Proton intensity modulated multi-objective optimization method based on synergy of multi-objective evolution and traditional optimization method and system thereof

Abstract

The application discloses a proton intensity modulated multi-objective optimization method based on cooperation of multi-objective evolution and traditional optimization method, and relates to the technical field of medical treatment. The application discloses a proton intensity modulated multi-objective optimization method based on cooperation of multi-objective evolution and traditional optimization method, and relates to the technical field of medical treatment. The application discloses a proton intensity modulated multi-objective optimization method based on cooperation of multi-objective evolution and traditional optimization method, and relates to the technical field of medical treatment.

Description

Technical Field

[0001] This invention relates to the field of cancer radiotherapy parameter optimization, and in particular to a proton intensity modulation multi-objective optimization method and system based on the synergy of multi-objective evolution and traditional optimization methods. Background Technology

[0002] Radiotherapy, one of the three major methods of cancer treatment, is required for over 70% of cancer patients. Protons are positively charged particles, and their main difference from photons and electron beams is that after entering the body, a proton forms a Bragg peak at the end of the depth-dose curve, preceded by a low-density flat region, and the dose drops sharply to almost zero after the peak. The depth of the Bragg peak can be adjusted with energy; therefore, by adjusting the energy emitted by the protons and appropriately expanding the peak width according to the tumor size, high-dose areas can be concentrated at different depths and sizes of tumors. Proton therapy is similar to targeted blasting, precisely delivering a targeted bomb to a specific tumor site to kill the tumor. Using proton therapy can increase the dose received by the tumor, improve the local tumor control rate, and has significantly fewer side effects on normal tissues compared to photon beams. As a new technology in radiotherapy, proton beams have received increasing attention and are being applied clinically. The traditional proton intensity-modulated radiotherapy (IMRT) plan design process involves the physician administering the desired dose to the target area based on the treatment goal and setting dose limits for organs at risk. Then, the physicist uses the planning system to set the target parameters for optimization, including how to achieve the dosage distribution required by the doctor and the corresponding weights. Then, using the optimization algorithm in the planning system, a treatment plan is obtained. The dosage parameters and weights are then adjusted, and the optimization is repeated multiple times to find the optimal treatment plan. It can be seen that a satisfactory treatment plan is often obtained through repeated trial-and-error, which is time-consuming and laborious.

[0003] Therefore, there is an urgent need to provide a novel proton intensity modulation multi-objective optimization method and system based on the synergy of multi-objective evolution and traditional optimization methods to solve the above problems. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a proton intensity modulation multi-objective optimization method and system based on the synergy of multi-objective evolution and traditional optimization methods. By utilizing the multi-objective evolutionary characteristics of evolutionary algorithms and the fast characteristics of traditional optimization methods, multiple non-dominated solutions that meet the requirements can be obtained quickly, avoiding the process of repeated parameter adjustment by the planner, thus saving time and effort.

[0005] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is: to provide a proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods, comprising the following steps:

[0006] S1: Data Input: Import the contribution matrix A, the location sequence of sampling points in each region of interest B, input the doctor's expected dose to the target area, the dose constraint limits for surrounding organs at risk, and the priority of each setting, and calculate the sampling points in a certain region of interest. i The dose received;

[0007] S2: Constructing a multi-objective optimization model: The doctor's expected dose to the target area is transformed into a multi-objective optimization problem with three optimization objectives, namely, the doctor's given prescription dose includes a uniform dose to each target area. Maximum dose Minimum dose This is transformed into a multi-objective optimization model with three objectives.

[0008] S3: Construct a single-objective optimization model: Transform the doctor's expected dose to the target area and the dose constraints on surrounding organs at risk into a single-objective optimization problem through a weighted approach, and use it as the objective function for optimization by traditional optimization methods;

[0009] S4: Multi-objective optimization that combines multi-objective evolution with traditional optimization methods in a co-evolutionary manner.

