Bone soft tissue tumor repair prosthesis printing method based on multi-agent decision-making system

Abstract

The embodiment of the invention discloses a bone soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system. The method comprises the following steps: generating tumor knowledge map information according to preset tumor grade screening information, preset tumor grading standard information, acquired tumor-related material information and tumor-related material information; performing change processing on the initialized preset tumor decision intelligent agents to obtain a tumor decision intelligent agent group; generating task processing distribution information according to preset tumor record information and the tumor decision agent group; according to the tumor decision intelligent agent group, the task processing distribution information and the tumor knowledge graph information, prosthesis printing reference information is generated; according to the prosthesis printing reference information, generating a tumor material reconstruction image; and according to the tumor material reconstruction image, controlling a printing device to print a tumor repair prosthesis. According to the implementation mode, the accuracy of the generated prosthesis printing reference information can be improved, and waste of biocompatible materials for 3D printing is reduced.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and specifically to a method for printing prostheses for repairing bone and soft tissue tumors based on a multi-agent decision-making system. Background Technology

[0002] Printed tumor repair prostheses can be customized implantable medical devices for tumor patients with tissue defects after tumor resection (e.g., patients with bone or soft tissue tumors). They can replace diseased tissues (such as bones, joints, and soft tissues) destroyed by the tumor or surgically removed, thus maintaining the mechanical stability of the body structure. The typical method for printing tumor repair prostheses is as follows: using an AI-assisted printing system, based on the accuracy requirements at the data level, directly analyzes the case to generate prosthesis printing reference information, and then generates images based on the prosthesis printing reference information for printing.

[0003] However, in practice, it has been found that when using the aforementioned AI-assisted printing system for printing tumor repair prostheses, the following technical problems often arise: AI-assisted printing systems fail to reference literature with high levels of evidence, integrate key molecular pathological features of tumors, and generate prosthesis printing reference information based solely on the accuracy of data-level output results. This results in low accuracy of the generated prosthesis printing reference information. Printing prostheses based on this low-accuracy reference information leads to dimensional deviations, missing or redundant key structures, rendering the printed prostheses unusable and wasting biocompatible materials used in 3D printing.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the present disclosure concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a method, apparatus, and computer-readable medium for printing prostheses for repairing bone and soft tissue tumors based on a multi-agent decision-making system to address one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a method for printing a prosthesis for bone and soft tissue tumor repair based on a multi-agent decision-making system. The method includes: responding to the detection of a printing instruction for a printing device; generating tumor grade score information based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information; generating tumor knowledge graph information based on the tumor grade score information and the tumor-related material information; modifying an initialized preset tumor decision-making agent to obtain a tumor decision-making agent group; generating task processing allocation information based on preset tumor record information and the tumor decision-making agent group; generating prosthesis printing reference information based on the tumor decision-making agent group, the task processing allocation information, and the tumor knowledge graph information; generating a tumor material reconstruction image based on the prosthesis printing reference information; and controlling the printing device to print a tumor repair prosthesis corresponding to the tumor material reconstruction image based on the tumor material reconstruction image.

[0008] Secondly, some embodiments of this disclosure provide a bone and soft tissue tumor repair prosthesis printing device based on a multi-agent decision-making system, comprising: a first generation unit configured to generate tumor grade score information based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information in response to detecting printing instruction information for a printing device; a second generation unit configured to generate tumor knowledge graph information based on the tumor grade score information and the tumor-related material information; a modification unit configured to modify the initialized preset tumor decision agents to obtain a tumor decision agent group; a third generation unit configured to generate task processing allocation information based on preset tumor record information and the tumor decision agent group; a fourth generation unit configured to generate prosthesis printing reference information based on the tumor decision agent group, the task processing allocation information, and the tumor knowledge graph information; a fifth generation unit configured to generate tumor material reconstruction images based on the prosthesis printing reference information; and a control unit configured to control the printing device to print tumor repair prostheses corresponding to the material reconstruction images based on the tumor material reconstruction images.

[0009] Thirdly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

[0010] The above embodiments of this disclosure have the following beneficial effects: the bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system of some embodiments of this disclosure can improve the accuracy of the generated prosthesis printing reference information and reduce the waste of 3D printing biocompatible materials. Specifically, the reason for the low accuracy of the generated prosthesis printing reference information and the waste of 3D printing biocompatible materials is that the AI-assisted printing system does not refer to literature with a high level of evidence, does not integrate key molecular pathological features of the tumor, and the logic for generating the prosthesis printing reference information only pursues the accuracy of the data-level output results, resulting in low accuracy of the generated prosthesis printing reference information. Printing a prosthesis based on the low accuracy of the prosthesis printing reference information results in deviations in the size of the printed prosthesis, missing or redundant key structures, making the printed prosthesis unusable and wasting 3D printing biocompatible materials. Based on this, the bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system of some embodiments of this disclosure can obtain the tumor grade score corresponding to the tumor-related material information obtained above through the above-mentioned preset tumor grade screening information. Then, a tumor knowledge graph is constructed based on the tumor grade scores of the aforementioned tumor-related materials. This allows for the generation of prosthesis printing reference information based on materials with higher tumor grade scores within the tumor-related materials information, thereby improving the accuracy of the generated prosthesis printing reference information. Prosthesis printing can also be performed based on the prosthesis printing reference information generated by the aforementioned tumor decision-making intelligence group. The various tumor decision-making intelligence agents within this group have different functions and can process the pre-set tumor record information from a multidisciplinary perspective, thus meeting practical needs rather than merely pursuing the accuracy of data-level output results. This improves the accuracy of the generated prosthesis printing reference information and allows for prosthesis printing based on highly accurate reference information, reducing the waste of biocompatible materials in 3D printing. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0012] Figure 1 This is a flowchart of some embodiments of the method for printing prostheses for repairing bone and soft tissue tumors based on a multi-agent decision-making system according to the present disclosure; Figure 2 This is a schematic diagram of some embodiments of the bone and soft tissue tumor repair prosthesis printing device based on a multi-agent decision-making system according to the present disclosure; Figure 3This is a schematic diagram of an application scenario suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0013] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0014] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0015] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0016] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0017] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0018] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Figure 1 A flowchart 100 illustrating some embodiments of the bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system according to this disclosure is shown. This bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system includes the following steps: Step 101: In response to the detection of printing instruction information for the printing device, generate tumor grade score information based on preset tumor grade screening information, preset tumor grading standard information and acquired tumor-related material information.

[0020] In some embodiments, in response to detecting printing instruction information for a printing device, the execution entity (e.g., a computing device) of the bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision system can generate tumor grade score information based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information. The printing instruction information can represent a request to initiate the printing device to print a prosthesis corresponding to preset tumor record information. Here, the specific type of printing device is not limited and can be adjusted according to actual needs. For example, it can be an industrial-grade metal 3D printer. The preset tumor record information can represent a structured JSON-formatted medical record uploaded by a target user. Here, the target user is not specifically limited; for example, the target user can be a patient with bone and soft tissue tumors. The acquired tumor-related material information can represent various text data related to the tumor. The tumor can represent bone and soft tissue tumors. Here, the specific type of text data in the various text data is not limited. For example, text data can represent tumor-related literature, gene mutation data, gene expression profiles, and clinical guidelines. The preset tumor grade screening information can include various material score information. The material score information in the above-mentioned material score information can represent the correspondence between the level and the score of the text data included in the acquired tumor-related material information. The level corresponding to the text data can be material score information and material mapping score information. The score can represent the score of the corresponding text data. For example, the material score information can be: "The level corresponding to the text data: Material score information is Level 1 (highest level) and material mapping score information is empty, score: 0.6". The above-mentioned preset tumor grading standard information can include first evidence-based grading standard information and second evidence-based grading standard information. The first evidence-based grading standard information can be the GRADE system. The second evidence-based grading standard information can be the Oxford Centre for Evidence Grading (OCEBM). The first evidence-based grading standard information can represent the correspondence between preset material type information and material score information. For example, the preset material type information can be: the type of treatment intervention effect evaluation, then the material score information is Level 2 (medium level). The second evidence-based grading standard information can represent the correspondence between preset material type information and material mapping score information. For example, if the preset material type information is: the type of treatment intervention effect evaluation, then the material mapping score information is Level 1 (highest reliability) material. The aforementioned tumor grade score information can characterize the score of each text data in the aforementioned tumor-related material information. For example, the score can be 0.6.

