Multi-objective plan optimization method and device based on different machine types and cancer types

Abstract

The application discloses a multi-target plan optimization method and device based on different models and cancer types, and relates to the technical field of medical science, wherein the method comprises the following steps: collecting an input data set, wherein the input data set at least comprises image data of a target object, a cancer type, and a radiotherapy device model; obtaining a dose target item set corresponding to the cancer type and a physical constraint parameter set corresponding to the radiotherapy device model; generating a multi-target optimization model according to the dose target item set and the physical constraint parameter set; iteratively solving the multi-target optimization model to generate a Pareto frontier solution set; and determining a target radiotherapy plan according to a candidate solution in the Pareto frontier solution set that simultaneously satisfies a clinical preference condition and a device kinematics constraint condition. The application solves the technical problems that the existing radiotherapy plan optimization method has poor generalization ability in different clinical scenarios and the generated plan is not executable on a target device.

Description

Technical Field

[0001] This application relates to the field of medical technology, and more specifically, to a multi-objective planning optimization method and apparatus based on different machine models and cancer types. Background Technology

[0002] In the field of radiotherapy planning optimization, physicians need to strike a balance between several conflicting objectives, such as ensuring that the tumor target area receives a sufficient dose while minimizing the radiation dose to surrounding normal tissues. Current radiotherapy planning systems generally employ single-objective weighted summation methods or constraint-based optimization methods to generate plans.

[0003] However, parameters optimized for a specific patient or cancer type are difficult to apply to other cases. Extensive manual parameter adjustments are required for different clinical scenarios, a time-consuming process heavily reliant on the operator's experience. Furthermore, the generated radiotherapy plan cannot be directly executed on some radiotherapy devices, necessitating secondary modifications to meet the device's physical limitations.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a multi-objective radiotherapy planning optimization method and apparatus based on different radiotherapy models and cancer types, in order to at least solve the technical problems of poor generalization ability of existing radiotherapy planning optimization methods in different clinical scenarios and the inability of the generated plans to be executed on the target equipment.

[0006] According to one aspect of the embodiments of this application, a multi-objective radiotherapy planning optimization method based on different radiotherapy machine models and cancer types is provided, comprising: collecting an input dataset, the input dataset including at least image data of the target object, cancer type, and radiotherapy machine model; obtaining a dose target item set corresponding to the cancer type and a physical constraint parameter set corresponding to the radiotherapy machine model, wherein the dose target item set includes at least one dose target item, each dose target item being used to quantify the clinical dosimetric performance parameters that the radiotherapy plan should achieve in the target area and organs at risk in the image data; the physical constraint parameter set includes at least one physical constraint parameter, each physical constraint parameter being used to characterize the radiotherapy machine's performance during operation. The system identifies mechanical or dosimetric constraints. A multi-objective optimization model is generated based on the dose target itemset and the physical constraint parameter set. The objective function of the multi-objective optimization model is a weighted summation of the dose target items in the dose target itemset. The constraints of the multi-objective optimization model include treating the physical constraint parameters in the physical constraint parameter set as hard constraints on device feasibility. The multi-objective optimization model is iteratively solved to generate a Pareto front solution set. The Pareto front solution set contains multiple non-dominated candidate solutions, each corresponding to a set of radiotherapy planning parameters. Based on the candidate solutions in the Pareto front solution set that simultaneously satisfy clinical preference conditions and device kinematic constraints, the target radiotherapy plan is determined.

[0007] According to another aspect of the embodiments of this application, a multi-objective radiotherapy planning optimization device based on different radiotherapy machine models and cancer types is also provided, comprising: a dataset acquisition unit for acquiring an input dataset, the input dataset including at least image data of the target object, cancer type, and radiotherapy machine model; and an acquisition unit for acquiring a dose target item set corresponding to the cancer type and a physical constraint parameter set corresponding to the radiotherapy machine model, wherein the dose target item set includes at least one dose target item, each dose target item being used to quantify the clinical dosimetric performance parameters that the radiotherapy plan should achieve in the target area and organs at risk in the image data; and the physical constraint parameter set includes at least one physical constraint parameter, each physical constraint parameter being used to characterize the mechanical or dosimetric performance of the radiotherapy machine during operation. The system comprises the following components: a model generation unit, which generates a multi-objective optimization model based on the dose target itemset and the physical constraint parameter set. The objective function of the multi-objective optimization model is a weighted summation of the dose target items in the dose target itemset. The constraints of the multi-objective optimization model include treating the physical constraint parameters in the physical constraint parameter set as hard constraints for device feasibility. A model processing unit iteratively solves the multi-objective optimization model to generate a Pareto front solution set. The Pareto front solution set contains multiple non-dominated candidate solutions, each corresponding to a set of radiotherapy planning parameters. A planning determination unit determines the target radiotherapy plan based on the candidate solutions in the Pareto front solution set that simultaneously satisfy clinical preference conditions and device kinematic constraints.

[0008] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein when the computer program is executed, the device on which the computer-readable storage medium is located executes the above-described multi-objective plan optimization method based on different machine models and cancer types.

[0009] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the above-described multi-objective program optimization method based on different machine models and cancer types.

[0010] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program or instructions, which, when executed by a processor, implement the above-described multi-objective program optimization method based on different machine types and cancer types.

[0011] In this embodiment, the radiotherapy planning system explicitly includes cancer type and radiotherapy equipment model when collecting the input dataset. This allows subsequent steps to automatically obtain the corresponding dose target itemset based on the cancer type, avoiding repeated manual parameter tuning for different cancer types. Furthermore, obtaining the corresponding physical constraint parameter set based on the equipment model facilitates subsequent hard constraints on equipment feasibility. An optimization objective function is constructed based on the dose target itemset obtained from the cancer type, and the physical constraint parameter set obtained from the equipment model serves as the hard constraint. The equipment physical constraints are directly embedded into the optimization model, ensuring that any candidate solutions violating these constraints are eliminated during the solution process, thus guaranteeing that the generated candidate solutions meet equipment executability requirements. Iteratively solving the multi-objective optimization model generates a Pareto front solution set, where each candidate solution is a non-dominated solution that satisfies the equipment hard constraints. Candidate solutions that simultaneously satisfy clinical preferences and equipment kinematic constraints are selected from the Pareto front solution set as the target radiotherapy plan. Since all candidate solutions already satisfy the equipment hard constraints, only kinematic constraint verification is required during the selection process. Thus, through the collaborative mechanism of cancer type-driven target automatic matching and machine type-driven hard constraint embedding, the optimization process adapts to different cancer types and always solves within the device constraints, thereby solving the technical problems of poor generalization ability of existing radiotherapy planning optimization methods in different clinical scenarios and the inability of the generated plans to be executed on the target devices. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0013] Figure 1 This is a schematic diagram of an optional multi-objective planning optimization method based on different machine models and cancer types, according to an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of an optional multi-objective planning optimization device based on different machine models and cancer types, according to an embodiment of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] According to the embodiments of this application, a method embodiment of a multi-objective plan optimization method based on different machine models and cancer types is provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0018] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0019] According to the embodiments of this application, a radiotherapy planning system can be used as the execution subject of the multi-objective planning optimization method based on different machine models and cancer types in the embodiments of this application. The system can be a software system or an embedded system combining software and hardware. Of course, the execution subject of the method in the embodiments of this application can also be other forms of execution subject, such as devices, equipment, etc. It should be known by those skilled in the art that this application does not particularly limit the specific form of the execution subject.

[0020] Figure 1 This is a schematic diagram illustrating a multi-objective treatment optimization method based on different machine models and cancer types according to embodiments of this application, such as... Figure 1 As shown, the method includes the following steps:

[0021] Step S101: Collect the input dataset. The input dataset includes at least the image data of the target object, the type of cancer, and the model of the radiotherapy equipment.

[0022] For example, the target object's imaging data can refer to computed tomography (CT) images or magnetic resonance imaging (MRI) images, providing anatomical information about the target object; the cancer type can refer to the category of cancer the target object has, such as non-small cell lung cancer or prostate cancer; and the radiotherapy equipment model can refer to the specific model of the linear accelerator performing the radiotherapy. The radiotherapy planning system collects the above input dataset, enabling the system to match corresponding clinical dose targets (such as target coverage and organ-at-risk limits) based on the cancer type and determine the physical limitations of the equipment (such as multi-leaf collimator blade width and maximum dose rate) based on the radiotherapy equipment model. By collecting cancer type and radiotherapy equipment model data, the radiotherapy planning system can provide personalized objective function templates and equipment constraint parameters for subsequent multi-objective optimization models, which is beneficial for improving the adaptability and automatic generation efficiency of radiotherapy plans in different clinical scenarios.

[0023] In some embodiments, the radiotherapy planning system can read target image data, target delineation files, and organ-at-risk delineation files from a hospital information system or radiotherapy information system via a medical digital imaging and communication standard interface. The radiotherapy planning system also receives operator input of cancer type and radiotherapy equipment model from the user interface, such as selecting the cancer name and accelerator model via a drop-down menu. This embodiment centralizes the acquisition of image data, cancer type, and equipment model, simplifying the data input process, reducing manual input errors, and ensuring the data integrity for subsequent multi-objective optimization model construction.

[0024] In other embodiments, the radiotherapy planning system can establish a data connection with an electronic medical record system or a radiotherapy record and verification system to automatically extract the target patient's cancer diagnosis information and the information of the radiotherapy equipment to be used. The radiotherapy planning system queries the electronic medical record system for the cancer type based on the target patient's identification and reads the radiotherapy equipment model specified in the current treatment appointment. This embodiment can reduce manual intervention, facilitate the automatic collection of input datasets, reduce the workload of operators, and improve the automation level of radiotherapy planning optimization and clinical work efficiency.

