A method and apparatus for matching medical terms

Abstract

Embodiments of the present disclosure provide a method and device for matching medical terms. The method comprises: obtaining a medical term to be matched included in clinical text; obtaining a plurality of standard medical term clusters, wherein each of the plurality of standard medical term clusters has a representative semantic vector, and wherein the representative semantic vector is a semantic vector having an average value of a plurality of first semantic vectors corresponding to the standard medical term cluster; comparing the medical term to be matched with the representative semantic vector to determine a candidate medical term cluster from the plurality of standard medical term clusters; sending, to a large language model, a standard medical term included in the candidate medical term cluster, the medical term to be matched, and a prompt word; receiving a matching result from the large language model; and determining a target standard medical term corresponding to the medical term to be matched based on the matching result.

Description

Technical Field

[0001] This disclosure relates to the field of data processing, and more specifically, to a method and apparatus for matching medical terms. Background Technology

[0002] To standardize medical data and enhance its utilization value, non-standard medical terminology needs to be normalized. For example, in clinical practice and research, the same clinical entity often has multiple expressions; for instance, "lung cancer," "lung cancer," and "malignant lung tumor" can all refer to the same concept. Failure to normalize these medical terms will lead to data fragmentation and information silos, severely impacting subsequent applications. For instance, in clinical decision-making and multi-center collaboration, inconsistent medical terminology can cause misunderstandings, hindering accurate communication and information sharing. Furthermore, in research data analysis, scattered and differentiated medical terminology makes subsequent statistical analysis, data mining, and correlation analysis difficult, significantly limiting the depth and breadth of disease trend analysis and efficacy comparisons. Moreover, when medical information systems are interconnected, mismatched medical terminology can become an obstacle to data exchange and integration.

[0003] Therefore, a method is desired that can match or map raw medical terms in clinical texts to standard medical terms in order to achieve more efficient and accurate terminology normalization. Summary of the Invention

[0004] Embodiments of this disclosure provide a method and apparatus for matching medical terms.

[0005] Embodiments of this disclosure provide a method for matching medical terms, comprising: processing clinical text to obtain medical terms to be matched included in the clinical text; obtaining multiple standard medical term clusters, wherein the multiple standard medical term clusters are determined by clustering multiple standard medical terms included in a standard medical term set, and each of the multiple standard medical term clusters has a representative semantic vector, wherein the representative semantic vector is the centroid semantic vector of the standard medical term cluster corresponding to the representative semantic vector, and the centroid semantic vector is obtained by: vectorizing multiple standard medical terms included in the standard medical term set to obtain multiple first semantic vectors, and determining the semantic vector having the average value of the multiple first semantic vectors corresponding to the multiple standard medical terms included in the corresponding standard medical term cluster as the representative semantic vector. Semantic vectors; comparing the medical term to be matched with the representative semantic vectors of each of the plurality of standard medical term clusters, to determine a first predetermined number of standard medical term clusters from the plurality of standard medical term clusters whose similarity to the medical term to be matched with the plurality of representative semantic vectors of the plurality of standard medical term clusters is the highest as candidate medical term clusters; sending the standard medical terms included in the candidate medical term clusters, the medical term to be matched, and prompt words to the large language model, wherein the prompt words instruct the large language model to select candidate standard medical terms corresponding to the medical term to be matched from the standard medical terms included in the candidate medical term clusters; receiving from the large language model a matching result including the candidate standard medical terms corresponding to the medical term to be matched; and based on the matching result, determining the target standard medical term corresponding to the medical term to be matched.

[0006] According to the matching method of embodiments of the present disclosure, the step of processing clinical text to obtain medical terms to be matched included in the clinical text includes: processing the clinical text using a transformer-based bidirectional encoder representation model to obtain medical terms to be matched included in the clinical text.

[0007] According to the matching method of embodiments of the present disclosure, the transformer-based bidirectional encoder representation model is trained based on previous clinical text and the standard medical terminology set, and the medical terms to be matched included in the previous clinical text are labeled with corresponding tags.

[0008] According to the matching method of embodiments of the present disclosure, the medical term to be matched has a first label among a plurality of labels; and the clustering of the plurality of standard medical terms included in the standard medical term set includes: clustering the standard medical terms included in the standard medical term set that have the first label to determine the plurality of standard medical term clusters.