[0021] Optionally, prior to step 101, the aforementioned executing entity may employ a Python-based distributed web crawler framework and distributed task queue system to crawl the latest tumor-related literature, gene mutation data, gene expression profiles, and clinical guidelines from preset databases and websites as the acquired tumor-related material information. Here, the specific types of the aforementioned preset databases and websites are not limited. For example, preset databases may include PubMed, the Cochrane Library, The Cancer Genome Atlas (TCGA), and the COSMIC public genomics database. For example, preset websites may include NCCN and the ESMO clinical guidelines website.

[0022] In some optional implementations of certain embodiments, in response to detecting printing instruction information for a printing device, the aforementioned execution entity can generate tumor grade score information based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information through the following steps: The first step involves generating material grade scores based on the first evidence-based grading standard information, in response to the determination that the preset material type information of the aforementioned tumor-related material information meets the preset type screening criteria. The preset material type information can characterize the research type corresponding to the aforementioned tumor-related material information. The preset type screening criteria can be that the preset material type information is used for evaluating the effectiveness of treatment interventions, for developing clinical guidelines, or for systematic reviews / meta-analyses. The material grade scores can be first-level, second-level, third-level, or fourth-level scores. The first-level score information can characterize the tumor-related material information as grade one (highest grade). The second-level score information can characterize the tumor-related material information as grade two (intermediate grade). The third-level score information can characterize the tumor-related material information as grade three (lower grade). The fourth-level score information can characterize the tumor-related material information as grade four (lowest grade). In practice, the implementing entity can determine the material grade scores corresponding to the preset material type information from the first evidence-based grading standard information.

[0023] The second step involves generating material mapping scores based on the second evidence-based grading criteria, in response to the determination that the preset material type information for the aforementioned tumor-related materials does not meet the preset type screening conditions. These material mapping scores can be a first, second, third, fourth, or fifth mapping score. The first mapping score indicates that the tumor-related material information is a Level 1 (highest reliability) material. The second mapping score indicates that the tumor-related material information is a Level 2 (medium reliability) material. The third mapping score indicates that the tumor-related material information is a Level 3 (lower reliability) material. The fourth mapping score indicates that the tumor-related material information is a Level 4 (very low reliability) material. The fifth mapping score indicates that the tumor-related material information is a text of expert opinion. In practice, the implementing entity can determine the material mapping score information corresponding to the preset material type information from the second evidence-based grading criteria.

[0024] The third step involves generating tumor grade score information corresponding to the aforementioned tumor-related material information based on the material grade score information and the aforementioned material mapping score information. In practice, the executing entity can determine the grade scores corresponding to the aforementioned material grade score information and the aforementioned material mapping score information from the aforementioned preset tumor grade screening information as the tumor grade score information corresponding to the aforementioned tumor-related material information. It should be noted that both the aforementioned material grade score information and the aforementioned material mapping score information can be empty.

[0025] Optionally, in response to determining that the aforementioned preset material type information meets the preset type screening conditions, the aforementioned executing entity may also generate material grade score information based on the aforementioned first evidence-based grading standard information through the following steps: The first step involves generating initial material grade score information based on the first evidence-based grading standard information, in response to the determination that the aforementioned preset material type information meets the aforementioned preset type screening conditions. This initial material grade score information can be either a first-class material grade score information or a second-class material grade score information. The first-class material grade score information can characterize the aforementioned tumor-related material information as grade one (the highest grade). The second-class material grade score information can characterize the aforementioned tumor-related material information as grade four (the lowest grade). In practice, the implementing entity can determine the material grade score information corresponding to the aforementioned preset material type information from the aforementioned first evidence-based grading standard information as the initial material grade score information.

[0026] The second step involves modifying the aforementioned first-category material grade score information to obtain the modified first-category material grade score information, which is then used as the first-category target material grade score information. The preset downgrade constraint can be that the aforementioned first-category material grade score information corresponds to preset downgrade factor information. This preset downgrade factor information may include, but is not limited to, at least one of the following: risk of bias, inconsistency, indirectness, imprecision, and publication bias. In practice, upon detecting that the aforementioned first-category material grade score information corresponds to preset downgrade factor information, the executing entity can downgrade the aforementioned first-category material grade score information to obtain the downgraded first-category material grade score information, which is then used as the first-category target material grade score information. For example, downgrade processing can characterize the grade of the aforementioned tumor-related material information from level one (highest grade) to level two (medium grade). It should be noted that each downgrade processing is performed at a lower level.

[0027] The third step involves modifying the second-class material grade score information to obtain the modified second-class material grade score information, which is then used as the second-class target material grade score information, in response to the determination that the second-class material grade score information meets the preset upgrade constraints. The preset upgrade constraints may include preset upgrade factor information corresponding to the second-class material grade score information. These preset upgrade factors may include, but are not limited to, at least one of the following: a large magnitude of effect, a dose-response relationship, or all plausible confounding factors accounted for. In practice, in response to the detection that the second-class material grade score information corresponds to preset upgrade factor information, the executing entity can upgrade the second-class material grade score information to obtain the upgraded second-class material grade score information, which is then used as the second-class target material grade score information. For example, the upgrade process can represent an upgrade of the tumor-related material information grade from level two (intermediate level) to level one (highest level). It should be noted that each upgrade process is performed at a higher level.

[0028] Fourth, based on the initial material grade score information, the first type of target material grade score information, and the second type of target material grade score information, determine the material grade score information. In practice, in response to the determination that the second type of material grade score information meets the preset upgrade constraint conditions, the second type of target material grade score information is determined as the material grade score information. In response to the determination that the first type of material grade score information meets the preset downgrade constraint conditions, the first type of target material grade score information is determined as the material grade score information.

[0029] Step 102: Generate tumor knowledge graph information based on tumor grade score information and tumor-related material information.

[0030] In some embodiments, the executing entity can generate tumor knowledge graph information based on the tumor grade score information and the tumor-related material information. The tumor knowledge graph information can represent a tumor-related knowledge graph that includes the tumor grade score information and molecular marker attributes. The molecular marker attributes can represent the properties of the molecular markers. For example, molecular marker attributes can include genetic attributes.

[0031] In some optional implementations of certain embodiments, the aforementioned executing entity may generate tumor knowledge graph information based on the aforementioned tumor grade score information and the aforementioned tumor-related material information through the following steps: The first step is to update the aforementioned tumor-related material information to obtain target tumor-related material information. This target tumor-related material information can be characterized by text data whose scores are greater than a preset grade score. The specific value of the preset grade score is not limited and can be adjusted according to actual needs. For example, the preset grade score can be 0.5. In practice, the executing entity can identify text data with scores greater than the preset grade score as target tumor-related material information.

[0032] The second step involves generating tumor material association information corresponding to the target tumor-related material information, based on a pre-trained tumor-related entity extraction model and the aforementioned target tumor-related material information. This tumor material association information can represent triples corresponding to the target tumor-related material information. The tumor material association information can include individual tumor-related entities, information about relationships between these entities, and causal information between them. The tumor-related entities can represent tumor type (e.g., osteosarcoma), anatomical location (e.g., distal femur), molecular markers (e.g., BRAF V600E mutation), resection method (e.g., wide resection), or prognostic indicators (e.g., 5-year survival rate). The information about relationships between these entities can represent non-causal associations based on medical consensus. These non-causal associations based on medical consensus can represent statistically significant associations between two or more medical phenomena, but according to current medical consensus, a direct causal relationship cannot be confirmed. The relationships between the aforementioned tumor-related entities may include, but are not limited to, at least one of the following: classification relationships (e.g., "osteosarcoma IS_A malignant tumor"), location relationships (e.g., "tumor LOCATED_IN femur"), and causal relationships (e.g., "gene mutation IS_DIAGNOSTIC_MARKER_OF tumor"). The causal information between the aforementioned tumor-related entities may be causal effects confirmed by clinical studies. The causal information between the aforementioned tumor-related entities may include, but is not limited to, at least one of the following: the relationship between treatment method and treatment outcome (e.g., "chemotherapy REDUCES tumor volume"), and the relationship between treatment method and its drawbacks (e.g., "radiotherapy CAUSES skin damage"). The aforementioned pre-trained tumor-related entity extraction model may be a large-scale language model jointly fine-tuned based on biomedical and cancer genomics principles. The training method for the aforementioned pre-trained tumor-related entity extraction model may be batch training. The aforementioned pre-trained tumor-related entity extraction model may be a large-scale language model that takes the aforementioned target tumor-related material information as input and outputs tumor material association information. Here, the specific type of the large-scale language model mentioned above is not limited. For example, a large-scale language model could be Llama-2-70B. Llama-2-70B could include one input embedding layer, 80 Transformer layers, and one output layer (LM Head). Each of the 80 Transformer layers could include one self-attention layer, one feedforward neural network (FFN), two layer normalization layers (LayerNorm), and two residual connections.In practice, the aforementioned implementing entity can input the target tumor-related material information into the pre-trained tumor-related entity extraction model to obtain tumor material association information.