[0010] S4.1: Randomly initialize the population in the multi-objective algorithm according to the values ​​in the contribution matrix A;

[0011] S4.2: The population undergoes crossover mutation. During the mutation process, the traditional optimization method is used with a certain probability. If the traditional optimization method is used for mutation, proceed to step S4.3.

[0012] S4.3: Update the weights based on the current population information. Traditional optimization methods use the new weights to optimize a randomly obtained solution in the population and add the optimized solution to the population of the multi-objective algorithm.

[0013] S4.4: Select a population based on its fitness. The fitness of the population is obtained from the three optimization objectives transformed in step S2. The population with the smaller fitness value will enter the next generation.

[0014] S4.5: If the current iteration count reaches the set maximum iteration count, the optimization ends; otherwise, proceed to step S4.2 to continue optimization.

[0015] S5: Automatically selects several representative dose distributions corresponding to the non-dominated solutions obtained from co-evolution, including isodose lines, dose clouds, and dose-volume histograms, and provides them to users for selection in a visual manner.

[0016] In a preferred embodiment of the present invention, the contribution matrix stores the influence of each proton pencil beam of unit intensity on each calculated sampling point in the patient's body under each irradiation field in the treatment plan, that is, the dose deposition in the patient's body when a proton beam of unit intensity is irradiated.

[0017] In a preferred embodiment of the present invention, the location sequence records the location index of each sampling point in each region of interest.

[0018] In a preferred embodiment of the present invention, sampling points within a region of interest are calculated. i The method for determining the dosage is as follows:

[0019] (1)

[0020] The dose effects of all pencil beams on this sampling point are summed to obtain Where m is the number of pencil bundles, a ij Let x be an element in matrix A, recording the dose of the j-th pencil beam of unit intensity to sampling point i. ij This refers to the intensity to be optimized, and the intensity of all pencil beams is obtained through optimization. x This ensures that the dosage at all sampling points reaches the optimal therapeutic effect.

[0021] In a preferred embodiment of the present invention, step S2 specifically includes the following steps:

[0022] The doctor-prescribed dose includes a uniform dose to each target area. Maximum dose minimum dose The target is transformed into an optimization objective using the following formula:

[0023] (2)

[0024] (3)

[0025] (4)

[0026] Where n is the number of voxels in a region of interest. This is the result calculated in step S1, indicating that the i-th voxel is at an intensity of x. ij The dosage size; It is the minimum dose given by the doctor, when all voxels in the region of interest are greater than or equal to At that time, meet the doctor's requirements. =0; This is the maximum dose given by the doctor, when all voxels in the region of interest are less than or equal to... At that time, meet the doctor's requirements. =0; It is a uniform dose, only when the point dose in the region of interest is uniform. hour, The condition is met, so the value is 0.

[0027] Based on the relationships between the objectives, the above objectives and dosage constraints are categorized and transformed into a multi-objective optimization model with three objectives:

[0028] (5)

[0029] Here, Targets is the set of all target regions, and the z-th target region is denoted as , The maximum allowable dose for the z-th target region. The minimum allowable dose for the z-th target region. For the z-th target region, the desired uniform dose is... The simulated dose is calculated for the z-th target region at the j-th point.

[0030] In a preferred embodiment of the present invention, in step S3, the objective function for single-objective optimization is:

[0031] (6)

[0032] Here, Targets is the set of all target regions, and the z-th target region is denoted as , The maximum allowable dose for the z-th target region. The minimum allowable dose for the z-th target region. For the z-th target region, the desired uniform dose is... Let $\frac{ ... Represents the z-th organ at risk, and the maximum dose to the z-th organ at risk is expressed as... Different areas of interest Different values; This represents the weight of the region of interest in the optimization objective; different regions of interest have different weights.

[0033] In a preferred embodiment of the present invention, in step S4.2, if the traditional optimization method is not used for mutation during the mutation process, then the polynomial mutation method is used for mutation.

[0034] Furthermore, the traditional optimization methods include, but are not limited to, gradient method, conjugate gradient method, and Newton's method.