[0009] According to the matching method of the embodiments of this disclosure, the clustering of multiple standard medical terms included in the standard medical terminology set includes: clustering the multiple first semantic vectors to obtain the multiple standard medical term clusters.

[0010] According to the matching method of the embodiments of this disclosure, the step of comparing the medical term to be matched with representative semantic vectors of a plurality of standard medical term clusters to determine candidate medical term clusters from the plurality of standard medical term clusters includes: vectorizing the medical term to be matched to obtain a second semantic vector; determining a plurality of representative semantic vectors corresponding to the plurality of standard medical term clusters based on the plurality of first semantic vectors; determining a plurality of first similarities between the second semantic vector and the plurality of representative semantic vectors; and determining the standard medical term cluster corresponding to the representative semantic vector having the highest first predetermined number of first similarities as the candidate medical term cluster.

[0011] According to the matching method of embodiments of the present disclosure, the vectorization of a plurality of standard medical terms included in the standard medical terminology set and the vectorization of the medical term to be matched are performed by a transformer-based bidirectional encoder representation model.

[0012] According to the matching method of the embodiments of this disclosure, the clustering of the plurality of first semantic vectors is performed using the K-means method.

[0013] According to the matching method of embodiments of the present disclosure, the large language model is trained based on previous clinical texts and the standard medical terminology set, and the medical terms to be matched included in the previous clinical texts are labeled with corresponding target standard medical terms.

[0014] The matching method according to embodiments of the present disclosure further includes: determining a plurality of second similarities between the medical term to be matched and standard medical terms included in the candidate medical term cluster; and sending the plurality of second similarities to a large language model.

[0015] The matching method according to embodiments of the present disclosure further includes: determining a context associated with the medical term to be matched, wherein the context associated with the medical term to be matched includes at least one of a first predetermined length of text in the clinical text preceding the medical term to be matched and a second predetermined length of text in the clinical text following the medical term to be matched; and sending the context associated with the medical term to be matched to a large language model.

[0016] According to the matching method of embodiments of this disclosure, the matching result further includes the reasoning of the large language model in selecting the candidate standard medical terms from the candidate medical term cluster.

[0017] According to the matching method of embodiments of the present disclosure, determining the target standard medical term based on the matching result includes: receiving a confirmation result for the candidate standard medical term, wherein the confirmation result indicates at least one of accepting the candidate standard medical term, rejecting the candidate standard medical term, and modifying the candidate standard medical term; and determining the target standard term based on the confirmation result and the candidate standard medical term.

[0018] According to the matching method of embodiments of the present disclosure, wherein obtaining the medical terms to be matched included in the clinical text is performed by a transformer-based bidirectional encoder representation model, and the method further includes: in response to the confirmation result indicating acceptance of the candidate standard medical terms, retraining the transformer-based bidirectional encoder representation model using the accepted candidate standard medical terms and the medical terms to be matched.

[0019] According to the matching method of an embodiment of the present disclosure, the method further includes: modifying the prompt word in response to receiving a second predetermined number of indications to reject the candidate standard medical terms and modifying at least one of the candidate standard medical terms in the confirmation result for the medical term to be matched.

[0020] Embodiments of this disclosure provide a matching apparatus for medical terms, comprising: a text processing module configured to process clinical text to obtain medical terms to be matched included in the clinical text; and an obtaining module configured to obtain multiple standard medical term clusters, wherein the multiple standard medical term clusters are determined by clustering multiple standard medical terms included in a standard medical term set, and each of the multiple standard medical term clusters has a representative semantic vector, wherein the representative semantic vector is the centroid semantic vector of the standard medical term cluster corresponding to the representative semantic vector, and the centroid semantic vector is obtained by: vectorizing multiple standard medical terms included in the standard medical term set to obtain multiple first semantic vectors, and determining the representative semantic vector as the semantic vector having the average value of the multiple first semantic vectors corresponding to the multiple standard medical terms included in the corresponding standard medical term cluster; and comparing the modules. The system comprises: a module configured to compare the medical term to be matched with representative semantic vectors of each of the plurality of standard medical term clusters, and to determine a first predetermined number of standard medical term clusters from the plurality of standard medical term clusters that have the highest similarity between the medical term to be matched and the plurality of representative semantic vectors of the plurality of standard medical term clusters as candidate medical term clusters; a sending module configured to send the standard medical terms included in the candidate medical term clusters, the medical term to be matched, and a prompt word to the large language model, wherein the prompt word instructs the large language model to select a candidate standard medical term corresponding to the medical term to be matched from the standard medical terms included in the candidate medical term clusters; a receiving module configured to receive from the large language model a matching result including the candidate standard medical terms corresponding to the medical term to be matched; and a determining module configured to determine a target standard medical term corresponding to the medical term to be matched based on the matching result.

