Medical term standardization system, method and terminal

Abstract

According to the medical term standardization system and method and the terminal, deep semantic comprehension and feature enhancement are carried out on original non-standard terms through a pre-trained medical term semantic comprehension model, a semantic standardization mapping result is generated, a retrieval model is enhanced through a pre-trained knowledge graph, and the semantic standardization mapping result is obtained. The method comprises the following steps: performing similarity retrieval on original non-standard terms to generate map standardized mapping results, and further performing weighted fusion on the two standardized mapping results through a pre-trained interpretable fusion model to obtain a target standardized mapping result of the original non-standard terms and an interpretable term standardized report. Therefore, the standardization accuracy of original non-standard terms is improved, the reasoning ability of the system is improved, the robustness and practicability of the system in real and complex medical scenes are enhanced, the interpretability and clinical trust of the system are enhanced, and the strict clinical-level application requirements of the medical field for safety and reliability are met.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and in particular to a medical terminology standardization system, method and terminal. Background Technology

[0002] In the era of medical informatics and big data analytics, achieving semantic interoperability of medical data across systems and institutions is fundamental to improving the quality of healthcare services and supporting clinical research and intelligent decision-making. The core of semantic interoperability lies in the standardization of medical terminology, accurately and consistently mapping non-standard terms freely written by clinicians or derived from different vendors' dictionaries in various medical systems such as Hospital Information Systems (HIS), Laboratory Information Systems (LIS), and Electronic Medical Records (EMR) to internationally or domestically recognized standard terminology systems, such as the LOINC medical terminology standard system and the SNOMED CT medical terminology standard system. However, achieving efficient and accurate standardization of medical terminology in a hospital environment still faces significant challenges, and existing technical solutions have many limitations.

[0003] Currently, existing technical solutions for standardizing medical terminology mainly focus on the following three aspects:

[0004] Firstly, early medical terminology standardization techniques primarily relied on rule-based models to obtain standard terms that matched non-standard terminology strings. This approach largely depended on predefined thesaurus databases for string comparison. While adequate for handling regular synonyms like "high blood pressure" and "hypertension," it failed to effectively cover the numerous non-standard expressions, abbreviations, misspellings, and colloquial descriptions present in clinical practice. Consequently, it lacked flexibility and struggled to cope with the complexity and diversity of clinical texts. For example, mapping "heart attack" to "myocardial infarction" requires deeper semantic understanding rather than literal matching. Although subsequent introductions of part-of-speech tagging and N-gram models improved feature extraction capabilities, these techniques remained fundamentally based on literal feature matching, lacking a deep understanding of the clinical context of non-standard terms (such as specific diseases, patient histories, and treatment scenarios), resulting in low standardization accuracy in complex contexts.

[0005] Secondly, existing technical solutions for standardizing medical terminology also include the use of models based on general machine learning or deep learning. For example, deep adversarial learning can be used to generate a candidate set of standard terms for non-standard terms and then reorder them to improve the model's generalization ability; or self-supervised and active learning can be combined to reduce the model's dependence on labeled data. However, deep learning models typically need to be trained on massive amounts of general text, and their inherent "world knowledge" lacks structured and authoritative medical expertise, facing bottlenecks due to a lack of knowledge and interpretability in the medical field. When encountering non-standard terms related to rare diseases, complex comorbidities, or the latest treatments, the model is prone to "illusions," outputting seemingly reasonable but factually incorrect standardized results. More importantly, the decision-making process of such "black box" models is opaque. In rigorous medical scenarios, clinical experts cannot verify their mapping logic, making it difficult to establish trust and hindering practical implementation.

[0006] Finally, in recent years, retrieval-enhanced generative techniques have been introduced into the medical vertical field to combine the generative capabilities of large language models with external knowledge for the standardization of medical terminology. However, standard retrieval-enhanced generative frameworks mainly rely on vector similarity retrieval, which has limitations in the medical field: the retrieval results may be inaccurate, and are limited by the length of the context window, making it difficult to incorporate comprehensive differential diagnostic relationship networks or drug interaction relationship networks. Although the latest medical terminology standardization technology attempts to introduce knowledge graphs into the classic retrieval-enhanced generative framework to optimize the retrieval and understanding process, thereby improving the accuracy of question answering, and achieving advanced performance in medical question answering tasks, this technology has not specifically optimized for the core reasoning task of accurately mapping massive, heterogeneous, and dynamically growing non-standard terms to standard terms in hospitals, resulting in insufficient reasoning ability and interpretability.

[0007] In summary, existing medical terminology standardization technologies still have the following shortcomings, which severely restrict the in-depth mining and application of the value of medical data:

[0008] ① Superficial understanding of the semantics of clinical context, over-reliance on literal features;

[0009] ② The lack of in-depth integration and guidance of authoritative and structured medical knowledge leads to low standardization accuracy and poor reliability;

[0010] ③ It was not specifically designed for terminology mapping reasoning tasks, and its reasoning ability and interpretability do not meet the requirements for clinical applications. Summary of the Invention

[0011] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a medical terminology standardization system, method and terminal to solve the technical problems of low standardization accuracy, poor reliability and insufficient reasoning ability and interpretability of existing medical terminology standardization technologies.

[0012] To achieve the above and other related objectives, a first aspect of this application provides a medical terminology standardization system, comprising: a data acquisition module, used to acquire raw medical text data to be standardized, and to filter one or more raw non-standard terms to be standardized, generating terminology feature data for each raw non-standard term; a semantic understanding module, connected to the data acquisition module, used to perform deep semantic understanding and feature enhancement on each raw non-standard term based on a pre-trained medical terminology semantic understanding model and according to each terminology feature data, generating a semantic standardization mapping result for each raw non-standard term; a graph retrieval module, connected to the semantic understanding module, used to perform similarity retrieval on each raw non-standard term based on a pre-trained knowledge graph enhanced retrieval model and according to each semantic standardization mapping result, generating a graph standardization mapping result for each raw non-standard term; and a decision fusion module, connected to the semantic understanding module and the graph retrieval module, used to generate a target standardization mapping result and an interpretable terminology standardization report for each raw non-standard term based on a pre-trained interpretable fusion model and according to each semantic standardization mapping result and each graph standardization mapping result.

[0013] In some embodiments of the first aspect of this application, the semantic understanding module includes: a context query unit, connected to the data acquisition module and connected to an external patient data center, for obtaining one or more patient medical record records in which each original non-standard term appears from the patient data center based on term feature data, and generating term context information for each original non-standard term based on the department information and record type of each patient medical record record; a formatting unit, connected to the context query unit, for formatting each term context information based on a predefined term structure template to generate term structure data for each original non-standard term; and a semantic understanding unit, connected to the formatting unit, for generating a contextualized semantic vector, one or more first candidate standard terms, and reasoning paths for each first candidate standard term for each original non-standard term based on the medical term semantic understanding model and the term structure data, as the semantic standardization mapping result for each original non-standard term.

