An automatic planning method and system based on dose prediction and parameter optimization

Abstract

The application discloses an automatic planning method and system based on dose prediction and parameter optimization, and the method comprises the following steps: performing a planning CT scan on a patient to obtain a CT image, and obtaining organ contouring data according to the CT image; inputting the CT image and the organ contouring data into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of an organ at risk, and obtaining an average dose of the organ at risk according to the three-dimensional dose distribution prediction; determining a target function of an irradiation field distribution model based on the organ contouring data and the average dose of the organ at risk, to design an initial radiotherapy plan, and optimizing parameters of the target function based on a particle swarm and genetic hybrid algorithm to obtain a parameter optimization result; evaluating the parameter optimization result, updating the parameters of the target function according to an evaluation result, and solving the target function according to a parameter updating result to obtain an optimal radiotherapy plan. The application realizes the function that an intelligent algorithm automatically completes parameter optimization in a radiotherapy plan, and is convenient for improving the quality and efficiency of the plan.

Description

Technical Field

[0001] This invention relates to the field of radiotherapy treatment planning systems, and in particular to an automatic planning method and system based on dose prediction and parameter optimization. Background Technology

[0002] Currently, the mainstream research directions for automated radiotherapy planning technology fall into two main categories: (1) using computer simulation of the physicist's design thinking to achieve automated plan design. (2) Knowledge-based planning (KBP), which aims to improve the quality and efficiency of IMRT plans by learning from high-quality clinical planning databases. The automated breast cancer planning function in the commercial treatment planning system RayStation integrates a heuristic optimization algorithm into TPS. This algorithm can effectively simulate all the steps and decisions of manual plan formulation, quickly and automatically generate clinically usable IMRT plans, and improve the quality and efficiency of plan formulation. The RapidPlan™ system developed by Varian Medical Systems uses machine learning algorithms to train a database to obtain a DVH prediction model. Physicists can use this model to predict the DVH distribution range of new plans and quickly obtain objective function settings such as volume-dose relationship and weighting factors.

[0003] Current radiotherapy planning systems rely heavily on manual adjustment of target volume and organ-at-risk parameters, resulting in low levels of intelligence and automation. The optimization results of commonly used radiotherapy planning systems largely depend on the experience and skills of the planner, and the optimization process is time-consuming. Physicists need to spend a significant amount of time finding the most reasonable dose-volume limits for the target volume and organs at risk, as well as the corresponding weights. In research on deep learning networks predicting three-dimensional dose distribution, most studies have only achieved dose distribution prediction; research on using predicted dose to generate treatment plans is scarce. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the purpose of this invention is to propose an automated planning method based on dose prediction and parameter optimization. This addresses the problems of time-consuming radiotherapy planning and the high dependence of plan quality on the physicist's experience. Based on the three-dimensional dose distribution predicted by a deep learning network, this invention utilizes a radiotherapy planning system and its scripting application programming interface to achieve intelligent algorithms that automatically optimize parameters, thereby improving the quality and efficiency of the planning process.

[0006] Another objective of this invention is to propose an automated planning system based on dose prediction and parameter optimization.

[0007] To achieve the above objectives, this invention proposes an automated planning method based on dose prediction and parameter optimization, comprising:

[0008] A planned CT scan is performed on the patient to obtain CT images, and organ delineation data is obtained based on the CT images;

[0009] The CT images and organ delineation data are input into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of the organs at risk, and the average dose of the organs at risk is obtained based on the three-dimensional dose distribution prediction.

[0010] The objective function of the irradiation field distribution model is determined based on the organ delineation data and the average dose of the organs at risk, in order to design an initial radiotherapy plan. The parameters of the objective function are then optimized based on a particle swarm optimization and genetic hybrid algorithm to obtain the parameter optimization results.

[0011] The optimization results of the parameters are evaluated, the parameters of the objective function are updated based on the evaluation results, and the optimal radiotherapy plan is obtained by solving the objective function based on the updated parameter results.