[0021] Embodiments of this disclosure provide a matching device for medical terms, comprising: one or more processors; and one or more memories storing a computer-executable program that, when executed by the processor, performs the method described above.

[0022] Embodiments of this disclosure provide a computer-readable storage medium having computer-executable instructions stored thereon, which, when executed by a processor, are used to implement the method described above.

[0023] Embodiments of this disclosure provide a computer program product, including a computer program or instructions, wherein the computer program or instructions, when executed by a processor, implement the method described above.

[0024] The methods, apparatus, devices, storage media, and computer program products for matching medical terms provided in the embodiments of this disclosure can improve the processing efficiency of medical terms, enhance the accuracy and coverage of medical term matching, and improve the sustainability and adaptability of medical term matching. The methods, apparatus, devices, storage media, and computer program products for matching medical terms provided in the embodiments of this disclosure combine the advantages of both "clustering initial screening" and "large language model verification," achieving simultaneous improvement in efficiency and accuracy, and reducing invalid verification of the large language model.

[0025] Specifically, in terms of improving the processing efficiency of medical terminology, this disclosure effectively overcomes the efficiency bottlenecks of traditional manual mapping or simple rule matching methods, which suffer from slow processing speed and poor real-time performance when dealing with massive, multi-center medical terminology data. The automated matching mechanism constructed through this disclosure enables high-speed processing of large-scale clinical text data, thereby meeting the real-time and high-throughput requirements for terminology standardization or normalization in clinical and research settings.

[0026] To enhance the accuracy and coverage of medical terminology matching, this disclosure overcomes the key shortcomings of traditional methods that rely on literal matching and struggle to handle semantically similar terms, neologisms, and complex expressions. By introducing large-scale models and contextual analysis, it significantly improves the accuracy of medical terminology matching and its generalization coverage of diverse medical texts.

[0027] In terms of improving the sustainability and adaptability of medical terminology matching, this disclosure fundamentally changes the static maintenance model of traditional terminology systems that relies on periodic manual review and input. By establishing a dynamic update mechanism with self-learning capabilities, the system can continuously absorb newly emerging terminology knowledge and automatically optimize the matching model, thereby forming a medical terminology engine that can intelligently adapt to the rapid evolution of clinical practice and significantly reduce long-term maintenance costs. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some exemplary embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0029] Figure 1 This is a schematic diagram illustrating a scenario for a matching method for medical terms according to an embodiment of the present disclosure;

[0030] Figure 2 This is a flowchart illustrating a matching method for medical terms according to an embodiment of the present disclosure;

[0031] Figure 3This is a flowchart illustrating a matching method for medical terms according to an embodiment of the present disclosure;

[0032] Figure 4 This is a schematic diagram of a matching device for medical terms according to an embodiment of the present disclosure;

[0033] Figure 5 A schematic diagram of a matching device for medical terminology according to an embodiment of the present disclosure is shown;

[0034] Figure 6 A schematic diagram of the architecture of an exemplary computing device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0036] In this specification and accompanying drawings, steps and elements that are substantially the same or similar are indicated by the same or similar reference numerals, and repeated descriptions of these steps and elements are omitted. Furthermore, in the description of this disclosure, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance or order.

[0037] In embodiments of this disclosure, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0039] To facilitate the description of this disclosure, the following concepts related to this disclosure are introduced.

[0040] The medical terminology matching method disclosed herein can be based on artificial intelligence (AI). Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. By studying the design principles and implementation methods of various intelligent machines, artificial intelligence enables the medical terminology matching method disclosed herein to match or map non-standard, highly differentiated original medical terms in clinical texts to standard medical terms for subsequent processing.

