of the planned sub-objective and z is the sub-objective function value;

[0094] The weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization

[0095] Calculating prediction bias

[0096] Calculating training loss

[0097] Update model parameters η is the learning rate;

[0098] Repeat the above steps until the weight search network model parameters converge.

[0099] As a specific implementation of this embodiment, see Figure 3 As shown, a multi-objective optimization method for a radiotherapy plan in this embodiment includes:

[0100] Step S101, generate N groups of anchor plans through a multi-objective optimization model, and the optimization goal of the i-th group of anchor plans is to minimize Ψ f i (d(Ψ))+ηTV(Ψ);

[0101] Step S201, generate a set of balancing plans, the optimization goal of the balancing plan is

[0102] Step S301: construct a weighted search network model according to the original plan. Input In={(w,z):w=(w1,…,w n ),z=(z1,…,z n )}, obtain the sub-objective weight value Out=Net for the next round of multi-objective strength optimization through the network search model θ (In) = π;

[0103] Step S401, generate an auxiliary plan based on the sub-objective weight values ​​of the next round of multi-objective strength optimization, and the optimization goal of the auxiliary plan is to minimize Ψ w1f1(d(Ψ))+…+w n f n (d(Ψ))+η TV (Ψ), where π=(w1,…,w n ) T ;

[0104] Step S501, repeat the above steps S301 to S401 until the multi-objective optimization model generates a Pareto solution corresponding to the Hausdorff distance d between the inner approximate surface and the outer approximate surface of the Pareto surface. Haus (Z in ,Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0105] In this embodiment, a multi-objective optimization method for radiotherapy planning is provided. First, the weight of the sub-objective of the next round of optimization is predicted according to the known Pareto solution through the weight search network model. Then, the single-objective intensity optimization problem with the new weight is optimized to update the Pareto solution. The above steps are repeated until the Hausdorff distance d between the inner approximation surface and the outer approximation surface is equal to d. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0106] This embodiment also provides a multi-objective optimization device for radiotherapy inverse intensity modulation plan for implementing the multi-objective optimization method, including:

[0107] Multi-objective optimization model module, including multi-objective optimization model, used for multi-objective intensity optimization of inverse intensity modulation plan for radiotherapy, and generating Pareto solution;

[0108] A weight search network model module, which inputs the Pareto solution currently generated by the multi-objective optimization model into the weight search network model, and the weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution;

[0109] Optimize the single-objective strength optimization problem with sub-objective weights through a multi-objective optimization model to update the current Pareto solution;

[0110] The multi-objective optimization model module and the weight search network model module are executed cyclically until the multi-objective optimization model generates a Pareto solution that forms the Hausdorff distance d on the upper and lower bounds of the Pareto surface in space. Haus (Z in ,Z out )<threshold, where threshold>0 is a predefined threshold parameter.

[0111] A multi-objective optimization device for inverse intensity modulation planning of radiotherapy in this embodiment proposes a multi-objective optimization method based on deep learning, and creatively uses the deep learning method to solve the constraint weight search problem in the multi-objective optimization method of inverse planning of radiotherapy. The generated plan solution is dominant, which saves the time of planners while generating more plan combinations.

[0112] This embodiment also provides a computer-readable storage medium storing a computer-executable program. When the computer-executable program is executed, the multi-objective optimization method for a radiotherapy plan is implemented.

[0113] The computer-readable storage medium described in this embodiment may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, an apparatus, or a device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0114] This embodiment also provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer executable program, and when the computer program is executed by the processor, the processor executes the multi-objective optimization method for a radiotherapy plan.

[0115] The electronic device is presented in the form of a general-purpose computing device. The processor may be one or more than one and work in coordination. The present invention does not exclude distributed processing, that is, the processor may be dispersed in different physical devices. The electronic device of the present invention is not limited to a single entity, but may also be the sum of multiple physical devices.

[0116] The memory stores a computer executable program, which is usually a machine-readable code. The computer-readable program can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least part of the steps in the method.

