Medical data evaluation and correction system, method, medium, program product and terminal based on large language model and multi-agent cooperation

Abstract

This application provides a medical data evaluation and correction system, method, medium, program product, and terminal based on a large language model and multi-agent collaboration. The system performs quality analysis on the medical data to obtain quality assessment results; performs semantic analysis on the medical data to obtain semantic recognition results; analyzes the quality assessment results and semantic recognition results to obtain a preliminary correction scheme corresponding to the medical data; adjusts the preliminary correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme; and corrects the medical data based on the final correction scheme to obtain corrected medical data. This effectively overcomes the technical shortcomings of traditional methods that rely on fixed rules and suffer from stagnant error rates.

Description

Technical Field

[0001] This application relates to the field of medical data evaluation and correction technology, and in particular to medical data evaluation and correction systems, methods, media, program products and terminals based on large language models and multi-agent collaboration. Background Technology

[0002] With the accelerated development of healthcare informatization, hospital information systems (HIS), electronic health records (EHR), and regional healthcare platforms have accumulated massive amounts of structured and unstructured medical data. However, data quality issues severely restrict their application in high-value scenarios such as clinical decision support, real-world research, and AI model training.

[0003] Current mainstream medical data quality governance technologies have the following core flaws:

[0004] (1) Lack of dynamic adaptability: Existing systems rely heavily on preset rule bases (such as ICD-10 international disease classification coding consistency verification and field non-empty constraints), which cannot cope with new diagnosis and treatment items, coding standard updates or regional differences in clinical habits, resulting in rules becoming outdated quickly.

[0005] (2) No feedback learning loop: The correction results only apply to a single data repair and do not form an iterative mechanism of governance, feedback, learning and optimization. Similar errors recur, resulting in low governance efficiency. Moreover, complex issues such as boundary cases and semantic conflicts still require expert review on a case-by-case basis, which is difficult to support the governance needs of millions of data points per day.

[0006] (3) Insufficient adaptation of TCM data: TCM data has a high degree of semantic coupling (e.g., the syndrome of "spleen deficiency and dampness" should correspond to "pale and swollen tongue, white and greasy coating, and soft and slow pulse"). However, the general treatment tools have not integrated authoritative knowledge systems such as "Classification and Code of TCM Diseases and Syndromes" and "Chinese Materia Medica Formulas", lack TCM-specific knowledge graph support, and cannot identify logical contradictions between "syndrome-symptom-drug-pulse" (e.g., "Eighteen Incompatible Chinese Medicine Combinations").

[0007] Although recent studies have attempted to incorporate machine learning for anomaly detection (such as Isolation Forest and Autoencoders), these remain "open-loop" systems, capable of only single-dimensional anomaly identification, lacking semantic understanding, unable to handle the complex logical relationships within traditional Chinese medicine, and lacking the ability to continuously learn from expert feedback. Furthermore, existing agent systems are primarily used for task scheduling or dialogue generation, and have not yet been applied to closed-loop evolutionary scenarios for medical data quality governance.

[0008] Therefore, there is an urgent need for a new medical data quality governance system that integrates the semantic understanding capabilities of large language models, multi-agent collaborative decision-making mechanisms, and autonomous evolutionary architecture to achieve the data governance goals of high quality, low manual labor, and sustainable evolution, and to solve the core pain points of traditional technologies such as static nature, lack of closed loop, high dependence on manual labor, and insufficient adaptation to traditional Chinese medicine. Summary of the Invention

[0009] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a medical data evaluation and correction system, method, medium, program product and terminal based on a large language model and multi-agent collaboration, in order to solve the deficiencies of the existing systems.

[0010] To achieve the above and other related objectives, a first aspect of this application provides a medical data evaluation and correction system based on a large language model and multi-agent collaboration. The system includes: a data acquisition module for acquiring medical data generated in a medical information system; a multi-agent evaluation module comprising: a quality assessment agent for performing quality analysis on the medical data to obtain quality assessment results; a semantic recognition agent for performing semantic analysis on the medical data to obtain semantic recognition results; a data correction agent for analyzing the quality assessment results and semantic recognition results to obtain a preliminary correction scheme corresponding to the medical data; a consensus coordination agent for adjusting the preliminary correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme, and correcting the medical data based on the final correction scheme to obtain corrected medical data; and a self-optimization module connected to the multi-agent evaluation module for updating the multi-agent evaluation module based on a large language model to obtain an updated multi-agent evaluation module.

[0011] In some embodiments of the first aspect of this application, the self-optimization module includes: a screening agent, used to perform comprehensive analysis based on quality assessment results, semantic recognition results, preliminary correction schemes, and final correction schemes, to screen out medical data that meets preset screening rules and their corresponding preliminary and final correction schemes; an expert review agent, used to manually adjust the screened medical data and their corresponding preliminary and final correction schemes to obtain the screened medical data and their corresponding expert feedback data, constituting a training dataset; and an incremental update agent, used to incrementally update the multi-agent evaluation module using a large language model combined with the training dataset.