[0041] The matching method for medical terms disclosed herein can be based on Natural Language Processing (NLP). NLP is an important field within computer science and artificial intelligence. It studies various theories and methods that enable effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language, that is, the language people use in daily life, and thus it is closely related to linguistic research. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and term matching.

[0042] The matching method for medical terms disclosed herein can utilize the Bidirectional Encoder Representations from Transformers (BERT) model. The BERT model is a pre-trained language model based on a Transformer encoder structure. The BERT model possesses "bidirectional" contextual understanding capabilities. Through masked language modeling (MLM) and next-sentence prediction (NSP) tasks, it can be pre-trained on large amounts of text, enabling the BERT model to deeply fuse contextual information from both sides of words, thereby capturing deep representations at the lexical, syntactic, and semantic levels. After pre-training, the powerful general semantic understanding capabilities of the BERT model can be rapidly transferred through fine-tuning for specific downstream tasks (e.g., text classification, entity recognition, semantic similarity calculation, and terminology matching), significantly improving task performance. Therefore, BERT provides a robust infrastructure for natural language processing, driving substantial progress in semantic understanding technology.

[0043] The human-computer dialogue method disclosed herein can be based on a Large Language Model (LLM). A LLM is a language model in the field of NLP that is trained on a large corpus of language data using an autoregressive approach to analyze and process language text, aiming to understand and generate human language. Trained on massive amounts of text data, LLMs can perform a wide range of tasks, including text summarization, translation, and sentiment analysis. A key characteristic of LLMs is their massive scale; they can contain billions of parameters to help them learn complex patterns in language text data. Therefore, in the medical terminology matching method according to this disclosure, a trained LLM can be applied to the information processing in the medical terminology matching process to provide efficient and accurate medical terminology matching, thereby improving the user experience.

[0044] In summary, the solutions provided by the embodiments of this disclosure involve technologies such as artificial intelligence, natural language processing, and large language models. The embodiments of this disclosure will be further described below with reference to the accompanying drawings.

[0045] When matching non-standard medical terms to be processed or matched to standard medical terms, a rule-based dictionary mapping method can be used. This method typically relies on a pre-built standard terminology dictionary, mapping the non-standard medical terms to be processed or matched to the standard dictionary through exact string matching or regular expression rules. Although this method is relatively simple and direct to implement, it has several obvious drawbacks in practical applications.

[0046] For example, string-matching-based mapping methods lack adaptability to the variability of terminology. This method struggles to effectively handle newly emerging terms not included in existing dictionaries, structurally complex expressions, and semantically similar but literally different terms (e.g., "cauliflower-like mass" and "squamous cell carcinoma" may refer to the same clinical entity, but literal matching cannot achieve a correct mapping), resulting in limited mapping coverage and accuracy.

[0047] Furthermore, existing methods rely heavily on manual comparison and review. Operators must compare and match non-standard terms against authoritative standard terminology databases one by one. This labor-intensive model faces significant bottlenecks. First, its processing efficiency is severely inadequate. When faced with massive clinical text data—thousands to tens of thousands of cases in a single center, and millions in multiple centers—the daily processing capacity is typically less than a thousand records, failing to meet the practical needs of rapid matching of massive amounts of data. Moreover, the consistency of matching results is difficult to guarantee. Due to subjective differences in the clinical connotations of terms among different reviewers, different standardization judgments may be made for the same non-standard term, leading to inconsistent data standardization quality and affecting the reliability of subsequent data analysis.

[0048] Even when using automatic comparison for term matching, the process of comparing each medical term to be matched with a massive number of standard medical terms in each matching process results in an extremely high computational load. When there are a large number of terms to be matched, this step-by-step comparison will cause the matching speed to be too slow.

[0049] Furthermore, many terminology matching systems generally lack efficient and automated dynamic updating and maintenance mechanisms. When new medical terms emerge in clinical practice or existing terms evolve, it is necessary to reorganize domain experts for manual review, judgment, and entry. This process is not only cumbersome and time-consuming, but also costly in terms of manpower and finances. As a result, the terminology standard system cannot keep up with the rapid development of clinical practice and medical knowledge, and the burden of system iteration and updates is heavy, making it difficult to maintain the timeliness and practicality of the terminology system.