[0117] The memory includes a volatile memory, such as a random access memory unit (RAM) and / or a cache memory unit, and may also be a non-volatile memory, such as a read-only memory unit (ROM).

[0118] It should be understood that the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as display screens, and some electronic devices also include human-computer interaction elements such as buttons and keyboards. As long as the electronic device can execute the computer-readable program in the memory to implement the method of the present invention or at least part of the steps of the method, it can be considered as an electronic device covered by the present invention.

[0119] Through the above description of the implementation mode, it is easy for those skilled in the art to understand that the present invention can be implemented by hardware capable of executing a specific computer program, such as the system of the present invention, and the electronic processing unit, server, client, mobile phone, control unit, processor, etc. contained in the system. The present invention can also be implemented by computer software that executes the method of the present invention, such as control software executed by a microprocessor, an electronic control unit, a client, a server, etc. However, it should be noted that the computer software that executes the method of the present invention is not limited to being executed by one or a specific hardware entity, and it can also be implemented in a distributed manner by unspecified specific hardware. For computer software, the software product can be stored in a computer-readable storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.), and can also be distributed and stored on the network, as long as it can enable the electronic device to execute the method according to the present invention.

[0120] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the above specific implementation methods. Therefore, any modification or equivalent replacement of the present invention; and all technical solutions and improvements thereof that do not depart from the spirit and scope of the invention are included in the scope of the claims of the present invention.

Claims

1. A multi-objective optimization method for radiotherapy planning, characterized in that: include: Step S1, establishing a multi-objective optimization model for multi-objective intensity optimization of inverse intensity modulation plan for radiotherapy and generating a Pareto solution; Step S2, inputting the Pareto solution currently generated by the multi-objective optimization model into the weight search network model, and the weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution; Step S3, optimizing the single-objective strength optimization problem with sub-objective weights through a multi-objective optimization model to update the current Pareto solution; Step S4, looping through steps S2 to S3 until the multi-objective optimization model generates a Pareto solution that forms a Hausdorff distance d between the upper and lower bounds of the Pareto surface in space. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

2. A multi-objective optimization method for radiotherapy planning according to claim 1, characterized in that: The optimization form of establishing the multi-objective optimization model in step S1 is: subject to c j (d(Ψ))≤0,j=1,...,m Ψ ≥ 0 Where Ψ is the radiation intensity value, d is the radiation dose, and f i and c i They are all convex functions, n ≥ 2.

3. A multi-objective optimization method for radiotherapy planning according to claim 2, characterized in that: The optimization form of establishing the multi-objective optimization model in step S1 is: subject to c j (d(Ψ))≤0,j=1,...,m Ψ ≥ 0 Among them, w i is the search parameter, which remains unchanged in each round of optimization. TV(Ψ) is the total variation function, which is used as a regularization term to balance the intensity and characteristic structure, and has the form Where B represents the intensity surface index set, IJ b represents the set of indices of the intensity plane b, Represents the radiation intensity at point (i, j) on plane b.

4. A multi-objective optimization method for radiotherapy planning according to claim 2 or 3, characterized in that: The weight search network model in step S2 predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution, including: The Pareto solution currently generated by the multi-objective optimization model forms a Pareto surface in space; The upper and lower bounds of the Pareto surface are estimated using discrete points on the Pareto surface. The upper bound of the Pareto surface is estimated using the inner approximate surface Z. in = {P T λ+Q T μ:λ,μ≥0,e T λ=1}, and the lower bound of the Pareto surface is represented by the outer approximate surface z out ={z:Wz≥r}, where P=f(D) is the matrix of the current Pareto solution, Q is the matrix of the polar directions of the current Pareto solution space, each row of W is the optimization target weight w, and each element of r is the dot product of the function weight and the target function value; The weight search network model searches for the most appropriate target weight for shortening Z in and Z out Hausdorff distance V is for Z out Vertex collection; Said The solution formula is subject to Pπ≥ρe, Qπ≥0， e T π≤1, π≥0。 5. A multi-objective optimization method for radiotherapy planning according to claim 4, characterized in that: According to the The solution formula is, if Then the corresponding hyperplane H(π, ρ) = {z: πz = ρ, π ≥ 0} is Z in π is the sub-objective weight of the next round of multi-objective strength optimization.