[0012] The automated planning method based on dose prediction and parameter optimization according to embodiments of the present invention may also have the following additional technical features:

[0013] Further, the step of inputting the CT images and organ delineation data into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of the organ at risk includes: acquiring samples of CT images and organ delineation data; inputting the samples into a trained 3D VGG-U-Net network model for training to obtain a trained 3D VGG-U-Net network model; inputting the CT images and organ delineation data into the trained 3D VGG-U-Net network model to perform relationship construction operations between the CT images, organ delineation data, and dose distribution to obtain relationship construction results; wherein, the relationship construction operations include: convolution operations, batch normalization operations, deconvolution operations, pooling operations, copying and stitching operations; and obtaining a three-dimensional dose distribution prediction of the organ at risk based on the relationship construction results.

[0014] Further, the parameter optimization result obtained by optimizing the objective function based on the particle swarm optimization and genetic hybrid algorithm includes: obtaining the initial position and initial velocity of all particles in the particle swarm; calculating the fitness value of each particle according to the fitness function; comparing the fitness value of each particle's current position with the fitness value of the particle's current best position; updating the fitness value of the particle's current best position according to the first comparison result; comparing the fitness value of each particle's current position with the fitness values ​​of all current best positions of all particles; updating the fitness values ​​of all current best positions of all particles according to the second comparison result; if the fitness values ​​of all current best positions of all particles do not meet the clinical prescription dose target and the organ average dose target, adjusting the target area weight factor and the average dose of the organ at risk, and randomly assigning the value to a particle; updating the velocity and position of each particle according to a preset formula; intersecting each particle with a randomly selected particle at a random point with a first preset probability; mutating each particle at a random point with a second preset probability; and taking the fitness value of all current best positions of all particles as the global optimal solution.

[0015] Furthermore, the method further includes: presetting the number and angle of each illumination field in the illumination field distribution model, the initial illumination field weight, and the weights of the objective function and the target area weight.

[0016] Furthermore, the method further includes: optimizing the subfields and subfield weight values ​​of each irradiation field based on the parameter optimization of the objective function, to obtain subfield optimization results and subfield weight value optimization results.

[0017] To achieve the above objectives, another aspect of the present invention proposes an automated planning system based on dose prediction and parameter optimization, comprising:

[0018] The data acquisition module is used to perform a planned CT scan on the patient to obtain CT images, and to obtain organ delineation data based on the CT images;

[0019] The dose prediction module is used to input the CT image and the organ delineation data into the trained 3DVGG-U-Net network model to obtain the three-dimensional dose distribution prediction of the organ at risk, and to obtain the average dose of the organ at risk based on the three-dimensional dose distribution prediction.

[0020] The parameter optimization module is used to determine the objective function of the irradiation field distribution model based on the organ delineation data and the average dose of the organs at risk, so as to design an initial radiotherapy plan, and to optimize the parameters of the objective function based on a particle swarm optimization and genetic hybrid algorithm to obtain the parameter optimization result.

[0021] The planning output module is used to evaluate the parameter optimization results, update the parameters of the objective function based on the evaluation results, and solve the objective function based on the updated parameters to obtain the optimal radiotherapy plan.

[0022] The automatic planning method and system based on dose prediction and parameter optimization of this invention utilizes deep learning to limit the parameter search range, thereby replacing physicists in automatically optimizing dosimetric parameters and improving the efficiency and quality of radiotherapy planning.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0025] Figure 1 This is a flowchart of an automated planning method based on dose prediction and parameter optimization according to an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of the 3D VGG-U-Net network model structure according to an embodiment of the present invention;

[0027] Figure 3 This is a flowchart of automatic parameter optimization based on dose distribution prediction according to an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of an automatic planning system based on dose prediction and parameter optimization according to an embodiment of the present invention. Detailed Implementation

[0029] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] The following description, with reference to the accompanying drawings, describes an automated planning method and system based on dose prediction and parameter optimization according to embodiments of the present invention.

[0032] Figure 1 This is a flowchart of an automatic planning method based on dose prediction and parameter optimization according to an embodiment of the present invention.

[0033] like Figure 1 As shown, the method includes:

[0034] S1, Perform a planned CT scan on the patient to obtain CT images, and obtain organ delineation data based on the CT images;

[0035] S2, input the CT image and the organ delineation data into the trained 3D VGG-U-Net network model to obtain the three-dimensional dose distribution prediction of the organ at risk, and obtain the average dose of the organ at risk based on the three-dimensional dose distribution prediction.