[0050] Figure 1 This is a schematic diagram illustrating a scenario for a matching method for medical terms according to an embodiment of the present disclosure.

[0051] like Figure 1 As shown, a user can send user input (e.g., a request to obtain information from the server, or a question from the user to the server) or upload medical terms to be matched to the server via a user terminal, so that the server can respond to the user input. Subsequently, the server can return the generated machine response information to the user terminal via the network. Optionally, the user terminal may specifically include smartphones, tablets, laptops, desktop computers, in-vehicle terminals, wearable devices, etc., but is not limited to these. The user terminal may also be a client with a browser or various applications (including system applications and third-party applications) installed. Optionally, the network may be an Internet of Things (IoT) based on the Internet and / or telecommunications networks, which can be a wired network or a wireless network. For example, it may be an electronic network capable of information exchange, such as a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), or a cellular data communication network. The user terminal and the server can be directly or indirectly connected through wired or wireless communication methods, which is not limited herein. Optionally, the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Of course, it should be understood that, in addition to... Figure 1Beyond the interactive scenarios shown, human-computer dialogue can be implemented solely on the user terminal side, and this disclosure does not impose any limitations on this. For example, a functional module for matching medical terms can be deployed on the user terminal side, enabling the user terminal to independently implement the matching method for medical terms without relying on a server.

[0052] The development of LLM-based intelligent assistants is a significant research direction in AI and NLP. LLM, as a deep learning-based NLP model, has wide applications, including language generation, text classification, information retrieval, and terminology matching. Currently, LLM-based intelligent assistant development schemes primarily involve inputting relevant knowledge during LLM training and training the model through autoregressive tasks to enable intelligent question-answering capabilities. In this approach, LLM is considered a general intelligent model capable of learning and simulating human language abilities, thereby achieving natural language understanding and generation. Therefore, LLM-based intelligent assistant development schemes have significant research and application value, especially for intelligent assistants in specific domains (e.g., terminology matching, legal, medical, customer service, etc.), making them a very promising application solution.

[0053] As an example, in the field of medical terminology matching, AI-assisted terminology matching assistants, as a new type of matching tool, can assist users in matching medical terms through human-computer dialogue, thereby providing users with convenient and efficient terminology matching services.

[0054] Figure 2 This is a flowchart illustrating a method for matching medical terms according to embodiments of the present disclosure. Figure 2 As shown, method 2000 may include steps S2100 to S2600.

[0055] In step S2100, the clinical text is processed to obtain the medical terms to be matched included in the clinical text.

[0056] According to at least one embodiment of this disclosure, clinical text generated by medical service institutions such as hospitals and medical applications can be obtained. The clinical text can be processed, for example, using a machine learning model to obtain the medical terms to be matched included in the clinical text. For example, the clinical text may include a large number of medical terms to be matched. The medical terms to be matched may be highly variable, non-standard medical terms input by operators such as doctors.

[0057] In step S2200, multiple standard medical term clusters are obtained, wherein the multiple standard medical term clusters are determined by clustering multiple standard medical terms included in the standard medical term set, and each of the multiple standard medical term clusters has a representative semantic vector.

[0058] According to at least one embodiment of this disclosure, a standard medical terminology set can be obtained. The standard medical terminology set may include the International Classification of Diseases (ICD) (e.g., the ICD Oncology Special Issue), and the National Comprehensive Cancer Network Clinical Practice Guidelines for Oncology, but this disclosure is not limited thereto. The standard medical terminology set may include multiple standard medical terms. The standard medical terms can be clustered to obtain multiple standard medical terminology clusters. According to at least one embodiment of this disclosure, the standard medical terms can be clustered in advance before performing medical terminology matching operations. Multiple standard medical terminology clusters may each have a representative semantic vector. According to at least one embodiment of this disclosure, a representative semantic vector of a single standard medical terminology cluster can comprehensively represent the semantics of the standard medical terms included in that cluster. For example, the representative semantic vector may be the semantic vector corresponding to a specific standard medical term included in the corresponding standard medical terminology cluster, but this disclosure is not limited thereto. For example, the representative semantic vector may be obtained by processing the semantic vectors corresponding to all standard medical terms included in the corresponding standard medical terminology cluster.