[0014] In some embodiments of the first aspect of this application, the medical terminology semantic understanding model is obtained by training a large language model based on a pre-built standard terminology vector database.

[0015] In some embodiments of the first aspect of this application, the graph retrieval module includes: a graph retrieval unit connected to the semantic understanding module, configured to perform graph similarity retrieval on each original non-standard term based on a pre-constructed medical terminology knowledge graph and multiple first candidate standard terms in each semantic standardization mapping result, to obtain a first similar standard term and its reasoning path for each original non-standard term; a first vector retrieval unit connected to the semantic understanding module, configured to perform vector similarity retrieval on each original non-standard term based on a pre-constructed graph node vector database and contextualized semantic vectors in each semantic standardization mapping result, to obtain a second similar standard term and its reasoning path for each original non-standard term; wherein the graph node vector database is constructed based on multiple standard term nodes of the medical terminology knowledge graph; and a similarity comparison unit connected to the graph retrieval unit and the first vector retrieval unit, configured to perform similarity comparison on the first similar standard term and the second similar standard term for each original non-standard term, to obtain a second candidate standard term and its reasoning path for each original non-standard term.

[0016] In some embodiments of the first aspect of this application, the graph retrieval module further includes: a second vector retrieval unit, which is connected to the semantic understanding module and the similarity comparison unit respectively, for performing vector similarity retrieval on each original non-standard term based on a pre-built standard term vector database and according to the contextualized semantic vector in each semantic standardization mapping result, to obtain a third similar standard term for each original non-standard term, so that the similarity comparison unit can perform similarity comparison on the first similar standard term, the second similar standard term, and the third similar standard term for each original non-standard term, to obtain a second candidate standard term and its reasoning path for each original non-standard term.

[0017] In some embodiments of the first aspect of this application, the decision fusion module includes: an inference alignment unit, connected to the semantic understanding module and the graph retrieval module, for aligning multiple first candidate standard terms in each semantic standardization mapping result and second candidate standard terms in each graph standardization mapping result to obtain one or more third candidate standard terms for each original non-standard term and the inference path of each third candidate standard term; a feature extraction unit, connected to the inference alignment unit, for extracting features from each third candidate standard term to obtain semantic understanding features, graph retrieval features, and vector retrieval features of each third candidate standard term; and a confidence arbitration unit, connected to the feature extraction unit, for evaluating the confidence of the semantic understanding features, graph retrieval features, and vector retrieval features of each third candidate standard term based on the interpretable fusion model to obtain the semantic understanding confidence, graph retrieval confidence, vector retrieval confidence, and comprehensive confidence of each third candidate standard term, and obtaining the target standard term for each original non-standard term based on predefined confidence arbitration rules, and generating the target standardization mapping result and interpretable term standardization report for each original non-standard term.

[0018] In some embodiments of the first aspect of this application, the medical terminology standardization system further includes: a data update module connected to the decision fusion module, used to update each original non-standard term in the original medical text data according to the standardization mapping results of each target, and generate standardized medical text data.

[0019] In some embodiments of the first aspect of this application, the medical terminology standardization system further includes: a human-computer interaction module, which is connected to the data acquisition module, the decision fusion module, and the data update module, respectively, for providing a visual human-computer interaction interface to collect the raw medical text data input by the user and send it to the data acquisition module, and to receive the interpretable terminology standardization report generated by the decision fusion module for visual display, and to receive the standardized medical text data generated by the data update module for output to the user.

[0020] To achieve the above and other related objectives, a second aspect of this application provides a method for standardizing medical terminology. The method includes: acquiring original medical text data to be standardized, and selecting one or more original non-standard terms to be standardized, generating terminology feature data for each original non-standard term; based on a pre-trained medical terminology semantic understanding model, performing deep semantic understanding and feature enhancement on each original non-standard term according to the terminology feature data, generating a semantic standardization mapping result for each original non-standard term; based on a pre-trained knowledge graph-enhanced retrieval model, performing similarity retrieval on each original non-standard term according to the contextualized semantic vectors in each semantic standardization mapping result and multiple first candidate standard terms, generating a knowledge graph standardization mapping result for each original non-standard term; and based on a pre-trained interpretable fusion model, generating a target standardization mapping result and an interpretable terminology standardization report for each original non-standard term according to the semantic standardization mapping results and the knowledge graph standardization mapping results.

[0021] To achieve the above and other related objectives, a third aspect of this application provides a medical terminology standardization terminal, the medical terminology standardization terminal comprising: a memory and a processor; the memory for storing a computer program; and the processor for executing the computer program stored in the memory, so that the terminal performs the functions of the medical terminology standardization system described in any of the above embodiments.

[0022] As described above, this application provides a medical terminology standardization system, method, and terminal. It utilizes a pre-trained medical terminology semantic understanding model to perform deep semantic understanding and feature enhancement on original non-standard terms, generating semantic standardization mapping results. Then, it employs a pre-trained knowledge graph-enhanced retrieval model to perform similarity retrieval on the original non-standard terms, generating graph-standardized mapping results. Finally, a pre-trained interpretable fusion model weightedly fuses the two standardization mapping results to obtain the target standardization mapping results for the original non-standard terms and an interpretable terminology standardization report. This application offers the following advantages: it improves the standardization accuracy of original non-standard terms, enhances the system's reasoning ability, strengthens the system's robustness and practicality in real and complex medical scenarios, and further enhances the system's interpretability and clinical trust, meeting the stringent clinical application requirements for safety and reliability in the medical field. This solves the technical problems of low standardization accuracy, poor credibility, and insufficient reasoning ability and interpretability in existing medical terminology standardization technologies. Attached Figure Description

[0023] Figure 1 The diagram shown is a structural schematic of a medical terminology standardization system according to one embodiment of this application.

[0024] Figure 2The diagram shown is a structural schematic of the semantic understanding module in one embodiment of this application.

[0025] Figure 3 The diagram shown is a schematic representation of the map retrieval module in one embodiment of this application.

[0026] Figure 4 The diagram shown is a schematic representation of the map retrieval module in another embodiment of this application.

[0027] Figure 5 The diagram shown is a structural schematic of the decision fusion module in one embodiment of this application.

[0028] Figure 6 The diagram shown is a structural schematic of a medical terminology standardization system according to another embodiment of this application.

[0029] Figure 7 The diagram shown is a flowchart of a medical terminology standardization method according to one embodiment of this application.

[0030] Figure 8 The diagram shown is a structural schematic of a medical terminology standardization terminal in one embodiment of this application. Detailed Implementation

[0031] 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.

[0032] In the embodiments of this application, terms such as "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the terms "first candidate standard term" and "second candidate standard term" are merely used to distinguish different candidate standard terms and do not limit their order. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or execution order, and that terms such as "first" and "second" do not necessarily imply that they are different.