[0033] The third step involves identifying the aforementioned tumor grade score information and source attribute information as metadata. The source attribute information can be a unique identifier representing the source of each text data in the target tumor-related material information. This unique identifier may include, but is not limited to, at least one of the following: Digital Object Identifier (DOI), PubMed Unique Identifier (PMID), or Clinical Guideline Number. This metadata information can be used to describe the target tumor-related material information.

[0034] The fourth step involves generating a tumor knowledge graph based on the aforementioned tumor material association information and metadata. In practice, the executing entity first binds the metadata and tumor material association information using a named graph. Then, the bound data is stored in a graph database to obtain the constructed knowledge graph as the tumor knowledge graph information. The specific type of graph database is not limited here; for example, it could be Neo4j.

[0035] In addressing the technical problems mentioned above, when using technical solutions to solve the application scenario of printing repair materials for advanced rare bone and soft tissue malignancies, the following technical problem often arises: Treatment strategies and targeted drugs for rare bone and soft tissue tumors are updated rapidly, resulting in a large and complex body of knowledge. Relying solely on a static knowledge base to build a knowledge graph to generate low-accuracy prosthesis printing reference information leads to the generation of tumor material reconstruction images based on this low-accuracy reference information. This results in dimensional deviations, missing or redundant key structures in the printed prosthesis, rendering it unusable and wasting biocompatible 3D printing materials. Considering the following requirements for this application scenario—adapting to rare and complex printing needs—we have decided to adopt the following solution: Optionally, after step 102 above, the executing entity may update the tumor knowledge graph information through the following steps: The first step involves determining an incremental update data package in response to the detection that the aforementioned tumor knowledge graph information meets the periodic triggering conditions, or the detection of event triggering information corresponding to the aforementioned tumor knowledge graph information. The event triggering information can be a change in the preset authoritative guideline version (e.g., NCCN / ESMO guidelines) corresponding to the aforementioned tumor knowledge graph information, the release of a new drug by a regulatory agency, or the publication of a paper with an impact factor higher than 20 by a medical journal. The periodic triggering conditions can be that the time interval between the last update of the tumor knowledge graph information and the current time is greater than a preset time interval. The preset time interval can represent one week. The incremental update data package can represent various tumor-related text data (e.g., full text of documents, guideline PDFs, gene mutation records) and their metadata (source URL, publication time, document type DOI, etc.) that have changed within the preset time interval. In practice, the executing entity can crawl data from the preset database and the preset website to obtain the various tumor-related text data that have changed within the preset time interval. Then, the various tumor-related text data that have changed within the preset time interval and their metadata are determined as the incremental update data package. It should be noted that the method for crawling the above incremental update data packets is the same as the method for crawling the above tumor-related material information, and will not be repeated here.

[0036] The second step involves generating a temporary tumor knowledge graph based on the incremental update data package. This temporary tumor knowledge graph information represents the updated tumor knowledge graph information. In practice, the executing entity can perform graph construction processing on the incremental update data package to obtain the temporary tumor knowledge graph information. It should be noted that the graph construction process is the same as steps one through four in step 102 above, and will not be repeated here.

[0037] The third step involves aligning and comparing the aforementioned temporary tumor knowledge graph information and the aforementioned tumor knowledge graph information to obtain contradictory statements. These contradictory statements characterize the points of contradiction in the definition of the same entity within the temporary and existing tumor knowledge graph information. The specific content of these contradictory statements is not limited; for example, they could be: "Tumor knowledge graph information: 'Pazopanib', 'TREATS', 'Advanced chordoma'; Temporary tumor knowledge graph information: 'Pazopanib', 'NO_EFFICACY_FOR', 'Advanced chordoma'". In practice, the executing entity can use Cypher subgraph matching queries to perform entity alignment and relation comparison processing on the temporary and existing tumor knowledge graph information to obtain the contradictory statements.

[0038] The fourth step involves generating a conflict resolution list based on the aforementioned contradictory statements and the conflict resolution rule base. The conflict resolution rule base represents a pre-defined set of rules, including triggering conditions and execution actions. The specific content of the triggering conditions and execution actions is not limited here; for example, the triggering condition could be a tumor grade score information difference Δ ≥ 0.3, and the execution action could be the text data corresponding to the highest tumor grade score. The conflict resolution list can include various conflict resolution information. Each conflict resolution information represents the correspondence between the contradictory statements and the execution actions. In practice, the executing entity can input the contradictory statements into the Drools rule engine based on the conflict resolution rule base to obtain the conflict resolution list.

[0039] Fifth, based on the conflict resolution list information, the tumor knowledge graph information is updated to obtain the updated tumor knowledge graph information. In practice, the executing entity can update the tumor knowledge graph information based on the conflict resolution list information using database transaction management and automated operation and maintenance scripts to obtain the updated tumor knowledge graph information.

[0040] The above-described technical solution, as an inventive point of this disclosure, addresses technical problem two: "the printed prosthesis exhibits dimensional deviations, missing or redundant key structures, and significant waste of biocompatible 3D printing materials." The reasons for these dimensional deviations, missing or redundant key structures, and significant waste of biocompatible 3D printing materials are as follows: Treatment strategies and targeted drugs for rare tumors are rapidly evolving, resulting in a large and complex body of knowledge. Relying solely on a static knowledge base to construct a knowledge graph generates low-accuracy reference information for prosthesis printing. This leads to the generation of tumor material reconstruction images based on this low-accuracy reference information, resulting in dimensional deviations, missing or redundant key structures, rendering the printed prosthesis unusable and wasting biocompatible 3D printing materials. Solving these factors can reduce dimensional deviations, missing or redundant key structures, and minimize the waste of biocompatible 3D printing materials. To achieve this effect, the disclosed method for printing prostheses for bone and soft tissue tumor repair based on a multi-agent decision-making system triggers the acquisition of newly added or changed tumor-related data in the medical field through periodic and event-triggered conditions, determining incremental update data packages. By comparing the updated tumor knowledge graph information with the unupdated tumor knowledge graph information, contradictory statements are obtained. The contradictory points are corrected according to the conflict resolution rule base and rule engine, thus updating the tumor knowledge graph information. This reduces erroneous data in the tumor knowledge graph information. Prosthesis printing reference information is generated based on the dynamically updated tumor knowledge graph information, improving the accuracy of the generated reference information and consequently improving the accuracy of tumor material reconstruction images generated based on the reference information. Furthermore, this reduces the occurrence of dimensional deviations, missing or redundant key structures in the printed prostheses, and minimizes the waste of biocompatible materials in 3D printing.

[0041] Step 103: Modify the initialized preset tumor decision agent to obtain a tumor decision agent group.

[0042] In some embodiments, the aforementioned executing entity can modify the initialized preset tumor decision-making agent to obtain a tumor decision-making agent group. The tumor decision-making agents in this group can represent large-scale language models possessing basic tumor domain knowledge and dialogue capabilities. The preset tumor decision-making agents can be open-source large-scale language models. The training method for the preset tumor decision-making agents can be batch training. The preset tumor decision-making agents can be large-scale language models that take second-type tumor-related material data information and preset tumor data instruction information as input, and output tumor-related question-and-answer pairs. The model structure of the preset tumor decision-making agents is the same as the structure of the pre-trained tumor-related entity extraction model, and will not be repeated here. The initialized preset tumor decision-making agents can be large-scale language models with basic tumor domain knowledge and dialogue capabilities after supervised fine-tuning. For example, the large-scale language model can be Llama-2-70B. It should be noted that the aforementioned multi-agent decision-making system can be composed of the aforementioned tumor decision-making agent group and the aforementioned tumor knowledge graph information.

[0043] In addressing the aforementioned technical problems in the process of adopting technical solutions, specifically for the application scenario of printing prostheses for re-excision of tumor recurrence / metastasis, the following technical problem often arises: When the system generates reference information for prosthesis printing through an intelligent agent, it directly parses large amounts of structured and unstructured data and constructs a large number of instruction-response pairs for supervised fine-tuning and initialization of the intelligent agent. This results in a long time consumption for generating reference information for prosthesis printing, which in turn prolongs the printing time. Furthermore, tumor recurrence / metastasis leads to a low degree of matching between the printed prosthesis and the patient's real-time anatomical state, resulting in deviations in the size and orientation of the printed prosthesis, rendering it unusable and wasting biocompatible 3D printing materials. Considering the following requirements for this application scenario: adaptability to short printing timeframes and adaptability to easily changeable structures of the prosthesis to be filled, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned execution entity may initialize the aforementioned preset tumor decision-making agent through the following steps: The first step is to identify the text data in the target tumor-related material information that meet the preset selection criteria based on the aforementioned tumor grade score information as a reference material information group. The preset selection criteria can be that the tumor grade score corresponding to the text data in the target tumor-related material information is greater than a preset score threshold. The preset score threshold can be 0.7. In practice, the executing entity can identify the text data in the target tumor-related material information whose corresponding grade score is greater than 0.7 as the reference material information group.

