Personalized joint reverse optimization method and device based on multi-agent collaborative design

Abstract

The application discloses a personalized joint reverse optimization method and device based on multi-agent collaborative design, and relates to the technical field of artificial joint optimization. The method comprises the following steps: based on an artificial joint structure-material field knowledge graph, according to clinical constraint conditions, a material design agent is used for material retrieval; based on the clinical constraint conditions, according to a material candidate list and a historical data set, a structure design agent is used for reverse optimization; according to the material candidate list, optimization structure parameters are used for compliance verification by a verification agent; according to material properties and verification structure parameters, a simulation agent is used for forward prediction; based on the material candidate list, according to simulation indexes, an optimization agent is used for material comprehensive sorting; and a traceable evidence chain is constructed. The application is an efficient and accurate personalized joint reverse optimization method based on patient information constraints, combined with a knowledge graph, multi-agents and a large language model.

Description

Technical Field

[0001] This invention relates to the field of artificial joint optimization design technology, and in particular to a personalized joint reverse optimization method and device based on multi-agent collaborative design. Background Technology

[0002] Artificial joint replacement surgery is an effective treatment for end-stage joint diseases. The design quality of the artificial joint prosthesis directly affects the patient's postoperative quality of life, prosthesis lifespan, and complication rate. Traditional artificial joint design methods have the following problems: Material selection lacks systematic decision support. Current material selection methods mainly rely on physician experience and manufacturer recommendations, lacking systematic decision support based on multi-dimensional evidence. Material selection needs to comprehensively consider multiple dimensions such as biocompatibility, mechanical properties, wear resistance, corrosion resistance, and processing feasibility. Traditional methods are insufficient to achieve multi-objective synergistic optimization.

[0003] The design of structural parameters lacks personalized adaptation. Existing prosthesis structural parameters are mainly designed based on statistical averages, failing to fully consider the impact of individual patient differences on prosthesis performance. Studies have shown that personalized structural parameters can significantly reduce wear rate and the risk of loosening.

[0004] Co-optimization of materials and structure is challenging. Material selection and structural parameter design are interdependent, with different material combinations resulting in different optimal structural parameters. Traditional design methods separate material selection from structural design, making it difficult to achieve co-optimization of materials and structure.

[0005] Design decisions lack interpretability and traceability. Existing design decision-making processes lack evidence chain tracing mechanisms, making it difficult to explain why specific material combinations and structural parameters were chosen, which is detrimental to clinical decision support and iterative design optimization.

[0006] Knowledge is fragmented and difficult to utilize effectively. The field of artificial joints has a wealth of knowledge resources, including literature, standards, and clinical data, but this knowledge is scattered from different sources and lacks a unified knowledge representation and reasoning mechanism, making it difficult to effectively support design decisions.

[0007] The existing system lacks the ability to enhance large language models, making it unable to dynamically generate personalized recommendation reasons and decision explanations based on patient characteristics, thus failing to meet clinicians' needs for explainable artificial intelligence.

[0008] In recent years, knowledge graph technology has been widely used in the medical field, enabling structured representation of knowledge and semantic reasoning. Multi-agent systems can decompose, collaborate, and optimize complex tasks. Breakthroughs in large language model technology have provided new possibilities for intelligent medical decision-making.

[0009] In the existing technology, there is a lack of an efficient and accurate personalized joint reverse optimization method that combines knowledge graphs, multi-agent systems, and large language models based on patient information constraints. Summary of the Invention

[0010] To address the technical problems of existing technologies, such as lack of systematic decision support for material selection; lack of personalized adaptation in structural parameter design; difficulty in co-optimizing materials and structures; lack of interpretability and traceability in design decisions; fragmented knowledge that is difficult to utilize effectively; and lack of large-scale model intelligent enhancement capabilities, this invention provides a personalized joint reverse optimization method and apparatus based on multi-agent collaborative design. The technical solution is as follows: On the one hand, a personalized joint back-optimization method based on multi-agent collaborative design is provided. This method is implemented by a personalized joint back-optimization device and includes: Acquire multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints; construct a knowledge graph of artificial joint structure-materials based on the pre-set Neo4j graph database, multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints. Obtain individual patient characteristics; based on the individual patient characteristics, use a clinical reasoning agent to perform reasoning mapping to obtain clinical constraints and clinical reasoning source identification information; Based on the knowledge graph of artificial joint structure-materials, and according to clinical constraints, a materials design agent is used to perform material retrieval to obtain a list of candidate materials, material properties, and material retrieval source identification information. Obtain historical simulation datasets; based on clinical constraints, and according to the material candidate list and historical datasets, use a structural design agent to perform reverse optimization to obtain optimized structural parameters and structural optimization iteration information; Based on the material candidate list, the optimized structural parameters are verified using a verification agent to obtain the verified structural parameters and verification result identification information. Based on material properties and verified structural parameters, a simulation agent is used to make positive predictions, and simulation indices and simulation prediction confidence are obtained. Based on the material candidate list, and according to simulation indicators, an optimization agent is used to comprehensively rank the materials to obtain the optimal design scheme and the ranking criteria identification information. Based on the source identification information of clinical reasoning, source identification information of material retrieval, structural optimization iteration information, verification result identification information, simulation prediction confidence level and ranking basis identification information, a traceable evidence chain is constructed to obtain evidence information for joint design decisions.

[0011] On the other hand, a personalized joint back-optimization device based on multi-agent cooperative design is provided. This device is applied to a personalized joint back-optimization method based on multi-agent cooperative design. The device includes: The knowledge graph construction module is used to acquire multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints; based on the preset Neo4j graph database, it constructs a knowledge graph of artificial joint structure-materials based on multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints. The clinical constraint conversion module is used to obtain individual patient characteristic information; based on the individual patient characteristic information, a clinical reasoning agent is used to perform reasoning mapping to obtain clinical constraint conditions and clinical reasoning source identification information; The material candidate retrieval module is used to perform material retrieval based on the knowledge graph of artificial joint structure-materials and according to clinical constraints, using a material design agent to obtain a list of material candidates, material properties, and material retrieval source identification information; The structural parameter optimization module is used to acquire historical simulation datasets; based on clinical constraints, and according to the material candidate list and historical datasets, a structural design agent is used to perform reverse optimization to obtain optimized structural parameters and structural optimization iteration information. The parameter compliance verification module is used to perform compliance verification on optimized structural parameters based on the material candidate list using a verification agent, and obtain the verified structural parameters and verification result identification information. The simulation index prediction module is used to make positive predictions using a simulation agent based on material properties and verification structure parameters, and to obtain simulation indexes and simulation prediction confidence. The optimal solution decision module is used to perform comprehensive material ranking based on the material candidate list and simulation indicators using an optimization agent to obtain the optimal design scheme and the ranking basis identification information. The evidence chain tracing module is used to construct a traceable evidence chain based on clinical reasoning source identification information, material retrieval source identification information, structural optimization iteration information, verification result identification information, simulation prediction confidence level, and ranking basis identification information to obtain joint design decision evidence information.