[0033] To address the problems mentioned above, this application provides a medical terminology standardization system, method, and terminal. The system aims to standardize original non-standard terms using a pre-trained medical terminology semantic understanding model and a knowledge graph-enhanced retrieval model, respectively. Furthermore, it fuses the two standardized mapping results using a pre-trained interpretable fusion model to obtain a more accurate target standardization mapping result for the original non-standard terms, as well as an interpretable terminology standardization report. This addresses the existing technical problems of low standardization accuracy, poor reliability, and insufficient reasoning ability and interpretability.

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application are further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application.

[0035] Before providing a further detailed description of this application, the nouns and terms used in the embodiments of this application are explained, and the nouns and terms used in the embodiments of this application shall be interpreted as follows:

[0036] <1> HIS (Hospital Information System): A comprehensive information system covering the main business and management of a hospital. It is mainly used to process information in various aspects of hospital administration, finance, medical care, nursing, and logistics. Its core objective is to achieve the digitization, paperless operation, and automation of hospital information.

[0037] <2> LIS (Laboratory Information System): A professional information management system specifically designed for hospital laboratories, testing laboratories, or third-party testing organizations. It is responsible for handling all data, information, and processes related to laboratory testing work.

[0038] <3> EMR (Electronic Medical Record): A digital record of individual patient medical information created, stored, and used within a hospital.

[0039] <4> LOINC (Logical Observation Identifiers Names and Codes) is a globally recognized identifier system for standardizing clinical and laboratory observation indicators. It provides a unified and clear "identity number" and "standard name" for various observation results generated in medical testing, measurement, and surveys.

[0040] <5> SNOMED CT (Systematized Nomenclature of Medicine – Clinical Terms): A systematized set of clinical medical terms that provides a unique, unambiguous numerical identifier and a standardized term for any concept that may be used in clinical practice (such as disease, symptom, procedure, drug, body structure, etc.).

[0041] <6> N-gram: A sequence model used for text or speech processing.

[0042] <7> NLP (Natural Language Processing): The core goal of natural language processing is to enable computers to understand, interpret, manipulate, and generate human language, thereby achieving effective communication between humans and machines.

[0043] <8> ChatGLM (Chat Generative Language Model): A general language model for dialogue generation.

[0044] <9> BioBERT (Biomedical Bidirectional Encoder Representations from Transformers): A pre-trained language model in the biomedical field.

[0045] <10> BERT (Bidirectional Encoder Representations from Transformers): A bidirectional encoder representation based on Transformers.

[0046] <11> RGA (Retrieval-Augmented Generation): A retrieval-enhanced generation framework, a technique that uses externally retrieved information to enhance text generation.

[0047] <12> API (Application Programming Interface): An application programming interface is a pre-defined service menu and interface method for communication between software programs.

[0048] like Figure 1 The diagram illustrates the structure of a medical terminology standardization system according to an embodiment of this application. The medical terminology standardization system in this embodiment mainly includes: a data acquisition module, a semantic understanding module, a graph retrieval module, and a decision fusion module.

[0049] Specifically, the data acquisition module is used to acquire the original medical text data to be standardized, and to filter one or more original non-standard terms to be standardized, and generate term feature data for each original non-standard term.

[0050] The original medical text data can be a hospital's internal dictionary file. This internal dictionary file is a set of internal coding and standard mapping systems established and maintained by the hospital to standardize data within its information system; it is essentially the hospital's internal standard mapping table. The internal dictionary file can be in CSV format and includes fields such as "original medical terminology," "related department," and "target medical terminology."

[0051] In this embodiment, the method of selecting one or more original non-standard terms from the original medical text data, such as the hospital dictionary file, includes: traversing all original medical terms in the hospital dictionary file and generating a medical term feature vector for each original medical term; based on a pre-built standard term vector database, performing equal-value matching on each medical term feature vector, and taking one or more original medical terms that do not match as original non-standard terms to be standardized.

[0052] It should be noted that the standard terminology vector database is constructed based on standard terms from medical textbooks, academic papers, clinical guidelines, drug instructions, medical literature, and historical standard terminology standardization records.

[0053] In one embodiment, the data acquisition module can also acquire multimodal medical data, including text data, video data, audio data and / or image data, such as medical record text, examination reports, teaching videos, doctor's voice, etc. The multimodal medical data can be converted into medical document data by using automatic speech recognition technology or optical character recognition technology, and then one or more original non-standard terms to be standardized can be screened out.

[0054] Based on the original medical text data, terminology feature data is generated for each original non-standard term. Specifically, the terminology feature data includes, but is not limited to: an original non-standard term and its key features, including, but not limited to: departmental information and / or synonymous non-standard terms.

[0055] It should be understood that medical terms may have different meanings in different application scenarios. Based on the key characteristics of the original non-standard terms, especially departmental information, their application scenarios can be effectively distinguished, which helps to accurately understand and standardize the original non-standard terms.

[0056] like Figure 1As shown, the semantic understanding module is connected to the data acquisition module. The semantic understanding module is used to perform deep semantic understanding and feature enhancement on each original non-standard term based on a pre-trained medical terminology semantic understanding model and the feature data of each term, generating a semantic standardization mapping result for each original non-standard term.

[0057] In one embodiment, such as Figure 2 As shown, the semantic understanding module includes a context query unit, a formatting unit, and a semantic understanding unit connected in sequence. The context query unit is also connected to the data acquisition module and to an external patient data center.

[0058] Specifically, the context query unit is used to obtain one or more patient medical records in which each original non-standard term appears from the patient data center based on the term feature data, and to generate term context information for each original non-standard term based on the department information and record type of each patient medical record.

[0059] The patient data center refers to a single, complete, accurate, and shareable view of a patient's full-cycle health data, which is formed by comprehensively collecting, cleaning, integrating, standardizing, storing, and managing patient-related data scattered in various independent business systems (such as HIS, LIS, EMR, etc.) within the hospital, with the patient's unique identification as the core.

[0060] In other embodiments, the context query unit may also be connected to an external electronic medical record system to obtain one or more patient medical record records in which each original non-standard term appears from the electronic medical record system, and generate terminology context information for each original non-standard term based on the department information and record type of each patient medical record record.

[0061] In one specific embodiment, the context query unit may use a scheduler connected to the patient data center and / or the electronic medical record system via a network to query and obtain the terminology context information for each original non-standard term. For example, based on the department information "Laboratory Department" for the original non-standard term "complete blood count (five categories)," the scheduler initiates a query to the patient data center and / or the electronic medical record system to obtain the patient's medical record for the original non-standard term "complete blood count (five categories)" as "The patient complains of fatigue, and today's 'complete blood count (five categories)' showed elevated white blood cells," thereby obtaining the context information for the original non-standard term "complete blood count (five categories)."

[0062] The terminology context information includes, but is not limited to: an original non-standard term and one or more patient medical records for that original non-standard term, the department information and record type of each patient medical record.