[0044] The second step involves generating first-type and second-type tumor-related material data based on a pre-defined automatic tumor data parsing tool and the aforementioned reference material information group. The pre-defined automatic tumor data parsing tool can be a tool used to parse the reference material information in the aforementioned reference material information group. The type of the pre-defined automatic tumor data parsing tool is not limited and can be adjusted according to actual needs. For example, the pre-defined automatic tumor data parsing tool could be Logstash. The first-type tumor-related material data can represent each reference material information in the aforementioned reference material information group that is structured data. The second-type tumor-related material data can represent each reference material information in the aforementioned reference material information group that is unstructured data. In practice, the executing entity can input the aforementioned reference material information group into the pre-defined automatic tumor data parsing tool to obtain the first-type and second-type tumor-related material data.

[0045] The third step involves generating tumor data association information corresponding to the aforementioned first-category tumor-related material data information, based on a preset tumor data selection method and a pre-trained tumor-related entity recognition model, in response to the determination that the data is not empty. The preset tumor data selection method can be a rule-based relation extraction method. For example, this method could be based on regular expression rules. The tumor data association information can be a subject-relationship-object triple extracted from the aforementioned first-category tumor-related material data information. The specific content of the triple is not limited; for example, it could be "Subject: Total vertebrectomy; Relation: Significantly reduced; Object: Local recurrence rate". The pre-trained tumor-related entity recognition model can be a large-scale language model for recognizing tumor-related entities. The training method for the pre-trained tumor-related entity recognition model can be batch training. The pre-trained tumor-related entity recognition model can be a model that takes the first-category tumor-related material data information as input and outputs medical entities from that data. The structure of the pre-trained tumor-related entity recognition model is the same as that of the pre-trained tumor-related entity extraction model, and will not be repeated here. In practice, firstly, the execution entity can identify medical entities in the first type of tumor-related material data information using the pre-trained tumor-related entity recognition model, and then extract the identified medical entities using a rule-based extraction relationship method to obtain tumor data association information.

[0046] The fourth step involves, in response to the determination that the aforementioned second-category tumor-related material data information is not empty, generating tumor-related question-and-answer pairs corresponding to the aforementioned second-category tumor-related material data information based on preset tumor data instruction information. The preset tumor data instruction information can be instructions to answer according to different preset prompts. These preset prompts can include, but are not limited to, at least one of the following: "generating question-and-answer pairs based on medical literature," "questions containing tumor types and clinical scenarios," and "answers citing original text evidence." The tumor-related question-and-answer pairs can represent each question-and-answer pair generated according to the preset tumor data instruction information corresponding to the aforementioned second-category tumor-related material data information. Each question-and-answer pair can represent a question and its corresponding answer. In practice, the executing entity can input the aforementioned second-category tumor-related material data information and the preset tumor data instruction information into a large-scale language model to obtain the tumor-related question-and-answer pairs corresponding to the aforementioned second-category tumor-related material data information. For example, the large-scale language model can be Llama-2-70B.

[0047] The fifth step involves initializing the initial tumor decision-making agent based on the preset tumor data verification method, the aforementioned tumor data association information, the aforementioned tumor-related question-and-answer pair information, the preset tumor data processing method, and the preset parameter fine-tuning method, thereby obtaining the initialized preset tumor decision-making agent.

[0048] In some optional implementations of certain embodiments, the aforementioned execution entity can initialize the initial preset tumor decision-making agent by following the steps of: a preset tumor data verification method, the aforementioned tumor data association information, the aforementioned tumor-related question-and-answer pair information, a preset tumor data processing method, and a preset parameter fine-tuning method, thereby obtaining the initialized preset tumor decision-making agent: The first step involves generating tumor-related instruction-answer pairs based on a preset tumor data verification method, the aforementioned tumor data association information, and the aforementioned tumor-related question-and-answer pair information. These tumor-related instruction-answer pairs represent various question-and-answer pairs corresponding to the aforementioned target tumor-related material information. The preset tumor data verification method can be a method of verifying the tumor-related instruction-answer pairs according to preset verification content. The preset verification content may include, but is not limited to, any of the following: the character length of the questions in the question-and-answer pair must be greater than 10; the questions in the question-and-answer pair must contain at least one tumor-related entity; the answers in the question-and-answer pair must contain the identifier "[REF]" for associating evidence sources; and the tumor grade score information of the answers in the question-and-answer pair must be greater than 0.7. In practice, firstly, the executing entity can use a distributed processing architecture built with a MapReduce parallel processing algorithm to update the tumor-related question-and-answer pair information based on the aforementioned tumor data association information and the aforementioned preset tumor data verification method, obtaining the updated tumor-related question-and-answer pair information. Then, a simple random sampling algorithm is used to sample the updated tumor-related question-and-answer pair information to obtain the extracted question-and-answer pairs. Then, the review results for each extracted question-and-answer pair are obtained. These results indicate whether the question-and-answer pair passed or failed the review. Finally, the question-and-answer pairs that passed the review are identified as tumor-related instruction response pairs. It should be noted that the obtained review results can be those received from human input. The specific method of obtaining these results is not limited. For example, the obtained review results could be those input by a tumor expert.

[0049] The second step involves initializing the initial preset tumor decision-making agent based on the tumor-related instruction response information, the preset tumor data processing method, and the preset parameter fine-tuning method, resulting in the initialized preset tumor decision-making agent. The preset tumor data processing method represents the initialization of the initial preset tumor decision-making agent as a combination of data partitioning strategy and acceleration technology. The acceleration technology represents a method that addresses the performance bottleneck of training the initial preset tumor decision-making agent by combining parallel computing, hardware acceleration, and algorithm optimization. For example, the performance bottleneck might be that the training time of the initial preset tumor decision-making agent exceeds the preset training time. The specific value of the preset training time is not limited and can be adjusted according to actual needs. For example, the preset training time could be 8 hours. The preset parameter fine-tuning method can be a parameter-efficient fine-tuning technique. For example, a parameter-efficient fine-tuning technique could be LoRA (Low-Rank Adaptation). In practice, the executing entity can use the data partitioning strategy and acceleration technology to adjust the initial preset tumor decision-making agent based on the tumor-related instruction response information, resulting in an adjusted initial preset tumor decision-making agent. Finally, the LoRA parameter fine-tuning technique is used to fine-tune the parameters of the adjusted initial preset tumor decision-making agent for initialization, resulting in the initialized preset tumor decision-making agent. For example, the adjustment steps for the initial preset tumor decision-making agent can be: "Initial adjustment of the initial preset tumor decision-making agent: Adjust the initial preset tumor decision-making agent based on 1,000 question-answer pairs in the tumor-related instruction response information. Then, optimization of the initial preset tumor decision-making agent: Adjust the initial preset tumor decision-making agent based on 10,000 question-answer pairs in the tumor-related instruction response information." The aforementioned initial preset tumor decision-making agent can represent a large-scale language model with different roles, abilities, and behavioral rules that possess dialogue capabilities and basic knowledge in the tumor domain. For example, the initial preset tumor decision-making agent can be the initial Llama-2-70B. The structure of the aforementioned initial preset tumor decision-making agent can be the same as the structure of the pre-trained tumor-related entity extraction model, and will not be elaborated further here. The training method for the aforementioned initial preset tumor decision-making agent can be batch training.