[0036] S3, Based on the organ delineation data and the average dose of the organs at risk, determine the objective function of the irradiation field distribution model to design an initial radiotherapy plan, and optimize the parameters of the objective function based on a particle swarm optimization and genetic hybrid algorithm to obtain the parameter optimization results;

[0037] S4. Evaluate the parameter optimization results, update the parameters of the objective function based on the evaluation results, and solve the objective function based on the updated parameters to obtain the optimal radiotherapy plan.

[0038] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0039] Understandably, a Treatment Planning System (TPS) is a computer software system that assists physicists in designing treatment plans. TPS supports plans such as IMRT, VMAT, brachytherapy, and proton therapy, and also supports various computational mathematical models. Some TPS systems provide researchers with an Application Programming Interface (API), allowing developers to read patient data, dose matrices, and treatment plans using C# or Python scripts, and also allowing modifications to radiotherapy plans and their transmission to the TPS system.

[0040] As an example, the method of the present invention may include the following steps:

[0041] (1) Plan a CT scan to obtain the patient’s outer contour, internal anatomical structure and tumor volume based on high-quality CT images.

[0042] (2) Target area and organs at risk delineation: Doctors or physicists delineate the anatomical structures of interest layer by layer based on CT images.

[0043] (3) Treatment planning: Physicians set the number and angle of irradiation fields, initial irradiation field weights, and objective function and its weight values ​​in the TPS system based on experience. Then, the computer determines the subfields and their weight values ​​through inverse optimization based on the objective function. Treatment planning and other steps are carried out simultaneously. Physicians will modify parameters such as the objective function weight values ​​based on the treatment plan evaluation results, and then perform dose calculation and treatment plan evaluation again. This process is repeated until the treatment plan meets clinical requirements. This process is the core of the treatment planning workflow.

[0044] (4) Dosage calculation: The dosage is calculated using the convolution superposition algorithm or the Monte Carlo method.

[0045] (5) Evaluation and optimization: After the plan is completed, the physicist evaluates the plan based on evaluation functions such as dose-volume histogram. If it does not meet clinical needs, the design and optimization will continue until the best treatment plan is found.

[0046] It is understood that by automatically optimizing the objective function parameters in step (3) above, the present invention can effectively reduce manual time, narrow the differences in treatment plans between different hospitals and different physicists, thereby improving the quality of radiotherapy plans and thus bringing about an improvement in the efficacy and quality of life for cancer patients.

[0047] Furthermore, samples of CT images and organ delineation data are acquired and input into a trained 3D VGG-U-Net network model to obtain a trained 3D VGG-U-Net network model. The CT images and organ delineation data are then input into the trained 3D VGG-U-Net network model to construct the relationship between CT images, organ delineation data, and dose distribution, thus obtaining the relationship construction result. Based on the relationship construction result, a three-dimensional dose distribution prediction of organs at risk is obtained.

[0048] Specifically, multiple samples of CT images and organ delineation data are acquired for model training. Convolutional Neural Networks (CNNs) are one of the most representative structures in deep learning networks and are widely used in the field of medical image processing. This invention proposes a novel network structure, 3D VGG-U-Net, which automatically learns the relationship between CT images, organ delineation data, and dose distribution, achieving voxel-level three-dimensional dose distribution prediction. The network structure diagram is shown below. Figure 2As shown, 3*3*3Conv refers to a convolution operation with a kernel size of 3*3*3; BN refers to batch normalization; ReLU refers to the activation function; 2*2*2Deconv refers to a deconvolution operation with a kernel size of 2*2*2; 2*2*2Max pooling refers to a pooling operation with a kernel size of 2*2*2; and Copy&Connect refers to copy and connect operations.

[0049] Furthermore, the basic principle of Particle Swarm Optimization (PSO) is to simulate the foraging behavior of birds: Imagine a food source in a certain area, and a flock of birds searches for this food. They don't know the exact location of the food, but they know how far they are from the food source and which bird is closest to it. The optimal search strategy is to search around the bird closest to the food. In the PSO algorithm, each potential solution to the optimization problem can be likened to a bird, called a "particle." The particles update themselves based on the current position of the best particle and the positions of the historical best particles, searching in space until they find the spatially optimal solution.