[0035] In a preferred embodiment of the present invention, step S4.3, the weight update method includes the following steps:

[0036] S4.3.1: Calculate the clinical target value according to the following formula:

[0037] (7)

[0038] in, It is the average of k clinical goals. It is the clinical target value of the k-th target on the i-th voxel, and N is the population size;

[0039] S4.3.2: The weight update formula is:

[0040] (8)

[0041] in, It is the primary factor determining the weight value, and is the priority of the k-th objective. It is obtained from step S1, and the degree of violation of the priority constraint affects the final weight value.

[0042] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is: to provide a proton intensity-modulated multi-objective optimization system based on the synergy of multi-objective evolution and traditional optimization methods, comprising:

[0043] The data input module is used to import the contribution matrix A, the location sequence B of sampling points in each region of interest, the doctor's expected dose to the target area, the dose constraint limits for surrounding organs at risk, and the priority of each setting, and to calculate the sampling points within a certain region of interest. i The dose received;

[0044] The multi-objective optimization model building module is used to transform the doctor's expected dose to the target area into a multi-objective optimization problem with three optimization objectives: the doctor's given prescription dose includes a uniform dose to each target area. Maximum dose Minimum dose This is transformed into a multi-objective optimization model with three objectives.

[0045] The single-objective optimization model construction module is used to transform the doctor's expected dose to the target area and the dose constraint limit to the surrounding organs at risk into a single-objective optimization problem through a weighted approach, which serves as the objective function for optimization by traditional optimization methods.

[0046] A multi-objective optimization module based on co-evolution is used for co-evolutionary multi-objective optimization using multi-objective evolution and traditional optimization methods, including:

[0047] S4.1: Randomly initialize the population in the multi-objective algorithm according to the values ​​in the contribution matrix A;

[0048] S4.2: The population undergoes crossover mutation. During the mutation process, the traditional optimization method is used with a certain probability. If the traditional optimization method is used for mutation, proceed to step S4.3.

[0049] S4.3: Update the weights based on the current population information. Traditional optimization methods use the new weights to optimize a randomly obtained solution in the population and add the optimized solution to the population of the multi-objective algorithm.

[0050] S4.4: Select a population based on its fitness. The fitness of the population is obtained from the three optimization objectives transformed in step S2. The population with the smaller fitness value will enter the next generation.

[0051] S4.5: If the current iteration count reaches the set maximum iteration count, the optimization ends; otherwise, proceed to step S4.2 to continue optimization.

[0052] The results filtering and display module is used to automatically filter out several representative dose distributions corresponding to the non-dominated solutions obtained from co-evolution, including isodose lines, dose clouds, and dose-volume histograms, and provide them to users for selection in a visual manner.

[0053] The beneficial effects of this invention are:

[0054] (1) The method and system described in this invention utilize the multi-objective evolutionary characteristics of evolutionary algorithms and the fast characteristics of traditional optimization methods to quickly obtain multiple non-dominated solutions that meet the requirements, avoiding the process of repeated parameter adjustments by the planners, thus saving time and effort.

[0055] (2) The present invention uses a proton intensity modulation multi-objective optimization method that combines multi-objective evolution and traditional optimization methods for cancer patient data. Unlike traditional optimization methods, it does not require users to adjust weights based on their experience. Users only need to input data to obtain the desired solution. In addition, compared with traditional methods that can only obtain one solution at a time, this method can obtain multiple non-dominated solutions at once, which contributes to the optimization of the intensity of cancer radiotherapy. Attached Figure Description

[0056] Figure 1 This is a flowchart of the proton intensity modulation multi-objective optimization method based on the synergy of multi-objective optimization and traditional optimization methods described in this invention;

[0057] Figure 2 This is a structural block diagram of the proton intensity modulation multi-objective optimization system based on the synergy of multi-objective optimization and traditional optimization methods;

[0058] Figure 3 These are comparison charts of the results obtained by the method described in this invention and the dose-volume histograms obtained by conventional methods in two examples. Detailed Implementation