[0050] The first to fifth steps of the above technical solution serve as an inventive point of this disclosure, solving the aforementioned technical problem three: "The generation of prosthesis printing reference information is time-consuming, tumor recurrence / metastasis results in a low degree of matching between the printed prosthesis and the patient's real-time anatomical state, and a large amount of 3D printing biocompatible material is wasted." The factors that cause the generation of prosthesis printing reference information to be time-consuming, the low degree of matching between the printed prosthesis and the patient's real-time anatomical state due to tumor recurrence / metastasis, and the large amount of 3D printing biocompatible material wasted are often as follows: When the system generates prosthesis printing reference information through the intelligent agent, it directly parses a large amount of structured and unstructured data and constructs a large number of instruction-response pairs for supervised fine-tuning and initialization of the intelligent agent. This results in a long generation time for the prosthesis printing reference information, which in turn leads to a long printing time. Furthermore, tumor recurrence / metastasis results in a low degree of matching between the printed prosthesis and the patient's real-time anatomical state, leading to deviations in the size and posture of the printed prosthesis, rendering it unusable, and wasting 3D printing biocompatible material. If the above factors are addressed, it is possible to shorten the time required to generate reference information for prosthesis printing, improve the matching degree between the printed prosthesis and the patient's real-time anatomical state caused by tumor recurrence / metastasis, and reduce the waste of biocompatible materials in 3D printing. To achieve this, this disclosure, based on pre-constructed tumor-related instruction response information, adjusts the initial preset tumor decision-making agent by introducing the aforementioned preset tumor data processing method, resulting in an adjusted initial preset tumor decision-making agent. Then, the aforementioned preset parameter fine-tuning method is used to fine-tune the parameters of the adjusted initial preset tumor decision-making agent for initialization, resulting in an initialized preset tumor decision-making agent. The aforementioned preset tumor data processing method can be characterized as a combination of data partitioning strategy and acceleration technology for initializing the initial preset tumor decision-making agent. The aforementioned preset parameter fine-tuning method can be a parameter efficient fine-tuning technique. This shortens the time required to generate reference information for prosthesis printing, thereby reducing the time spent printing the prosthesis, reducing the low matching degree between the printed prosthesis and the patient's real-time anatomical state caused by tumor recurrence / metastasis, and further reducing the waste of biocompatible materials in 3D printing.

[0051] In some optional implementations of certain embodiments, the aforementioned execution entity can generate a tumor decision-making agent group based on the initialized preset tumor decision-making agent through the following steps: The first step involves generating preference decision result ranking information based on a preset set of tumor problem information and the initialized preset tumor decision-making agent. The preset tumor problem information in the preset set represents pre-acquired tumor-related clinical questions. These clinical questions can represent statements asking how to treat tumors. The preference decision result ranking information represents the ranked question-and-answer pairs corresponding to the preset set of tumor problem information. Each question-and-answer pair can have a one-to-one correspondence with the preset tumor problem information in the preset set; the question in the question-and-answer pair can be the corresponding preset tumor problem information, and the answer can represent the answer to the corresponding preset tumor problem information. In practice, firstly, the executing agent can input the preset set of tumor problem information into the initialized preset tumor decision-making agent to obtain the corresponding question-and-answer pairs. Then, the target ranking information corresponding to each question-and-answer pair is obtained as the preference decision result ranking information. This target ranking information represents the ranked question-and-answer pairs obtained by the target object. The target object can be an expert in the field of oncology. For example, an expert could be a spinal tumor surgeon.

[0052] The second step involves generating a preference decision signal generation model based on the preference decision result ranking information and ranking loss function information. The ranking loss function can be a function that allows the preference decision signal generation model to learn the relative distances between different samples. For example, the ranking loss function can be a list-based ranking loss function. In practice, the executing entity can use the preference decision result ranking information and ranking loss function information to perform batch training on the initial preference decision signal generation model, obtaining the batch-trained initial preference decision signal generation model as the preference decision signal generation model. This preference decision signal generation model can be a natural language understanding model that takes question-answer pairs as input and preference decision reward signal information as output. For example, the natural language understanding model can be DeBERTa-v3-large. DeBERTa-v3-large can include one input embedding layer (DebertaV2Embeddings), 24 Transformer encoding layers (Encoder), and one output layer (Classifier). Each of the 24 Transformer coding layers (i.e., DebertaV2 Layer) mentioned above can include a disentangled self-attention layer, a feedforward neural network (FFN) layer, three layer normalization layers (LayerNorm), two residual connection layers, and a related linear transformation layer. The initial preference decision signal generation model has the same structure as the preference decision signal generation model mentioned above, and will not be described again here.

[0053] The third step is to determine a decision reference model based on the initialized preset tumor decision agent. This decision reference model can be the same as the initialized preset tumor decision agent. In practice, the executing entity can use a copy of the initialized preset tumor decision agent as the decision reference model. The copy of the initialized preset tumor decision agent can represent a model instance generated based on the initialized preset tumor decision agent, with parameters identical to those of the initialized preset tumor decision agent.

[0054] The fourth step is to generate a tumor decision-making agent group based on the above decision reference model and the preset tumor decision-making agent after initialization.

[0055] Optionally, the aforementioned execution entity can be further configured to generate a tumor decision-making agent group based on the aforementioned decision reference model and the aforementioned initialized preset tumor decision-making agent through the following steps: The first step, based on the target preset tumor problem information set and the preset tumor decision-making agent initialized above, is to perform the following training steps: The first sub-step involves generating target tumor question-answer pairs based on the initialized preset tumor decision-making agent and the target preset tumor question information set that meets the preset question selection criteria. The preset question selection criteria can be target preset tumor question information in the target preset tumor question information set that has not been used in the training steps. The target preset tumor question information in the target preset tumor question information set can represent tumor-related questions. The target tumor question-answer pairs can represent each question-answer pair corresponding to the target preset tumor question information set. In practice, firstly, the executing agent can randomly sample an unused target preset tumor question from the target preset tumor question information set. Then, the randomly sampled target preset tumor question information is input into the initialized preset tumor decision-making agent to obtain the target tumor question-answer pairs.

[0056] The second sub-step involves generating a preference decision reward signal based on the target tumor question-and-answer pair information and the preference decision signal generation model. This preference decision reward signal represents a score for the target tumor question-and-answer pair information. For example, the preference decision reward signal could be 60 points. In practice, the executing entity can input the target tumor question-and-answer pair information into the preference decision signal generation model to obtain the preference decision reward signal. It should be noted that the obtained preference decision reward signal can serve as a reward signal for subsequent decisions, representing the degree of fit between the generated target tumor question-and-answer pair information and the target tumor-related material information. The higher the score represented by the preference decision reward signal, the higher the degree of fit between the answers in the target tumor question-and-answer pair information and the actual medical knowledge and expression quality of the tumor.

[0057] The third sub-step involves generating answer-pair penalty constraint information based on the aforementioned decision reference model and the initialized preset tumor decision agent. This answer-pair penalty constraint information can be the KL divergence between the first logarithmic probability and the second logarithmic probability. The first logarithmic probability represents the logarithmic probability of the decision reference model generating an answer corresponding to the aforementioned target preset tumor question information. The second logarithmic probability represents the logarithmic probability of the initialized preset tumor decision agent generating an answer corresponding to the aforementioned target preset tumor question information. In practice, firstly, the executing agent can determine the first and second logarithmic probabilities by calculating and summing the conditional probabilities token by token. Finally, the KL divergence between the obtained first and second logarithmic probabilities is determined as the answer-pair penalty constraint information.

[0058] The fourth sub-step involves modifying the initialized preset tumor decision-making agent based on the aforementioned preference decision reward signal information and the aforementioned response penalty constraint information, resulting in a modified preset tumor decision-making agent. This modified preset tumor decision-making agent can be a large-scale language model possessing basic knowledge of the tumor domain and conversational capabilities, capable of balancing efficacy, risk, and expert values. In practice, the executing agent can use a reinforcement learning from human feedback (RLHF) algorithm to update the parameters of the initialized preset tumor decision-making agent according to preset optimization objective conditions. The specific content of the parameters of the initialized preset tumor decision-making agent is not limited here; for example, the parameters can be those of a LoRA adapter. The preset optimization objective conditions can be that the aforementioned preference decision reward signal information is greater than a preset fit, and the aforementioned response penalty constraint information is less than a preset KL divergence value. The specific value of the preset KL divergence value is not specifically limited; for example, the preset KL divergence value can be 0.5. The specific value of the preset fit is not limited; for example, the preset fit can be 98%. It should be noted that the preset optimization target condition is set to reduce the possibility that the preset tumor decision-making agent after initialization may generate target tumor question answers with low fit with the target tumor-related material information due to pursuing higher preference decision reward signal information.

[0059] Optionally, the above training steps also include: In the first step, in response to the determination that the above-mentioned preference decision reward signal information and the above-mentioned answer penalty constraint information do not meet the preset optimization objective conditions, the modified preset tumor decision agent is used as the initialized preset tumor decision agent, and the above training steps are executed again.