[0025] Step S102: Obtain the dose target item set corresponding to the cancer type and the physical constraint parameter set corresponding to the radiotherapy equipment model. The dose target item set includes at least one dose target item, each dose target item is used to quantify the clinical dosimetric performance parameters that the radiotherapy plan should achieve in the target area and organs at risk in the imaging data. The physical constraint parameter set includes at least one physical constraint parameter, each physical constraint parameter is used to characterize the mechanical or dosimetric limitation parameters of the radiotherapy equipment during operation.

[0026] For example, dose targets can refer to mathematical expressions of clinical dosimetric indicators such as target volume dose coverage and organ-at-risk limits, based on clinical prescriptions and radiotherapy guidelines. Examples include the volume percentage of the target volume receiving at least the prescribed dose or the maximum dose limit for organs at risk. Physical constraint parameters can refer to inherent mechanical or dosimetric limitations of the radiotherapy equipment during operation, such as the blade width of a multi-leaf collimator, the maximum dose rate of an accelerator, or the minimum rotation angle interval of the gantry. The radiotherapy planning system retrieves the corresponding dose target set from a predefined mapping table based on the acquired cancer type. For example, for prostate cancer, the dose target set could include targets such as D95 ≥ 78 Gy and rectal V70 < 15%. Simultaneously, the radiotherapy planning system loads the corresponding physical constraint parameter set from the equipment database based on the radiotherapy equipment model, such as a blade width of 2.5 mm and a maximum dose rate of 600 MU / min. By dynamically acquiring the corresponding dose target itemset and physical constraint parameter set as input parameters, the radiotherapy planning system can automatically adapt to different clinical scenarios and equipment characteristics, which helps to improve the targeting of the optimization model and the executability of the generated radiotherapy plan on the target equipment.

[0027] In some embodiments, the radiotherapy planning system has a built-in cancer type-target mapping table. This table records dose targets and their clinical reference values ​​for various cancer types. When the operator selects a cancer type or the radiotherapy planning system automatically identifies it, the system automatically loads the set of dose targets corresponding to that cancer type based on the mapping table. This set includes indicators such as minimum target dose, target coverage, maximum dose to organs at risk, and volumetric dose. Simultaneously, the system reads the physical constraint parameters corresponding to the current radiotherapy equipment model from the equipment parameter library. These parameters include the blade width of the multi-leaf collimator, the maximum blade velocity, and the maximum dose rate of the accelerator. This method of loading dose targets and equipment parameters via a mapping table avoids the need for operators to manually input numerous dose targets and equipment parameters, thus reducing workload and the probability of errors.

[0028] In other embodiments, the radiotherapy planning system dynamically generates a set of dose target items using an artificial intelligence model. Specifically, the system inputs the cancer type, target stage information, and anatomical features into a pre-trained target item prediction model. The model outputs a set of personalized dose target items and their clinical reference values. For patients with the same cancer type but different stages, the model recommends target coverage targets with varying degrees of aggressiveness. Simultaneously, the system retrieves the latest physical constraint parameters in real-time from a cloud-based equipment database based on the radiotherapy equipment model. This method of dynamically generating dose target items using an artificial intelligence model further enables individualized customization of dose targets, improving the clinical adaptability and treatment efficacy of radiotherapy planning.

[0029] Step S103: Generate a multi-objective optimization model based on the dose target item set and the physical constraint parameter set. The objective function of the multi-objective optimization model is a function that sums the dose target items in the dose target item set by weights. The constraints of the multi-objective optimization model include using each physical constraint parameter in the physical constraint parameter set as a hard constraint on equipment feasibility.

[0030] For example, a multi-objective optimization model can refer to a set of mathematical expressions used to describe the radiotherapy planning optimization problem, including the objective function and constraints. The objective function is a function obtained by weighted summation of each dose objective item in the dose objective item set, such as the weighted sum of the target dose coverage objective item and the organ at risk limit objective item. Equipment feasibility hard constraints can refer to the various physical constraint parameters in the physical constraint parameter set, such as blade width, maximum dose rate, and minimum gantry rotation angle interval, directly used as inviolable restrictions during the optimization solution process. The radiotherapy planning system generates a multi-objective optimization model based on the dose objective item set and the physical constraint parameter set, so that the optimization objective function of this model can reflect the clinician's preference ranking of different dose objective items, while the constraints facilitate the actual executability of the solved radiotherapy plan parameters on the target radiotherapy equipment. By embedding physical constraint parameters as hard constraints into the optimization model, the radiotherapy planning system can directly eliminate candidate solutions that violate equipment physical constraints during the optimization process, which helps improve the equipment executability and clinical acceptability of the final generated plan.

[0031] In some embodiments, the radiotherapy planning system converts each dose objective in the dose objective set into a mathematical expression about the dose distribution vector. For example, the target coverage objective is represented as the squared difference between the prescribed dose coverage volume percentage and the target value, and the organ-at-risk limit objective is represented as a penalty function between the actual dose and the limit threshold. The radiotherapy planning system assigns a weight coefficient to each mathematical expression, which is adaptively determined based on the cancer type and the radiotherapy equipment model. The radiotherapy planning system sums all the weighted mathematical expressions to construct the optimization objective function. The radiotherapy planning system uses the set of physical constraint parameters as the solution constraints for the optimization objective function, resulting in a multi-objective optimization model. This embodiment integrates multiple clinical objectives through weighted summation, which simplifies the formulation of the optimization problem and facilitates the use of mature algorithms for solution.

[0032] In other embodiments, when generating a multi-objective optimization model, the radiotherapy planning system incorporates a subset of parameters from the physical constraint parameter set, such as the multi-leaf collimator blade width and the accelerator maximum dose rate, as optimizable variables rather than fixed values. The system participates in the iterative solution of the multi-objective optimization model within a range not exceeding the corresponding equipment physical limits. This allows the optimization process to simultaneously adjust the machine hop count vector and some equipment parameters to find a better dose distribution. This embodiment allows the optimization algorithm to fine-tune the physical constraint parameters within the equipment's allowable range, which helps to expand the solution space search range and improve the dose conformity and treatment efficiency of the final radiotherapy plan.

[0033] Step S104: Iteratively solve the multi-objective optimization model to generate a Pareto front solution set, wherein the Pareto front solution set contains multiple non-dominated candidate solutions, and each candidate solution corresponds to a set of radiotherapy planning parameters.

[0034] For example, the Pareto front solution set can refer to a set of candidate solutions in a multi-objective optimization problem that are not mutually dominant, where each candidate solution corresponds to a set of radiotherapy planning parameters, such as the machine hop count vector. A non-dominated candidate solution can be defined as follows: in a multi-objective optimization problem, for two candidate solutions A and B, if candidate solution A performs no worse than candidate solution B in all dimensions of the optimization objective function, and is superior to candidate solution B in at least one dimension, then candidate solution A dominates candidate solution B; a solution not dominated by any other candidate solution is a non-dominated solution. The radiotherapy planning system iteratively solves the multi-objective optimization model, continuously optimizing the candidate solution population through evolutionary algorithms or the alternating direction multiplier method, gradually approaching the Pareto front, and finally generating a Pareto front solution set containing multiple non-dominated candidate solutions. By generating the Pareto front solution set, the radiotherapy planning system can provide clinicians with multiple mutually balanced alternative radiotherapy planning schemes. Each scheme has different trade-offs between target coverage and protection of organs at risk, which helps doctors choose the most appropriate treatment plan based on the patient's specific situation, improving the level of individualized treatment.

[0035] In some embodiments, during the iterative solution of a multi-objective optimization model by the radiotherapy planning system, a population containing multiple candidate solutions can be initialized first, with each candidate solution representing a set of radiotherapy planning parameters. The radiotherapy planning system performs crossover and mutation operations on the candidate solutions in the current population to generate offspring candidate solutions. The radiotherapy planning system verifies whether each newly generated candidate solution satisfies the hard constraints of equipment feasibility in the physical constraint parameter set, and eliminates candidate solutions that violate any hard constraints. The radiotherapy planning system merges the parent candidate solutions that satisfy the hard constraints with the offspring candidate solutions, and sorts them hierarchically according to the non-dominated relationship in multi-objective optimization, retaining candidate solutions with higher non-dominated levels. When the change in the non-dominated solution set is lower than a preset threshold or reaches a preset maximum number of iterations in consecutive iterations, the radiotherapy planning system stops iterating and uses the current set of non-dominated candidate solutions as the Pareto front solution set. This embodiment utilizes the global search capability of evolutionary algorithms to effectively explore complex high-dimensional solution spaces and generate uniformly distributed Pareto fronts.

[0036] In other embodiments, the radiotherapy planning system can employ the alternating direction multiplier method to iteratively solve the multi-objective optimization model. The system decomposes the multi-objective optimization model into multiple sub-problems, each corresponding to an optimization objective. By introducing Lagrange multipliers, the system separates the coupling constraints of each sub-problem, alternately updating the decision variables and Lagrange multipliers of each sub-problem, gradually converging to the Pareto front. In each iteration, the system solves each sub-problem in parallel, which improves computational efficiency. When the change in decision variables between two adjacent iterations falls below a preset threshold or reaches a preset maximum number of iterations, the system stops iterating and uses the candidate solution set obtained in the current iteration as the Pareto front solution set. This embodiment reduces the difficulty of solving the original problem through decomposition and coordination strategies, which helps to accelerate convergence while ensuring solution quality, making it suitable for large-scale radiotherapy planning optimization problems.

[0037] Step S105: Determine the target radiotherapy plan based on the candidate solutions in the Pareto front solution set that simultaneously satisfy the clinical preference conditions and the device kinematic constraints.