[0063] The formatting unit is used to format the context information of each term based on a predefined term structure template, and generate term structure data for each original non-standard term.

[0064] In one specific embodiment, the formatting unit may also use a scheduler to concatenate the terminology context information of each original non-standard term according to a predefined terminology structure template to generate terminology structure data for each original non-standard term.

[0065] It should be understood that the terminology structure template is used to specify the data format of the terminology structure data, ensuring that the data input to the pre-trained medical terminology semantic understanding model maintains a uniform format. The terminology structure template is customized by the user according to their needs, and this application does not specifically limit it.

[0066] In a preferred embodiment, the terminology structure template is: "[Instruction] Please analyze the medical terminology in the following clinical scenario.\n[Terminology] {Original non-standard term}\n[Department where it appears] {Department information}\n[Context information] {Patient medical record}\n[Record type] {Record type}\nPlease execute: 1. Generate the most likely standard terminology name for this medical term in this scenario; 2. List key synonyms; 3. Briefly explain the reasoning basis."

[0067] The semantic understanding unit is used to generate, based on the medical terminology semantic understanding model and according to the structured data of each term, a contextualized semantic vector for each original non-standard term, one or more first candidate standard terms, and the reasoning path of each first candidate standard term, as the semantic standardization mapping result for each original non-standard term.

[0068] In one embodiment, the medical terminology semantic understanding model is obtained by training a large language model based on a pre-built standard terminology vector database. The model training method mainly includes the following steps.

[0069] ① Introduce the standard terminology vector database and use self-supervised learning to train a large language model, enabling the large language model to learn the language patterns, factual knowledge, and conceptual relationships of standard medical terms, thereby constructing a medical large language model.

[0070] ② Obtain multiple non-standard terms and label each non-standard term with its actual standard term to construct a term training set and a term validation set.

[0071] ③ Input the terminology training set into the medical big language model to perform semantic understanding and standardization processing on each non-standard term in the terminology training set, generate one or more predicted standard terms, compare each predicted standard term with the real standard term of each non-standard term, and use an iterative optimization algorithm to optimize the medical big language model to obtain a converged initial medical terminology semantic understanding model.

[0072] ④ Input each non-standard term and its corresponding real standard term from the term validation set into the initial medical term semantic understanding model, verify the model performance of the initial medical term semantic understanding model, update the model hyperparameters, and obtain an optimized medical term semantic understanding model.

[0073] In a preferred embodiment, the large language model used to train the medical terminology semantic understanding model may be a ChatGLM model, a BioBERT model, or a bidirectional encoder model.

[0074] It should be understood that the ChatGLM model is based on the GLM (General Language Model) architecture, combining autoregressive and autoencoder pre-training techniques. During the pre-training phase, through a "fill-in-the-blank" task, the model learns both contextual understanding and long text generation simultaneously. Furthermore, the ChatGLM model supports various efficient fine-tuning methods, enabling it to be adapted to vertical domains, such as the medical field, at a lower cost, thus making the model's understanding and application of medical terminology more accurate.

[0075] The BioBERT model is a pre-trained language model for the biomedical field, specifically designed and pre-trained for biomedical texts. By pre-training on massive amounts of biomedical literature, the model first learns "biomedical language," and then fine-tunes it according to the model task. This allows the model to internalize the unique language patterns, terminology system, and knowledge connections of the medical field, focusing on deep semantic understanding of standard medical terminology.

[0076] The bidirectional encoder model is an encoder that utilizes a bidirectional attention mechanism to construct a deep contextual understanding model. Simultaneously, an independent rule-based model or a small-scale sequence-to-sequence model can be configured to extract key information from the context of non-standard terms, such as "department information," "patient gender," and "acute / chronic," to assist the bidirectional encoder model in performing deep semantic understanding of non-standard terms and generating predicted standard terms.

[0077] However, it should be noted that the user can choose the type of large language model and the training method used in the medical terminology semantic understanding model according to their needs, and this application does not limit the specifics.

[0078] The trained medical terminology semantic understanding model is used to input the structured data of each original non-standard term, generating one or more first candidate standard terms and their reasoning paths. The generated first candidate standard terms include not only the standard term name but also its unique identifier, such as the LOINC code. The reasoning path of the first candidate standard term includes, but is not limited to: the terminology explanation corresponding to the original non-standard term and the reasoning basis for inferring the first candidate standard term. For example, for the original non-standard term "complete blood count (five categories)," based on the medical terminology semantic understanding model, a first candidate standard term "complete blood cell count (CBC) and five differentials" is generated, along with the reasoning path "In the laboratory requisition form, 'complete blood count (five differentials)' usually refers to a comprehensive item involving counting white blood cells, red blood cells, platelets, and white blood cell typing."

[0079] Furthermore, the hidden states of the last layer of the medical terminology semantic understanding model are extracted, and the obtained feature vectors are used as contextualized semantic vectors for each original non-standard term in the current context. Each contextualized semantic vector is a numerical vector representation of the original non-standard term obtained after modification with terminology context information. It not only includes the semantics of the corresponding original non-standard term itself, but also incorporates contextual information such as patient medical records and treatment scenarios. Compared with general static word vectors or sentence vectors, it has greater clinical discriminative power, which helps to ensure the accuracy and reliability of subsequent standard term retrieval and similarity calculation.

[0080] In one embodiment, the semantic understanding unit can also generate one or more synonymous non-standard terms for each original non-standard term based on the medical terminology semantic understanding model and the structured data of each term, to assist in subsequent standard term retrieval and match more accurate standard terms. For example, for the original non-standard term "blood routine (five categories)", based on the medical terminology semantic understanding model, two synonymous non-standard terms "blood cell analysis" and "blood picture" are generated.

[0081] This application, through the semantic understanding module, queries the terminology context information of the original non-standard terms and utilizes the contextual modeling capability of the medical terminology semantic understanding model to place the original non-standard terms in a specific clinical scenario to understand their true semantics. This overcomes the limitation of traditional medical terminology standardization methods that only compare isolated terminology strings, resulting in literal matching. As a result, it can deeply understand the original non-standard terms and output more accurate and clinically targeted first candidate standard terms and contextualized semantic vectors.

[0082] like Figure 1As shown, the graph retrieval module is connected to the semantic understanding module. The graph retrieval module is used to perform similarity searches on each original non-standard term based on a pre-trained knowledge graph-enhanced retrieval model, according to the semantic standardization mapping results, and to generate a graph standardization mapping result for each original non-standard term.

[0083] Existing technical solutions for standardizing medical terminology also incorporate retrieval-enhanced generation (RGA) technology. It should be understood that the Retrieval-Enhanced Generation (RGA) framework is a technical framework combining information retrieval and text generation. A typical RGA consists of two main stages: indexing and retrieval-generation. The indexing stage involves collecting all relevant documents and segmenting long documents into smaller, semantically complete "text chunks"; using an embedding model to convert each text chunk into a high-dimensional vector; and storing these text chunks and their corresponding vectors in a vector database. The retrieval-generation stage involves converting the user's question into a vector using the embedding model, performing a similarity search in the vector database to find the few text chunks most similar to the question vector; using the retrieved relevant text chunks as "context" or "reference," concatenating them with the user's question to form an enhanced prompt, which is then input into a large language model to generate the final answer.