[0060] The second step involves generating a group of tumor decision-making agents based on the modified preset tumor decision-making agents, their framework settings, and configuration information, in response to the determination that the aforementioned preference decision-making reward signal information and the aforementioned response penalty constraint information satisfy the preset optimization objective conditions. Specifically, the preset agent framework settings can represent the AutoGen automation framework. The preset agent configuration information can represent the modified preset tumor decision-making agent's LLM configuration, dedicated system prompts, and registered utility functions. The dedicated system prompts can represent statements defining the roles, capabilities, tools, and behavioral guidelines of the tumor decision-making agents. For example, the dedicated system prompt could be "The role of the tumor decision-making agent is: a world-class spinal tumor surgeon with 20 years of clinical experience." The registered utility functions can be tool classes built using Python code for interacting with the Neo4j graph database. Each tumor decision-making agent can represent an agent built using Python code that possesses tumor-related knowledge and conversational capabilities. The aforementioned tumor decision-making agents may include, but are not limited to, at least one of the following: a bone tumor surgeon agent, a molecular pathologist agent, a radiologist agent, an oncology / radiotherapy physician agent, a coordinator agent, a rehabilitation medicine physician agent, a prognostic prediction agent, and a report generation agent. The input to the bone tumor surgeon agent may be a question about bone tumor surgery, and the output may be a text response to that question. The input to the molecular pathologist agent may be a question about molecular pathology, and the output may be a text response to that question. The input to the radiologist agent may be a question about the extent of tumor invasion, and the output may be a text response to that question. The input to the oncology / radiotherapy physician agent may be a question about tumor management through oncology / radiotherapy, and the output may be a text response to that question. The input to the rehabilitation medicine physician agent may be a question about postoperative rehabilitation, and the output may be a text response to that question. The input to the prognostic prediction agent may be a question about prognosis, and the output may be a text response to that question. The input to the coordinator agent can be preset tumor record information, and the output is the task to be performed by each agent, which is then distributed to the corresponding agents. The task to be performed can represent the question that the agent needs to answer. The model structure of each tumor decision-making agent is the same as the structure of the pre-trained tumor-related entity extraction model, and will not be repeated here. The training method for each of the tumor decision-making agents can be batch training.In practice, the aforementioned execution entities can be based on the Microsoft AutoGen automation framework and Python code. By configuring LLM parameters, defining exclusive system prompts, and registering tool functions for the modified preset tumor decision-making agents, they can construct a group of tumor decision-making agents with different roles, abilities, and behavioral guidelines, each possessing dialogue capabilities and basic knowledge in the field of oncology.

[0061] Step 104: Generate task processing allocation information based on preset tumor record information and tumor decision-making intelligent agent group.

[0062] In some embodiments, the aforementioned executing entity can generate task processing allocation information based on preset tumor record information and the aforementioned tumor decision-making intelligent agent group. The task processing allocation information can characterize the steps of the execution task for each of the aforementioned tumor decision-making intelligent agents. In practice, firstly, the aforementioned executing entity can input the received preset tumor record information into the aforementioned coordinator intelligent agent to obtain various execution tasks. Each execution task can characterize the task that the tumor decision-making intelligent agent needs to perform. For example, the task that the tumor decision-making intelligent agent needs to perform could be responding to questions about bone tumor surgery. Next, the aforementioned execution tasks are transformed using a template-based task generation method. Finally, the transformed execution tasks are determined as the task processing allocation information.

[0063] Step 105: Generate reference information for prosthesis printing based on the tumor decision-making intelligent agent group, task processing allocation information, and tumor knowledge graph information.

[0064] In some embodiments, the executing entity can generate prosthesis printing reference information based on the tumor decision-making intelligence group, the task processing allocation information, and the tumor knowledge graph information. The prosthesis printing reference information can characterize the text data that needs to be referenced when printing a prosthesis corresponding to the preset tumor record information. The prosthesis printing reference information may include the text of the responses output by each tumor decision-making intelligence agent in the tumor decision-making intelligence group, excluding the report-generating intelligence agent. It should be noted that the text of the AI ​​agent's response to the aforementioned reference information for prosthesis printing can support reference data for printing the prosthesis. For example, the input of the radiologist's AI agent could be the question: "Assess the extent of tumor invasion with MRI / CT showing bone destruction in the proximal tibia," and the response text could be: "Determine that the tumor invades the proximal 1 / 3 of the tibia + 2cm of surrounding soft tissue (Source: 10.3899 / jrheum.210345, Evidence Level 0.85)." The input of the bone tumor surgeon's AI agent could be the question: "Assess the surgical resection boundaries and vascular protection plan for the tumor's adjacent neurovascular structures (distal femoral osteosarcoma, surrounding the popliteal artery and vein)," and the response text could be: "Recommend wide resection of the distal femoral tumor + artificial prosthesis replacement (Source: 10.1002 / jor.25567, Evidence Level 0.9)."

[0065] It should be noted that the bone and soft tissue tumor repair prosthesis printing method disclosed herein based on a multi-agent decision-making system is for non-therapeutic purposes. Specifically, it can be used to reduce the dimensional deviation of the printed prosthesis, reduce the situation of missing or redundant key structures, reduce the waste of 3D printing biocompatible materials due to unusable printed prostheses, and the generated prosthesis printing reference information is used for non-therapeutic purposes. Specifically, the generated prosthesis printing reference information can improve the accuracy of prosthesis printing.

[0066] In the process of adopting technical solutions to address the aforementioned technical problems, the following technical problem often arises: When applying technical solutions to solve the technical problems in the background, specifically for the application scenario of printing functional reconstruction prostheses for palliative surgery in advanced tumors, the following technical problem often occurs: When the system directly analyzes task processing and allocation information through various intelligent agents, it generates a large number of redundant analysis results. This necessitates filtering these redundant results, resulting in a long processing time for generating prosthesis printing reference information based on the filtered results. Consequently, printing the prosthesis based on this reference information is time-consuming and lacks timeliness. Considering the following requirements for this application scenario—adapting to short-time printing needs—we have decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned executing entity can generate prosthesis printing reference information based on the aforementioned tumor decision-making intelligent agent group, the aforementioned task processing allocation information, and the aforementioned tumor knowledge graph information through the following steps: The first step involves generating query statements and retrieval request information based on the aforementioned tumor decision-making intelligent agent group and task allocation information. The query statements represent the query statements from the Cypher graph database constructed by the tumor decision-making intelligent agent group, corresponding to the aforementioned tumor knowledge graph information. The retrieval request information represents the request generated by the tumor decision-making intelligent agent group to invoke a standardized tool. This standardized tool can be a database interaction function implemented in Python. In practice, the executing entity can input the task allocation information into the tumor decision-making intelligent agent group to obtain the query statements and retrieval request information.

[0067] The second step involves generating potential tumor decision-making outcome information based on the tumor inference framework settings, the aforementioned tumor decision-making intelligent agent group, and the aforementioned tumor knowledge graph information. This potential outcome information characterizes the result of processing the tumor based on the textual responses output by the tumor decision-making intelligent agent group. The tumor inference framework settings serve as a mind map framework guiding the tumor decision-making intelligent agent group to iteratively query the tumor knowledge graph information and perform counterfactual analysis. In practice, firstly, the executing entity can input the preset tumor record information into the coordinator intelligent agent. Then, using the tumor inference framework settings, it can control the tumor decision-making intelligent agent group to query and process the tumor knowledge graph information, conduct collaborative analysis, and evaluate and determine the potential tumor decision-making outcome information corresponding to the tumor knowledge graph information through counterfactual analysis.

[0068] The third step involves generating initial prosthesis printing reference information based on the aforementioned preset intelligent agent framework settings, the aforementioned tumor decision-making intelligent agent group, the aforementioned tumor knowledge graph information, the aforementioned query retrieval statement information, the aforementioned retrieval call request information, and the aforementioned potential tumor decision-making result information. These initial prosthesis printing reference information represent the text data that the initially generated prosthesis corresponding to the aforementioned preset tumor record information needs to reference. In practice, firstly, the executing entity can use the aforementioned preset intelligent agent framework settings to control the aforementioned tumor decision-making intelligent agent group to query the aforementioned tumor knowledge graph information and obtain query results. These query results represent the text data corresponding to the aforementioned preset tumor record information found in the aforementioned tumor knowledge graph information. Then, the query results are input into the aforementioned tumor decision-making intelligent agents to obtain the initial prosthesis printing reference information for each prosthesis.

[0069] The fourth step involves generating tumor prognostic assessment information based on the preset tumor staging system information and the aforementioned tumor decision-making intelligent agent group. The preset tumor staging system information can be any tumor staging system. This system may include, but is not limited to, at least one of the following: Enneking staging, TNM staging. The tumor prognostic assessment information characterizes the impact of each initial prosthesis printing reference in the aforementioned initial prosthesis printing reference information on the prognosis. In practice, the executing entity can control the aforementioned tumor decision-making intelligent agent group to determine the prognostic impact of each initial prosthesis printing reference information in the aforementioned initial prosthesis printing reference information in conjunction with the preset tumor staging system information. Finally, the obtained prognostic impacts are determined as the tumor prognostic assessment information.