[0038] For example, clinical preference conditions can refer to priority rules set by physicians based on the patient's specific condition and treatment goals, such as prioritizing spinal cord protection or ensuring the lowest possible dose to the target area. Equipment kinematic constraints can refer to the mechanical feasibility limitations of the radiotherapy equipment during dynamic execution, such as the maximum speed of a multi-leaf collimator or the continuity requirement of the gantry rotation angle. The radiotherapy planning system filters candidate solutions from the Pareto front solution set that simultaneously satisfy both clinical preference conditions and equipment kinematic constraints, and determines the selected candidate solutions as the target radiotherapy plan. By simultaneously considering clinical preference and equipment kinematic feasibility, the radiotherapy planning system can ensure that the final selected radiotherapy plan not only meets the physician's treatment plan dosimetrically but also can be executed accurately and stably on the actual equipment, thus reducing secondary adjustments due to plan unfeasibility.

[0039] In some embodiments, the radiotherapy planning system receives a user-input priority ranking of organs at risk, such as the spinal cord having the highest priority and the parotid gland the next. The system then weights and scores candidate solutions in the Pareto front solution set according to this priority ranking, selecting the highest-scoring candidate solution. The system further verifies whether the candidate solution meets the device's kinematic constraints, for example, by checking whether the displacement between adjacent control points of the multi-leaf collimator blades exceeds the allowable range of the blades' maximum velocity. If the conditions are met, the system identifies the candidate solution as the target radiotherapy plan. This embodiment, by prioritizing clinical preferences and then verifying device feasibility, facilitates the rapid identification of feasible plans that align with the physician's treatment intentions.

[0040] In other embodiments, the radiotherapy planning system uses both clinical preference conditions and equipment kinematic constraints as hard filtering conditions. The system iterates through each candidate solution in the Pareto front solution set, eliminating solutions that violate any equipment kinematic constraints (e.g., gantry rotation angle intervals less than the minimum allowable interval) or clinical preference conditions (e.g., the lowest target dose being below the clinical lower limit). If the number of remaining candidate solutions is greater than one, the system selects the optimal solution as the target radiotherapy plan based on the target dose homogeneity index. This embodiment, through rigorous screening under dual conditions, ensures that the final output radiotherapy plan meets both clinical acceptability and equipment executability requirements, thus improving the safety and reliability of the plan output.

[0041] In some optional embodiments, a multi-objective optimization model is generated based on a dose target item set and a physical constraint parameter set, including: obtaining the clinical dosimetric performance parameters corresponding to each dose target item in the dose target item set, including target volume, organ at risk volume, prescribed dose value, and dose-volume constraint threshold; converting each dose target item into a mathematical expression about a dose distribution vector, each mathematical expression being used to quantify the degree of deviation between the current dose distribution and the clinical dosimetric performance parameter corresponding to the dose target item; assigning a weight coefficient to each mathematical expression, the weight coefficient being adaptively determined according to the cancer type and radiotherapy equipment model; constructing an optimization objective function by summing all mathematical expressions scaled by the weight coefficients, wherein the total value of the optimization objective function is the objective to be minimized in the multi-objective optimization model; and using the physical constraint parameter set as the solution constraint condition for the optimization objective function to obtain the multi-objective optimization model, wherein the optimizable variables of the multi-objective optimization model include the machine hop count vector and at least a portion of the parameters in the physical constraint parameter set, and these parameters participate in the iterative solution of the multi-objective optimization model within a range not exceeding the corresponding equipment physical upper limit.

[0042] For example, the dose distribution vector can refer to the spatial distribution set of radiation doses absorbed by each voxel within the target body. The mathematical expression can refer to converting each dose target item into a penalty function with respect to the dose distribution vector; the value of the penalty function quantifies the deviation between the current dose distribution and the clinical target. The weighting coefficient can refer to a value adaptively determined based on the cancer type and radiotherapy equipment model, used to adjust the relative importance of different dose target items in the optimization objective function. The radiotherapy planning system obtains the clinical dosimetric performance parameters corresponding to each dose target item in the dose target item set, such as target volume, organ-at-risk volume, prescribed dose value, and dose-volume constraint threshold. It converts each dose target item into a mathematical expression with respect to the dose distribution vector, assigns an adaptively determined weighting coefficient to each mathematical expression, and then sums all the weighted mathematical expressions to construct the optimization objective function. By setting the total value of the optimization objective function as the target to be minimized, the radiotherapy planning system can transform the clinician's dosimetric requirements for the radiotherapy plan into a mathematical optimization problem, which is beneficial for automatically searching for plan parameters that meet clinical goals using numerical optimization algorithms. Using the set of physical constraint parameters as solution constraints ensures that the optimization process always operates within the physical limitations of the equipment. Including the machine hop count vector and at least a portion of the parameters in the physical constraint parameter set (e.g., the maximum speed of the multi-leaf collimator blades or the upper limit of the actual dose rate) as optimizable variables, and allowing these parameters to participate in iterative solutions within their corresponding physical limits, further expands the solution space. This facilitates the discovery of better dose distributions beyond fixed equipment parameter settings.

[0043] For example, the objective function of a multi-objective optimization model is expressed in a weighted summation form, and its mathematical expression can be:

[0044]

[0045] in, Indicates the first Each target dose objective (e.g., target dose homogeneity or target dose coverage) with respect to the dose distribution vector The mathematical expression, Indicates the first Individual dose targets for organs at risk (e.g., maximum dose or volumetric dose) with respect to the dose distribution vector The mathematical expression, and These are the corresponding weighting coefficients. The objective function aims to minimize the dose conflict between the target area and organs at risk while meeting clinical dosimetric requirements.

[0046] It should be noted that the objective function of the aforementioned multi-objective optimization model only reflects the dose-related optimization objective, while the equipment physical constraint parameters in the physical constraint parameter set (such as the blade width of the multi-leaf collimator, the maximum dose rate of the accelerator, and the minimum rotation angle interval of the gantry) are not reflected in the objective function of the multi-objective optimization model. Each physical constraint parameter in the physical constraint parameter set serves as a constraint condition for the multi-objective optimization model, i.e., as a hard constraint on equipment feasibility, and is enforced during the solution process. When solving the multi-objective optimization model, the radiotherapy planning system sets each physical constraint parameter in the physical constraint parameter set (such as blade width, maximum dose rate, and minimum rotation angle interval of the gantry) as an inviolable hard constraint. Any candidate solution that violates any physical constraint parameter in the physical constraint parameter set will be eliminated. Using the physical constraint parameters in the physical constraint parameter set as hard constraints on equipment feasibility ensures that the optimized radiotherapy plan is actually executable on the target radiotherapy equipment, avoiding secondary modifications due to plan infeasibility.

[0047] In some embodiments, the radiotherapy planning system uses a quadratic penalty function to convert the dose target into a mathematical expression. For the target dose coverage target, the system calculates the difference between the prescribed dose coverage volume percentage and the target value, and squares the difference as a penalty value. For the organ-at-risk dose limit target, the system calculates the amount by which the actual dose exceeds the limit, and squares the excess as a penalty value. The system multiplies all penalty values ​​by their corresponding weighting coefficients and sums them to construct the optimization objective function. The system uses parameters such as leaf width and maximum dose rate as hard constraints in the physical constraint parameter set, which are not allowed to be violated during optimization iterations. This embodiment, through the combination of the quadratic penalty term and hard constraints, helps guide the optimization process to quickly converge to a feasible solution region that satisfies clinical goals and equipment limitations.

[0048] In other embodiments, the radiotherapy planning system incorporates a subset of parameters from the physical constraint parameter set as optimizable variables into the solution process of the objective function. For example, the system sets the maximum movement velocity of the multi-leaf collimator blades and the upper limit of the actual dose rate used by the accelerator as optimizable variables, allowing these parameters to be adjusted along with the machine hop count vector within a range not exceeding the equipment's physical limits. In each iteration, the system simultaneously updates parameters such as the machine hop count vector and blade movement velocity, enabling the optimization algorithm to find the optimal dynamic execution strategy within the equipment's performance limits. This embodiment, by synchronously optimizing planning parameters and equipment execution parameters, facilitates the generation of superior dose distributions that are unattainable under conventional fixed equipment parameters, thereby improving the treatment efficiency and dose conformity of the plan.

[0049] In some optional embodiments, the physical constraint parameter set includes a linear relationship between the dose distribution vector and the machine hop number vector. The physical constraint parameter set is used as a constraint condition for solving the objective function to obtain a multi-objective optimization model, which includes: pre-determining a dose influence matrix based on the radiotherapy equipment model, where each element of the dose influence matrix represents the dose contributed by a unit machine hop number to a voxel in a beam direction; multiplying the machine hop number vector by the dose influence matrix to obtain a dose distribution vector, where the machine hop number vector represents the numerical set of machine hop numbers at each control point of the multi-leaf collimator or in each beam direction, used to characterize the radiation output of the accelerator; and embedding the dose distribution vector as a hard constraint condition into the multi-objective optimization model.

[0050] For example, the dose influence matrix can refer to a two-dimensional matrix pre-calculated based on the radiotherapy equipment model. Each element in the dose influence matrix represents the dose value contributed by a unit machine hop count to a voxel in the target body under a specific beam direction. The machine hop count vector can refer to the set of machine hop count values ​​for each control point of the multi-leaf collimator or for each beam direction, used to characterize the radiation output by the accelerator. The dose distribution vector can be obtained by multiplying the machine hop count vector by the dose influence matrix. Each element in the dose distribution vector represents the dose value absorbed by the corresponding voxel. The radiotherapy planning system embeds the dose distribution vector as a hard constraint into the multi-objective optimization model, ensuring that during the optimization process, the dose distribution generated by the machine hop count vector corresponding to any candidate solution must be strictly equal to the product result and must not deviate. By using the linear relationship between the dose distribution vector and the machine hop count vector as a hard constraint, the radiotherapy planning system helps to ensure the physical accuracy of dose calculation, avoids the optimization algorithm from generating physically unrealizable dose distributions, and helps to improve the dose verification pass rate of the final radiotherapy plan in actual execution.