[0012] On the other hand, a personalized joint reverse optimization device is provided, the personalized joint reverse optimization device comprising: a processor; a memory storing computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, any one of the methods in the above-described personalized joint reverse optimization method based on multi-agent cooperative design is implemented.

[0013] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement any of the above-described personalized joint reverse optimization methods based on multi-agent cooperative design.

[0014] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: This invention proposes a personalized joint back optimization method based on multi-agent collaborative design. By using multi-agent collaboration, it achieves joint optimization of material selection and structural parameters, solving the problem of separating material selection and structural design in traditional methods. Combined with the differential evolution back optimization algorithm, it realizes integrated collaborative solution of structure and materials, improving the synergy and optimality of the design scheme. By analyzing individual patient characteristics through a clinical reasoning agent, and generating artificial joint prosthesis structural parameters that meet individual patient target constraints through a reverse optimization method, the prosthesis design has been transformed from "statistical averaging" to "personalized adaptation", which can significantly reduce wear rate and loosening risk. Multi-dimensional evidence retrieval and reasoning based on knowledge graphs provide a scientific basis for design decisions, avoiding the limitations of traditional methods that rely on experience and intuition, and improving the scientificity and reliability of design decisions. By using an evidence chain tracing mechanism, the evidence source and reasoning process of each design decision are recorded, supporting the interpretability of design decisions, making it easier for clinicians to understand and accept the design plan, and facilitating doctor-patient communication and informed consent; through complete execution trajectory recording and evidence chain construction, it is possible to trace back from the final plan to the original evidence, achieving the traceability of design decisions, and facilitating design iteration optimization and quality traceability. This invention utilizes a large-scale model deployed based on vLLM and its linkage enhancement capabilities. Leveraging the natural language understanding and logical reasoning abilities of a large language model, it provides large-scale model enhancement capabilities for clinical reasoning agents, material design agents, and verification agents, making design decisions more intelligent and human-centered, and generating natural language explanations and recommendation rationale. This invention is a highly efficient and accurate personalized joint back-optimization method based on patient information constraints, combining knowledge graphs, multi-agent systems, and a large language model. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1This is a flowchart of a personalized joint reverse optimization method based on multi-agent collaborative design provided by an embodiment of the present invention; Figure 2 This is a block diagram of a personalized joint reverse optimization device based on multi-agent collaborative design provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a personalized joint reverse optimization device provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] This invention provides a personalized joint reverse optimization method based on multi-agent collaborative design. This method can be implemented by a personalized joint reverse optimization device, which can be a terminal or a server. Figure 1 The flowchart shown is for a personalized joint back-optimization method based on multi-agent cooperative design. The processing flow of this method may include the following steps: S1. Obtain multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints; Based on the preset Neo4j graph database, construct a knowledge graph of artificial joint structure-materials based on multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints. Optionally, based on a pre-defined Neo4j graph database, a knowledge graph of artificial joint structure and materials is constructed using multi-source heterogeneous data in the field of artificial joints and BFO ontology reference documents, including: Based on the pre-defined Neo4j graph database and the BFO ontology reference document, the top-level category is defined using Neo4j node tags; Based on a pre-defined expert knowledge base for artificial joints and a pre-defined Neo4j graph database, modules are constructed according to top-level categories to obtain the artificial joint domain module ontology; the artificial joint domain module ontology includes anatomy module, material module, implant structure module, process module, and data traceability module; Based on the pre-defined Neo4j graph database, a four-layer relational architecture is configured according to the top-level categories and BFO ontology reference documents to obtain domain relation types. The domain relation types include the first layer structural skeleton relation, the second layer process-evidence relation, the third layer clinical semantic relation, and the fourth layer UMLS extended relation. Based on the pre-set Neo4j graph database, and according to the multi-source heterogeneous data in the artificial joint field, entity data instantiation is performed on the module ontology of the artificial joint field to obtain knowledge graph entity instances. Based on the domain relationship type, and according to the multi-source heterogeneous data in the artificial joint domain, the entity instances of the knowledge graph are instantiated with inter-entity association relationships to obtain knowledge graph relationship instances; Construct a knowledge graph for the artificial joint structure-materials domain based on entity instances and relationship instances of the knowledge graph.

[0023] In one feasible implementation, the resources required for constructing the knowledge graph in this invention include the Neo4j graph database (version 5.x and above) and the BFO ontology reference document. Neo4j node labels are used to define six top-level categories: AnatomicalEntity, MaterialEntity, Artifact, Process, InformationEntity, and Quality. A lightweight alignment relationship with the BFO ontology is established through node attribute constraints, clarifying the hierarchical structure and attribute constraints of each category.

[0024] Based on a pre-built knowledge base of experts in artificial joint fields such as anatomy, materials science, and implant structure, the following modules are created in Neo4j: anatomy (defining nodes such as bone structure, articular surfaces, and soft tissue), materials (defining metal, ceramic, and polymer materials and their performance properties), implant structure (defining structural entities such as the head, liner, stem, and conical neck and their parameters), process (defining processes such as material selection, structural design, simulation evaluation, and clinical surgery), and data sourcing (defining evidence entities such as literature, standards, experimental, and clinical data and their sourcing attributes).

[0025] In Neo4j, a four-layer relational architecture is configured using the CREATE statement: The first layer, the structural skeleton relation (part_of, has_property, composed_of), describes the structural composition between entities; the second layer, the process-evidence relation (has_input, has_output, supported_by), describes the process input and output and the supporting evidence; the third layer, the clinical semantic relation (indicated_for, contraindicated_with, suitable_for), describes clinical indications and contraindications; and the fourth layer, the UMLS extension relation (mapped_to, equivalent_to), implements semantic mapping with the Unified Medical Language System.

[0026] Multi-source heterogeneous data in the field of artificial joints includes PubMed literature databases, material performance databases (such as MATWEB), and artificial joint clinical registry data. Materials, structures, and clinical entities are extracted from literature using literature extraction tools; material mechanical properties and wear resistance data are imported from material performance databases; and patient characteristics and follow-up results are imported from clinical registry data. After being processed into CSV format, each entity is written into a graph database using Neo4j batch import tools (neo4j-admin import or LOAD CSV), with each node labeled with its data source, collection time, and confidence level.