[0070] The fifth step involves generating prosthesis printing reference information corresponding to the preset tumor record information, based on the aforementioned tumor decision-making intelligent agent group, the initial prosthesis printing reference information, the preset prosthesis printing reference information generation instruction, and the tumor prognosis assessment information. The preset prosthesis printing reference information generation instruction can be an instruction that controls the report generation intelligent agent in the tumor decision-making intelligent agent group to summarize the initial prosthesis printing reference information. For example, the summary instruction could represent "Please integrate the initial prosthesis printing reference information and tumor prognosis assessment information into a formal, coherent text data for prosthesis printing reference." In practice, firstly, the coordinator intelligent agent included in the tumor decision-making intelligent agent group can send the collected initial prosthesis printing reference information, the tumor prognosis assessment information, and the preset prosthesis printing reference information generation instruction to the report generation intelligent agent included in the tumor decision-making intelligent agent group. Finally, the report generation intelligent agent included in the tumor decision-making intelligent agent group integrates and formats the initial prosthesis printing reference information and the tumor prognosis assessment information, outputting the final text data for printing the prosthesis corresponding to the preset tumor record information as the prosthesis printing reference information. The input to the aforementioned report-generating agent can be the text output by other tumor decision-making agents in the aforementioned tumor decision-making agent group, and the output can be the text data required for prosthesis printing. The aforementioned other tumor decision-making agents can represent each tumor decision-making agent in the aforementioned tumor decision-making agent group other than the aforementioned report-generating agent.

[0071] The first to fifth steps of the above technical solution serve as an inventive point of this disclosure, solving the fourth technical problem mentioned above: "long printing time and poor timeliness of prostheses." Factors leading to excessive storage resource consumption and consequently poor timeliness of system alerts often include: when the system directly analyzes task processing allocation information through various agents, it generates a large number of redundant analysis results. This necessitates filtering these redundant results, resulting in a long time required to generate prosthesis printing reference information based on the filtered results. Consequently, printing the prosthesis based on this reference information is time-consuming and lacks timeliness. Solving these factors can shorten the printing time of prostheses and improve timeliness. To achieve this, this disclosure first generates query and retrieval statement information and retrieval call request information corresponding to each tumor decision-making agent in the aforementioned tumor decision-making agent group. Then, by introducing the aforementioned tumor reasoning framework, the reasoning process of tumor analysis can be explicitly modeled through a graph structure, supporting multi-branch parallel exploration. It can also automatically merge similar analysis nodes, strengthen key conclusions, and weaken low-level branches, significantly reducing the generation of redundant information. This can shorten the time required to print the prosthesis, and further shorten the time required to print the prosthesis based on the prosthesis printing reference information, thereby improving the timeliness of prosthesis printing.

[0072] Step 106: Generate a tumor material reconstruction image based on the prosthesis printing reference information.

[0073] In some embodiments, the executing entity can generate a tumor material reconstruction image based on the aforementioned prosthesis printing reference information. This tumor material reconstruction image can represent the medical image upon which the tumor repair prosthesis for the target user is based. The aforementioned tumor repair prosthesis can represent a prosthesis used to repair the site after resection of a bone or soft tissue tumor in the target user. Here, the type of medical image is not limited; for example, the medical image can be a 3D CT image. It should be noted that the aforementioned tumor repair prosthesis can be the aforementioned bone or soft tissue tumor repair prosthesis.

[0074] In some optional implementations of certain embodiments, the aforementioned execution entity can generate a tumor material reconstruction image based on prosthesis printing reference information through the following steps: The first step involves generating tumor resection information based on the aforementioned prosthesis printing reference information and preset tumor images. This tumor resection information can characterize the extent of the bone defect after tumor resection and the biological anatomical structures to be avoided. The preset tumor images can represent medical images of the tumor site of the target user, such as 3D CT images. The specific content of the aforementioned bone defect extent and biological anatomical structures is not limited here; for example, the bone defect extent can be "gap extent: 6 cm long, 2 cm wide, 3 cm high," and the biological anatomical structure can be a blood vessel. In practice, firstly, the aforementioned prosthesis printing reference information can be processed using Named Entity Recognition (NER) and relation extraction methods to obtain extracted entities. For example, each entity can represent the distal femur, osteosarcoma, wide resection, popliteal artery and vein, and artificial prosthesis replacement. Then, the relationships between these entities are created using a relation extraction model. Finally, based on a surgical knowledge graph constructed using ICD-9-CM-3, the tumor resection information is determined according to the relationships between the entities and the preset tumor images.

[0075] The second step is to determine the repair material parameters based on the tumor resection information. These parameters characterize the parameters of the prosthetic material that needs to be used to fill the tumor after resection. The specific type of these parameters is not limited here; for example, the parameters could be the dimensions of the prosthetic material. In practice, the executing entity can determine the repair material parameters based on the tumor resection information using a three-dimensional geometric parameter extraction and mapping algorithm.

[0076] The third step involves modifying the aforementioned preset tumor image to obtain the modified preset tumor image as the target preset tumor image. In practice, firstly, the executing entity can preprocess the preset tumor image using a combination of grayscale thresholding and region growing algorithms to obtain tumor region annotation information. Then, through image cropping, the tumor region annotation information is removed from the preset tumor image, resulting in the modified preset tumor image as the target preset tumor image. The aforementioned tumor region annotation information can characterize the image region where the tumor exists in the preset tumor image.

[0077] The fourth step involves generating a reconstructed tumor image based on the aforementioned repair material parameters and the target pre-defined tumor image. In practice, firstly, the executing entity can use parametric modeling combined with geometric contour matching to convert the repair material parameters into a three-dimensional prosthetic structure adapted to the target pre-defined tumor image. Then, through spatial registration and geometric fusion methods, the three-dimensional prosthetic structure is aligned and shaped with the target pre-defined tumor image to obtain the reconstructed tumor image.

[0078] Step 107: Based on the tumor material reconstruction image, control the printing device to print out the tumor repair prosthesis corresponding to the tumor material reconstruction image.

[0079] In some embodiments, the executing entity can control a printing device to print a tumor repair prosthesis corresponding to the tumor material reconstruction image based on the tumor material reconstruction image. Here, the specific types of the printing device and the tumor repair prosthesis are not limited. For example, if the tumor repair prosthesis is an implantable metal prosthesis, then the printing device can be an industrial-grade metal 3D printer.

[0080] The above embodiments of this disclosure have the following beneficial effects: the bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system of some embodiments of this disclosure can improve the accuracy of the generated prosthesis printing reference information and reduce the waste of 3D printing biocompatible materials. Specifically, the reason for the low accuracy of the generated prosthesis printing reference information and the waste of 3D printing biocompatible materials is that the AI-assisted printing system does not refer to literature with a high level of evidence, does not integrate key molecular pathological features of the tumor, and the logic for generating the prosthesis printing reference information only pursues the accuracy of the data-level output results, resulting in low accuracy of the generated prosthesis printing reference information. Printing a prosthesis based on the low accuracy of the prosthesis printing reference information results in deviations in the size of the printed prosthesis, missing or redundant key structures, making the printed prosthesis unusable and wasting 3D printing biocompatible materials. Based on this, the bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system of some embodiments of this disclosure can obtain the tumor grade score corresponding to the tumor-related material information obtained above through the above-mentioned preset tumor grade screening information. Then, a tumor knowledge graph is constructed based on the tumor grade scores of the aforementioned tumor-related materials. This allows for the generation of prosthesis printing reference information based on materials with higher tumor grade scores within the tumor-related materials information, thereby improving the accuracy of the generated prosthesis printing reference information. Prosthesis printing can also be performed based on the prosthesis printing reference information generated by the aforementioned tumor decision-making intelligence group. The various tumor decision-making intelligence agents within this group have different functions and can process the pre-set tumor record information from a multidisciplinary perspective, thus meeting practical needs rather than merely pursuing the accuracy of data-level output results. This improves the accuracy of the generated prosthesis printing reference information and allows for prosthesis printing based on highly accurate reference information, reducing the waste of biocompatible materials in 3D printing.