[0051] In some embodiments, the radiotherapy planning system pre-calculates and stores corresponding dose influence matrices for different radiotherapy equipment models, such as models with different multi-leaf collimator blade widths. When the radiotherapy planning system obtains the radiotherapy equipment model information, it automatically loads the dose influence matrix matching that model. During the optimization iteration process, the radiotherapy planning system performs matrix multiplication on the machine hop count vector in each candidate solution with the loaded dose influence matrix to obtain the corresponding dose distribution vector. The radiotherapy planning system uses this dose distribution vector as input to calculate the mathematical expressions for each dose target item and as a hard constraint for equipment feasibility verification, such as checking whether the dose distribution meets the clinical requirements for target coverage. This embodiment, through the model-matched dose influence matrix, helps to ensure the accuracy of dose calculation and the executability of the plan on the target equipment.

[0052] In other embodiments, the radiotherapy planning system uses some elements of the dose influence matrix as optimizable variables. Specifically, the system allows fine-tuning of elements in the dose influence matrix related to the transmission factor of the multi-leaf collimator or the gantry rotation speed within a preset range. During the optimization iteration, the system synchronously updates the machine hop count vector and adjustable dose influence matrix elements, making the dose distribution vector calculation more closely reflect the dosimetric characteristics of the actual equipment. This embodiment, by introducing adaptive adjustment of the dose influence matrix, helps compensate for differences in dose calculation models between different devices, improving the portability and consistency of the generated plan across different devices.

[0053] In some optional embodiments, the weights of each dose target item in the optimization objective function are determined as follows: Cancer type, radiotherapy equipment model, and anatomical features extracted from image data are obtained as context parameters; based on the context parameters, the weight value corresponding to each dose target item is retrieved from a pre-set weight mapping rule base, or the weight value corresponding to each dose target item is determined through a pre-trained weight prediction model. The weight prediction model is a neural network model trained under supervised learning, using the cancer type, radiotherapy equipment model, and anatomical features of historical radiotherapy plans as training samples and the weight vector actually used in the historical radiotherapy plans as training labels; the weight value corresponding to each dose target item is used as the coefficient corresponding to that dose target item in the optimization objective function.

[0054] For example, context parameters can refer to a combination of input information used to characterize the features of the current planning optimization scenario, including cancer type (e.g., non-small cell lung cancer), radiotherapy equipment type (e.g., a specific model of linear accelerator), and anatomical features extracted from image data (e.g., target volume, spatial relationship of organs at risk). Weight values ​​can refer to the coefficients of each dose target in the optimization objective function, used to adjust the importance of that target relative to other targets. The radiotherapy planning system acquires the cancer type, radiotherapy equipment type, and anatomical features extracted from image data as context parameters. Based on the context parameters, it queries a pre-defined weight mapping rule base to obtain the weight value corresponding to each dose target, or determines the weight value corresponding to each dose target through a pre-trained weight prediction model. This weight prediction model is a neural network model trained through supervised learning, using the cancer type, radiotherapy equipment type, and anatomical features of historical radiotherapy plans as training samples and the weight vectors actually used in those historical radiotherapy plans as training labels. The radiotherapy planning system uses the weight value corresponding to each dose target as the coefficient corresponding to that dose target in the optimization objective function. By adaptively determining weighting coefficients based on cancer type, equipment model, and anatomical features, the radiotherapy planning system can automatically adapt the optimization objective function to different clinical scenarios and equipment characteristics, avoiding repeated manual parameter adjustments and improving the efficiency of plan optimization and the consistency of plan quality.

[0055] In some embodiments, the radiotherapy planning system incorporates a weight mapping rule base, which records the correspondence between various cancer types, such as prostate cancer, lung cancer, and head and neck cancer, and the weight values ​​of each dose target item. Once the radiotherapy planning system obtains the cancer type, it directly queries the weight mapping rule base to retrieve a set of predefined weight values ​​based on the cancer type. For example, for prostate cancer, the target volume dose coverage target item has a higher weight, while the rectal and bladder limit target items have a medium weight. The radiotherapy planning system directly assigns the retrieved weight values ​​to the corresponding dose target item coefficients in the optimization objective function. This embodiment quickly matches weights through a rule base, has low computational load, and fast response speed, making it suitable for cancer types where weights have mature clinical experience.

[0056] In other embodiments, the radiotherapy planning system uses a pre-trained weight prediction neural network model to determine weight values. The system encodes cancer type, radiotherapy equipment model, and anatomical features extracted from imaging data, such as target volume and spatial relationships of organs at risk, into an input feature vector. This feature vector is fed into the weight prediction neural network model, and after forward propagation, outputs a weight vector, where each element corresponds to a weight value for a dose target term. This weight prediction neural network model has been trained on a large amount of historical radiotherapy planning data and can learn a non-linear mapping from clinical and anatomical features to optimal weights. The output weight vector is used as the coefficient of the corresponding dose target term in the optimization objective function. This embodiment can learn more complex and individualized weight allocation strategies than a fixed rule base, which is beneficial for generating radiotherapy plans that better suit the specific anatomical characteristics of patients.

[0057] In some optional embodiments, the dose target includes at least one or more of the following targets: target area dose coverage target, for defining the volume percentage of the target area receiving not less than the prescribed dose; target area dose uniformity target, for defining the ratio or difference between the maximum dose and the minimum dose within the target area; organ at risk dose limit target, for defining the maximum received dose value set for organs at risk; and organ at risk volume dose target, for defining the maximum volume percentage of the organ at risk receiving more than a specific dose threshold.

[0058] For example, target volume dose coverage targets can refer to clinical indicators used to limit the percentage of the target volume receiving at least the prescribed dose, such as requiring at least 95% of the target volume to reach the prescribed dose. Target volume dose homogeneity targets can refer to clinical indicators used to limit the ratio or difference between the maximum and minimum doses within the target volume, such as requiring the ratio of the maximum dose to the prescribed dose within the target volume to not exceed a certain upper limit. Organ-at-risk dose limit targets can refer to clinical indicators used to limit the maximum dose value received by a single organ at risk, such as requiring the maximum dose to the spinal cord to not exceed 45 Gy. Organ-at-risk volume dose targets can refer to clinical indicators used to limit the percentage of the maximum volume of organs at risk receiving doses exceeding a specific dose threshold, such as requiring the rectum to receive less than 15% of the volume exceeding 70 Gy. The radiotherapy planning system incorporates the above dose targets into a multi-objective optimization model, enabling the optimization process to simultaneously consider adequate target irradiation, dose homogeneity within the target volume, peak dose limitation for organs at risk, and low-dose volume control for organs at risk. By setting various types of dose targets, the radiotherapy planning system can comprehensively depict the dosimetric requirements of clinicians for radiotherapy plans, which helps guide optimization algorithms to generate radiotherapy plans that both meet the treatment needs of the target area and protect normal tissues.

[0059] In some embodiments, the radiotherapy planning system automatically selects the types of dose targets to be activated based on the cancer type. For head and neck cancers, the system prioritizes target coverage targets, target dose homogeneity targets, and dose-limiting and volumetric dose-response targets for multiple organs at risk, such as the spinal cord, parotid gland, and optic nerve. For prostate cancer, the system focuses on target coverage targets, as well as dose-limiting and volumetric dose-response targets for the rectum and bladder. This embodiment, by adaptively selecting target types based on cancer type, allows the optimization model to focus on the dosimetric parameters of greatest clinical concern for that cancer type, thereby improving planning optimization efficiency and clinical acceptability.

[0060] In other embodiments, the radiotherapy planning system configures multiple sub-targets for each dose objective to achieve finer control. For example, the target dose coverage objective may include multiple sub-targets at different prescribed dose levels, such as high-dose coverage and low-dose coverage. The organ-at-risk dose limit objective may include volumetric constraints at multiple dose levels, such as V10, V20, and V30, which correspond to different dose threshold volume percentages. The radiotherapy planning system incorporates all sub-targets as independent dose objectives into the optimization model and assigns their respective weight coefficients. This embodiment, through finer dosimetric constraints, facilitates the generation of radiotherapy plans with steeper dose distributions and higher target conformity, while better controlling the dose gradient of organs at risk.

[0061] In some optional embodiments, the physical constraint parameters include at least one or more parameters: the blade width of the multi-leaf collimator, used to limit the minimum spatial resolution of beam modulation; the maximum dose rate of the accelerator, used to limit the maximum dose output per unit time; the minimum rotation angle interval of the gantry, used to limit the angle sampling accuracy of the beam direction; and the maximum movement speed of the multi-leaf collimator blades, used to limit the upper limit of the change in blade position between adjacent control points.

[0062] For example, the leaf width of a multi-leaf collimator can refer to the thickness of each leaf in the collimator perpendicular to the leaf movement direction. This leaf width can limit the minimum spatial resolution of beam modulation, which is beneficial for determining the fineness of dose distribution that can be achieved in the radiotherapy plan. The maximum dose rate of the accelerator can refer to the maximum radiation dose that the accelerator can output per unit time, which directly affects the execution efficiency of the treatment plan and the achievable range of dose rate modulation in dynamic intensity-modulated therapy. The minimum rotation angle interval of the gantry can refer to the minimum angle step that a ring or C-shaped gantry can stably stop during rotation, which is beneficial for limiting the angle sampling accuracy of the beam direction and affecting the optimization degree of freedom of the rotational intensity-modulated plan. The maximum movement speed of the multi-leaf collimator leaves can refer to the upper limit of the change in leaf position between adjacent control points, which is beneficial for limiting the physically achievable range of leaf movement in dynamic intensity-modulated therapy. By embedding the above physical constraint parameters as hard constraints on equipment feasibility into the multi-objective optimization model, the radiotherapy planning system can ensure that the optimized plan parameters are actually executable on the target radiotherapy equipment, avoiding the inability to implement the plan or execution deviation due to violation of equipment physical limitations, which is beneficial for improving the reliability of the plan in clinical application.