[0059] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0060] Please see Figure 1 and Figure 2 The embodiments of the present invention include:

[0061] A proton intensity-modulated multi-objective optimization method based on multi-objective evolution and gradient method cooperation includes the following steps:

[0062] S1: Data Input: Import the contribution matrix A, the location sequence of sampling points in each region of interest B, input the doctor's expected dose to the target area, the dose constraint limits for surrounding organs at risk, and the priority of each setting, and calculate the sampling points in a certain region of interest. i The dose received;

[0063] Among them, the contribution matrix A stores the influence of each proton pencil beam of unit intensity on each calculated sampling point in the patient's body under each irradiation field in the treatment plan, that is, the dose deposition in the patient's body when a proton beam of unit intensity is irradiated; the position sequence B records the position index of each sampling point in each region of interest.

[0064] During treatment plan optimization, only sampling points within the region of interest (ROI) are included. The ROI comprises the treatment target area and organs at risk. Target areas, denoted as Targets, include three types: PTV, GTV, and CTV, with potentially more than one of each. Organs at risk, denoted as OARs, are critical organs that may be irradiated during treatment, and the dose they receive must not exceed the tolerable dose.

[0065] Points within a region of interest i The dose received can be expressed as:

[0066] (1)

[0067] The dose effects of all pencil beams on this sampling point are summed to obtain Where m is the number of pencil bundles, a ij Let x be an element in matrix A, recording the dose of the j-th pencil beam of unit intensity to sampling point i, while x... ij This refers to the intensity to be optimized, and the intensity of all pencil beams is obtained through optimization. x This ensures that the dosage at all sampling points achieves the optimal therapeutic effect.

[0068] S2: Constructing a multi-objective optimization model: The doctor's expected dose to the target area is transformed into a multi-objective optimization problem with three optimization objectives, namely, the doctor's given prescription dose includes a uniform dose to each target area. Maximum dose Minimum dose This is transformed into a multi-objective optimization model with three objectives.

[0069] This patent covers multi-objective evolutionary algorithms, not limited to any single one. Any algorithm capable of obtaining a non-dominated solution through population evolution falls within the scope of this patent. The evolutionary algorithm sets the population size to n, the maximum number of iterations to maxGen, and initializes the solution set based on the contribution matrix. The evolutionary process is as follows: select n solutions to enter the next generation based on their fitness. The steps are: first, the population uses simulated binary crossover to generate offspring solutions, then polynomial mutation is used for mutation, and finally, the positions of the solutions in the population are adjusted according to the shift policy. Select the n solutions with the lowest fitness from the current population to enter the next generation. Repeat the above steps until the maximum number of iterations is reached.

[0070] In this example, the following multi-objective evolutionary algorithm is used, and the specific steps include:

[0071] The doctor-prescribed dose includes a uniform dose to each target area. Maximum dose minimum dose The target is transformed into an optimization objective using the following formula:

[0072] (2)

[0073] (3)

[0074] (4)

[0075] Where n is the number of voxels in a region of interest. This is the result calculated in step S1, indicating that the i-th voxel is at an intensity of x. ij The dosage size; It is the minimum dose given by the doctor, when all voxels in the region of interest are greater than or equal to At that time, meet the doctor's requirements. =0; This is the maximum dose given by the doctor, when all voxels in the region of interest are less than or equal to... At that time, meet the doctor's requirements. =0; It is a uniform dose, only when the point dose in the region of interest is uniform. hour, The condition is met, so the value is 0.

[0076] Since target areas often contain multiple targets, and each includes at least one of the aforementioned maximum dose, minimum dose, and uniform dose, the above targets and dose constraints on organs at risk are classified according to the relationships between the targets, and transformed into a multi-objective optimization model with three objectives:

[0077] (5)

[0078] Here, Targets is the set of all target regions, and the z-th target region is denoted as , The maximum allowable dose for the z-th target region. The minimum allowable dose for the z-th target region. For the z-th target region, the desired uniform dose is... The simulated dose is calculated for the z-th target region at the j-th point.