[0027] Based on multi-source heterogeneous data in the field of artificial joints, this study establishes relationships between entities through entity linking and semantic matching algorithms, uses UMLS semantic network for concept disambiguation, and employs Cypher rule reasoning and graph query to verify the integrity of relationships. Finally, a knowledge graph is constructed and stored in the Neo4j graph database for multi-agent systems to retrieve and call using Cypher statements.

[0028] S2. Obtain individual patient characteristic information; based on the individual patient characteristic information, use a clinical reasoning agent to perform reasoning mapping to obtain clinical constraints and clinical reasoning source identification information; In one feasible implementation, the clinical reasoning agent employs a few-shot prompting algorithm, injecting historical patient-constraint mappings as example prompts to guide the large model in generating clinical constraint weights that match patient characteristics.

[0029] Input parameters include patient age (numerical, ranging from 18 to 100 years), weight (numerical, ranging from 40 to 100 kg), activity level (enumerated, high / medium / low), diagnosis (text), and diagnosis (list). The output is a clinical constraint data structure.

[0030] The clinical reasoning source identifier information is used to record the generation source of clinical constraints, including five fields: objective weight source, large language model recommended content, large language model original output, large language model call status, and error information. This identifier information serves as the starting node of the evidence chain, and together with patient information and clinical constraints, it constitutes a complete evidentiary unit for clinical decision-making, supporting the reverse tracing of design decisions.

[0031] The objective weight source identifies the source of the final decision, and its value can be a rule, a large language model, or a combination of both. When patient information completely matches the rule conditions, it is marked as a rule; when the rule and the large language model's suggestion conflict, the rule output is given priority and the objection is recorded, and it is marked as a combination; when the rule does not cover the entire process, it is marked as a large language model.

[0032] The large language model records the adjustment suggestions and evaluation reasons generated for the recommended content, saves the complete JSON response of the original output of the large language model, records the call status as success, failure or timeout, and records the specific reasons for call failure in the error message.

[0033] Among them, the clinical reasoning agent, the material design agent, the verification agent, and the optimization agent are built based on the Tongyi Qianwen big model; the structural design agent is built based on the differential evolution algorithm; and the simulation agent is built based on the extreme random tree regression model.

[0034] In one feasible implementation, the present invention constructs a multi-agent system comprising six specialized agents. The clinical reasoning agent converts patient information into clinical constraints, including risk level, target lifespan, and maximum permissible wear volume. The materials design agent generates multiple sets of material candidate combinations, retrieves material evidence from a knowledge graph, calculates a comprehensive material score, and outputs the optimal material candidate solution. The structural design agent supports both forward prediction and backward optimization modes; the backward optimization mode optimizes and generates optimal structural parameters based on patient information and target constraints. The verification agent verifies the completeness, legality, and consistency of material and structural candidates. The simulation agent predicts pressure and wear indicators based on an extreme random tree model. The optimization agent performs multi-objective comprehensive ranking based on material scores and simulation results.

[0035] The clinical reasoning agent receives patient information (age, weight, activity level, diagnosis, type of surgery, complications) and outputs clinical constraints including risk level (high_load / standard / low_load), target service life (years), and maximum permissible wear volume (mm³). The materials design agent outputs candidate material information including liner material, ball head material, handle material, friction pair type, number of pieces of evidence, evidence score, and overall score. The structural design agent outputs structural parameters including ball head diameter (mm), liner thickness (mm), radial clearance (mm), outward inclination angle (degrees), conical neck base diameter (mm), neck length (mm), eccentricity (mm), handle width (mm), and femoral handle modulus (GPa). The verification agent outputs verification results including verification status (true / false) and an error list. The simulation agent outputs simulation indices corresponding to the structural parameters. The optimization agent outputs a multi-objective comprehensive ranking result based on material scores and simulation results.

[0036] This invention provides large-scale model enhancement capabilities for clinical reasoning agents, materials design agents, verification agents, and optimization agents. The clinical reasoning agent utilizes a large language model to convert patient information into clinical constraints, including risk level, target lifespan, and maximum permissible wear volume. Output fields include `objective_weight_source` (identifying the weight source), `llm_recommendation` (recording the recommendations generated by the large model), `llm_output` (recording the original output of the large model), and `llm_status` and `llm_error` (recording the large model's call status and error information). The materials design agent uses the large model to generate material recommendation reasons. The verification agent uses the large model to perform design verification and anomaly detection, generating natural language explanations of the verification results.

[0037] This invention adopts a technical approach that combines prompt engineering with structured output constraints. It designs a unique system prompt template for each agent and uses JSON Schema to constrain the output format of the large model, ensuring that the output can be parsed by downstream programs.

[0038] The large language model service is deployed using the vLLM framework, integrating various open-source large language models and providing standardized text generation and dialogue generation interfaces. The Agent enhancement mode provides large model enhancement capabilities for clinical reasoning agents, material design agents, and validation agents, including intelligent generation of clinical constraint weights, generation of material recommendation reasons, design validation, and intelligent anomaly detection. A decision source identification mechanism explicitly marks the decision source in the output of each Agent, including the fields `objective_weight_source`, `llm_recommendation`, `llm_output`, `llm_status`, and `llm_error`. The orchestration layer's intelligent decision-making includes routing decisions, Agent selection, constraint coordination, result aggregation, anomaly interpretation, and large model failure degradation handling. vLLM's continuous batching and PagedAttention mechanisms are used to improve inference throughput, key-value caching reduces repetitive inference latency, and frequently called prompt templates are cached to reduce token consumption.

[0039] Clinical constraints include the patient's activity risk level, the lifespan of the artificial joint, and the maximum permissible wear volume of the artificial joint.

[0040] In one feasible implementation, the clinical constraint data obtained in this invention includes patient activity risk level (enumerated type, high_load / standard / low_load), artificial joint lifespan (numerical type, 10-30 years), and maximum allowable wear volume of artificial joint (numerical type, mm³), presented in the form of a structured dictionary, and applied to the target constraint calculation of subsequent back optimization.

[0041] The target constraints, including maximum peak pressure, maximum mean pressure, and maximum cervical pressure, are dynamically calculated based on the patient's weight, activity level, and risk level. A base value of 0.000003 mm³ is used, with adjustments made sequentially based on risk level, weight, and activity level.

[0042] The risk level adjustment rules are as follows: standard remains unchanged, fragile / high load is multiplied by 0.7; weight adjustment rules are as follows: ≥90 kg is multiplied by 0.7, 75-90 kg is multiplied by 0.85, 65-75 kg is multiplied by 1, <65 kg is multiplied by 1.2; activity level adjustment rules are as follows: high is multiplied by 0.6, medium is multiplied by 0.8, low is multiplied by 1.3.