[0084] However, the RGA model mainly relies on vector similarity retrieval, which has limitations in the medical field: its retrieval results may be inaccurate and are limited by the length of the context window, making it difficult to incorporate comprehensive differential diagnostic relationship networks or drug interaction relationship networks.

[0085] In this embodiment, this application introduces a knowledge graph to optimize the retrieval and understanding process. A knowledge graph-enhanced retrieval model is obtained by training an RGA model based on the knowledge graph. According to the contextualized semantic vector of each original non-standard term output by the semantic understanding module and each first candidate standard term, a similarity search is performed on each original non-standard term to generate a graph-normalized mapping result for each original non-standard term.

[0086] The graph standardization mapping result includes: the second candidate standard term of the original non-standard term and the reasoning path of the second candidate standard term.

[0087] It should be noted that the specific model structure and training method of the knowledge graph enhanced retrieval model can be selected by the user according to their needs, and this application does not limit them.

[0088] In one embodiment, such as Figure 3 As shown, the graph retrieval module includes a graph retrieval unit, a first vector retrieval unit, and a similarity comparison unit. The graph retrieval unit and the first vector retrieval unit are respectively connected to the semantic understanding module and respectively connected to the similarity comparison unit.

[0089] Specifically, the graph retrieval unit is used to perform graph similarity retrieval on each original non-standard term based on a pre-constructed medical terminology knowledge graph and multiple first candidate standard terms in each semantic standardization mapping result, to obtain the first similar standard term and its reasoning path for each original non-standard term.

[0090] In one embodiment, the medical terminology knowledge graph is constructed based on a medical terminology standard system (such as the LOINC medical terminology standard system and / or the SNOMED CT medical terminology standard system), local diagnosis and treatment guidelines, and historical standard terminology standardization records. This forms a structured relationship network covering core medical entities such as diseases, symptoms, tests, drugs, and surgeries, and is stored in a graph database. This effectively utilizes and integrates multi-source heterogeneous medical knowledge, providing structured knowledge support for medical terminology standardization and improving the knowledge coverage and reasoning depth of the medical terminology standardization system.

[0091] In this context, the nodes of the medical terminology knowledge graph represent standard terminology concepts, and the attributes of these standard terminology concepts include, but are not limited to, the standard terminology name, the unique identifier of the standard terminology (such as the LOINC code), and the terminology explanation; the edges of the medical terminology knowledge graph represent the relationships between standard terminology concepts, such as belonging relationships, synonym relationships, measurement relationships, and inclusion relationships.

[0092] In this embodiment, the method of performing graph similarity retrieval on the original non-standard terms based on the medical terminology knowledge graph includes: using each of the first candidate standard terms of the original non-standard terms output by the semantic understanding module as query terms, searching for one or more standard term nodes with string matching in the medical terminology knowledge graph, or searching for one or more standard term nodes connected by edges with synonym relationships; selecting the standard term node with the highest similarity as the first similar standard term of the original non-standard terms, and generating its reasoning path.

[0093] In one embodiment, the method of performing graph similarity retrieval on the original non-standard terms based on the medical terminology knowledge graph further includes: using each synonymous non-standard term of the original non-standard term output by the semantic understanding module as a query term, searching for one or more standard term nodes with string matching in the medical terminology knowledge graph, or searching for one or more standard term nodes connected by edges of synonym relationships; combining each standard term node obtained with each first candidate standard term as a query term, selecting the standard term node with the highest similarity as the first similar standard term of the original non-standard term, and generating its reasoning path.

[0094] For example, for the original non-standard term "complete blood count (five categories)," the first candidate standard term "complete blood cell count (CBC) and five-part differential" output by the medical terminology semantic understanding model, along with the synonymous non-standard terms "blood cell analysis" and "blood picture," can be used as query terms. The system searches the medical terminology knowledge graph for one or more standard term nodes that match the strings or are connected by edges with synonym relationships. The standard term node with the highest similarity, "LOINC57021-8 Complete Blood Cell Count (CBC) and Automatic Blood Segmentation," is selected as the first similar standard term for the original non-standard term "complete blood count (five categories)." The inclusion relationship edges of this standard term node also connect to multiple sub-standard term nodes, including: white blood cells, red blood cells, platelet count, etc.

[0095] The first vector retrieval unit is used to perform vector similarity retrieval on each original non-standard term based on a pre-built graph node vector database and according to the contextualized semantic vectors in each semantic standardization mapping result, to obtain the second similar standard term and its reasoning path for each original non-standard term.

[0096] In this embodiment, the graph node vector database is constructed based on multiple standard terminology nodes of the medical terminology knowledge graph. Specifically, graph node vectors are generated for each standard terminology node in the medical terminology knowledge graph, and the graph node vector database is constructed based on each graph node vector.

[0097] In one embodiment, the method of performing vector similarity retrieval on the original non-standard terms based on the graph node vector database includes: performing an approximate nearest neighbor search on the contextualized semantic vector of the original non-standard terms output by the semantic understanding module and each graph node vector in the graph node vector database to obtain one or more graph node vectors with vector matching, and selecting the standard term node corresponding to the graph node vector with the highest similarity as the second similar standard term of the original non-standard terms, while generating its reasoning path.

[0098] The similarity comparison unit is used to compare the first similar standard term and the second similar standard term for each original non-standard term to obtain the second candidate standard term and its reasoning path for each original non-standard term.

[0099] This application, through the knowledge graph-enhanced retrieval model, achieves dual-path retrieval of graph retrieval and vector retrieval. It not only introduces knowledge graphs into the retrieval enhancement generation framework, solving the technical problem of the lack of authoritative medical knowledge in general RGA models and optimizing the retrieval and understanding process of RGA models, but also generates reasoning paths. Through the explicit relational paths of knowledge graphs, it improves the accuracy and interpretability of medical terminology standardization, providing authoritative and visualized evidence chains for graph standardization mapping results, thereby solving the problem of opaque reasoning processes in existing technologies.

[0100] In one embodiment, the dual-path retrieval based on the knowledge graph-enhanced retrieval model can also be used in parallel with retrieval from a pre-built standard term vector database. Specifically, such as... Figure 4 As shown, the graph retrieval module further includes a second vector retrieval unit. This second vector retrieval unit, connected to both the semantic understanding module and the similarity comparison unit, is used to perform vector similarity retrieval on each original non-standard term based on a pre-constructed standard term vector database and the contextualized semantic vectors in each semantic standardization mapping result. This yields a third similar standard term for each original non-standard term, which is then used by the similarity comparison unit to compare the first, second, and third similar standard terms for each original non-standard term, obtaining a second candidate standard term and its reasoning path for each original non-standard term. Thus, after vector retrieval based on graph node vectors, it further retrieves semantically similar but potentially not yet included standard term description fragments from the latest literature from the standard term vector database as supplementary evidence to obtain a second candidate standard term with higher similarity.