[0063] In some embodiments, the radiotherapy planning system can automatically load the corresponding blade width value from the equipment parameter library based on the radiotherapy equipment model, such as 2.5 mm or 5 mm. During the optimization iteration process, when calculating the dose distribution, the radiotherapy planning system uses this blade width as the minimum beam modulation unit, which helps ensure that the generated machine hop vector can be actually executed by the equipment's multi-leaf collimator. Simultaneously, the radiotherapy planning system uses the maximum dose rate as a constraint, limiting the dose rate corresponding to the machine hops allocated at each beam direction or control point to not exceeding this upper limit. This embodiment automatically loads physical constraints through equipment model matching, which helps ensure that the optimization model is strictly consistent with the physical characteristics of the target equipment, improving the clinical executability of the generated plan.

[0064] In other embodiments, the radiotherapy planning system uses the minimum gantry rotation angle interval and the maximum blade movement speed as dynamic feasibility verification conditions during the optimization process. For rotational intensity-modulated radiotherapy (IMRT) plans, when generating candidate solutions, the system checks whether the angle interval of each beam direction is less than the minimum gantry rotation angle interval; if so, it adjusts the angle sampling to meet the equipment positioning accuracy. For dynamic IMRT plans, when optimizing blade movement trajectories, the system helps ensure that the displacement of blades between adjacent control points does not exceed the maximum blade movement speed multiplied by the control point time interval, and automatically adjusts the blade sequence or adds intermediate control points when violations occur. This embodiment, through dynamic verification and correction, helps ensure that the generated plan does not trigger the equipment's movement limit alarm during dynamic execution, improving the smoothness and safety of treatment implementation.

[0065] In some optional embodiments, the step of iteratively solving the multi-objective optimization model to generate a Pareto front solution set includes: initializing a population containing multiple candidate solutions, each candidate solution being a set of radiotherapy planning parameters; performing at least one iterative solution operation on the current population until the convergence condition is met; and using the final set of non-dominated candidate solutions as the Pareto front solution set, wherein each iterative solution operation is used to generate a new generation of population.

[0066] For example, a candidate solution can refer to a set of numerical values ​​for a radiotherapy planning parameter, such as a machine hop count vector, a beam angle sequence, or a multi-leaf collimator blade position sequence. A population can refer to a set of multiple candidate solutions, where each candidate solution represents a sampling point in the optimization space. The radiotherapy planning system initializes a population containing multiple candidate solutions, each candidate solution being a set of radiotherapy planning parameters, such as a randomly generated machine hop count vector. The radiotherapy planning system performs at least one iterative solution operation on the current population. Each iterative solution operation generates a new generation of the population, for example, through crossover, mutation, or other operations to produce new candidate solutions. In each generation, the candidate solutions are non-dominated and their diversity is maintained. The radiotherapy planning system repeats the iterative solution operation until a convergence condition is met, such as reaching a preset maximum number of iterations or the change in the non-dominated solution set being less than a preset threshold. The final set of non-dominated candidate solutions is then used as the Pareto front solution set. By searching for the Pareto front through iterative evolution of the population, the radiotherapy planning system can simultaneously explore candidate solutions in different regions of the optimization target space, which is beneficial for discovering multiple mutually beneficial high-quality planning schemes and avoiding getting trapped in local optima.

[0067] In some embodiments, the radiotherapy planning system generates initial candidate solutions by combining random generation with prior knowledge during population initialization. Based on clinical experience, the system sets reasonable numerical ranges for the radiotherapy planning parameters in each candidate solution; for example, the upper limit of machine jumps is referenced from historical data of similar plans. Then, multiple sets of parameters are randomly generated within these numerical ranges as the initial population. After initialization, the system enters an iterative solution loop. This embodiment constrains the initial solution space through prior knowledge, ensuring the population is distributed within promising regions from the early stages of optimization, which helps accelerate convergence and improve optimization efficiency.

[0068] In other embodiments, the radiotherapy planning system employs an elite-preserving strategy during iterative solution development. In each generation, the system directly copies the non-dominated candidate solutions (i.e., solutions on the Pareto front) from the current population to the next generation without crossover or mutation. The remaining non-elite candidate solutions undergo crossover and mutation to generate offspring candidate solutions, and then the elite solutions are merged with the offspring candidate solutions to form a new generation. This embodiment, by preserving the optimal solution in each generation, prevents the loss of discovered excellent candidate solutions during evolution, ensuring the algorithm converges to the true Pareto front and improving the quality of the solution set.

[0069] In some optional embodiments, each iteration of the solution process includes: in each generation iteration, performing crossover and mutation operations on candidate solutions in the current population to generate offspring candidate solutions; for each newly generated candidate solution, verifying whether the candidate solution satisfies the hard constraints on equipment feasibility, and eliminating candidate solutions that violate any hard constraints on equipment feasibility; merging parent candidate solutions that satisfy the hard constraints on equipment feasibility with offspring candidate solutions, and sorting them hierarchically according to the non-dominance relationship in multi-objective optimization, wherein the non-dominance relationship represents: for two candidate solutions, if the performance of the first candidate solution is not inferior to the second candidate solution in all dimensions of the optimization objective function, and is superior to the second candidate solution in at least one dimension. The first candidate solution dominates the second candidate solution; solutions not dominated by any other candidate solutions are non-dominated solutions; the candidate solution with the smallest non-dominated level value is preferentially retained, where the non-dominated level represents the ranking level obtained by recursively stripping the current non-dominated solutions, and the candidate solution with the smallest level value has the highest dominance level; when the number of non-dominated solutions in the same level exceeds the population size, candidate solutions to be retained are selected according to a preset diversity maintenance strategy, where the diversity maintenance strategy includes: a method based on reference point association, which preferentially retains the candidate solution with the smallest distance from the preset reference point, or a method based on crowding distance comparison, which preferentially retains the candidate solution with the largest crowding distance, so as to cover all regions of the Pareto front.

[0070] For example, crossover can refer to the process of exchanging some parameters of two parent candidate solutions to generate offspring candidate solutions. Mutation can refer to the process of randomly perturbing some parameters of a single candidate solution to increase population diversity. Hard constraints on equipment feasibility can refer to the physical limitations on equipment specified in the set of physical constraint parameters, such as blade width, maximum dose rate, maximum blade speed, etc. Non-dominated relations can be used to compare the merits of two candidate solutions. Non-dominated hierarchy can refer to the ranking hierarchy obtained by recursively stripping the current non-dominated solutions, with the candidate solution with the smallest hierarchy value having the highest dominance level. Diversity maintenance strategies can refer to methods used to select which candidate solutions to retain when the number of non-dominated solutions in the same hierarchy exceeds the population size, such as a method based on reference point association (prioritizing the candidate solution with the smallest distance to the preset reference point) or a method based on crowding distance comparison (prioritizing the candidate solution with the largest crowding distance). In each iteration, the radiotherapy planning system performs crossover and mutation operations on candidate solutions in the current population to generate offspring candidate solutions. For each newly generated candidate solution, it verifies whether the candidate solution satisfies the hard constraints of equipment feasibility, and eliminates candidate solutions that violate any hard constraints. Parent candidate solutions that satisfy the hard constraints are merged with offspring candidate solutions and sorted hierarchically according to non-dominated relationships. Candidate solutions with the smallest non-dominated hierarchical values ​​are preferentially retained. When the number of non-dominated solutions in the same hierarchical level exceeds the population size, a diversity maintenance strategy is used to select retained candidate solutions to cover all regions of the Pareto front. Through the above hierarchical sorting and diversity maintenance mechanisms, the radiotherapy planning system can maintain population diversity and convergence during iteration, which is beneficial for generating a uniformly distributed solution set that approximates the true Pareto front.

[0071] In some embodiments, in each iteration, the radiotherapy planning system first performs simulated binary crossover and polynomial mutation operations on candidate solutions in the current population to generate a offspring population with the same size as the parent population. The system then checks each newly generated offspring candidate solution against hard constraints on device feasibility, such as checking if any component in the machine hop count vector exceeds the maximum dose rate limit, or if the multi-leaf collimator blade sequence meets the maximum blade speed limit. For any offspring candidate solution that violates any hard constraint, the system directly eliminates it and randomly selects a candidate solution from the parent population to fill the offspring population, maintaining the population size. Subsequently, the system merges the parent population with the corrected offspring population, performs fast non-dominated sorting, and divides the merged population into multiple non-dominated levels, with the first level being the optimal set of non-dominated solutions. The system prioritizes adding non-dominated solutions from the first level to the next generation population; if the number of solutions in the first level is insufficient, it adds them to the next level sequentially until the population size limit is reached. When the final layer needs to be truncated, the radiotherapy planning system employs a reference point association-based method, selecting the most uniformly distributed solution based on a pre-set set of reference points. This embodiment, through hard constraint verification and reference point association strategies, facilitates ensuring that all candidate solutions in each generation of the population are physically feasible, and ultimately achieves a uniformly distributed Pareto front.

[0072] In other embodiments, the radiotherapy planning system uses differential evolution operations instead of simulated binary crossover and polynomial mutation in each iteration. For each candidate solution in the current population, the system randomly selects three different candidate solutions, generates mutated individuals using a difference vector, and then crossovers these with the original candidate solutions to generate offspring candidate solutions. After performing the same hard constraint verification and elimination filling operations on the offspring candidate solutions, the system merges the parent and offspring and performs a non-dominated sort. At the final truncation level, the system uses a crowding distance-based comparison method to calculate the sum of distances between each candidate solution and its neighboring solutions across all objective function dimensions, prioritizing the retention of candidate solutions with larger crowding distances, i.e., those with sparser surrounding solutions. This embodiment enhances the global search capability of the population through the differential evolution mutation strategy, while maintaining diversity based on crowding distance helps preserve the uniformity of the Pareto front, making it suitable for scenarios with high objective function computational complexity.