[0012] In some embodiments of the first aspect of this application, the process of incrementally updating the multi-agent evaluation module using a large language model combined with a training dataset includes: training the quality evaluation agent and data correction agent based on a horizontal federated learning framework and using encrypted gradient aggregation to obtain updated quality evaluation agent and updated data correction agent; training the semantic recognition agent based on a natural language processing algorithm to obtain updated semantic recognition agent; and training the consensus coordination agent based on a preset reward function and using a reinforcement learning model to obtain updated consensus coordination agent.

[0013] In some embodiments of the first aspect of this application, the process of performing quality analysis on the medical data includes: performing multi-dimensional quality scoring on the medical data based on a pre-trained machine learning model and preset judgment rules to obtain quality assessment results.

[0014] In some embodiments of the first aspect of this application, the process of performing semantic analysis on the medical data includes: using a semantic matching algorithm based on a preset clinical medical knowledge graph to identify semantic conflicts in the medical data and generate corresponding semantic recognition results.

[0015] In some embodiments of the first aspect of this application, the process of analyzing the quality assessment results and semantic recognition results includes: analyzing medical data based on a conditional variational autoencoder to obtain a preliminary correction scheme corresponding to the medical data.

[0016] To achieve the above and other related objectives, a second aspect of this application provides a method for medical data evaluation and correction based on a large language model and multi-agent collaboration. The method includes: acquiring medical data generated in a medical information system; performing quality analysis on the medical data to obtain quality evaluation results; performing semantic analysis on the medical data to obtain semantic recognition results; analyzing the quality evaluation results and semantic recognition results to obtain a preliminary correction scheme corresponding to the medical data; adjusting the preliminary correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme, and correcting the medical data based on the final correction scheme to obtain corrected medical data; and updating the multi-agent evaluation module based on a large language model to obtain an updated multi-agent evaluation module.

[0017] To achieve the above and other related objectives, a third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for medical data evaluation and correction based on a large language model and multi-agent collaboration.

[0018] To achieve the above and other related objectives, a fourth aspect of this application provides a computer program product comprising computer program code, which, when executed on a computer, enables the computer to implement the aforementioned medical data evaluation and correction method based on a large language model and multi-agent collaboration.

[0019] To achieve the above and other related objectives, a fifth aspect of this application provides an electronic terminal, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the aforementioned medical data evaluation and correction method based on a large language model and multi-agent collaboration.

[0020] As described above, the medical data evaluation and correction system, method, medium, program product, and terminal based on large language models and multi-agent collaboration of this application have the following beneficial effects:

[0021] (1) This application constructs a feedback-driven closed-loop evolutionary system through multi-agent collaboration and self-optimization mechanisms, enabling the medical data evaluation and correction capabilities to continuously improve during operation. The system can reduce the data error rate in the initial stage and continuously optimize the correction effect through incremental learning in subsequent operation, overcoming the technical defects of traditional medical data governance systems that rely on fixed rules and whose error rate stagnates for a long time.

[0022] (2) This application can continuously improve the accuracy and consistency of automatic correction results through a multi-agent negotiation mechanism and an expert feedback-driven self-optimization module, so that the system’s reliance on human expert review gradually decreases over time. Under the premise of ensuring medical safety and accuracy, it can effectively reduce the cost of human participation of clinical experts and improve the overall efficiency of medical data governance.

[0023] (3) This application can adapt to changes in medical data patterns and updates in clinical norms without manual modification of rules. Especially in medical scenarios with complex semantics and professional rules, such as traditional Chinese medicine, it can achieve accurate governance through semantic understanding and knowledge graphs. At the same time, the system adopts federated learning and process traceability mechanisms to ensure the privacy and security of medical data, and to achieve auditability and traceability of the model update process, thus meeting the requirements of medical data supervision.

[0024] (4) This application can accurately identify semantic conflicts unique to TCM data by using a built-in TCM clinical knowledge graph. The governance effect is better than that of general medical data governance systems and is accurately adapted to TCM scenarios. Attached Figure Description

[0025] Figure 1The diagram shown is a structural schematic of a medical data evaluation and correction system based on a large language model and multi-agent collaboration, according to an embodiment of this application.

[0026] Figure 2 The diagram shown is a schematic representation of a self-optimization module in one embodiment of this application.

[0027] Figure 3 The diagram shown is a flowchart illustrating a medical data evaluation and correction method based on a large language model and multi-agent collaboration, according to an embodiment of this application.