[0079] S3: Constructing a Single-Objective Optimization Model: The doctor's expected dose to the target area and the dose constraints on surrounding organs at risk are transformed into a single-objective optimization problem through a weighted approach; the objective function of the single-objective optimization is:

[0080] (6)

[0081] Here, Targets is the set of all target regions, and the z-th target region is denoted as , The maximum allowable dose for the z-th target region. The minimum allowable dose for the z-th target region. For the z-th target region, the desired uniform dose is... Let $\frac{ ... Represents the z-th organ at risk, and the maximum dose to the z-th organ at risk is expressed as... Different areas of interest Different values; This represents the weight of the region of interest in the optimization objective; different regions of interest have different weights.

[0082] In single-objective optimization engineering, the single-objective optimization model is invoked with a certain probability p. The weights of the single-objective optimization algorithm are updated according to the population information of the multi-objective evolutionary algorithm. The single-objective optimization algorithm calculates the final solution by weighted summation of all objectives and dose constraints of organs at risk. The obtained solution is then added to the population of the multi-objective optimization algorithm.

[0083] S4: Multi-objective optimization with co-evolution of multi-objective evolution and traditional optimization methods. The main steps of co-evolution include:

[0084] S4.1: Randomly initialize the population in the multi-objective algorithm according to the values ​​in the contribution matrix A;

[0085] S4.2: The population undergoes crossover mutation. During the mutation process, the traditional optimization method is used with a certain probability. If the traditional optimization method is used, proceed to step S4.3; if the traditional optimization method is not used, the conventional method of multi-objective optimization, namely the polynomial mutation method, is used.

[0086] Preferably, the crossover mutation employs a simulated binary crossover method.

[0087] S4.3: Update the weights based on the current population information. Traditional optimization methods use the new weights to optimize a randomly selected solution in the population, and then add the optimized solution to the population of the multi-objective algorithm. The weight update method includes the following steps:

[0088] S4.3.1: Calculate the clinical target value according to the following formula:

[0089] (7)

[0090] in, It is the average of k clinical goals. It is the clinical target value of the k-th target on the i-th voxel, and N is the population size;

[0091] S4.3.2: The weight update formula is:

[0092] (8)

[0093] in, It is the primary factor determining the weight value, and is the priority of the k-th objective. It is obtained from step S1, and the degree of violation of the priority constraint affects the final weight value.

[0094] S4.4: Select a population based on its fitness. The fitness of the population is obtained from the three optimization objectives transformed in step S2, namely formulas (2), (3), and (4). The population with the smaller fitness value enters the next generation.

[0095] S4.5: If the current iteration count reaches the set maximum iteration count, the optimization ends; otherwise, proceed to step S4.2 to continue optimization.

[0096] Due to the characteristics of the proton beam Bragg peak—releasing most of its energy in the target area with minimal damage to surrounding organs—a multi-objective optimization model containing only the target dose objective is optimized based on a multi-objective evolutionary algorithm to accelerate the solution rate of the multi-objective function. This reduces the number of objectives to be solved by the evolutionary algorithm, speeds up the solution process, and makes it easier to find good solutions. During the mutation phase of the evolutionary algorithm, a traditional optimization method is used for rapid optimization with a certain probability. The weights of the objectives in the traditional optimization method are given based on the current population information, and the weight adjustment process is as shown in steps S4.3.1 and S4.3.2. The optimal solution obtained by the rapid optimization using the traditional optimization method is then used as an individual in the evolutionary algorithm to participate in the population evolution. Based on the survival-of-the-fittest characteristic of the evolutionary algorithm, the population will evolve towards the optimal direction, and through continuous iteration, the population will eventually find multiple non-dominated solutions that meet the requirements.

[0097] Furthermore, the traditional optimization methods include, but are not limited to, gradient method, conjugate gradient method, Newton's method, etc.