[0101] In a preferred embodiment, the knowledge graph retrieval module can also output the standard term nodes matched by each retrieval unit, their similarity, and reasoning path, and generate retrieval logs for users to view and review. The review results can also be used to train and adjust the knowledge graph-enhanced retrieval model, thereby improving the model's reasoning ability and enabling the model to more accurately reason for the second candidate standard term for each original non-standard term.

[0102] like Figure 1 As shown, the decision fusion module is connected to both the semantic understanding module and the graph retrieval module. The decision fusion module is used to generate a target standardization mapping result and an interpretable term standardization report for each original non-standard term based on a pre-trained interpretable fusion model, according to the semantic standardization mapping results and the graph standardization mapping results.

[0103] In one embodiment, such as Figure 5As shown, the decision fusion module includes: an inference alignment unit, a feature extraction unit, and a confidence arbitration unit connected in sequence. Wherein, as... Figure 5 As shown, the inference alignment unit is also connected to the semantic understanding module and the graph retrieval module.

[0104] The inference alignment unit is used to align multiple first candidate standard terms in each semantic standardization mapping result and second candidate standard terms in each graph standardization mapping result to obtain one or more third candidate standard terms for each original non-standard term and the inference path of each third candidate standard term. That is, duplicate candidate standard terms are removed, and the remaining multiple different candidate standard terms are respectively used as third candidate standard terms.

[0105] The feature extraction unit is used to extract features from each third candidate standard term to obtain semantic understanding features, graph retrieval features, and vector retrieval features for each third candidate standard term.

[0106] The semantic understanding features include: the probability value or confidence score of the third candidate standard term output by the medical terminology semantic understanding model; the graph retrieval features include: the number of paths retrieved by the third candidate standard term in the medical terminology knowledge graph, the path type weight, and the core degree of the standard term node itself; the vector retrieval features include: the cosine similarity between the contextualized semantic vector of the third candidate standard term and the graph node vector and the standard term vector.

[0107] The confidence arbitration unit is used to evaluate the confidence of the semantic understanding features, graph retrieval features, and vector retrieval features of each third candidate standard term based on the interpretable fusion model, and obtain the semantic understanding confidence, graph retrieval confidence, vector retrieval confidence, and comprehensive confidence of each third candidate standard term. Based on the predefined confidence arbitration rules, the unit obtains the target standard term for each original non-standard term and generates the target standardization mapping result and interpretable term standardization report for each original non-standard term.

[0108] In one embodiment, the interpretable fusion model can be obtained by training an interpretable ensemble learning model. It should be understood that an ensemble learning model refers to constructing a powerful "strong learner" by combining the predictions of multiple "weak learners," such as random forests or gradient boosting trees. This inevitably increases the complexity of the model, making it difficult to interpret. The interpretable fusion model employs ensemble learning based on interpretable base learners or uses ex-post interpretation techniques, thereby effectively explaining the model's prediction logic. This allows the model to maintain the high performance of ensemble learning while providing clinical experts with reliable and understandable decision-making support.

[0109] In another embodiment, the interpretable fusion model can also be trained based on a credibility arbitrator. It should be understood that the credibility arbitrator is used to perform a multi-dimensional, quantifiable credibility assessment of the raw output generated by the intelligent model, and to determine, based on the assessment results, whether to deliver directly, correct, or issue a warning.

[0110] Based on the interpretable fusion model, weights are dynamically assigned to the semantic understanding features, graph retrieval features, and vector retrieval features of each third candidate standard term. The dual evidence from deep semantic understanding and structured knowledge reasoning is weighted and fused to achieve confidence evaluation of the semantic understanding features, graph retrieval features, and vector retrieval features of the third candidate standard term. The semantic understanding confidence, graph retrieval confidence, vector retrieval confidence, and comprehensive confidence of each third candidate standard term are output. Based on predefined confidence arbitration rules, the target standard term for each original non-standard term is obtained.

[0111] In one embodiment, the confidence level arbitration rule includes: preferentially selecting the third candidate standard term with the highest confidence level in graph retrieval as the target standard term; however, if the confidence level of the graph retrieval is lower than a preset confidence level threshold, selecting the third candidate standard term with the highest overall confidence level as the target standard term. It should be noted that the confidence level arbitration rule can be defined by the user according to their needs, and this application does not specifically limit it.

[0112] Based on the confidence arbitration rule, the optimal third candidate standard term is selected as the target standard term for each original non-standard term, and the target standardization mapping result and interpretable term standardization report for each original non-standard term are generated according to the reasoning path of the third candidate standard term.

[0113] The target standardization mapping result includes: the target standard terminology of the original non-standard terminology and the reasoning path of the target standard terminology.

[0114] The interpretable terminology standardization report includes: the original non-standard terms and their semantic standardization mapping results, graph standardization mapping results, and target standardization mapping results. This visualizes the complete reasoning path from the original non-standard terms to the target standard terms and the confidence level of each candidate standard term in the reasoning path. This enhances the interpretability of the medical terminology standardization system, enabling the establishment of trust with clinical experts and facilitating its widespread adoption. Simultaneously, the interpretable terminology standardization report also facilitates the identification and correction of potential knowledge gaps in the system, providing a transparent channel for continuous system optimization and meeting the stringent clinical application requirements for safety and reliability in the medical field.

[0115] In a preferred embodiment, the decision fusion module can also output the similarity, reasoning path, semantic understanding features, graph retrieval features, vector retrieval features, semantic understanding confidence, graph retrieval confidence, vector retrieval confidence, and comprehensive confidence of each third candidate standard term and its corresponding original non-standard term, and generate an arbitration log for users to view and review. The review results can also be used to train and adjust the medical terminology semantic understanding model, the knowledge graph enhanced retrieval model, and the interpretable fusion model, and to optimize the confidence arbitration rules, thereby improving the reasoning ability of the medical terminology standardization system and enabling the system to make optimal and reliable medical terminology standardization decisions.

[0116] In one embodiment, such as Figure 6 As shown, the medical terminology standardization system also includes a data update module that connects to the decision fusion module.

[0117] Specifically, the data update module, connected to the decision fusion module, is used to update each original non-standard term in the original medical text data according to the standardization mapping results of each target, and generate standardized medical text data.

[0118] For example, if the original medical text data acquired by the data acquisition module is a hospital's internal dictionary file, the data update module can update the "target medical term" field corresponding to each original non-standard term in the internal dictionary file according to the target standard term of one or more original non-standard terms to be standardized in the internal dictionary file, thereby obtaining the updated internal dictionary file and realizing the maintenance of the internal dictionary file.