[0028] Figure 4 The diagram shown is a structural schematic of an electronic terminal according to an embodiment of this application. Detailed Implementation

[0029] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0030] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0031] <1> LLM: Large Language Model (LLM) is a deep learning model trained on massive amounts of text that can understand and generate human language.

[0032] <2> Agent: An intelligent agent is a software entity that can perceive the environment, make autonomous decisions, and perform actions to achieve its goals.

[0033] <3> C-VAE: Conditional Variational Autoencoder (C-VAE) is a generative model that learns the implicit representation of data given additional conditional information.

[0034] <4> Semantic matching algorithm: A technical method to achieve accurate content association by calculating the semantic similarity between texts.

[0035] <5> K-means clustering algorithm: An unsupervised classification method that divides data into K compact clusters based on distance iteration.

[0036] <6> Horizontal federated learning framework: A collaborative framework in which multiple data owners train models locally and achieve joint learning by exchanging parameters in a cryptographic manner.

[0037] <7> Natural Language Processing (NLP) algorithms are a set of computational methods that enable computers to parse, understand, and generate human language.

[0038] <8> Reinforcement learning models: a machine learning paradigm that optimizes decision-making strategies through trial and error and interaction with the environment, using reward signals.

[0039] To facilitate understanding of the embodiments of this application, firstly, in conjunction with Figure 1 Detailed explanation. Figure 1 This diagram illustrates the structure of a medical data evaluation and correction system 100 based on a large language model and multi-agent collaboration, according to an embodiment of the present invention. The system 100 includes: a data acquisition module 101, a multi-agent evaluation module 102, and a self-optimization module 103. The data acquisition module 101 is connected to the multi-agent evaluation module 102, and the multi-agent evaluation module 102 is connected to the self-optimization module 103. The system can be deployed on a cloud server, a local server, or a hybrid deployment environment combining cloud and local deployments, and is used for automated evaluation, correction, and continuous optimization of medical data generated in a medical information system.

[0040] The data acquisition module 101 is used to acquire medical data generated in the medical information system.

[0041] The medical information system may include, but is not limited to: Hospital Information System (HIS), Electronic Medical Record System (EMR), Laboratory Information System (LIS), Picture Archiving and Communication System (PACS), and regional medical platform.

[0042] In one embodiment, the data acquisition module may acquire medical data through at least one of the following methods: calling the data interface provided by the medical information system, parsing the data file exported by the medical information system, monitoring the data change log of the medical information system, etc., without specific limitations.

[0043] The medical data, categorized by data type, can be structured, semi-structured, or unstructured, including but not limited to medical records, examination and test results, medical orders, medical progress notes, and related text data. Furthermore, based on data content, the medical data also includes general medical data, such as electronic medical records and laboratory reports; and data specific to traditional Chinese medicine, such as TCM diagnosis records, herbal prescriptions, and tongue and pulse information. After necessary format conversion and preprocessing, the acquired medical data is sent to the multi-agent evaluation module for further processing.

[0044] The multi-agent evaluation module 102 includes: a quality assessment agent, used to perform quality analysis on the medical data to obtain quality assessment results; a semantic recognition agent, used to perform semantic analysis on the medical data to obtain semantic recognition results; a data correction agent, used to analyze the quality assessment results and semantic recognition results to obtain a preliminary correction scheme corresponding to the medical data; and a consensus coordination agent, used to adjust the preliminary correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme, and to correct the medical data based on the final correction scheme to obtain corrected medical data.

[0045] It should be noted that the aforementioned agents are not simple rule modules or single model instances, but rather intelligent processing units built on the capabilities of large language models, possessing autonomous analysis, reasoning, and collaboration abilities.

[0046] Furthermore, each agent assumes a different functional role in the system. In specific implementations, each agent can be considered an independent intelligent processing entity, which includes at least: an input interface for receiving medical data or intermediate results from other agents; semantic understanding and reasoning capabilities based on a large language model; task execution and result generation; and a communication interface for information interaction with other agents. Through the above structural design, each agent can achieve information sharing and collaborative decision-making while maintaining functional division of labor, thereby forming a multi-agent collaborative processing system to further collaboratively complete the initial assessment and correction of data quality.

[0047] In this embodiment, the semantic understanding and reasoning capabilities of each agent are primarily achieved by introducing a large language model. Specifically, each agent, when performing its corresponding task, can invoke a pre-deployed large language model to perform semantic understanding and reasoning analysis on the input medical data. The large language model can be a general pre-trained language model, or a language model that has undergone domain adaptation or fine-tuning in the medical field.

[0048] Specifically, the multi-agent evaluation module includes at least: a quality assessment agent; a semantic recognition agent; a data correction agent; and a consensus coordination agent. These agents are logically independent but functionally collaborative, working together to evaluate and correct medical data.