[0050] The PSO algorithm works as follows: First, N particles are initialized in a D-dimensional search space. Each particle is randomly assigned an initial position and initial velocity, both of which are D-dimensional. In the t-th iteration, the fitness values ​​of the particles are calculated to determine the currently found optimal position pbest and the optimal position gbest found so far. In the t+1-th iteration, each particle updates its velocity and position according to the following formula:

[0051] v t+1 =w·v t +c1·r1·(pbest t -x t )+c2·r2·(gbest t -x t (1-1)

[0052] x t+1 =x t +v t+1 (1-2)

[0053] In the formula, v t+1 v represents the velocity at the (t+1)th iteration. t Let x be the velocity at the t-th iteration. t+1 Let x be the position at the (t+1)th iteration. tThe position is at the t-th iteration. w is the inertia factor, a non-negative number, which adjusts the search range in the solution space; c1 and c2 are acceleration constants, which adjust the maximum step size in the pbest and gbest directions respectively, determining the influence of individual and collective experience on the particle's trajectory. In this invention, c1 = c2 = 1.49445. r1 and r2 are random numbers between 0 and 1, increasing the randomness of particle flight.

[0054] The basic principle of the Genetic Algorithm (GA) is to simulate the natural evolution process: each potential solution to an optimization problem can be compared to a chromosome, multiple chromosomes form a population, and chromosomes evolve from one population to a new population through crossover and mutation, and continue to evolve until the stopping condition is met.

[0055] The PSO algorithm requires few parameter adjustments and is simple in structure and easy to implement, but it lacks dynamic speed adjustment and is prone to getting trapped in local optima. This invention proposes an improved particle swarm optimization algorithm, namely the Genetic Particle Swarm Optimization (GAPSO) algorithm, which introduces the crossover and mutation operation from the GA algorithm into the PSO algorithm, reducing the probability of getting trapped in local optima and improving algorithm performance. The process of the improved particle swarm optimization algorithm of this invention is as follows:

[0056] (1) Initialize the particle swarm to N particles. Within the search space, assign random initial positions and initial velocities to the N particles.

[0057] (2) Calculate the fitness value of each particle based on the fitness function;

[0058] (3) For each particle, compare its current fitness value with the fitness value of its best position so far, pbest. If the fitness value of the current position is better, then update its pbest with the current position.

[0059] (4) For each particle, compare its current fitness value with the fitness value of the best position of all particles so far. If the fitness value of the current position is better, then update gbest with the current position.

[0060] (5) If gbest does not meet the clinical prescription dose target and organ average dose target, the target weight factor is increased by 10, the organ average dose at risk is decreased by 1.0 Gy, and the value is randomly assigned to a particle.

[0061] (6) Update the velocity and position of each particle according to formulas (1-1) and (1-2);

[0062] (7) Each particle crosses with a randomly selected particle at a random point with a 14% probability, and each particle mutates at a random point with a 0.06% probability.

[0063] (8) If the termination condition is not met, return to step 2. If the termination condition is met, the process ends. gbest is the global optimal solution.

[0064] Furthermore, this invention inputs CT images and organ delineation information into a trained 3D VGG-U-Net network to predict the three-dimensional dose distribution. Based on the dose distribution, the average dose value of organs can be obtained. These dose values ​​can be used to limit the search range of parameters during optimization, improving the optimization speed. Then, using the API interface provided by TPS, relevant data on the tumor target area and organs at risk are output. A hybrid algorithm combining particle swarm optimization and genetic algorithms is used to automatically optimize dosimetric parameters such as optimization degree and average dose, replacing the manual adjustment process by physicists. Finally, the optimized parameters are input into the planning system to generate a clinically acceptable radiotherapy plan. The optimization flowchart is as follows. Figure 3 As shown.

[0065] The automatic planning method based on dose prediction and parameter optimization according to embodiments of the present invention utilizes deep learning to limit the parameter search range, thereby replacing physicists in automatically optimizing dosimetric parameters and improving the efficiency and quality of radiotherapy planning.

[0066] To achieve the above embodiments, such as Figure 4 As shown, this embodiment also provides an automatic planning system based on dose prediction and parameter optimization. The system 10 includes: a data acquisition module 100, a dose prediction module 200, a parameter optimization module 300, and a planning output module 400.

[0067] The data acquisition module 100 is used to perform a planned CT scan on the patient to obtain CT images and to obtain organ delineation data based on the CT images.