[0081] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a bone and soft tissue tumor repair prosthesis printing device based on a multi-agent decision-making system. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0082] like Figure 2 As shown, the bone and soft tissue tumor repair prosthesis printing device 200 based on a multi-agent decision system in some embodiments includes: a first generation unit 201, a second generation unit 202, a modification unit 203, a third generation unit 204, a fourth generation unit 205, a fifth generation unit 206, and a control unit 207. The system comprises the following components: a first generation unit 201, configured to generate tumor grade score information in response to a detected printing instruction for a printing device, based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information; a second generation unit 202, configured to generate tumor knowledge graph information based on the tumor grade score information and the tumor-related material information; a modification unit 203, configured to modify the initialized preset tumor decision-making agent to obtain a tumor decision-making agent group; a third generation unit 204, configured to generate task processing allocation information based on preset tumor record information and the tumor decision-making agent group; a fourth generation unit 205, configured to generate prosthesis printing reference information based on the tumor decision-making agent group, the task processing allocation information, and the tumor knowledge graph information; a fifth generation unit 206, configured to generate a tumor material reconstruction image based on the prosthesis printing reference information; and a control unit 207, configured to control the printing device to print a tumor repair prosthesis corresponding to the material reconstruction image based on the tumor material reconstruction image.

[0083] It is understandable that the units described in the bone and soft tissue tumor repair prosthesis printing device 200 based on a multi-agent decision-making system are similar to those in the reference system. Figure 1 The steps described in the method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the bone and soft tissue tumor repair prosthesis printing device 200 based on a multi-agent decision-making system and the units contained therein, and will not be repeated here.

[0084] Figure 3 This is a schematic diagram illustrating an application scenario of a bone and soft tissue tumor repair prosthesis printing method based on a multi-agent decision-making system according to some embodiments of the present disclosure.

[0085] exist Figure 3In the application scenario, firstly, in response to the detection of printing instruction information 302 for the printing device, the computing device 301 generates tumor grade score information 303 based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information. Secondly, based on the tumor grade score information 303 and the tumor-related material information 304, tumor knowledge graph information 305 is generated. Then, the initialized preset tumor decision-making agent is modified to obtain a tumor decision-making agent group 306. Next, based on preset tumor record information and the tumor decision-making agent group 306, task processing allocation information 307 is generated. Next, based on the tumor decision-making agent group 306, the task processing allocation information 307, and the tumor knowledge graph information 305, prosthesis printing reference information 308 is generated. Then, based on the prosthesis printing reference information 308, a tumor material reconstruction image 309 is generated. Finally, based on the tumor material reconstruction image 309, the printing device 310 is controlled to print a tumor repair prosthesis 310 corresponding to the material reconstruction image 309.

[0086] It should be noted that the aforementioned computing device 301 can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed within the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0087] It should be understood that Figure 3 The number of computing devices shown is merely illustrative. Any number of computing devices can be used depending on implementation needs.

[0088] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0089] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0090] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0091] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0092] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor, and for example, can be described as: a first generation unit, a second generation unit, a modification unit, a third generation unit, a fourth generation unit, a fifth generation unit, and a control unit. The names of these units do not necessarily limit the unit itself; for example, the first generation unit can also be described as "a unit that, in response to detecting printing instruction information for a printing device, generates tumor grade score information based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information."

[0093] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0094] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for printing prostheses for repairing bone and soft tissue tumors based on a multi-agent decision-making system, comprising: In response to the detection of printing command information for the printing device, tumor grade score information is generated based on preset tumor grade screening information, preset tumor grading standard information and acquired tumor-related material information; Based on the tumor grade score information and the tumor-related material information, a tumor knowledge graph information is generated; Modify the initialized preset tumor decision-making agent to obtain a tumor decision-making agent group; Based on the preset tumor record information and the tumor decision-making intelligent agent group, task processing allocation information is generated; Based on the tumor decision-making intelligent agent group, the task processing allocation information, and the tumor knowledge graph information, generate prosthesis printing reference information; Based on the prosthesis printing reference information, a tumor material reconstruction image is generated; Based on the tumor material reconstruction image, the printing device is controlled to print out a tumor repair prosthesis corresponding to the material reconstruction image.

2. The method according to claim 1, wherein, The preset tumor grading standard information includes first evidence-based grading standard information and second evidence-based grading standard information; and In response to detecting a printing instruction for the printing device, tumor grade score information is generated based on preset tumor grade screening information, preset tumor grading standard information, and acquired tumor-related material information, including: In response to determining that the preset material type information of the tumor-related material information meets the preset type screening conditions, material grade score information is generated according to the first evidence-based grading standard information; In response to the determination that the preset material type information of the tumor-related material information does not meet the preset type screening conditions, material mapping score information is generated according to the second evidence-based grading standard information; Based on the material grade score information and the material mapping score information, tumor grade score information corresponding to the tumor-related material information is generated.

3. The method according to claim 2, wherein, In response to determining that the preset material type information of the tumor-related material information meets the preset type screening conditions, the material grade score information is generated according to the first evidence-based grading standard information, including: In response to determining that the preset material type information meets the preset type screening conditions, initial material grade score information is generated according to the first evidence-based grading standard information, wherein the initial material grade score information is either the first type of material grade score information or the second type of material grade score information; In response to determining that the grade score information of the first type of material meets the preset downgrade constraint conditions, the grade score information of the first type of material is modified to obtain the modified grade score information of the first type of material as the grade score information of the first type of target material. In response to determining that the second type of material grade score information meets the preset upgrade constraint conditions, the second type of material grade score information is modified to obtain the modified second type of material grade score information as the second type of target material grade score information; The material grade score information is determined based on the initial material grade score information, the first type of target material grade score information, and the second type of target material grade score information.

4. The method according to claim 1, wherein, The step of generating a tumor knowledge graph based on the tumor grade score information and the tumor-related material information includes: The tumor-related material information is updated to obtain the target tumor-related material information; Based on the pre-trained tumor-related entity extraction model and the target tumor-related material information, tumor material association information corresponding to the target tumor-related material information is generated; The tumor grade score information and source attribute information are identified as metadata information; Based on the tumor material association information and the metadata information, a tumor knowledge graph is generated.

5. The method according to claim 1, wherein, The modification process of the initialized preset tumor decision-making agent to obtain a tumor decision-making agent group includes: Based on the preset tumor problem information set and the initialized preset tumor decision-making agent, generate preference decision result ranking information; Based on the preference decision result ranking information and ranking loss function information, a preference decision signal generation model is generated. Based on the initialized preset tumor decision-making agent, a decision reference model is determined; Based on the decision reference model and the initialized preset tumor decision agent, a tumor decision agent group is generated.

6. The method according to claim 5, wherein, The method further includes: Based on the target preset tumor problem information set and the initialized preset tumor decision-making agent, perform the following training steps: Based on the decision reference model and the target preset tumor question information in the target preset tumor question information set that meets the preset question selection conditions, target tumor question answer pair information is generated; Based on the target tumor question answer pair information and the preference decision signal generation model, generate preference decision reward signal information; Based on the decision reference model and the initialized preset tumor decision agent, generate answer-pair penalty constraint information; Based on the preference decision reward signal information and the answer pair penalty constraint information, the initialized preset tumor decision agent is modified to obtain the modified preset tumor decision agent.

7. The method according to claim 6, wherein, The training steps also include: In response to determining that the preference decision reward signal information and the answer to the penalty constraint information do not meet the preset optimization target conditions, the modified preset tumor decision agent is used as the initialized preset tumor decision agent, and the training step is executed again. In response to determining that the preference decision reward signal information and the answer to the penalty constraint information satisfy the preset optimization target conditions, each tumor decision agent is generated as a tumor decision agent group based on the modified preset tumor decision agent, preset agent framework setting information and preset agent configuration information.

8. A prosthesis printing device for bone and soft tissue tumor repair based on a multi-agent decision-making system, comprising: The first generation unit is configured to generate tumor grade score information in response to the detection of printing instruction information for the printing device, based on preset tumor grade screening information, preset tumor grading standard information and acquired tumor-related material information. The second generation unit is configured to generate tumor knowledge graph information based on the tumor grade score information and the tumor-related material information; The modification unit is configured to modify the initialized preset tumor decision agent to obtain a tumor decision agent group. The third generation unit is configured to generate task processing allocation information based on preset tumor record information and the tumor decision-making intelligent agent group. The fourth generation unit is configured to generate prosthesis printing reference information based on the tumor decision-making intelligent agent group, the task processing allocation information, and the tumor knowledge graph information. The fifth generation unit is configured to generate a tumor material reconstruction image based on the prosthesis printing reference information; The control unit is configured to control the printing device to print out a tumor repair prosthesis corresponding to the material reconstruction image based on the tumor material reconstruction image.

9. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images