[0119] In one embodiment, such as Figure 6 As shown, the medical terminology standardization system further includes a human-computer interaction module that connects the data acquisition module, the decision fusion module, and the data update module.

[0120] Specifically, the human-computer interaction module is used to provide a visual human-computer interaction interface to collect the raw medical text data input by the user and send it to the data acquisition module, and to receive the interpretable terminology standardization report generated by the decision fusion module for visual display, and to receive the standardized medical text data generated by the data update module for output to the user.

[0121] In a specific embodiment, the human-computer interaction module may adopt an interactive WEB page. On this interactive WEB page, users can upload the original medical text data and specify to select one or more original non-standard terms to generate a medical term standardization request and send it to the data acquisition module. Preferably, the medical term standardization request can be sent to the data acquisition module through an API grid. The medical term standardization request may adopt the JSON format, including but not limited to: one or more original non-standard terms, as well as the department information and / or synonymous non-standard terms of each original non-standard term.

[0122] On this interactive WEB page, the target standard term of each original non-standard term and the unique identification code of this target standard term, such as the LOINC code, can also be clearly displayed, and an expandable "view basis" area is provided to display the interpretable term standardization report of each original non-standard term, so as to provide users with accurate, credible and transparent medical term standardization results.

[0123] This interactive WEB page can also display the standardized medical text data and provide a download entry.

[0124] For traditional medical term standardization technical solutions based on string matching or general vector similarity, when dealing with abbreviations, common sayings, typos and ambiguous original non-standard terms, the standardization accuracy rate is usually lower than 70%. Through the semantic understanding module of this application, the original non-standard terms are placed in rich contexts such as specific patient medical record records and department information for in-depth semantic understanding, significantly improving the discrimination of semantic representations. This application also combines the graph retrieval module. Based on the knowledge graph enhanced retrieval model, it can effectively distinguish homonymous standard terms, such as "glucose screening" (diabetes screening) and "Down syndrome screening", etc. Through simulation tests, the standardization accuracy rate of this application for emergency medical records and pathology reports containing a large number of non-standard expressions can be stably increased to over 92%, an increase of more than 20 percentage points compared with traditional medical term standardization technical solutions, greatly enhancing the robustness and practicability of the system in real and complex medical scenarios.

[0125] Moreover, when encountering rare diseases or newly emerging non-standard terms, the system can not only make reasonable inferences through the generalization ability of the medical term semantic understanding model, but more importantly, through the multi-hop relationship reasoning of the knowledge graph enhanced retrieval model, and associate with known standard term concepts along paths such as "etiology - symptoms - examinations", so as to make well-grounded inferences. This way of combining "memory (knowledge graph) and reasoning (large language model)" significantly reduces the probability of generating "hallucinations" in the professional medical field compared with methods that simply rely on model parameter memory or shallow retrieval, and improves the processing ability for long-tail terms.

[0126] It should 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 or functional module, exist as separate physical entities, or be divided into more functional modules. The integrated modules or units described above can be implemented in hardware or as software functional modules.

[0127] like Figure 7 The diagram illustrates a flowchart of a medical terminology standardization method according to an embodiment of this application. The medical terminology standardization method in this embodiment mainly includes the following steps.

[0128] Step S1: Obtain the original medical text data to be standardized, and filter one or more original non-standard terms to be standardized, and generate term feature data for each original non-standard term.

[0129] Step S2: Based on the pre-trained medical terminology semantic understanding model, perform deep semantic understanding and feature enhancement on each original non-standard term according to the feature data of each term, and generate the semantic standardization mapping result of each original non-standard term.

[0130] Step S3: Based on the pre-trained knowledge graph-enhanced retrieval model, according to the contextualized semantic vectors in each semantic standardization mapping result and multiple first candidate standard terms, perform similarity retrieval on each original non-standard term and generate the graph standardization mapping result for each original non-standard term.

[0131] Step S4: Based on the pre-trained interpretable fusion model, generate the target standardization mapping result and the interpretable term standardization report for each original non-standard term according to the semantic standardization mapping results and the graph standardization mapping results.

[0132] It should be understood that the medical terminology standardization method embodiment and the medical terminology standardization system embodiment provided in the above embodiments belong to the same concept. For details of each step, please refer to the visual method of each functional module in the system embodiment, which will not be repeated here.

[0133] Figure 8 This is a schematic diagram of the structure of the medical terminology standardization terminal 800 provided in this application embodiment. For example... Figure 8As shown, the medical terminology standardization terminal 800 includes: at least one processor 801, a memory 802, at least one network interface 803, and a user interface 805. The various components in the terminal are coupled together via a bus system 804. It is understood that the bus system 804 is used to implement communication between these components. In addition to a data bus, the bus system 804 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 8 The general will label all buses as bus systems.

[0134] The user interface 805 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0135] It is understood that memory 802 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 application are intended to include, but are not limited to, these and any other suitable categories of memory.

[0136] In this embodiment, the memory 802 is used to store various types of data to support the operation of the medical terminology standardization terminal 800. Examples of this data include: any executable program for operation on the medical terminology standardization terminal 800, such as operating system 8021 and application program 8022; operating system 8021 includes various system programs, such as framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. Application program 8022 may include various applications, such as media player, browser, etc., for implementing various application services. The medical terminology standardization method provided in this embodiment can be included in application program 8022.

[0137] The medical terminology standardization method disclosed in the above embodiments of this application can be applied to or implemented by the processor 801. The processor 801 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the medical terminology standardization method can be completed by the integrated logic circuits in the hardware of the processor 801 or by instructions in software form. The processor 801 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. The processor 801 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor 801 may be a microprocessor or any conventional processor, etc. The steps of the medical terminology standardization method provided in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in a memory. The processor reads information from the memory and, in conjunction with its hardware, completes the steps of the aforementioned method.

[0138] In an exemplary embodiment, the medical terminology standardization terminal 800 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to perform the aforementioned method.

[0139] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented using computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0140] In the embodiments provided in this application, the computer-readable and writable storage medium may include read-only memory, random access memory, EEPROM, CD-ROM or other optical disc storage devices, disk storage devices or other magnetic storage devices, flash memory, USB flash drive, portable hard drive, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a computer. Additionally, any connection may be appropriately referred to as a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. However, it should be understood that computer-readable and writable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but are intended for non-transient, tangible storage media. The disks and optical discs used in the application include compact optical discs (CDs), laser optical discs, optical discs, digital multifunction optical discs (DVDs), floppy disks, and Blu-ray discs, where disks typically copy data magnetically, while optical discs use lasers to copy data optically.