[0043] S3. Based on the knowledge graph of artificial joint structure-materials, and according to clinical constraints, a materials design agent is used to perform material retrieval to obtain a list of candidate materials, material properties, and material retrieval source identification information. Optionally, based on a knowledge graph of artificial joint structure-materials, and according to clinical constraints, a materials design agent is used to perform materials retrieval to obtain a list of candidate materials, material properties, and material retrieval source identification information, including: Determine the material screening constraints based on clinical requirements; Based on the knowledge graph of artificial joint structure-materials, according to the material selection constraint boundary, the Cypher query statement is used to retrieve material entities that meet the constraints and obtain a list of retrieved materials. Based on preset weights and the list of retrieved materials, a weighted scoring algorithm is used to calculate the comprehensive score of the candidate materials. Based on the preset material screening prompt template, and according to the search material list and the comprehensive score of candidate materials, the large language model is used to enhance the search and generate the material candidate list and the material search source identification information. Based on the knowledge graph of artificial joint structure-materials, entity attribute queries are performed according to the material candidate list to obtain material attributes.

[0044] In one feasible implementation, the input to the materials design agent consists of clinical constraints and patient context information. The clinical constraints include the prosthesis's mechanical load risk level, the prosthesis's designed service life, and the upper limit of the wear rate of the friction pair. The patient context information includes the patient's age, weight, activity level, diagnosis, bone quality, and surgical type, all encapsulated in JSON format. The materials design agent employs a Retrieval-Augmented Generation (RAG) algorithm, first retrieving relevant material evidence from a knowledge graph, and then injecting the retrieval results as context into the large model to generate reasons for material recommendations.

[0045] The material selection constraints were determined based on the prosthesis's mechanical load risk level and the upper limit of the friction pair's volumetric wear rate, as defined in clinical constraints. The retrieval based on the artificial joint structure-materials knowledge graph was achieved using Cypher queries.

[0046] The weighted scoring algorithm has preset weights including biocompatibility (0.3), wear resistance (0.3), mechanical properties (0.2), and bone ingrowth (0.2). The score for each material candidate is calculated as the sum of the products of each dimension's score and its corresponding weight. The scores for each dimension are derived from the attribute values ​​of the material entities in the knowledge graph and the confidence level of the associated evidence.

[0047] In the retrieval enhancement generation phase, the system uses the retrieved material list as a contextual injection prompt template, along with the comprehensive scores of candidate materials, and inputs it into the material design agent. The large language model generates reasons for material recommendations and generates a candidate material list, sorted in descending order of comprehensive scores. Each item in the candidate material list includes the lining material, ball head material, handle material, friction pair type, number of pieces of evidence, evidence score, and comprehensive score, presented in list form and passed to the structural design agent and optimization agent.

[0048] During the entity attribute query phase, the system queries the physical attributes of the corresponding material entity in the Material knowledge graph module based on the material name in the material candidate list.

[0049] The material retrieval source identification information records the knowledge graph retrieval query statement, the number of retrieved evidences and the average confidence level, the recommendation reasons of the large language model, the original output, the call status and error information.

[0050] S4. Obtain historical simulation datasets; based on clinical constraints, and according to the material candidate list and historical datasets, use a structural design agent to perform reverse optimization to obtain optimized structural parameters and structural optimization iteration information. Optionally, based on clinical constraints, and using a candidate material list and historical datasets, a structural design agent is used for reverse optimization to obtain optimized structural parameters and structural optimization iteration information, including: Based on clinical constraints, the parameter search boundary is determined according to the material candidate list, and a target optimization problem is established; Based on historical simulation datasets, a weighted scoring method was used to select samples and obtain the initial population. Based on the objective optimization problem, and according to the initial population and parameter search boundary, the differential evolution algorithm is used to perform iterative optimization search to obtain the optimized structure parameters and structure optimization iteration information.

[0051] In one feasible implementation, the structural design agent supports two modes: forward prediction and backward optimization. The backward optimization mode generates optimal structural parameters based on patient information and target constraints. Its input parameters are target constraints and material candidate schemes, and the output is a structural parameter data structure, including the ball head diameter (mm), liner thickness (mm), radial clearance (mm), outward inclination angle (degrees), cone neck bottom diameter (mm), neck length (mm), eccentricity (mm), stem width (mm), and femoral stem modulus (GPa), which is applied to the verification agent and the simulation agent.

[0052] In the reverse optimization process, the historical simulation data used were the artificial joint finite element simulation case library and the in vitro wear experiment database. Wear simulations with different structural parameters and material combinations were executed in batches through the parametric finite element simulation platform, while experimental data from the in vitro hip joint wear simulator were included as verification samples. The preprocessing steps included outlier removal (identifying and removing outliers using the isolated forest algorithm), data normalization (standardizing each index to the [0,1] interval using Min-Max), and feature engineering (extracting five key indicators: peak contact pressure of the liner, historical average contact pressure of the liner, cumulative wear volume of the liner, peak contact pressure of the conical neck, and cumulative mechanical wear volume of the conical neck). The storage format adopted was a structured JSON file, with each record containing a case identifier, structural parameters, material combination, simulation index, and collection timestamp. The update mechanism was to regularly add simulation cases and retrain the forward prediction model every month to ensure data timeliness.

[0053] A weighted scoring method was used to select the top ten best samples from historical simulation data as the initial population. The scoring formula is: Score = w1×Y1_norm + w2×Y2_norm + w3×Y3_norm + w4×Y4_norm + w5×Y5_norm + Penalty, where Y1_norm, Y2_norm, Y3_norm, Y4_norm, and Y5_norm are the normalized peak contact pressure of the liner, the historical average contact pressure of the liner, the cumulative wear volume of the liner, the peak contact pressure of the conical neck, and the cumulative mechanical wear volume of the conical neck, respectively; w1, w2, w3, w4, and w5 are the weights of Y1_norm, Y2_norm, Y3_norm, Y4_norm, and Y5_norm, respectively; and Penalty is the edge penalty term.

[0054] The differential evolution algorithm is used for global optimization search. During the iteration process, the positive prediction model (extreme random tree model) is called to evaluate the quality of candidate solutions and output the optimization parameters that meet the constraints, including the ball head diameter, liner thickness, radial clearance, outward inclination angle, cone neck bottom diameter, neck length, eccentricity, shank width, femoral shank modulus, and the predicted simulation indicators, including liner contact peak pressure, liner historical average contact pressure, liner cumulative wear volume, cone neck contact peak pressure, and cone neck cumulative mechanical wear volume.