[0049] In one embodiment of this application, the process of performing quality analysis on the medical data includes: performing multi-dimensional quality scoring on the medical data based on a pre-trained machine learning model and preset judgment rules to obtain quality assessment results.

[0050] A quality assessment agent is used to analyze the medical data at the quality level to obtain quality assessment results. Specifically, the quality assessment agent can analyze the medical data based on a pre-trained machine learning model, focusing on multiple preset quality dimensions. These preset quality dimensions include: data integrity, used to assess whether key information fields in the medical data are missing; data consistency, used to assess whether inconsistencies exist between different fields or records in the medical data; data logicality, used to assess whether the medical data conforms to medical logical relationships; and data accuracy, used to assess the degree to which the content of the medical data conforms to medical common sense or clinical standards.

[0051] Preferably, the pre-trained machine learning model is constructed using the isolated forest algorithm. The model consists of multiple decision trees, each of which is trained based on a randomly selected subset of data to output the scoring results of the multiple preset quality dimensions.

[0052] For example, the isolated forest model may include, for instance, 100 decision trees, with each decision tree corresponding to a data subset of 256 samples. It should be noted that the above parameters are merely exemplary settings and can be adjusted according to the data scale and system performance requirements in actual applications; this invention does not limit these adjustments.

[0053] Furthermore, during the aforementioned quality analysis process, the quality assessment agent can combine a large language model to understand the semantic structure of medical data, thereby helping to determine whether there are problems such as mismatch between the diagnosis conclusion and the pulse description, or potential conflicts in the compatibility of traditional Chinese medicine, thus improving the accuracy of quality assessment.

[0054] Furthermore, the quality assessment agent generates a quality assessment result based on the scoring results of multiple preset quality dimensions corresponding to the medical data. The quality assessment result can be normalized to a preset range (e.g., between 0 and 1). When the quality assessment result of the medical data is lower than a preset assessment threshold (e.g., 0.8), the quality assessment agent can mark the corresponding medical data as low-quality data, while medical data higher than the preset assessment threshold can be marked as high-quality data.

[0055] The quality assessment results and scoring reports output by the quality assessment agent serve as important inputs for subsequent multi-agent collaborative processing, thus providing a reliable basis for the automatic correction and optimization of medical data.

[0056] In one embodiment of this application, the process of performing semantic analysis on the medical data includes: using a semantic matching algorithm based on a preset clinical medical knowledge graph to identify semantic conflicts in the medical data and generate corresponding semantic recognition results.

[0057] Furthermore, the semantic recognition agent can run synchronously or in parallel with the quality assessment agent. By invoking a preset clinical medical knowledge graph, it performs structured parsing and semantic mapping on diagnostic information, syndrome descriptions, medication regimens, and related text content in medical data. The preset clinical medical knowledge graph may contain information such as basic concepts, syndrome relationships, drug contraindications, and compatibility rules in the field of traditional Chinese medicine.

[0058] Furthermore, the semantic recognition agent employs a semantic matching algorithm to compare and analyze the semantic relationships between different fields in the medical data during semantic analysis, in order to identify potential semantic conflicts or logical inconsistencies. The semantic recognition agent can combine the semantic understanding and reasoning capabilities of a large language model to comprehensively judge the contextual semantics of medical text, thereby improving the accuracy of identifying complex semantic conflicts.

[0059] Furthermore, the semantic recognition agent can accurately identify logical contradictions unique to traditional Chinese medicine data within the medical data. For example, when analyzing medication regimens, it can identify violations of traditional Chinese medicine contraindications, such as drug combinations with incompatible relationships like the "Eighteen Incompatibilities" or "Nineteen Antagonisms" in the same prescription, or conflicts between the "pale red tongue with white coating" syndrome and the "rapid pulse" pulse pattern.

[0060] Specifically, the semantic recognition agent performs semantic conflict identification on the medical data, and can identify contradictions and conflicts such as "Aconite and Pinellia (Eighteen Incompatibilities) in the same prescription" which have contraindications in traditional Chinese medicine, and marks the corresponding medical data as data with semantic conflicts, and generates semantic recognition results corresponding to the medical data.

[0061] The semantic recognition results can be used to characterize the semantic rationality of medical data and serve as an important input basis for the data correction agent and consensus coordination agent to make correction decisions, thereby supporting subsequent multi-agent collaborative correction processing.

[0062] In one embodiment of this application, the process of analyzing the quality assessment results and semantic recognition results includes: analyzing medical data based on a conditional variational autoencoder to obtain a preliminary correction scheme corresponding to the medical data.

[0063] Specifically, the data correction agent receives quality assessment results from the quality assessment agent corresponding to the same medical data, as well as semantic recognition results from the semantic recognition agent.