[0073] In some optional embodiments, the convergence condition is any of the following: the change in the non-dominated solution set is less than a preset threshold in N consecutive iterations, where N is an integer greater than 1; the preset maximum number of iterations is reached; and all candidate solutions satisfy the preset clinical dose constraint.

[0074] For example, the magnitude of change in the non-dominated solution set can refer to the degree of difference between the current generation and the previous generation of non-dominated solution sets during continuous iterations, such as by measuring the positional shift or the number of additions / reductions of candidate solutions in the target space. The maximum number of iterations can refer to the upper limit of the number of iterative solution loops preset by the radiotherapy planning system. The preset clinical dose constraints can refer to the dosimetric acceptance criteria for the target area and organs at risk determined in advance by the physician or treatment planning system, such as a target coverage of not less than 95% and a maximum spinal cord dose not exceeding 45 Gy. During the iterative solution process, the radiotherapy planning system monitors the magnitude of change in the non-dominated solution set in N consecutive iterations. When the magnitude of change is lower than a preset threshold, convergence is determined and iteration is stopped; or when the number of iterations reaches the preset maximum number of iterations, it is forcibly stopped; or when all candidate solutions meet the preset clinical dose constraints, convergence is determined and iteration is stopped. By setting multiple convergence conditions, the radiotherapy planning system can terminate optimization in a timely manner when the solution set quality has stabilized to avoid unnecessary computational overhead. At the same time, it can also ensure that the algorithm has a clear termination point in the worst case and exit early when the solution set meets clinical requirements, which is conducive to balancing planning quality and optimization efficiency.

[0075] In some embodiments, the radiotherapy planning system uses a convergence criterion that the change in the non-dominated solution set over five consecutive iterations is less than one percent. After each iteration, the system calculates the average Euclidean distance between the current non-dominated solution set and the non-dominated solution sets from the previous five iterations. This average Euclidean distance refers to the distance in the normalized target space. If the average distance is less than a preset distance threshold, such as 0.01, and this condition is met for five consecutive iterations, the system determines that the algorithm has converged and terminates the optimization. This embodiment avoids premature termination due to random fluctuations through continuous monitoring over multiple generations, thus ensuring the stability and reliability of the final solution set.

[0076] In other embodiments, the radiotherapy planning system uses the satisfaction of preset clinical dose constraints as a convergence criterion for all candidate solutions. After each iteration, the system iterates through all candidate solutions in the current population, checking whether the dose distribution corresponding to each candidate solution satisfies preset clinical dose constraints, such as minimum target dose and maximum dose to organs at risk. If all candidate solutions satisfy these clinical dose constraints, the system immediately terminates the iteration and outputs the current non-dominated solution set. This embodiment allows optimization to end earlier when a clinically acceptable level is reached, reducing unnecessary computation time and improving the clinical responsiveness of radiotherapy planning optimization.

[0077] In some optional embodiments, the radiotherapy planning system can also iteratively solve the multi-objective optimization model using the following methods: For example, the beam weights can be directly output using reinforcement learning to train the policy network. This involves modeling the multi-objective optimization problem as a Markov decision process, training the agent using deep reinforcement learning (such as deep Q-networks or policy gradient methods), and directly outputting the optimal machine hop vector or beam intensity distribution based on the current planning parameter state, thereby skipping the iterative optimization process and achieving real-time plan generation. Alternatively, Bayesian optimization can be used for efficient searching in the weight space. The weight coefficients of each dose objective item are used as hyperparameters, with planning quality evaluation indicators such as target coverage and dose to organs at risk as the objective function. A small number of iterative samples are performed in the weight space using a Gaussian process surrogate model and a sampling function (such as desired improvement) to quickly approximate the optimal weight combination, reducing the cost of manual parameter tuning. Or, for yet another example, a graph neural network can be constructed to encode anatomical relationships to guide the optimization direction. By treating the target area and each organ at risk as graph nodes, and the spatial proximity and dose dependence relationships between organs as edges, a graph neural network is used to extract high-order features of the anatomical structure. These high-order features are then used as prior knowledge to input into the optimization model, guiding the optimization algorithm to converge toward a clinically better region, thereby improving the quality of the plan and the efficiency of optimization.

[0078] In some optional embodiments, clinical preference criteria include: prioritization of organs at risk, or a trade-off coefficient between the lowest dose to the target area and the maximum dose to the organs at risk.

[0079] For example, prioritization can refer to the order of importance of protecting multiple organs at risk as set by the physician based on the patient's condition and treatment goals. For instance, spinal cord protection has a higher priority than parotid gland protection, and parotid gland protection has a higher priority than oral cavity protection. The trade-off coefficient can be a numerical parameter that represents a compromise between the lowest dose to the target area and the maximum dose to the organs at risk, reflecting the physician's relative emphasis on increasing target coverage versus reducing the dose to the organs at risk. The radiotherapy planning system uses the priority ranking of organs at risk or the trade-off coefficient between the lowest dose to the target area and the maximum dose to the organs at risk as clinical preference criteria to screen candidate solutions from the Pareto front solution set that best match the physician's treatment intentions. By introducing clinical preference criteria, the radiotherapy planning system can quantify the physician's subjective treatment experience into calculable screening rules, facilitating the automatic and rapid identification of the most suitable radiotherapy plan for the current patient from multiple Pareto optimal options, reducing the time spent on manually evaluating each option individually.

[0080] In some embodiments, the radiotherapy planning system receives a user-input list of priority organs, such as spinal cord over brainstem, brainstem over parotid gland, and parotid gland over oral cavity. The system compares candidate solutions in the Pareto front solution set according to priority: first, it compares spinal cord protection metrics, such as maximum dose, retaining the candidate solution with the lowest spinal cord dose; if multiple candidate solutions have comparable spinal cord doses, it further compares brainstem doses, and so on. The system then selects the final candidate solution as the target radiotherapy plan. This embodiment, through hierarchical priority comparison, ensures that high-priority organs are prioritized for protection, facilitating the generation of plans that align with physicians' clinical practices.

[0081] In other embodiments, the radiotherapy planning system receives a user-defined tradeoff coefficient (which reflects the relative importance of the minimum dose to the target area versus the maximum dose to organs at risk). The system weights and scores candidate solutions in the Pareto front solution set based on this tradeoff coefficient, selecting the highest-scoring candidate solution as the target radiotherapy plan. This embodiment, through a continuously adjustable tradeoff coefficient, allows for fine-tuning the balance between target coverage and organ at risk protection, which is beneficial for meeting the personalized preferences of different patients regarding treatment goals and side effects.

[0082] In some optional embodiments, the machine hop count vector is a function of at least one physical constraint parameter in the set of physical constraint parameters. The method further includes: during the iterative solution of the multi-objective optimization model, simultaneously optimizing the machine hop count vector and at least one physical constraint parameter, wherein the physical constraint parameters include one or more of the following: the maximum speed of the multi-leaf collimator blades and the upper limit of the actual dose rate of the accelerator; and each component of the machine hop count vector is constrained by the width of the multi-leaf collimator blades, the maximum dose rate of the accelerator, and the maximum speed of the multi-leaf collimator blades; the physical constraint parameters are adjusted within a range not exceeding the corresponding physical upper limit of the device so that the dose distribution corresponding to the finally generated candidate solution meets the clinical objectives.

[0083] For example, the machine hop count vector can refer to the numerical set of machine hop counts at each control point or in each beam direction of the multi-leaf collimator, used to characterize the radiation output of the accelerator. Physical constraint parameters can refer to variable mechanical or dosimetric limiting parameters of the radiotherapy equipment during operation, such as the maximum speed of the multi-leaf collimator blades or the upper limit of the actual dose rate used by the accelerator. Equipment physical upper limits can refer to the maximum values ​​of physical constraint parameters allowed by equipment design or safety specifications, such as the factory-set upper limit for the maximum speed of the blades. During the iterative solution of the multi-objective optimization model, the radiotherapy planning system simultaneously optimizes the machine hop count vector and at least one physical constraint parameter, such as the maximum speed of the multi-leaf collimator blades or the upper limit of the actual dose rate used by the accelerator. Each component of the machine hop count vector is constrained by physical constraint parameters such as the multi-leaf collimator blade width, the maximum dose rate of the accelerator, and the maximum speed of the multi-leaf collimator blades. These physical constraint parameters are adjusted within a range not exceeding the corresponding equipment physical upper limits to ensure that the dose distribution corresponding to the final generated candidate solution meets clinical objectives. By simultaneously optimizing the machine hop vector and physical constraint parameters, the radiotherapy planning system can explore better dynamic execution strategies within the performance limits allowed by the equipment. This helps to discover better dose distributions than those under conventional fixed parameter settings, thereby improving the treatment efficiency and dose conformity of the plan.

[0084] Alternatively, the dose distribution can be calculated as follows: ,in, It can represent the dose distribution vector. It can represent a dose-effect matrix based on a pre-determined model of radiotherapy equipment. This can be represented as a machine hop count vector. Physical constraint parameters such as blade width, machine hop count limit (i.e., accelerator maximum dose rate), and blade movement rate (i.e., maximum speed of the multi-leaf collimator blades) are used as optimizable variables, along with... Simultaneously participate in optimization, adjusting within the limits of the device's physical capacity.