[0098] In the optimization process of the multi-objective evolutionary algorithm, the gradient method is invoked with a certain probability p. When invoking the gradient method, the weights of each objective of the gradient method are first updated according to the current population information using formulas (7) and (8). The gradient method is then used for optimization, and the solution obtained by the gradient method is added to the multi-objective evolutionary algorithm population to guide the evolution of the multi-objective evolutionary algorithm and accelerate the search efficiency. The above steps are repeated until the maximum number of iterations is reached.

[0099] S5: Automatically selects several representative dose distributions corresponding to the non-dominated solutions obtained from co-evolution, including isodose lines, dose clouds, and dose-volume histograms, and provides them to users for selection in a visual manner.

[0100] Preferably, the default value is 10, but the range can also be given by the user from 0 to 20.

[0101] See Figure 2 The present invention also provides a proton intensity modulation multi-objective optimization system based on the synergy of multi-objective evolution and traditional optimization methods, comprising:

[0102] The data input module is used to import the contribution matrix A, the location sequence B of sampling points in each region of interest, the doctor's expected dose to the target area, the dose constraint limits for surrounding organs at risk, and the priority of each setting, and to calculate the sampling points within a certain region of interest. i The dose received;

[0103] The multi-objective optimization model building module is used to transform the doctor's expected dose to the target area into a multi-objective optimization problem with three optimization objectives: the doctor's given prescription dose includes a uniform dose to each target area. Maximum dose Minimum dose This is transformed into a multi-objective optimization model with three objectives.

[0104] The single-objective optimization model construction module is used to transform the doctor's expected dose to the target area and the dose constraint limits to surrounding organs at risk into a single-objective optimization problem through a weighted approach.

[0105] A multi-objective optimization module based on co-evolution is used for co-evolutionary multi-objective optimization using multi-objective evolution and traditional optimization methods, including:

[0106] S4.1: Randomly initialize the population in the multi-objective algorithm according to the values ​​in the contribution matrix A;

[0107] S4.2: The population undergoes crossover mutation. During the mutation process, the traditional optimization method is used with a certain probability. If the traditional optimization method is used for mutation, proceed to step S4.3.

[0108] S4.3: Update the weights based on the current population information. Traditional optimization methods use the new weights to optimize a randomly obtained solution in the population and add the optimized solution to the population of the multi-objective algorithm.

[0109] S4.4: Select a population based on its fitness. The fitness of the population is obtained from the three optimization objectives transformed in step S2. The population with the smaller fitness value will enter the next generation.

[0110] S4.5: If the current iteration count reaches the set maximum iteration count, the optimization ends; otherwise, proceed to step S4.2 to continue optimization.

[0111] The results filtering and display module is used to automatically filter out several representative dose distributions corresponding to the non-dominated solutions obtained from co-evolution, including isodose lines, dose clouds, and dose-volume histograms, and provide them to users for selection in a visual manner.

[0112] An implementation example was performed on a head and neck case using the method and system described in this invention. The results are shown in Table 1 below:

[0113] Table 1 Head and Neck Cases

[0114]

[0115] This table shows the region of interest (ROI) in a case of head and neck cancer. The CTV (Clinical Target Volume) is the clinical target volume, including the CTV and its surrounding area. The remaining areas represent organs at risk—normal organs that may be harmed by radiation. The numbers in the table represent the minimum, maximum, and uniform doses to be achieved in the ROI. After visualizing the results obtained by the method described in this invention, the results meet the physician's desired dose. Furthermore, compared with traditional single-objective optimization methods such as gradient methods, conjugate gradient methods, and quasi-Newton methods, the results obtained by the method described in this invention are superior to those obtained by the aforementioned methods.

[0116] Figure 3 This is a comparison of dose-volume histograms obtained using the proposed method and those obtained using the conventional method in two examples. The dashed line represents the result obtained using the proposed method, and the solid line represents the result obtained using the conventional method. As can be seen from the graphs, within the target area (CTV and CTV low-contraction 10 mm in the graph), the results obtained using the proposed method are significantly better than those obtained using the conventional method. In organs at risk (right parotid gland, submandibular gland, and CTV combination ring 25-35 mm in the graph), the results obtained using the proposed method are better than or similar to those obtained using the conventional method, and are consistent with the doses given in Table 1. Therefore, the proposed method is more effective than the conventional method.