[0064] Furthermore, the data correction agent performs correction analysis on medical data based on a conditional variational autoencoder (CVA) model. The input layer of the CVA model includes a medical data feature filtering module. After the quality assessment results and semantic recognition results are input into the CVA model, the module filters out features relevant to the quality assessment and removes redundant fields such as patient names and contact information that are irrelevant to the quality assessment. The output layer of the CVA model outputs preliminary correction schemes for abnormal medical data, such as missing value imputation or abnormal code correction. The output layer is also configured with correction constraints (such as TCM syndrome coding conforming to the "Classification and Code of TCM Diseases and Syndromes") to impose normative constraints on the correction results, ensuring that the preliminary correction scheme is limited to the legal scope stipulated in the "Classification and Code of TCM Diseases and Syndromes" and the "Regulations for the Management of Traditional Chinese Medicine Prescriptions," thereby guaranteeing the medical rationality and compliance of the output correction scheme.

[0065] Using the conditional variational autoencoder model, the data correction agent generates preliminary correction schemes for problematic medical data, provided that the quality assessment results and semantic constraints are met. These preliminary correction schemes may include suggestions for adjusting aspects such as syndrome descriptions, pulse information, and medication regimens.

[0066] For example, for medical data where there is a mismatch between symptoms and pulse, the data correction agent can correct the original "spleen deficiency and dampness + rapid pulse" to "spleen deficiency and dampness + soft and slow pulse," which is more in line with the logic of traditional Chinese medicine. For medication plans that violate the contraindications of traditional Chinese medicine, the data correction agent can automatically remove Pinellia ternata from the prescription that violates the "Eighteen Incompatibilities" rule, and recommend alternative herbs based on medical knowledge, such as tangerine peel.

[0067] Furthermore, the input layer feature dimension of the conditional variational autoencoder model can be 128, and the number of hidden layer nodes can be 64. It should be noted that the above parameters are merely exemplary settings, and can be adjusted according to the scale of medical data and system performance requirements in practical applications; this invention does not limit these adjustments.

[0068] In one embodiment of this application, the consensus coordination agent is used to adjust the preliminary correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme, and to correct the medical data based on the final correction scheme to obtain the corrected medical data.

[0069] The consensus coordination agent receives the same medical data, a preliminary correction plan from the data correction agent, and the quality assessment results and semantic recognition results from the quality assessment agent and semantic recognition agent. When the consensus coordination agent identifies significant discrepancies among the preliminary correction plan, quality assessment results, and semantic recognition results, such as a quantitative score difference exceeding 50%, the consensus coordination agent further coordinates the preliminary correction plan based on a multi-agent negotiation mechanism.

[0070] Specifically, the multi-agent negotiation mechanism assigns corresponding coordination weight parameters to the quality assessment agent, semantic recognition agent, and data correction agent based on the type, semantic complexity, and historical correction performance of the current medical data. In scenarios involving medical semantic conflicts or traditional Chinese medicine rule judgments, the weight ratio of the semantic recognition agent in the negotiation process is increased.

[0071] Furthermore, the consensus coordination agent calculates a weighted score based on the coordination weight parameters and corresponding quantitative scores of the quality assessment agent, semantic recognition agent, and data correction agent. When the weighted score meets the preset medical rule constraints, the preliminary correction scheme is confirmed as the final correction scheme, and the medical data is corrected based on the final correction scheme to obtain the corrected medical data.

[0072] In another embodiment of this application, when there are multiple preliminary correction schemes corresponding to the medical data, the candidate scheme that satisfies the preset medical rule constraints and has the highest weighted score is selected as the final correction scheme output.

[0073] By embedding a large language model into the decision-making and analysis process of a multi-agent system, the multi-agent system in this invention not only possesses the data processing capabilities of traditional machine learning models, but also has higher-level semantic understanding and reasoning capabilities, thereby improving the accuracy and intelligence level of medical data evaluation and correction.

[0074] like Figure 1 As shown, the self-optimization module 103 is connected to the multi-agent evaluation module 102 and is used to update the multi-agent evaluation module based on the large language model to obtain the updated multi-agent evaluation module.

[0075] In one embodiment of this application, the self-optimization module includes: a screening agent, used to perform comprehensive analysis based on quality assessment results, semantic recognition results, preliminary correction schemes, and final correction schemes, to screen out medical data that meets preset screening rules and their corresponding preliminary and final correction schemes; an expert review agent, used to manually adjust the screened medical data and their corresponding preliminary and final correction schemes to obtain the screened medical data and their corresponding expert feedback data, forming a training dataset; and an incremental update agent, used to incrementally update the multi-agent evaluation module using a large language model combined with the training dataset.

[0076] Specifically, such as Figure 3 As shown, the corresponding information generated during the correction of the same medical data includes: the original medical data sample, the quality assessment result, the semantic recognition result, the preliminary correction scheme, and the final correction scheme.