[0059] In step S2300, the medical term to be matched is compared with the representative semantic vectors of each of the multiple standard medical term clusters, so as to determine a first predetermined number of standard medical term clusters with the highest similarity between the medical term to be matched and the multiple representative semantic vectors of the multiple standard medical term clusters as candidate medical term clusters.

[0060] According to at least one embodiment of this disclosure, the medical term to be matched can be compared with the representative semantic vectors of each of a plurality of standard medical term clusters to determine multiple similarities between the medical term to be matched and the representative semantic vectors of each of the plurality of standard medical term clusters. A first predetermined number of standard medical term clusters with the highest similarity (e.g., 5, 10, 15, but this disclosure is not limited thereto) can be identified as candidate medical term clusters. That is, in the process of determining candidate medical term clusters, it is not necessary to compare the medical term to be matched with all standard medical terms in the standard medical term set. In this way, the computational load required in determining candidate medical term clusters and subsequent target standard medical terms can be significantly reduced.

[0061] In step S2400, the standard medical terms, the medical terms to be matched, and the prompt words included in the candidate medical term cluster are sent to the large language model. The prompt words instruct the large language model to select the candidate standard medical terms that correspond to the medical terms to be matched from the standard medical terms included in the candidate medical term cluster.

[0062] According to at least one embodiment of this disclosure, the standard medical terms included in the candidate medical terminology cluster determined in step S2300, the medical terms to be matched, and the prompt words can be sent to a large language model for a medical terminology matching task. In this way, the large language model can perform a matching operation on the standard medical terms included in the candidate medical terminology cluster based on the prompt words.

[0063] In step S2500, the matching results, including candidate standard medical terms corresponding to the medical terms to be matched, are received from the large language model.

[0064] According to at least one embodiment of this disclosure, a matching result generated by a large language model can be received from the large language model. For example, the matching result may include candidate standard medical terms corresponding to the medical term to be matched, determined by the large language model based on prompt words from the standard medical terms included in the candidate medical term cluster.

[0065] In step S2600, based on the matching results, the target standard medical term corresponding to the medical term to be matched is determined.

[0066] According to at least one embodiment of this disclosure, candidate standard medical terms included in the matching results can be directly identified as target standard medical terms. According to at least one embodiment of this disclosure, the candidate standard medical terms included in the matching results can be further analyzed by a human, such as a reviewer, to determine whether to identify at least one candidate standard medical term from the candidate standard medical terms included in the matching results as the target standard medical term.

[0067] The matching method for medical terms according to at least one embodiment of this disclosure can significantly improve the processing efficiency of medical terms and reduce the computational load required for the matching process. Firstly, the matching method for medical terms according to at least one embodiment of this disclosure does not require matching the medical term to be matched against all standard medical terms, but rather compares it against a representative semantic vector of a standard medical term cluster. In this way, the number of matches can be significantly reduced, thereby reducing the computational load.

[0068] According to at least one embodiment of this disclosure, multiple medical terms to be matched can be clustered, thereby reducing the number of matches and computational load. However, in the implementation of clustering multiple medical terms to be matched, since the differences between the medical terms to be matched can be significant, clustering standard medical terms may be more effective than clustering multiple medical terms to be matched. For example, clustering by standard medical terms can significantly reduce the initial screening error caused by semantic dispersion of medical terms to be matched. Furthermore, since the matching method for medical terms according to at least one embodiment of this disclosure clusters standard medical terms, its clustering effect is better, thus obtaining more accurate subsequent matching results.

[0069] Figure 3 This is a flowchart illustrating a method for matching medical terms according to embodiments of the present disclosure. Figure 3 As shown, method 3000 may include steps S3010 to S3100. Figure 3 Zhongyu Figure 2 Identical or similar parts will not be described again to avoid redundancy.

[0070] In step S3010, a standard medical terminology set can be obtained. The standard medical terminology set may include the International Classification of Diseases (ICD) (e.g., the ICD Oncology Special Issue), and the National Comprehensive Cancer Network Clinical Practice Guidelines for Oncology, but this disclosure is not limited thereto.