[0055] The structural optimization iteration information comes from the complete optimization search process of the differential evolution algorithm, including three parts: initial population generation record, iteration evolution log and convergence output result. It supports tracing back from the optimal structural parameters to the source of historical simulation data and the optimization search path.

[0056] The initial population generation record includes the population size, the weighted score of each sample, and the selection criteria; the iterative evolution log records the optimal individual structure parameters, mutation vectors, crossover results, and selection results of each generation of the population, forming a complete evolutionary trajectory from the initial population to the convergent optimal solution; the convergence output results include the termination reason, the final iteration generation, the optimal structure parameters, and constraint satisfaction analysis.

[0057] S5. Based on the material candidate list, optimize the structural parameters and use a verification agent to perform compliance verification to obtain the verification structural parameters and verification result identification information. In one feasible implementation, the input parameters of the verification agent are a list of candidate materials and optimized structural parameters, and the output is the verified structural parameters and verification result identification information, including verification status (Boolean) and error list, which is applied to anomaly interception and process control.

[0058] The validation process for the intelligent agent consists of two stages: rule validation and large language model enhancement validation. In the rule validation stage, the system loads the validation rule base and performs parameter range checks, material-structure compatibility checks, and clinical constraint consistency checks. Parameter range checks verify whether each structural parameter is within its allowable range. Material-structure compatibility checks verify whether the combination of candidate materials and structural parameters matches. Clinical constraint consistency checks verify whether the optimized structural parameters meet the prosthesis mechanical load risk level and the upper limit of friction pair volumetric wear rate in the clinical constraint conditions.

[0059] During the enhanced verification phase of the large language model, the system loads a dedicated thought chain prompt template to guide the reasoning and verification logic. Abnormal combinations that the large language model's rules cannot cover generate natural language alerts and risk warnings. After parsing the large language model's output using JSON Schema, the anomaly judgment results and explanatory text are extracted and integrated into the verification results.

[0060] When the verification status is true, the verified structural parameters are passed to the simulation agent for positive prediction. When the verification status is false, the system triggers an exception interception mechanism, interrupts the subsequent process, returns the error list to the structural design agent for parameter correction, or returns to the material design agent to re-search for materials.

[0061] S6. Based on the material properties and the verification structural parameters, use the simulation agent to make positive predictions and obtain simulation indicators and simulation prediction confidence. Optionally, based on material properties and validated structural parameters, a simulation agent is used to perform forward predictions to obtain simulation metrics and simulation prediction confidence levels, including: The material physical property vector and the verification structure parameter vector are obtained by vectorizing the material properties and the verification structure parameter vector. Based on the material physical property vector and the verification structure parameter vector, batch inference is performed using an extreme random tree regression model to obtain decision tree prediction results. The prediction results of the decision tree are averaged to obtain the simulation index; The variance is calculated based on the prediction results of the decision tree to obtain the confidence level of the simulation prediction.

[0062] In one feasible implementation, the positive prediction model used by the simulated intelligent agent is an ExtraTrees regression model. Its model architecture is an ensemble regressor composed of multiple decision trees. Each tree splits nodes using a random threshold strategy, and the final prediction value is the mean of the prediction results of all trees.

[0063] The model training method uses historical structural parameters as input features and simulation indicators as labels, and employs five-fold cross-validation for hyperparameter tuning. The model performance indicators are evaluated by the coefficient of determination R² and the mean square error MSE, requiring the predicted wear volume of the lining to have an R² of no less than 0.85. The application logic is to use the forward prediction model to quickly evaluate the quality of candidate solutions in each generation of the back-optimization differential evolution iteration, replacing the high-cost finite element simulation.

[0064] S7. Based on the material candidate list, according to the simulation indicators, use the optimization agent to perform comprehensive ranking of materials to obtain the optimal design scheme and the ranking basis identification information; In one feasible implementation, the input parameters of the optimization agent are material scores and simulation results, and the output is a list of ranked schemes, arranged in descending order of comprehensive score. Each item includes material combination, structural parameters, simulation index, and comprehensive score, which are applied to the recommendation and output of the final design scheme.

[0065] The overall score is obtained by weighted fusion of material scores and simulation indicators. A dedicated ranking prompt template is loaded, and candidate materials, simulation indicators, and weight configurations are input. This guides the large language model to output the reasons for recommending each scheme and risk warnings. The reasons for recommendation are parsed using JSON Schema and then assembled with the overall score, material combinations, structural parameters, and simulation indicators into a ranking list of schemes.

[0066] The sorting is based on the identification information recording the source of the comprehensive score calculation, the weight configuration version, the recommendation reason of the large language model, and the calling status. The optimal design scheme is passed to the evidence chain construction module to generate traceable joint design decision evidence information.

[0067] S8. Construct a traceable chain of evidence based on clinical reasoning source identification information, material retrieval source identification information, structural optimization iteration information, verification result identification information, simulation prediction confidence level, and ranking basis identification information to obtain joint design decision evidence information.

[0068] In one feasible implementation, the inputs to the evidence chain construction module are clinical reasoning source identifiers, material retrieval source identifiers, structural optimization iteration information, verification result identifiers, simulation prediction confidence levels, and ranking basis identifiers. These information originate from the execution outputs of the clinical reasoning agent, material design agent, structural design agent, verification agent, simulation agent, and optimization agent, respectively, covering the entire decision-making process from patient information input to the output of the optimal design scheme.

[0069] The algorithm analyzes the source identification information to extract the decision source type, large language model call status, knowledge graph retrieval records, and algorithm execution logs. Clinical reasoning source identification information includes objective weight source, large language model recommended content, original output, call status, and error information; material retrieval source identification information includes knowledge graph retrieval query, number of pieces of evidence, average confidence score, large language model recommendation reasons, and decision source; structural optimization iteration information includes initial population selection records, differential evolution iteration logs, convergence judgment criteria, and optimal structure parameter generation path; verification result identification information includes rule verification source, large language model verification source, mixed verification source, and anomaly judgment records; simulation prediction confidence includes the variance of each decision tree prediction result and the degree of deviation between the input vector and the training data distribution; ranking basis identification information includes the comprehensive score calculation source, weight configuration version, large language model recommendation reasons, and call status.

[0070] The evidence chain construction module categorizes the above information according to decision nodes, establishing a ternary association of evidence-knowledge-decision. Each design decision node is associated with a corresponding evidence entity, knowledge graph entity, and decision source identifier. Evidence types include documentary evidence, standard evidence, experimental evidence, and clinical experimental evidence, spanning all stages from material selection and structural design to simulation evaluation and comprehensive ranking. The association between evidence sources and the knowledge graph is as follows: each piece of evidence is instantiated as an information entity in the knowledge graph, associated with the corresponding material entity or implant structure entity through "reference" relationships, and linked to the specific design decision node through "support" relationships.