[0068] The dose prediction module 200 is used to input CT images and organ delineation data into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of the organs at risk, and to obtain the average dose of the organs at risk based on the three-dimensional dose distribution prediction.

[0069] The parameter optimization module 300 is used to determine the objective function of the irradiation field distribution model based on organ delineation information and the average dose of organs at risk, so as to design an initial radiotherapy plan, and to optimize the parameters of the objective function based on a particle swarm optimization and genetic hybrid algorithm to obtain the parameter optimization results.

[0070] The planning output module 400 is used to evaluate the parameter optimization results, update the parameters of the objective function based on the evaluation results, and solve the objective function based on the updated parameters to obtain the optimal radiotherapy plan.

[0071] Furthermore, the dose prediction module 200 described above is also used for:

[0072] Acquire samples of CT images and organ delineation data, input the samples into the 3D VGG-U-Net network model for training, and obtain the trained 3D VGG-U-Net network model.

[0073] CT images and organ delineation data are input into a trained 3D VGG-U-Net network model to perform relationship construction operations between CT images, organ delineation data, and dose distribution, resulting in relationship construction results. The relationship construction operations include: convolution operation, batch normalization operation, deconvolution operation, pooling operation, copying and stitching operation.

[0074] Based on the relationship construction results, a three-dimensional dose distribution prediction for organs at risk is obtained.

[0075] Furthermore, the aforementioned parameter optimization module 300 is also used for:

[0076] Obtain the initial position and initial velocity of all particles in the particle swarm, and calculate the fitness value of each particle according to the fitness function;

[0077] The fitness value of each particle's current position is compared with the fitness value of the particle's current best position, and the fitness value of the particle's current best position is updated according to the first comparison result; and the fitness value of each particle's current position is compared with the fitness values ​​of all current best positions of all particles, and the fitness values ​​of all current best positions of all particles are updated according to the second comparison result.

[0078] If the fitness values ​​of the best positions of all particles do not meet the clinical prescription dose target and the organ average dose target, then adjust the target weight factor and the average dose of the organs at risk, and randomly assign that value to a particle.

[0079] The velocity and position of each particle are updated according to a preset formula. Each particle is crossed with a randomly selected particle at a random point with a first preset probability. Each particle is mutated at a random point with a second preset probability. The fitness value of the best position of all particles is taken as the global optimal solution.

[0080] Furthermore, the system 10 also includes an initial setting module for presetting the number and angle of each illumination field in the illumination field distribution model, the initial illumination field weight, and the weights of the objective function and the target area.

[0081] Furthermore, the system 10 also includes a subfield optimization module, which optimizes the subfield and subfield weight values ​​of each irradiation field according to the parameter optimization of the objective function, and obtains the subfield optimization result and the subfield weight value optimization result.

[0082] The automatic planning system based on dose prediction and parameter optimization according to embodiments of the present invention utilizes deep learning methods to limit the parameter search range, thereby replacing physicists in automatically optimizing dosimetric parameters and improving the efficiency and quality of radiotherapy planning.