[0077] Furthermore, the screening agent comprehensively analyzes the quality assessment results, semantic recognition results, preliminary correction scheme, and final correction scheme to determine whether the corresponding medical data meets the preset screening rules. The screening rules may include, but are not limited to: significant differences between the medical data before and after correction, obvious adjustments between the preliminary and final correction schemes, and representativeness of the correction results in historical data.

[0078] Furthermore, in one specific embodiment, the screening agent is also used to identify new data patterns in the medical data generated during system operation. Specifically, the screening agent performs cluster analysis on the feature vectors of the medical data based on an unsupervised clustering algorithm to identify new data distribution patterns not covered by existing models. These new data distribution patterns refer to medical data that possesses structure but does not match the pre-defined rules such as the TCM knowledge graph within the system. In one implementation scenario, the unsupervised clustering algorithm is the K-means clustering algorithm.

[0079] When medical data and its corresponding preliminary and final correction schemes meet the screening rules, the screening agent includes them in the candidate training sample set, thereby avoiding interference from low-value or noisy data on model updates.

[0080] The expert review agent pushes the candidate training sample set and its corresponding preliminary correction scheme to the expert review platform. The expert review platform is equipped with an expert team composed of multiple clinical experts with different professional titles, including at least one chief physician and at least one attending physician.

[0081] In one specific embodiment, the expert review platform consists of 3 chief physicians and 5 attending physicians, used to conduct parallel reviews of the correction schemes. Based on their own clinical experience and the diagnostic and treatment guidelines of their respective medical institutions, the experts review the candidate training sample set and provide expert feedback data for each preliminary correction scheme. The expert feedback data includes at least one of the following types: agreeing with the correction scheme, modifying the correction scheme, or rejecting the correction scheme.

[0082] Furthermore, the expert review agent performs structured processing on the expert feedback data and its corresponding medical data, and sets credibility weights for the expert feedback data according to the expert's professional title (0.8 for chief physician, 0.5 for attending physician, and 0.3 for resident physician), and constructs a training dataset accordingly.

[0083] In one embodiment of this application, the process of incrementally updating the multi-agent evaluation module using a large language model combined with a training dataset includes: training the quality evaluation agent and data correction agent based on a horizontal federated learning framework and using encrypted gradient aggregation to obtain updated quality evaluation agent and updated data correction agent; training the semantic recognition agent based on a natural language processing algorithm to obtain updated semantic recognition agent; and training the consensus coordination agent based on a preset reward function and using a reinforcement learning model to obtain updated consensus coordination agent.

[0084] In one specific embodiment, the incremental update agent, based on a horizontal federated learning framework, establishes communication connections with federated learning nodes of at least two collaborating medical institutions to initiate a cross-institutional joint model training process. Each federated learning node incrementally trains the isolated forest model and the conditional variational autoencoder model locally using the training signal sets collected by the collaborating medical institutions.

[0085] During training, each federated learning node only uploads model gradients or parameter update information and does not share the training signal set. The gradient information is encrypted using homomorphic encryption during transmission to ensure data privacy and security during federated training.

[0086] The federated learning employs the FedAvg aggregation algorithm to perform a weighted average of model update results from different federated nodes to generate a globally updated model. This globally updated model is used to replace or update the original quality assessment agent and data correction agent to obtain updated quality assessment agents and updated data correction agents, thereby improving their generalization ability in different medical institution scenarios.

[0087] Furthermore, the incremental update agent, based on natural language processing algorithms, performs text analysis on the expert feedback data output by the expert review agent, extracting clinical diagnosis and treatment logic, drug compatibility constraints, and semantic judgment rules from it.

[0088] New semantic rules are automatically generated based on expert feedback, and the threshold parameters of existing rules are adjusted and updated. For example, multiple new clinical semantic judgment rules are generated, and the judgment thresholds of some existing rules are modified.

[0089] The updated rule set is synchronized to the semantic recognition agent, resulting in an updated semantic recognition agent, which is used to improve its accuracy in recognizing complex medical semantic conflicts.

[0090] Furthermore, the incremental update agent constructs a reward function based on a reinforcement learning model. This reward function comprehensively considers factors such as the effectiveness of medical data correction, the cost of manual intervention, and system operating efficiency. The specific reward function is as follows:

[0091] Reward = λ1*(1-data error rate) + λ2*(1-percentage of manual review) - λ3*(system response delay); (Formula 1)

[0092] Wherein, λ1=0.5, λ2=0.3, λ3=0.2, Reward is the reward value, and the data error rate, the proportion of manual review and the system response delay are all business indicators obtained by statistics during the operation of the system.