[0071] The confidence level is calculated by considering three factors: the weight of the evidence type, the timeliness of the evidence, and the authority of the evidence source. The formula is Conf = α × TypeWeight + β × RecencyScore + γ × AuthorityScore, where α, β, and γ are weighting coefficients, ranging from 0 to 1, and α + β + γ = 1. TypeWeight is assigned a value of 0.9 for documentary evidence, 1.0 for standard evidence, 0.8 for experimental evidence, and 0.95 for clinically tested evidence. RecencyScore is calculated based on the number of years since the evidence was published or collected, decaying exponentially. AuthorityScore is determined based on the impact factor of the journal from which the evidence was obtained or the qualifications of the experimental institution. The confidence level assessment criteria are: Conf ≥ 0.8 is considered high confidence, 0.6 ≤ Conf < 0.8 is considered medium confidence, and Conf < 0.6 is considered low confidence.

[0072] The evidence chain building module connects fragmented evidence into a complete chain, supporting tracing from the final solution back to the original foundational evidence. Joint design decision-making evidence information includes a list of decision nodes, evidence source mapping, confidence ratings, and traceability paths, stored in structured data format for use in design process auditing, quality traceability, and doctor-patient communication.

[0073] This invention proposes a personalized joint back optimization method based on multi-agent collaborative design. By using multi-agent collaboration, it achieves joint optimization of material selection and structural parameters, solving the problem of separating material selection and structural design in traditional methods. Combined with the differential evolution back optimization algorithm, it realizes integrated collaborative solution of structure and materials, improving the synergy and optimality of the design scheme. By analyzing individual patient characteristics through a clinical reasoning agent, and generating artificial joint prosthesis structural parameters that meet individual patient target constraints through a reverse optimization method, the prosthesis design has been transformed from "statistical averaging" to "personalized adaptation", which can significantly reduce wear rate and loosening risk. Multi-dimensional evidence retrieval and reasoning based on knowledge graphs provide a scientific basis for design decisions, avoiding the limitations of traditional methods that rely on experience and intuition, and improving the scientificity and reliability of design decisions. By using an evidence chain tracing mechanism, the evidence source and reasoning process of each design decision are recorded, supporting the interpretability of design decisions, making it easier for clinicians to understand and accept the design plan, and facilitating doctor-patient communication and informed consent; through complete execution trajectory recording and evidence chain construction, it is possible to trace back from the final plan to the original evidence, achieving the traceability of design decisions, and facilitating design iteration optimization and quality traceability. This invention utilizes a large-scale model deployed based on vLLM and its linkage enhancement capabilities. Leveraging the natural language understanding and logical reasoning abilities of a large language model, it provides large-scale model enhancement capabilities for clinical reasoning agents, material design agents, and verification agents, making design decisions more intelligent and human-centered, and generating natural language explanations and recommendation rationale. This invention is a highly efficient and accurate personalized joint back-optimization method based on patient information constraints, combining knowledge graphs, multi-agent systems, and a large language model.

[0074] Figure 2 This is a block diagram of a personalized joint reverse optimization device based on multi-agent cooperative design, provided by an embodiment of the present invention. This device is used in a personalized joint reverse optimization method based on multi-agent cooperative design. (Refer to...) Figure 2 The device includes a knowledge graph construction module 210, a clinical constraint conversion module 220, a material candidate retrieval module 230, a structural parameter optimization module 240, a parameter compliance verification module 250, a simulation index prediction module 260, an optimal solution decision-making module 270, and an evidence chain tracing module 280. Among them: The knowledge graph construction module 210 is used to acquire multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints; based on the preset Neo4j graph database, it constructs a knowledge graph in the field of artificial joint structure-materials according to the multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints. The clinical constraint conversion module 220 is used to acquire individual patient characteristic information; based on the individual patient characteristic information, a clinical reasoning agent is used to perform reasoning mapping to obtain clinical constraint conditions and clinical reasoning source identification information; The material candidate retrieval module 230 is used to perform material retrieval based on the knowledge graph of artificial joint structure-materials and according to clinical constraints, using a material design intelligent agent to obtain a list of material candidates, material properties and material retrieval source identification information; The structural parameter optimization module 240 is used to acquire historical simulation datasets; based on clinical constraints, and according to the material candidate list and historical datasets, a structural design agent is used to perform reverse optimization to obtain optimized structural parameters and structural optimization iteration information. The parameter compliance verification module 250 is used to perform compliance verification on the optimized structural parameters based on the material candidate list using a verification intelligent agent, and obtain the verified structural parameters and verification result identification information. The simulation index prediction module 260 is used to make positive predictions using a simulation agent based on material properties and verification structural parameters, and to obtain simulation indexes and simulation prediction confidence. The optimal solution decision module 270 is used to perform comprehensive material ranking based on the material candidate list and simulation indicators using an optimization agent to obtain the optimal design scheme and ranking basis identification information. The evidence chain tracing module 280 is used to construct a traceable evidence chain based on clinical reasoning source identification information, material retrieval source identification information, structural optimization iteration information, verification result identification information, simulation prediction confidence level and sorting basis identification information, so as to obtain joint design decision evidence information.

[0075] Optionally, the knowledge graph construction module 210 is further used for: Based on the pre-defined Neo4j graph database and the BFO ontology reference document, the top-level category is defined using Neo4j node tags; Based on a pre-defined expert knowledge base for artificial joints and a pre-defined Neo4j graph database, modules are constructed according to top-level categories to obtain the artificial joint domain module ontology; the artificial joint domain module ontology includes anatomy module, material module, implant structure module, process module, and data traceability module; Based on the pre-defined Neo4j graph database, a four-layer relational architecture is configured according to the top-level categories and BFO ontology reference documents to obtain domain relation types. The domain relation types include the first layer structural skeleton relation, the second layer process-evidence relation, the third layer clinical semantic relation, and the fourth layer UMLS extended relation. Based on the pre-set Neo4j graph database, and according to the multi-source heterogeneous data in the artificial joint field, entity data instantiation is performed on the module ontology of the artificial joint field to obtain knowledge graph entity instances. Based on the domain relationship type, and according to the multi-source heterogeneous data in the artificial joint domain, the entity instances of the knowledge graph are instantiated with inter-entity association relationships to obtain knowledge graph relationship instances; Construct a knowledge graph for the artificial joint structure-materials domain based on entity instances and relationship instances of the knowledge graph.