6. A multi-objective optimization method for radiotherapy planning according to claim 4, characterized in that: The weight search network model adopts a recurrent neural network, and the network input of the weight search network model has the form In={(w, z): w=(w1, ..., w n ), z=(z1,...,z n )}, where w is the target weight of the previous round of optimization that generated the current Pareto solution, z is the objective function value of the current Pareto solution, and the network output of the weighted search network model has the form Out = Net θ (In) = π, which represents the target weight of the next round of multi-objective strength optimization, and Net represents the weight search network.

7. A multi-objective optimization method for radiotherapy planning according to claim 6, characterized in that: The weight search network model is trained using a supervised learning method, and the training loss is Where p is the function value of the current Pareto solution, is the estimated target weight for the next round of multi-objective intensity optimization, Represents estimated Z in Support surface The support surface H(π, ρ) constructed by π and ρ corresponds to Z out The pole v is the distance Z in The biggest point.

8. A multi-objective optimization method for radiotherapy planning according to claim 7, characterized in that: The training process of the weight search network model includes: Calculate the inner approximate surface Z of the corresponding Pareto surface based on the optimization results of the previous rounds of the multi-objective optimization model in = {P T λ+Q T μ:λ,μ≥0,e T λ=1} and the outer approximate surface Z out ={z:Wz≥r}; Calculate V = {v: v is Z out vertices}; Calculate v′=argmax v∈V d(v,Z in ); Get the support surface H(π, v) corresponding to v′; Construct the network input of the weighted search network model In={(w, z): w=(w1, ..., w n ), z=(z1,...,z n )}, where w is the weight value of the planned sub-objective and z is the sub-objective function value; The weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization Calculating prediction bias Calculating training loss Update model parameters η is the learning rate; Repeat the above steps until the weight search network model parameters converge.

9. The multi-objective optimization method for radiotherapy planning according to claim 7, characterized in that: include: Step S101, generate N groups of anchor plans through a multi-objective optimization model, and the optimization goal of the i-th group of anchor plans is to minimize Ψ f i (d(Ψ))+ηTV(Ψ); Step S201, generate a set of balancing plans, the optimization goal of the balancing plan is Step S301, construct a weighted search network model according to the original plan, input In = {(w, z): w = (w1, ..., w n ), z=(z1,...,z n )}, obtain the sub-objective weight value Out=Net for the next round of multi-objective strength optimization through the network search model θ (In) = π; Step S401, generate an auxiliary plan based on the sub-objective weight values ​​of the next round of multi-objective strength optimization, and the optimization goal of the auxiliary plan is to minimize Ψ w1f1(d(Ψ))+…+w n f n (d(Ψ))+ηTV(Ψ), where π=(w1,...,w n ) T ; Step S501, repeat the above steps S301 to S401 until the multi-objective optimization model generates a Pareto solution corresponding to the Hausdorff distance d between the inner approximate surface and the outer approximate surface of the Pareto surface. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

10. A multi-objective optimization device for radiotherapy inverse intensity modulation plan implementing the multi-objective optimization method according to any one of claims 1 to 9, characterized in that: include: Multi-objective optimization model module, including multi-objective optimization model, used for multi-objective intensity optimization of inverse intensity modulation plan for radiotherapy, and generating Pareto solution; A weight search network model module, which inputs the Pareto solution currently generated by the multi-objective optimization model into the weight search network model, and the weight search network model predicts the sub-objective weights of the next round of multi-objective strength optimization of the multi-objective optimization model according to the current Pareto solution; Optimize the single-objective strength optimization problem with sub-objective weights through a multi-objective optimization model to update the current Pareto solution; The multi-objective optimization model module and the weight search network model module are executed cyclically until the multi-objective optimization model generates a Pareto solution that forms the Hausdorff distance d on the upper and lower bounds of the Pareto surface in space. Haus (Z in , Z out )<threshold, where threshold>0 is a predefined threshold parameter.

Images