[0085] In some embodiments, the radiotherapy planning system uses the maximum speed of the multi-leaf collimator blades as an optimizable variable, allowing this speed to be continuously adjusted between a physical upper limit (e.g., 50 mm / s) and a lower limit (e.g., 5 mm / s). During the optimization iteration, the radiotherapy planning system encodes the maximum blade speed along with the machine hop count vector into the candidate solutions, updating the maximum blade speed simultaneously with each update of the machine hop count vector. For each candidate solution, the radiotherapy planning system dynamically adjusts the positional variation constraints of the multi-leaf collimator blades between adjacent control points based on the maximum blade speed value in that candidate solution; that is, the actual blade speed cannot exceed the maximum speed set for that candidate solution. This embodiment, by optimizing the blade speed, can find the optimal match between blade speed and machine hop count for different plans while meeting the physical upper limit of the equipment, which is beneficial for shortening treatment time or reducing blade wear.

[0086] In other embodiments, the radiotherapy planning system uses the upper limit of the actual dose rate used by the accelerator as an optimizable variable, allowing this upper limit to be continuously adjusted between the device's maximum dose rate (e.g., 600 machine hops per minute) and a lower limit (e.g., 100 machine hops per minute). During the optimization iteration process, the radiotherapy planning system encodes the upper limit of the actual dose rate used together with the machine hop vector into candidate solutions. For each candidate solution, the radiotherapy planning system constrains the dose rate at each control point in the machine hop vector to not exceed the upper limit value based on the dose rate upper limit value in the candidate solution, and simultaneously calculates the feasibility of the beam exit time and blade motion trajectory corresponding to the upper limit of the dose rate. This embodiment, by optimizing the upper limit of the dose rate, can select a lower dose rate to reduce accelerator filament loss or a higher dose rate to shorten treatment time, while ensuring planning quality, thus meeting the preferences of different clinical scenarios for treatment efficiency or device lifespan.

[0087] See Figure 2According to another aspect of the embodiments of this application, a multi-objective radiotherapy plan optimization device based on different radiotherapy models and cancer types is also provided, comprising: a dataset acquisition unit for acquiring an input dataset, the input dataset including at least image data of the target object, cancer type, and radiotherapy equipment model; and an acquisition unit for acquiring a dose target item set corresponding to the cancer type and a physical constraint parameter set corresponding to the radiotherapy equipment model, wherein the dose target item set includes at least one dose target item, each dose target item being used to quantify the clinical dosimetric performance parameters that the radiotherapy plan should achieve in the target area and organs at risk in the image data; and the physical constraint parameter set includes at least one physical constraint parameter, each physical constraint parameter being used to characterize the mechanical or dose-dependent operation of the radiotherapy equipment. The system comprises: a model generation unit for generating a multi-objective optimization model based on the dose target itemset and the physical constraint parameter set, wherein the objective function of the multi-objective optimization model is a weighted summation function of each dose target item in the dose target itemset, and the constraints of the multi-objective optimization model include treating each physical constraint parameter in the physical constraint parameter set as a hard constraint on device feasibility; a model processing unit for iteratively solving the multi-objective optimization model to generate a Pareto front solution set, wherein the Pareto front solution set contains multiple non-dominated candidate solutions, each corresponding to a set of radiotherapy planning parameters; and a planning determination unit for determining the target radiotherapy plan based on the candidate solutions in the Pareto front solution set that simultaneously satisfy clinical preference conditions and device kinematic constraints.

[0088] Optionally, the model generation unit includes: a performance parameter acquisition subunit, used to acquire the clinical dosimetry performance parameters corresponding to each dose target item in the dose target item set, including target volume, organ at risk volume, prescribed dose value, and dose-volume constraint threshold; an expression transformation subunit, used to convert each dose target item into a mathematical expression about the dose distribution vector, each mathematical expression being used to quantify the degree of deviation between the current dose distribution and the clinical dosimetry performance parameters corresponding to the dose target item; a weight allocation subunit, used to assign a weight coefficient to each mathematical expression, the weight coefficient being adaptively determined according to the cancer type and radiotherapy equipment model; an objective function construction subunit, used to construct an optimization objective function by summing all mathematical expressions scaled by the weight coefficients, wherein the total value of the optimization objective function is the objective to be minimized in the multi-objective optimization model; and a model constraint subunit, used to use the set of physical constraint parameters as the solution constraints for the optimization objective function to obtain a multi-objective optimization model, wherein the optimizable variables of the multi-objective optimization model include the machine hop count vector and at least a portion of the parameters in the set of physical constraint parameters, and these parameters participate in the iterative solution of the multi-objective optimization model within a range not exceeding the corresponding physical upper limit of the equipment.

[0089] Optionally, the model constraint subunit includes: a dose influence matrix determination module, used to predetermine the dose influence matrix based on the radiotherapy equipment model, wherein each element of the dose influence matrix represents the dose contributed by a unit machine hop count to a voxel in a beam direction; a dose distribution calculation module, used to multiply the machine hop count vector with the dose influence matrix to obtain a dose distribution vector, wherein the machine hop count vector represents the numerical set of machine hop counts at each control point of the multi-leaf collimator or in each beam direction, used to represent the radiation output of the accelerator; and a hard constraint embedding module, used to embed the dose distribution vector as a hard constraint condition into the multi-objective optimization model.

[0090] Optionally, the acquisition unit includes: a context parameter acquisition subunit, used to acquire cancer type, radiotherapy equipment model, and anatomical structure features extracted from image data as context parameters; a weight value determination subunit, used to query a preset weight mapping rule base to obtain the weight value corresponding to each dose target item based on the context parameters, or to determine the weight value corresponding to each dose target item through a pre-trained weight prediction model, wherein the weight prediction model is a neural network model trained by supervised learning using cancer type, radiotherapy equipment model, and anatomical structure features of historical radiotherapy plans as training samples and the weight vector actually used in the historical radiotherapy plans as training labels; and a coefficient setting subunit, used to set the weight value corresponding to each dose target item as the coefficient corresponding to the dose target item in the optimization objective function.

[0091] Optionally, the acquisition unit is also configured to acquire one or more of the following dose targets: target area dose coverage target, used to limit the volume percentage of the target area receiving not less than the prescribed dose; target area dose uniformity target, used to limit the ratio or difference between the maximum dose and the minimum dose within the target area; organ at risk dose limit target, used to limit the maximum received dose value set for organs at risk; organ at risk volume dose target, used to limit the maximum volume percentage of the organ at risk receiving exceeding a specific dose threshold.

[0092] Optionally, the acquisition unit is also used to acquire one or more of the following physical constraint parameters: the blade width of the multi-leaf collimator, used to limit the minimum spatial resolution of beam modulation; the maximum dose rate of the accelerator, used to limit the maximum dose output per unit time; the minimum rotation angle interval of the gantry, used to limit the angle sampling accuracy of the beam direction; and the maximum movement speed of the multi-leaf collimator blades, used to limit the upper limit of the change in blade position between adjacent control points.

[0093] Optionally, the model processing unit includes: a population initialization subunit, used to initialize a population containing multiple candidate solutions, each candidate solution being a set of radiotherapy plan parameters; and an iterative solution subunit, used to perform at least one iterative solution operation on the current population until the convergence condition is met, and to use the final set of non-dominated candidate solutions as the Pareto front solution set, wherein each iterative solution operation is used to generate a new generation of population.

[0094] Optionally, the iterative solution subunit includes: a crossover and mutation module, used to perform crossover and mutation operations on candidate solutions in the current population in each iteration to generate offspring candidate solutions; a hard constraint verification module, used to verify whether each newly generated candidate solution satisfies the equipment feasibility hard constraint and eliminate candidate solutions that violate any equipment feasibility hard constraint; and a non-dominated sorting module, used to merge parent candidate solutions that satisfy the equipment feasibility hard constraint with offspring candidate solutions and sort them hierarchically according to the non-dominated relationship in multi-objective optimization, wherein the non-dominated relationship represents: for two candidate solutions, if the performance of the first candidate solution is not inferior to the second candidate solution in all dimensions of the optimization objective function, and is superior to the second candidate solution in at least one dimension. The first candidate solution dominates the second candidate solution; solutions not dominated by any other candidate solutions are non-dominated solutions; the hierarchy priority retention module is used to prioritize the retention of candidate solutions with the smallest non-dominated hierarchy value, wherein the non-dominated hierarchy represents the ranking hierarchy obtained by recursively stripping the current non-dominated solutions, and the candidate solution with the smallest hierarchy value has the highest dominance level; the diversity maintenance module is used to select and retain candidate solutions according to a preset diversity maintenance strategy when the number of non-dominated solutions in the same layer exceeds the population size, wherein the diversity maintenance strategy includes: a method based on reference point association, prioritizing the retention of candidate solutions with the smallest distance to the preset reference point, or a method based on crowding distance comparison, prioritizing the retention of candidate solutions with the largest crowding distance, so as to cover all regions of the Pareto front.

[0095] Optionally, the iterative solution subunit also includes a convergence condition judgment module, which is used to determine whether any of the following convergence conditions are met: the change amplitude of the non-dominated solution set in N consecutive iterations is lower than a preset threshold, where N is an integer greater than 1; the preset maximum number of iterations is reached; and all candidate solutions meet the preset clinical dose constraint conditions.

[0096] Optionally, the clinical preference criteria used in the planning unit may include: prioritization of organs at risk, or a trade-off coefficient between the lowest dose to the target area and the maximum dose to the organs at risk.

[0097] Optionally, the model processing unit further includes: a synchronous optimization subunit, used to synchronously optimize the machine hop count vector and at least one physical constraint parameter during the iterative solution of the multi-objective optimization model, wherein the physical constraint parameter includes one or more of the following: the maximum speed of the multi-leaf collimator blades and the upper limit of the actual dose rate of the accelerator; and each component of the machine hop count vector is constrained by the width of the multi-leaf collimator blades, the maximum dose rate of the accelerator, and the maximum speed of the multi-leaf collimator blades; the physical constraint parameters are adjusted within a range not exceeding the corresponding physical upper limit of the device so that the dose distribution corresponding to the finally generated candidate solution meets the clinical objectives.