[0076] Among them, the clinical reasoning agent, the material design agent, the verification agent, and the optimization agent are built based on the Tongyi Qianwen big model; the structural design agent is built based on the differential evolution algorithm; and the simulation agent is built based on the extreme random tree regression model.

[0077] Clinical constraints include the patient's activity risk level, the lifespan of the artificial joint, and the maximum permissible wear volume of the artificial joint.

[0078] Optionally, the material candidate retrieval module 230 is further configured to: Determine the material screening constraints based on clinical requirements; Based on the knowledge graph of artificial joint structure-materials, according to the material selection constraint boundary, the Cypher query statement is used to retrieve material entities that meet the constraints and obtain a list of retrieved materials. Based on preset weights and the list of retrieved materials, a weighted scoring algorithm is used to calculate the comprehensive score of the candidate materials. Based on the preset material screening prompt template, and according to the search material list and the comprehensive score of candidate materials, the large language model is used to enhance the search and generate the material candidate list and the material search source identification information. Based on the knowledge graph of artificial joint structure-materials, entity attribute queries are performed according to the material candidate list to obtain material attributes.

[0079] Optionally, the structural parameter optimization module 240 is further used for: Based on clinical constraints, the parameter search boundary is determined according to the material candidate list, and a target optimization problem is established; Based on historical simulation datasets, a weighted scoring method was used to select samples and obtain the initial population. Based on the objective optimization problem, and according to the initial population and parameter search boundary, the differential evolution algorithm is used to perform iterative optimization search to obtain the optimized structure parameters and structure optimization iteration information.

[0080] Optionally, the simulation index prediction module 260 is further used for: The material physical property vector and the verification structure parameter vector are obtained by vectorizing the material properties and the verification structure parameter vector. Based on the material physical property vector and the verification structure parameter vector, batch inference is performed using an extreme random tree regression model to obtain decision tree prediction results. The prediction results of the decision tree are averaged to obtain the simulation index; The variance is calculated based on the prediction results of the decision tree to obtain the confidence level of the simulation prediction.

[0081] This invention proposes a personalized joint back optimization method based on multi-agent collaborative design. By using multi-agent collaboration, it achieves joint optimization of material selection and structural parameters, solving the problem of separating material selection and structural design in traditional methods. Combined with the differential evolution back optimization algorithm, it realizes integrated collaborative solution of structure and materials, improving the synergy and optimality of the design scheme. By analyzing individual patient characteristics through a clinical reasoning agent, and generating artificial joint prosthesis structural parameters that meet individual patient target constraints through a reverse optimization method, the prosthesis design has been transformed from "statistical averaging" to "personalized adaptation", which can significantly reduce wear rate and loosening risk. Multi-dimensional evidence retrieval and reasoning based on knowledge graphs provide a scientific basis for design decisions, avoiding the limitations of traditional methods that rely on experience and intuition, and improving the scientificity and reliability of design decisions. By using an evidence chain tracing mechanism, the evidence source and reasoning process of each design decision are recorded, supporting the interpretability of design decisions, making it easier for clinicians to understand and accept the design plan, and facilitating doctor-patient communication and informed consent; through complete execution trajectory recording and evidence chain construction, it is possible to trace back from the final plan to the original evidence, achieving the traceability of design decisions, and facilitating design iteration optimization and quality traceability. This invention utilizes a large-scale model deployed based on vLLM and its linkage enhancement capabilities. Leveraging the natural language understanding and logical reasoning abilities of a large language model, it provides large-scale model enhancement capabilities for clinical reasoning agents, material design agents, and verification agents, making design decisions more intelligent and human-centered, and generating natural language explanations and recommendation rationale. This invention is a highly efficient and accurate personalized joint back-optimization method based on patient information constraints, combining knowledge graphs, multi-agent systems, and a large language model.

[0082] Figure 3 This is a schematic diagram of the structure of a personalized joint reverse optimization device provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the personalized joint reverse optimization device may include the above-mentioned Figure 2 The illustrated personalized joint inverse optimization device is based on multi-agent collaborative design. Optionally, the personalized joint inverse optimization device 310 may include a first processor 2001.

[0083] Optionally, the personalized joint reverse optimization device 310 may also include a memory 2002 and a transceiver 2003.

[0084] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0085] The following is combined with Figure 3 A detailed introduction to each component of the personalized joint reverse optimization device 310 is provided below: The first processor 2001 is the control center of the personalized joint reverse optimization device 310. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0086] Optionally, the first processor 2001 can perform various functions of the personalized joint reverse optimization device 310 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0087] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 3 CPU0 and CPU1 are shown in the diagram.

[0088] In a specific implementation, as one example, the personalized joint reverse optimization device 310 may also include multiple processors, for example... Figure 3 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0089] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0090] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and be used to reverse-optimize the interface circuitry of the device 310 via a personalized joint. Figure 3 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0091] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0092] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 3 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0093] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently and its interface circuitry can be optimized by reverse engineering the device 310 through a customized joint. Figure 3 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0094] It should be noted that, Figure 3 The structure of the personalized joint reverse optimization device 310 shown in the figure does not constitute a limitation on the router. The actual personalized joint reverse optimization device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0095] Furthermore, the technical effect of the personalized joint reverse optimization device 310 can be referred to the technical effect of the personalized joint reverse optimization method based on multi-agent collaborative design described in the above method embodiments, and will not be repeated here.

[0096] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or it may be any conventional processor, etc.

[0097] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DRRAM).

[0098] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0099] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0100] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0101] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0105] 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 network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, the functional units in the various embodiments of the present invention 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.

[0107] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a 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 invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A personalized joint back-optimization method based on multi-agent cooperative design, characterized in that, The method includes: Acquire multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints; construct a knowledge graph of artificial joint structure-materials based on the pre-set Neo4j graph database, multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints. Obtain individual patient characteristics; based on the individual patient characteristics, use a clinical reasoning agent to perform reasoning mapping to obtain clinical constraints and clinical reasoning source identification information; Based on the knowledge graph of artificial joint structure-materials, and according to clinical constraints, a materials design agent is used to perform material retrieval to obtain a list of candidate materials, material properties, and material retrieval source identification information. Obtain historical simulation datasets; based on clinical constraints, and according to the material candidate list and historical datasets, use a structural design agent to perform reverse optimization to obtain optimized structural parameters and structural optimization iteration information; Based on the material candidate list, the optimized structural parameters are verified using a verification agent to obtain the verified structural parameters and verification result identification information. Based on material properties and verified structural parameters, a simulation agent is used to make positive predictions, and simulation indices and simulation prediction confidence are obtained. Based on the material candidate list, and according to simulation indicators, an optimization agent is used to comprehensively rank the materials to obtain the optimal design scheme and the ranking criteria identification information. Based on the source identification information of clinical reasoning, source identification information of material retrieval, structural optimization iteration information, verification result identification information, simulation prediction confidence level and ranking basis identification information, a traceable evidence chain is constructed to obtain evidence information for joint design decisions.