[0093] The reinforcement learning model, based on the reward function, iteratively trains the weight allocation strategy of the consensus coordination agent in the multi-agent negotiation process. This allows the model to dynamically adjust the influence weights of each agent under different business scenarios, achieving optimization effects such as reduced data error rate, reduced manual review, and reduced system response latency. For example, in scenarios involving integrated traditional Chinese and Western medicine diagnosis and treatment, the reinforcement learning training results can increase the weight ratio of the semantic recognition agent in the negotiation process, thereby enhancing the system's ability to handle complex semantic relationships.

[0094] In another embodiment of this application, the self-optimization module periodically deploys the updated multi-agent evaluation module to the sandbox testing environment and performs performance evaluation based on historically acquired difficult case sets and newly collected test data.

[0095] Furthermore, the performance evaluation includes preset positive optimization criteria and negative degradation criteria. When the performance evaluation meets the positive optimization criteria, it is automatically released as a new version of the system, and the new version of the system supports gray-scale deployment, that is, it first covers 10% of medical data, and then is fully promoted after no abnormalities are found. When the performance evaluation meets the negative degradation criteria, the corresponding updated multi-agent evaluation module is determined to be negatively degraded, the release of the new version of the system is automatically terminated, and a rollback mechanism is initiated to restore the previous stable version of the system.

[0096] The negative degradation judgment index can be selected from any of the following: data error rate increase >3%, manual review volume increase >10%, and number of rule conflicts >5; the positive optimization judgment index can be selected from data error rate decrease ≥3%, manual review volume decrease ≥5%, and system response latency ≤100ms. It should be noted that the above parameters are only exemplary settings, and can be adjusted according to the data scale and system performance requirements in actual applications. This invention does not limit them.

[0097] The self-optimization module, based on the output of the multi-agent evaluation module, sequentially filters, manually reviews, and structures key samples in the medical data correction process through a screening agent, an expert review agent, and an incremental update agent. This forms a high-quality training dataset, providing a reliable data foundation for system updates. This invention realizes the transformation of the multi-agent evaluation and correction system from "static rule-driven" to "data and experience-driven," enabling the system to continuously adjust its behavior strategies according to changes in actual application scenarios, gradually reducing the proportion of human intervention and improving the automation level and consistency of medical data correction.

[0098] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" do not necessarily imply that they are different.

[0099] It should be noted that, in the embodiments of this application, the words "exemplary" or "for example" indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0100] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0101] Figure 3 This is a schematic block diagram illustrating a medical data evaluation and correction method based on a large language model and multi-agent collaboration, as provided in an embodiment of this application. The method is applied to a medical data evaluation and correction system based on a large language model and multi-agent collaboration as described above, and the method includes:

[0102] Step S31: Obtain medical data generated in the medical information system;

[0103] Step S32: Perform quality analysis on the medical data to obtain quality assessment results; perform semantic analysis on the medical data to obtain semantic recognition results; analyze the quality assessment results and semantic recognition results to obtain a preliminary correction scheme corresponding to the medical data; adjust the preliminary correction scheme using a multi-Agent negotiation mechanism to obtain a final correction scheme, and correct the medical data based on the final correction scheme to obtain corrected medical data;

[0104] Step S33: Update the multi-agent evaluation module based on the large language model to obtain the updated multi-agent evaluation module.

[0105] It should be understood that the specific process of each module performing the above-mentioned steps has been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

[0106] It should also be understood that the module division in the embodiments of this application is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0107] Figure 4This is a schematic block diagram of the electronic terminal provided in an embodiment of this application. Figure 4 As shown, the computer device includes at least one processor 401, a memory 402, at least one network interface 403, and a user interface 405. The various components in the device are coupled together via a bus system 404. It is understood that the bus system 404 is used to implement communication between these components. In addition to a data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 4 The general will label all buses as bus systems.

[0108] The user interface 405 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0109] It is understood that memory 402 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0110] In this embodiment of the invention, the memory 402 is used to store various types of data to support the operation of the electronic terminal 400. Examples of this data include any executable program for operation on the electronic terminal 400, such as the operating system 4021 and application programs 4022. The operating system 4021 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 4022 may contain various applications, such as media players, browsers, etc., for implementing various application services. The medical data evaluation and correction method based on a large language model and multi-agent collaboration provided in this embodiment of the invention can be included in the application program 4022.

[0111] The methods disclosed in the above embodiments of the present invention can be applied to processor 401, or implemented by processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 401 or by instructions in the form of software. The processor 401 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 401 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. General-purpose processor 401 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0112] In an exemplary embodiment, the electronic terminal 400 may be used to execute the aforementioned method by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs).

[0113] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to perform the method of any of the embodiments.

[0114] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code, which, when run on a computer, causes the computer to execute the medical data evaluation and correction method based on a large language model and multi-agent collaboration in any of the embodiments shown.