[0141] In summary, this application provides a medical terminology standardization system, method, and terminal. Through a pre-trained medical terminology semantic understanding model, it performs deep semantic understanding and feature enhancement on original non-standard terms, generating semantic standardization mapping results. Then, through a pre-trained knowledge graph-enhanced retrieval model, it performs similarity retrieval on the original non-standard terms, generating graph-standardized mapping results. Finally, through a pre-trained interpretable fusion model, it weightedly fuses the two standardization mapping results to obtain the target standardization mapping results for the original non-standard terms and an interpretable terminology standardization report. This application has the following beneficial effects: improving the standardization accuracy of original non-standard terms, enhancing the system's reasoning ability, strengthening the system's robustness and practicality in real and complex medical scenarios, further enhancing the system's interpretability and clinical trust, meeting the stringent clinical application requirements for safety and reliability in the medical field, thereby solving the technical problems of low standardization accuracy, poor credibility, and insufficient reasoning ability and interpretability in existing medical terminology standardization technologies.

[0142] Therefore, this application effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0143] 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 terminology standardization system, characterized in that, include: The data acquisition module is used to acquire the original medical text data to be standardized, and to filter one or more original non-standard terms to be standardized, and generate term feature data for each original non-standard term. The semantic understanding module, connected to the data acquisition module, is used to perform deep semantic understanding and feature enhancement on each original non-standard term based on the pre-trained medical term semantic understanding model and the feature data of each term, and generate the semantic standardization mapping result of each original non-standard term. The graph retrieval module, connected to the semantic understanding module, is used to enhance the retrieval model based on the pre-trained knowledge graph. According to the semantic standardization mapping results, it performs similarity retrieval on each original non-standard term and generates the graph standardization mapping result for each original non-standard term. The decision fusion module, which is connected to the semantic understanding module and the graph retrieval module respectively, is used to generate the target standardization mapping result and the interpretable term standardization report for each original non-standard term based on the pre-trained interpretable fusion model and the semantic standardization mapping results and the graph standardization mapping results.

2. The medical terminology standardization system according to claim 1, characterized in that, The semantic understanding module includes: The context query unit is connected to the data acquisition module and to an external patient data center. It is used to obtain one or more patient medical records in which each original non-standard term appears from the patient data center based on the term feature data, and to generate term context information for each original non-standard term based on the department information and record type of each patient medical record. The formatting unit, connected to the context query unit, is used to format the context information of each term based on a predefined term structure template, and generate term structure data for each original non-standard term. The semantic understanding unit, connected to the formatting unit, is used to generate, based on the medical terminology semantic understanding model and according to the structured data of each term, a contextualized semantic vector for each original non-standard term, one or more first candidate standard terms, and the reasoning path of each first candidate standard term, as the semantic standardization mapping result for each original non-standard term.

3. The medical terminology standardization system according to claim 1, characterized in that, The medical terminology semantic understanding model is obtained by training a large language model based on a pre-built standard terminology vector database.

4. The medical terminology standardization system according to claim 1, characterized in that, The map retrieval module includes: The graph retrieval unit, connected to the semantic understanding module, is used to perform graph similarity retrieval on each original non-standard term based on a pre-built medical terminology knowledge graph and multiple first candidate standard terms in each semantic standardization mapping result, to obtain the first similar standard term and its reasoning path for each original non-standard term. The first vector retrieval unit, connected to the semantic understanding module, is used to perform vector similarity retrieval on each original non-standard term based on a pre-built graph node vector database and according to the contextualized semantic vectors in each semantic standardization mapping result, to obtain the second similar standard term and its reasoning path for each original non-standard term; wherein, the graph node vector database is constructed based on multiple standard term nodes of the medical terminology knowledge graph; The similarity comparison unit is connected to the graph retrieval unit and the first vector retrieval unit, respectively, and is used to compare the first similar standard term and the second similar standard term for each original non-standard term to obtain the second candidate standard term and its reasoning path for each original non-standard term.

5. The medical terminology standardization system according to claim 4, characterized in that, The map retrieval module also includes: The second vector retrieval unit, connected to the semantic understanding module and the similarity comparison unit, is used to perform vector similarity retrieval on each original non-standard term based on a pre-built standard term vector database and the contextualized semantic vectors in each semantic standardization mapping result, to obtain the third similar standard term for each original non-standard term. The similarity comparison unit then performs similarity comparison on the first, second, and third similar standard terms of each original non-standard term to obtain the second candidate standard term and its reasoning path for each original non-standard term.

6. The medical terminology standardization system according to claim 1, characterized in that, The decision fusion module includes: The reasoning alignment unit is connected to the semantic understanding module and the graph retrieval module respectively. It is used to align multiple first candidate standard terms in each semantic standardization mapping result and second candidate standard terms in each graph standardization mapping result to obtain one or more third candidate standard terms for each original non-standard term and the reasoning path of each third candidate standard term. The feature extraction unit, connected to the inference alignment unit, is used to extract features from each third candidate standard term to obtain semantic understanding features, graph retrieval features, and vector retrieval features for each third candidate standard term. The confidence arbitration unit, connected to the feature extraction unit, is used to evaluate the confidence of the semantic understanding features, graph retrieval features, and vector retrieval features of each third candidate standard term based on the interpretable fusion model. It obtains the semantic understanding confidence, graph retrieval confidence, vector retrieval confidence, and comprehensive confidence of each third candidate standard term. Based on predefined confidence arbitration rules, it obtains the target standard term for each original non-standard term and generates the target standardization mapping result and interpretable term standardization report for each original non-standard term.

7. The medical terminology standardization system according to claim 1, characterized in that, Also includes: The data update module, connected to the decision fusion module, is used to update the original non-standard terms in the original medical text data according to the standardization mapping results of each target, and generate standardized medical text data.

8. The medical terminology standardization system according to claim 7, characterized in that, Also includes: The human-computer interaction module is connected to the data acquisition module, the decision fusion module, and the data update module, respectively. It is used to provide a visual human-computer interaction interface to collect the raw medical text data input by the user and send it to the data acquisition module, receive the interpretable terminology standardization report generated by the decision fusion module for visual display, and receive the standardized medical text data generated by the data update module for output to the user.

9. A method for standardizing medical terminology, characterized in that, include: Obtain the original medical text data to be standardized, and filter one or more original non-standard terms to be standardized, and generate term feature data for each original non-standard term. Based on a pre-trained medical terminology semantic understanding model, deep semantic understanding and feature enhancement are performed on each original non-standard term according to the feature data of each term, generating a semantic standardization mapping result for each original non-standard term. Based on the pre-trained knowledge graph augmented retrieval model, the similarity retrieval is performed on each original non-standard term according to the contextualized semantic vector in each semantic normalization mapping result and multiple first candidate standard terms, generating the graph normalization mapping result for each original non-standard term; Based on the pre-trained interpretable fusion model, the target standardization mapping result and the interpretable term standardization report for each original non-standard term are generated according to the semantic standardization mapping results and the graph standardization mapping results.

10. A medical terminology standardization terminal, characterized in that, include: Memory and processor; The memory is used to store computer programs; The processor is used to execute the computer program stored in the memory to enable the terminal to perform the functions of the medical terminology standardization system as described in any one of claims 1 to 8.

Images