2. The personalized joint reverse optimization method based on multi-agent cooperative design according to claim 1, characterized in that, The aforementioned knowledge graph, based on the pre-set Neo4j graph database and constructed using multi-source heterogeneous data in the field of artificial joints and BFO ontology reference documents, covers the artificial joint structure-materials domain, including: Based on the pre-defined Neo4j graph database and the BFO ontology reference document, the top-level category is defined using Neo4j node tags; Based on a pre-defined expert knowledge base for artificial joints and a pre-defined Neo4j graph database, modules are constructed according to top-level categories to obtain an artificial joint domain module ontology; the artificial joint domain module ontology includes an anatomy module, a material module, an implant structure module, a process module, and a data traceability module; Based on the pre-defined Neo4j graph database, a four-layer relational architecture is configured according to the top-level categories and BFO ontology reference documents to obtain domain relation types; the domain relation types include the first layer structural skeleton relation, the second layer process-evidence relation, the third layer clinical semantic relation, and the fourth layer UMLS extended relation; Based on the pre-set Neo4j graph database, and according to the multi-source heterogeneous data in the artificial joint field, entity data instantiation is performed on the module ontology of the artificial joint field to obtain knowledge graph entity instances. Based on the domain relationship type, and according to the multi-source heterogeneous data in the artificial joint domain, the entity instances of the knowledge graph are instantiated with inter-entity association relationships to obtain knowledge graph relationship instances; Construct a knowledge graph for the artificial joint structure-materials domain based on entity instances and relationship instances of the knowledge graph.

3. The personalized joint reverse optimization method based on multi-agent cooperative design according to claim 1, characterized in that, The clinical reasoning agent, the material design agent, the verification agent, and the optimization agent are constructed based on the Tongyi Qianwen large model; the structural design agent is constructed based on the differential evolution algorithm; and the simulation agent is constructed based on the extreme random tree regression model.

4. The personalized joint reverse optimization method based on multi-agent cooperative design according to claim 1, characterized in that, The clinical constraints include the patient's activity risk level, the lifespan of the artificial joint, and the maximum permissible wear volume of the artificial joint.

5. The personalized joint reverse optimization method based on multi-agent cooperative design according to claim 1, characterized in that, The aforementioned knowledge graph based on the artificial joint structure-materials domain, according to clinical constraints, utilizes a materials design agent to perform materials retrieval, obtaining a candidate list of materials, material properties, and material retrieval source identification information, including: Determine the material screening constraints based on clinical requirements; Based on the knowledge graph of artificial joint structure-materials, according to the material selection constraint boundary, the Cypher query statement is used to retrieve material entities that meet the constraints and obtain a list of retrieved materials. Based on preset weights and the list of retrieved materials, a weighted scoring algorithm is used to calculate the comprehensive score of the candidate materials. Based on the preset material screening prompt template, and according to the search material list and the comprehensive score of candidate materials, the large language model is used to enhance the search and generate the material candidate list and the material search source identification information. Based on the knowledge graph of artificial joint structure-materials, entity attribute queries are performed according to the material candidate list to obtain material attributes.

6. The personalized joint reverse optimization method based on multi-agent cooperative design according to claim 1, characterized in that, Based on clinical constraints, and using a candidate material list and historical datasets, a structural design agent performs reverse optimization to obtain optimized structural parameters and iterative optimization information, including: Based on clinical constraints, the parameter search boundary is determined according to the material candidate list, and a target optimization problem is established; Based on historical simulation datasets, a weighted scoring method was used to select samples and obtain the initial population. Based on the objective optimization problem, and according to the initial population and parameter search boundary, the differential evolution algorithm is used to perform iterative optimization search to obtain the optimized structure parameters and structure optimization iteration information.

7. The personalized joint reverse optimization method based on multi-agent cooperative design according to claim 1, characterized in that, The step of using a simulation agent to make positive predictions based on material properties and verified structural parameters to obtain simulation indicators and simulation prediction confidence includes: The material physical property vector and the verification structure parameter vector are obtained by vectorizing the material properties and the verification structure parameter vector. Based on the material physical property vector and the verification structure parameter vector, batch inference is performed using an extreme random tree regression model to obtain decision tree prediction results. The prediction results of the decision tree are averaged to obtain the simulation index; The variance is calculated based on the prediction results of the decision tree to obtain the confidence level of the simulation prediction.

8. A personalized joint reverse optimization device based on multi-agent cooperative design, wherein the personalized joint reverse optimization based on multi-agent cooperative design... The apparatus is used to implement the personalized joint reverse optimization method based on multi-agent cooperative design as described in any one of claims 1-7, characterized in that the apparatus comprises: The knowledge graph construction module is used to acquire multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints; based on the preset Neo4j graph database, it constructs a knowledge graph of artificial joint structure-materials based on multi-source heterogeneous data and BFO ontology reference documents in the field of artificial joints. The clinical constraint conversion module is used to obtain individual patient characteristic information; based on the individual patient characteristic information, a clinical reasoning agent is used to perform reasoning mapping to obtain clinical constraint conditions and clinical reasoning source identification information; The material candidate retrieval module is used to perform material retrieval based on the knowledge graph of artificial joint structure-materials and according to clinical constraints, using a material design agent to obtain a list of material candidates, material properties, and material retrieval source identification information; The structural parameter optimization module is used to acquire historical simulation datasets; based on clinical constraints, and according to the material candidate list and historical datasets, a structural design agent is used to perform reverse optimization to obtain optimized structural parameters and structural optimization iteration information. The parameter compliance verification module is used to perform compliance verification on optimized structural parameters based on the material candidate list using a verification agent, and obtain the verified structural parameters and verification result identification information. The simulation index prediction module is used to make positive predictions using a simulation agent based on material properties and verification structure parameters, and to obtain simulation indexes and simulation prediction confidence. The optimal solution decision module is used to perform comprehensive material ranking based on the material candidate list and simulation indicators using an optimization agent to obtain the optimal design scheme and the ranking basis identification information. The evidence chain tracing module is used to construct a traceable evidence chain based on clinical reasoning source identification information, material retrieval source identification information, structural optimization iteration information, verification result identification information, simulation prediction confidence level, and ranking basis identification information to obtain joint design decision evidence information.

9. A personalized joint reverse optimization device, characterized in that, The personalized joint reverse optimization device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images