[0098] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein when the computer program is executed, the device on which the computer-readable storage medium is located executes the above-described multi-objective plan optimization method based on different machine models and cancer types.

[0099] According to another aspect of the embodiments of this application, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the above-described multi-objective program optimization method based on different machine models and cancer types.

[0100] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program or instructions, which, when executed by a processor, implement the above-described multi-objective program optimization method based on different machine types and cancer types.

[0101] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0102] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0104] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0106] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0107] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A multi-objective treatment planning optimization method based on different machine types and cancer types, characterized in that, include: Collect an input dataset, which includes at least the image data of the target object, the type of cancer, and the model of the radiotherapy equipment. Obtain a set of dose targets corresponding to the cancer type and a set of physical constraint parameters corresponding to the radiotherapy equipment model. The set of dose targets includes at least one dose target, each dose target being used to quantify the clinical dosimetric performance parameters that the radiotherapy plan should achieve in the target area and organs at risk in the imaging data. The set of physical constraint parameters includes at least one physical constraint parameter, each physical constraint parameter being used to characterize the mechanical or dosimetric limitations of the radiotherapy equipment during operation. A multi-objective optimization model is generated based on the dose target itemset and the physical constraint parameter set. The objective function of the multi-objective optimization model is a function that sums the dose target items in the dose target itemset with weights. The constraints of the multi-objective optimization model include using the physical constraint parameters in the physical constraint parameter set as hard constraints on equipment feasibility. The multi-objective optimization model is iteratively solved to generate a Pareto front solution set, wherein the Pareto front solution set contains multiple non-dominated candidate solutions, and each candidate solution corresponds to a set of radiotherapy planning parameters; The target radiotherapy plan is determined based on the candidate solutions in the Pareto front solution set that simultaneously satisfy both clinical preference conditions and equipment kinematic constraints.

2. The method according to claim 1, characterized in that, A multi-objective optimization model is generated based on the dose target itemset and the physical constraint parameter set, including: Obtain the clinical dosimetric performance parameters corresponding to each dose target item in the dose target item set, including target volume, organ at risk volume, prescribed dose value, and dose-volume constraint threshold; Each dose target is converted into a mathematical expression about the dose distribution vector. Each mathematical expression is used to quantify the degree of deviation between the current dose distribution and the clinical dosimetric performance parameter corresponding to the dose target. Each mathematical expression is assigned a weighting coefficient, which is adaptively determined based on the cancer type and the radiotherapy equipment model. The optimization objective function is constructed by summing all the mathematical expressions after scaling by weight coefficients, wherein the total value of the optimization objective function is the objective that needs to be minimized in the multi-objective optimization model. The set of physical constraint parameters is used as the solution constraint condition for the optimization objective function to obtain the multi-objective optimization model. The optimizable variables of the multi-objective optimization model include the machine hop count vector and at least a portion of the parameters in the set of physical constraint parameters. These parameters participate in the iterative solution of the multi-objective optimization model within a range not exceeding the corresponding physical upper limit of the device.

3. The method according to claim 2, characterized in that, The physical constraint parameter set includes the linear relationship between the dose distribution vector and the machine hop count vector. Using this physical constraint parameter set as the solution constraint for the optimization objective function, the multi-objective optimization model is obtained, including: Based on the radiotherapy equipment model, a dose influence matrix is ​​predetermined, wherein each element of the dose influence matrix represents the dose contributed by a unit machine jump to a voxel in a beam direction; Multiplying the machine hop count vector with the dose influence matrix yields the dose distribution vector, where the machine hop count vector represents the set of machine hop count values ​​at each control point or in each beam direction of the multi-leaf collimator, and is used to characterize the radiation output of the accelerator. The dose distribution vector is embedded as a hard constraint in the multi-objective optimization model.

4. The method according to claim 1, characterized in that, The weights of each dose objective term in the optimization objective function are determined in the following way: The cancer type, the radiotherapy equipment model, and the anatomical features extracted from the image data are obtained as context parameters. Based on the context parameters, the weight value corresponding to each dose target item is obtained by querying a preset weight mapping rule base, or by determining the weight value corresponding to each dose target item through a pre-trained weight prediction model. The weight prediction model is a neural network model trained by supervised learning, using the cancer type, radiotherapy equipment model, and anatomical features of the historical radiotherapy plan as training samples and the weight vector actually used in the historical radiotherapy plan as training labels. The weight value corresponding to each dose target item is used as the coefficient corresponding to that dose target item in the optimization objective function.

5. The method according to claim 1, characterized in that, The dosage target includes at least one or more of the following targets: The target dose coverage parameter is used to define the volume percentage of the target area that receives at least the prescribed dose; The target dose uniformity objective is used to define the ratio or difference between the maximum dose and the minimum dose within the target area. At-risk organ dose limit target, used to limit the maximum acceptable dose value set for said at-risk organ; The volume target for at-risk organ is used to limit the maximum volume percentage of the at-risk organ that receives a dose exceeding a specific dose threshold.

6. The method according to claim 1, characterized in that, The physical constraint parameters include at least one or more parameters: The blade width of a multi-leaf collimator is used to define the minimum spatial resolution of beam modulation; The accelerator's maximum dose rate is used to limit the maximum dose output per unit time. The minimum rotation angle interval of the frame is used to limit the angle sampling accuracy of the beam direction; The maximum speed of the multi-leaf collimator blades is used to limit the upper limit of the change in blade position between adjacent control points.

7. The method according to claim 1, characterized in that, The steps for iteratively solving the multi-objective optimization model to generate the Pareto front solution set include: Initialize a population containing multiple candidate solutions, each candidate solution being a set of radiotherapy planning parameters; Perform at least one iterative solution operation on the current population until the convergence condition is met. The final set of non-dominated candidate solutions is taken as the Pareto front solution set. Each iterative solution operation is used to generate a new generation of population.

8. The method according to claim 7, characterized in that, Each iteration of the solution process includes: In each generation iteration, crossover and mutation operations are performed on the candidate solutions in the current population to generate offspring candidate solutions; For each newly generated candidate solution, verify whether the candidate solution satisfies the hard constraints on device feasibility, and eliminate candidate solutions that violate any of the hard constraints on device feasibility; The parent and child candidate solutions that satisfy the hard constraints of equipment feasibility are merged and sorted hierarchically according to the non-dominated relationship in multi-objective optimization. The non-dominated relationship is characterized as follows: for two candidate solutions, if the first candidate solution performs no worse than the second candidate solution in all dimensions of the optimization objective function and is better than the second candidate solution in at least one dimension, then the first candidate solution dominates the second candidate solution; the solution that is not dominated by any other candidate solution is a non-dominated solution. The candidate solution with the smallest non-dominated level value is retained first. The non-dominated level represents the sorting level obtained by recursively stripping the current non-dominated solution. The candidate solution with the smallest level value has the highest domination level. When the number of non-dominated solutions in the same layer exceeds the population size, candidate solutions are selected and retained according to a preset diversity maintenance strategy. The diversity maintenance strategy includes: a reference point association method that prioritizes retaining candidate solutions with the smallest distance to a preset reference point, or a crowding distance comparison method that prioritizes retaining candidate solutions with the largest crowding distance, so as to cover all regions of the Pareto front.

9. The method according to claim 7, characterized in that, The convergence condition is any one of the following: The change in the non-dominated solution set is less than a preset threshold in N consecutive iterations, where N is an integer greater than 1; The preset maximum number of iterations has been reached; All candidate solutions meet the pre-defined clinical dose constraints.

10. The method according to claim 1, characterized in that, The clinical preference criteria include: a priority ranking for the organs at risk, or a trade-off coefficient between the lowest dose to the target area and the maximum dose to the organs at risk.

11. The method according to claim 2, characterized in that, The machine hop count vector is used as a function of at least one physical constraint parameter in the physical constraint parameter set, and the method further includes: During the iterative solution of the multi-objective optimization model, the machine hop count vector and at least one physical constraint parameter are simultaneously optimized. The physical constraint parameters include one or more of the following: the maximum speed of the multi-leaf collimator blades and the upper limit of the actual dose rate of the accelerator; and each component of the machine hop vector is constrained by the width of the multi-leaf collimator blades, the maximum dose rate of the accelerator, and the maximum speed of the multi-leaf collimator blades; the physical constraint parameters are adjusted within the range not exceeding the corresponding physical upper limit of the device so that the dose distribution corresponding to the finally generated candidate solution meets the clinical goal.

12. A multi-objective treatment optimization device based on different machine models and cancer types, characterized in that, include: A dataset acquisition unit is used to acquire an input dataset, which includes at least the image data of the target object, the type of cancer, and the model of the radiotherapy equipment. An acquisition unit is configured to acquire a set of dose target items corresponding to the cancer type and a set of physical constraint parameters corresponding to the radiotherapy equipment model. The set of dose target items includes at least one dose target item, each used to quantify the clinical dosimetric performance parameters that the radiotherapy plan should achieve in the target area and organs at risk in the imaging data. The set of physical constraint parameters includes at least one physical constraint parameter, each used to characterize the mechanical or dosimetric limitations of the radiotherapy equipment during operation. The model generation unit is used to generate a multi-objective optimization model based on the dose target itemset and the physical constraint parameter set. The optimization objective function of the multi-objective optimization model is a function that performs a weighted summation of each dose target item in the dose target itemset. The constraint conditions of the multi-objective optimization model include using each physical constraint parameter in the physical constraint parameter set as a hard constraint on equipment feasibility. The model processing unit is used to iteratively solve the multi-objective optimization model to generate a Pareto front solution set, wherein the Pareto front solution set contains multiple non-dominated candidate solutions, and each candidate solution corresponds to a set of radiotherapy planning parameters; The planning unit is used to determine the target radiotherapy plan based on the candidate solutions in the Pareto front solution set that simultaneously satisfy the clinical preference conditions and the equipment kinematic constraints.

Images