[0117] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods, characterized in that, Includes the following steps: S1: Data Input: Import the contribution matrix A, the location sequence of sampling points in each region of interest B, input the doctor's expected dose to the target area, the dose constraint limits for surrounding organs at risk, and the priority of each setting, and calculate the sampling points in a certain region of interest. i The dose received; S2: Constructing a multi-objective optimization model: The doctor's expected dose to the target area is transformed into a multi-objective optimization problem with three optimization objectives, namely, the doctor's given prescription dose includes a uniform dose to each target area. Maximum dose Minimum dose This is transformed into a multi-objective optimization model with three objectives. S3: Construct a single-objective optimization model: Transform the doctor's expected dose to the target area and the dose constraints on surrounding organs at risk into a single-objective optimization problem through a weighted approach, and use it as the objective function for optimization by traditional optimization methods; S4: Multi-objective optimization that combines multi-objective evolution with traditional optimization methods in a co-evolutionary manner. S4.1: Randomly initialize the population in the multi-objective algorithm according to the values ​​in the contribution matrix A; S4.2: The population undergoes crossover mutation. During the mutation process, the traditional optimization method is used with a certain probability. If the traditional optimization method is used for mutation, proceed to step S4.

3. S4.3: Update the weights based on the current population information. Traditional optimization methods use the new weights to optimize a randomly obtained solution in the population and add the optimized solution to the population of the multi-objective algorithm. The weight update method includes the following steps: S4.3.1: Calculate the clinical target value according to the following formula: (7) in, It is the average of k clinical goals. It is the clinical target value of the k-th target on the i-th voxel, and N is the population size; S4.3.2: The weight update formula is: （8） in, It is the primary factor determining the weight value, and is the priority of the k-th objective. It is obtained from step S1, and the degree of violation of the priority constraint affects the final weight value. S4.4: Select a population based on its fitness. The fitness of the population is obtained from the three optimization objectives transformed in step S2. The population with the smaller fitness value will enter the next generation. S4.5: If the current iteration count reaches the set maximum iteration count, the optimization ends; otherwise, proceed to step S4.2 to continue optimization. The traditional optimization methods include gradient method, conjugate gradient method, and Newton's method; S5: Automatically selects several representative dose distributions corresponding to the non-dominated solutions obtained from co-evolution, including isodose lines, dose clouds, and dose-volume histograms, and provides them to users for selection in a visual manner.

2. The proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods as described in claim 1, characterized in that, The contribution matrix stores the effect of each proton pencil beam of unit intensity on each calculated sampling point in the patient's body under each irradiation field in the treatment plan, that is, the dose deposition in the patient's body when a proton beam of unit intensity is irradiated.

3. The proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods as described in claim 1, characterized in that, The location sequence records the location index of each sampling point in each region of interest.

4. The proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods according to claim 1, characterized in that, Calculate sampling points within a region of interest i The method for determining the dosage is as follows: (1) The dose effects of all pencil beams on this sampling point are summed to obtain Where m is the number of pencil bundles, a ij Let x be an element in matrix A, recording the dose of the j-th pencil beam of unit intensity to sampling point i. ij This refers to the intensity to be optimized, and the intensity of all pencil beams is obtained through optimization. x This ensures that the dosage at all sampling points reaches the optimal therapeutic effect.