[0083] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0084] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0085] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An automatic planning method based on dose prediction and parameter optimization, characterized in that, The method comprises the following steps: a CT image is obtained by performing a planned CT scan on a patient, and organ contouring data is obtained according to the CT image; the CT image and the organ contouring data are input into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of an organ at risk, and an average dose of the organ at risk is obtained according to the three-dimensional dose distribution prediction; a target function of an irradiation field distribution model is determined based on the organ contouring data and the average dose of the organ at risk to design an initial radiotherapy plan, and parameters of the target function are optimized based on a particle swarm and genetic hybrid algorithm to obtain a parameter optimization result; the parameter optimization result is evaluated, the parameters of the target function are updated according to the evaluation result, and the target function is solved according to the parameter update result to obtain an optimal radiotherapy plan; the CT image and the organ contouring data are input into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of an organ at risk, and an average dose of the organ at risk is obtained according to the three-dimensional dose distribution prediction; samples of the CT image and the organ contouring data are obtained, the samples are input into the 3D VGG-U-Net network model for training, and a trained 3D VGG-U-Net network model is obtained; the CT image and the organ contouring data are input into the trained 3D VGG-U-Net network model, and a relationship construction operation is performed to obtain a relationship construction result; wherein the relationship construction operation comprises convolution operation, batch normalization operation, deconvolution operation, pooling operation, copying and splicing operation; a three-dimensional dose distribution prediction of an organ at risk is obtained based on the relationship construction result; the parameter optimization result is obtained by optimizing the parameters of the target function based on the particle swarm and genetic hybrid algorithm, comprising: the initial position and initial speed of all particles of the particle swarm are obtained, and the fitness value of each particle is calculated according to the fitness function; the fitness value of the current position of each particle is compared with the fitness value of the current best position of the particle, and the fitness value of the current best position of the particle is updated according to the first comparison result; and the fitness value of the current position of each particle is compared with the fitness value of the best position of all current particles, and the fitness value of the best position of all current particles is updated according to the second comparison result; if the fitness value of the best position of all current particles does not meet the clinical prescription dose target and the organ average dose target, the target region weight factor and the average dose of the organ at risk are adjusted, and the value is randomly assigned to a particle; the speed and position of each particle are updated according to a preset formula, each particle and a randomly selected particle are crossed at a random point with a first preset probability, each particle is mutated at a random point with a second preset probability, and the fitness value of the best position of all current particles is taken as a global optimal solution; the method further comprises: presetting the number and angle of each irradiation field in the irradiation field distribution model, the initial irradiation field weight, and the weight of the target function and the target region weight. The method further comprises: optimizing the sub-field and the sub-field weight value of each of the irradiation fields according to the parameter optimization of the objective function, to obtain a sub-field optimization result and a sub-field weight value optimization result.

2. An automatic planning system based on dose prediction and parameter optimization, characterized in that, The method comprises: a data acquisition module configured to perform a planning CT scan on a patient to obtain a CT image, and obtain organ contouring data according to the CT image; a dose prediction module configured to input the CT image and the organ contouring data into a trained 3D VGG-U-Net network model to obtain a three-dimensional dose distribution prediction of an organ at risk, and obtain an average dose of the organ at risk according to the three-dimensional dose distribution prediction; a parameter optimization module configured to determine an objective function of an irradiation field distribution model based on the organ contouring data and the average dose of the organ at risk, to design an initial radiotherapy plan, and to optimize parameters of the objective function based on a particle swarm and genetic hybrid algorithm to obtain a parameter optimization result; a plan output module configured to evaluate the parameter optimization result, update parameters of the objective function according to an evaluation result, and solve the objective function according to a parameter update result to obtain an optimal radiotherapy plan; The dose prediction module is further configured to: obtain samples of CT images and organ contouring data, input the samples into the 3D VGG-U-Net network model for training, and obtain a trained 3D VGG-U-Net network model; input the CT image and the organ contouring data into the trained 3D VGG-U-Net network model, and perform a relationship construction operation between the CT image, the organ contouring data and a dose distribution to obtain a relationship construction result; wherein the relationship construction operation comprises a convolution operation, a batch normalization operation, a deconvolution operation, a pooling operation, a replication and a splicing operation; obtain a three-dimensional dose distribution prediction of an organ at risk based on the relationship construction result; The parameter optimization module is further configured to: obtain initial positions and initial velocities of all particles of a particle swarm, and calculate an adaptability value of each particle according to an adaptability function; compare the adaptability value of the current position of each particle with an adaptability value of the current best position of the particle, update the adaptability value of the current best position of the particle according to a first comparison result, and compare the adaptability value of the current position of each particle with an adaptability value of the best position of all current particles, and update the adaptability value of the best position of all current particles according to a second comparison result; if the adaptability value of the best position of all current particles does not satisfy a clinical prescription dose target and an organ average dose target, adjust a target region weight factor and an average dose of an organ at risk, and randomly assign the value to a certain particle; update the velocity and position of each particle according to a preset formula, cross each particle and a randomly selected particle at a random point with a first preset probability, mutate each particle at a random point with a second preset probability, and take the adaptability value of the best position of all current particles as a global optimal solution; The system further comprises an initial setting module configured to preset a number and an angle of each irradiation field in the irradiation field distribution model, an initial irradiation field weight, and a weight of the objective function and a target region weight. The system further comprises a sub-field optimization module configured to optimize the sub-field and the sub-field weight value of each of the irradiation fields according to the parameter optimization of the target function, to obtain a sub-field optimization result and a sub-field weight value optimization result.

Images