[0115] As used in this specification, the terms "component," "module," "system," etc., are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process and / or an execution thread, and components may be located on a single computer and / or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local and / or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, and / or a network, such as the Internet interacting with other systems via signals).

[0116] Those skilled in the art will recognize that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0117] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0121] In the above embodiments, the functions of each functional unit can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs, DVDs), or semiconductor media (e.g., solid-state disks, SSDs, etc.).

[0122] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0124] In summary, this application provides a medical data evaluation and correction system, method, medium, program product, and terminal based on a large language model and multi-agent collaboration. The system performs quality analysis on the medical data to obtain quality assessment results; performs semantic analysis on the medical data to obtain semantic recognition results; analyzes the quality assessment results and semantic recognition results to obtain a preliminary correction scheme corresponding to the medical data; adjusts the preliminary correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme; and corrects the medical data based on the final correction scheme to obtain corrected medical data. This effectively overcomes the technical shortcomings of traditional methods that rely on fixed rules and suffer from stagnant error rates.

[0125] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A medical data evaluation and correction system based on a large language model and multi-agent collaboration, characterized in that, The system includes: The data acquisition module is used to acquire medical data generated in the medical information system; A multi-agent evaluation module, connected to the data acquisition module, includes: A quality assessment agent is used to perform quality analysis on the medical data in order to obtain quality assessment results. A semantic recognition agent is used to perform semantic analysis on the medical data to obtain semantic recognition results; The data correction agent is used to analyze the quality assessment results and semantic recognition results to obtain a preliminary correction plan corresponding to the medical data. A consensus coordination agent is used to adjust the initial correction scheme using a multi-agent negotiation mechanism to obtain a final correction scheme, and to correct the medical data based on the final correction scheme to obtain the corrected medical data. A self-optimization module, connected to the multi-agent evaluation module, is used to update the multi-agent evaluation module based on a large language model to obtain an updated multi-agent evaluation module.

2. The medical data evaluation and correction system based on a large language model and multi-agent collaboration as described in claim 1, characterized in that, The self-optimization module includes: The screening agent is used to perform comprehensive analysis based on quality assessment results, semantic recognition results, preliminary correction plans, and final correction plans to screen out medical data that meet the preset screening rules and their corresponding preliminary and final correction plans. The expert review agent is used to manually adjust the selected medical data and its corresponding preliminary and final correction schemes to obtain the selected medical data and its corresponding expert feedback data, which constitute the training dataset. The incremental update Agent is used to incrementally update the multi-agent evaluation module using a large language model combined with the training dataset.

3. The medical data evaluation and correction system based on a large language model and multi-agent collaboration as described in claim 2, characterized in that, The process of incrementally updating the multi-agent evaluation module using a large language model combined with a training dataset includes: The quality assessment agent and data correction agent are trained based on a horizontal federated learning framework and cryptographic gradient aggregation to obtain updated quality assessment agents and updated data correction agents. The semantic recognition agent is trained based on a natural language processing algorithm to obtain an updated semantic recognition agent; The consensus coordination agent is trained based on a preset reward function and a reinforcement learning model to obtain an updated consensus coordination agent.

4. The medical data evaluation and correction system based on a large language model and multi-agent collaboration as described in claim 1, characterized in that, The process of performing quality analysis on the medical data includes: scoring the medical data in multiple dimensions based on a pre-trained machine learning model and preset judgment rules to obtain quality assessment results.

5. The medical data evaluation and correction system based on a large language model and multi-agent collaboration as described in claim 1, characterized in that, The process of performing semantic analysis on the medical data includes: using a semantic matching algorithm based on a preset clinical medical knowledge graph to identify semantic conflicts in the medical data and generate corresponding semantic recognition results.

6. The medical data evaluation and correction system based on a large language model and multi-agent collaboration as described in claim 1, characterized in that, The process of analyzing the quality assessment results and semantic recognition results includes: analyzing the medical data based on the conditional variational autoencoder to obtain a preliminary correction scheme corresponding to the medical data.

7. A method for medical data evaluation and correction based on a large language model and multi-agent collaboration, characterized in that, include: Acquire medical data generated in the medical information system; The medical data is subjected to quality analysis to obtain quality assessment results; Semantic analysis is performed on the medical data to obtain semantic recognition results; the quality assessment results and semantic recognition results are analyzed to obtain a preliminary correction scheme corresponding to the medical data; a multi-agent negotiation mechanism is used to adjust the preliminary correction scheme to obtain a final correction scheme, and the medical data is corrected based on the final correction scheme to obtain corrected medical data; The multi-agent evaluation module is updated based on the large language model to obtain the updated multi-agent evaluation module.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of claim 7.

9. A computer program product, characterized in that, The computer program product includes computer program code that, when run on a computer, causes the computer to perform the method as described in claim 7.

10. An electronic terminal, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method of claim 7.

Images