5. The proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods according to claim 1, characterized in that, The specific steps of step S2 include: The doctor-prescribed dose includes a uniform dose for each target area. Maximum dose minimum dose The target is transformed into an optimization objective using the following formula: （2） （3） （4） Where n is the number of voxels in a region of interest. This is the result calculated in step S1, indicating that the i-th voxel is at an intensity of x. ij The dosage size; It is the minimum dose given by the doctor, when all voxels in the region of interest are greater than or equal to At that time, meet the doctor's requirements. =0; This is the maximum dose given by the doctor, when all voxels in the region of interest are less than or equal to... At that time, meet the doctor's requirements. =0; It is a uniform dose, only when the point dose in the region of interest is uniform. hour, The condition is met, so the value is 0. Based on the relationships between the objectives, the above objectives and dosage constraints are categorized and transformed into a multi-objective optimization model with three objectives: （5） Here, Targets is the set of all target regions, and the z-th target region is denoted as , The maximum allowable dose for the z-th target region. The minimum allowable dose for the z-th target region. For the z-th target region, the desired uniform dose is... The simulated dose is calculated for the z-th target region at the j-th point.

6. The proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods according to claim 1, characterized in that, In step S3, the objective function for single-objective optimization is: Here, Targets is the set of all target regions, and the z-th target region is denoted as , The maximum allowable dose for the z-th target region. The minimum allowable dose for the z-th target region. For the z-th target region, the desired uniform dose is... Let $\frac{ ... Represents the z-th organ at risk, and the maximum dose to the z-th organ at risk is expressed as... Different areas of interest Different values; This represents the weight of the region of interest in the optimization objective; different regions of interest have different weights.

7. The proton intensity modulation multi-objective optimization method based on the synergy of multi-objective evolution and traditional optimization methods according to claim 1, characterized in that, In step S4.2, if the traditional optimization method is not used for mutation during the mutation process, then the polynomial mutation method is used for mutation.

8. A proton intensity-modulated multi-objective optimization system based on the synergy of multi-objective evolution and traditional optimization methods, characterized in that, include: The data input module is used to import the contribution matrix A, the location sequence B of sampling points in each region of interest, the doctor's expected dose to the target area, the dose constraint limits for surrounding organs at risk, and the priority of each setting, and to calculate the sampling points within a certain region of interest. i The dose received; The multi-objective optimization model building module is used to transform the doctor's expected dose to the target area into a multi-objective optimization problem with three optimization objectives: the doctor's given prescription dose includes a uniform dose to each target area. Maximum dose Minimum dose This is transformed into a multi-objective optimization model with three objectives. The single-objective optimization model construction module is used to transform the doctor's expected dose to the target area and the dose constraint limit to the surrounding organs at risk into a single-objective optimization problem through a weighted approach, which serves as the objective function for optimization by traditional optimization methods. A multi-objective optimization module based on co-evolution is used for co-evolutionary multi-objective optimization using multi-objective evolution and traditional optimization methods, including: S4.1: Randomly initialize the population in the multi-objective algorithm according to the values ​​in the contribution matrix A; S4.2: The population undergoes crossover mutation. During the mutation process, the traditional optimization method is used with a certain probability. If the traditional optimization method is used for mutation, proceed to step S4.

3. S4.3: Update the weights based on the current population information. Traditional optimization methods use the new weights to optimize a randomly selected solution in the population, and then add the optimized solution to the population of the multi-objective algorithm. The weight update method includes the following steps: S4.3.1: Calculate the clinical target value according to the following formula: (7) in, It is the average of k clinical goals. It is the clinical target value of the k-th target on the i-th voxel, and N is the population size; S4.3.2: The weight update formula is: （8） in, It is the primary factor determining the weight value, and is the priority of the k-th objective. It is obtained from step S1, and the degree of violation of the priority constraint affects the final weight value. S4.4: Select a population based on its fitness. The fitness of the population is obtained from the three optimization objectives transformed in step S2. The population with the smaller fitness value will enter the next generation. S4.5: If the current iteration count reaches the set maximum iteration count, the optimization ends; otherwise, proceed to step S4.2 to continue optimization. The traditional optimization methods include gradient method, conjugate gradient method, and Newton's method; The results filtering and display module is used to automatically filter out several representative dose distributions corresponding to the non-dominated solutions obtained from co-evolution, including isodose lines, dose clouds, and dose-volume histograms, and provide them to users for selection in a visual manner.

Images