[0071] In step S3020, the standard medical terms in the standard medical terminology set can be vectorized. For example, the standard medical terms in the standard medical terminology set can be vectorized using a BERT model. According to at least one embodiment of this disclosure, the BERT model can be trained based on previous clinical texts and the standard medical terminology set. For example, a Chinese or multilingual BERT model (such as a BERT base or RoBERTa) can be further pre-trained on a large amount of medical literature, medical textbooks, and clinical guidelines, thereby enabling the BERT model to further learn medical language.

[0072] Additionally or alternatively, a large amount of prior clinical text can be obtained in advance, and this prior clinical text may include a large number of medical terms to be matched. The obtained prior clinical text can be annotated, thereby labeling the medical terms to be matched within the prior clinical text with the corresponding target standard medical terms. After further training of the BERT model, the BERT model can be validated and deployed to perform subsequent semantic analysis tasks.

[0073] Additionally or alternatively, a large amount of prior clinical text can be obtained in advance, and this prior clinical text may include a large number of medical terms to be matched. The obtained prior clinical text can be annotated, thereby assigning corresponding tags to the medical terms to be matched included in the prior clinical text. Tags may include disease names, surgical names, drug names, medical device names, etc., and this disclosure is not limited to these. After further training the BERT model, the BERT model can be validated and deployed to perform subsequent semantic analysis tasks. Based on standard medical terminology sets (e.g., ICD and the US National Comprehensive Cancer Network Clinical Practice Guidelines for Oncology) and tags jointly defined by clinical experts, the BERT model can be used to achieve entity recognition (NER) of clinical text, thereby simultaneously completing the extraction and tag-based classification of the medical terms to be matched. In this way, the problems of incomplete coverage and chaotic classification of medical terms in traditional NER can be solved.

[0074] According to at least one embodiment of this disclosure, multiple first semantic vectors can be obtained by vectorizing multiple standard medical terms included in the standard medical terminology set using the BERT model. For example, for each of the multiple standard medical terms included in the standard medical terminology set, a corresponding first semantic vector can be obtained.

[0075] In step S3030, the standard medical terms in the standard medical terminology set can be clustered. For example, by clustering multiple first semantic vectors corresponding to multiple standard medical terms included in the standard medical terminology set, multiple standard medical term clusters can be obtained. According to at least one embodiment of this disclosure, clustering multiple first semantic vectors can be performed using the K-means method, but this disclosure is limited thereto. To reduce the number of subsequent comparisons, the number of standard medical terms included in each standard medical term cluster in the K-means method can be set to a large value.

[0076] According to at least one embodiment of this disclosure, standard medical terms in a standard medical terminology set can be grouped into multiple groups according to labels, and the standard medical terms included in each group can be further clustered to obtain one or more standard medical term clusters for each group. For example, the labels of standard medical terms in the standard medical terminology set can be manually determined by professionals with reference to information such as catalogs and versions. For example, the labels of standard medical terms in the standard medical terminology set can be determined using a BERT model. As mentioned above, labels can include disease names, surgical procedure names, drug names, medical device names, etc., and this disclosure is not limited thereto.

[0077] According to at least one embodiment of this disclosure, multiple representative semantic vectors corresponding to multiple standard medical terminology clusters can be determined based on multiple first semantic vectors. For example, for each standard medical terminology cluster, a representative semantic vector corresponding to that standard medical terminology cluster can be determined based on multiple first semantic vectors corresponding to the multiple standard medical terms included in that standard medical terminology cluster. According to at least one embodiment of this disclosure, the representative semantic vector can be the centroid semantic vector of the standard medical terminology cluster corresponding to the representative semantic vector. The centroid semantic vector can be a semantic vector having the average value of the multiple first semantic vectors corresponding to the multiple standard medical terms included in that standard medical terminology cluster. In this disclosure, the semantic vector can be a high-dimensional vector, but this disclosure is not limited thereto.

[0078] In step S3040, clinical text can be obtained. For example, as described above, clinical text generated by medical service institutions such as hospitals and medical applications can be obtained, and the clinical text may include a large number of medical terms to be matched.

[0079] In step S3050, the medical terms to be matched in the clinical text can be identified. According to at least one embodiment of this disclosure, the BERT model can be used to process the clinical text to obtain the medical terms to be matched included in the clinical text.

[0080] According to at least one embodiment of this disclosure, a BERT model can be used to determine the labels of medical terms to be matched. In this way, a BERT model can be used to identify the medical terms to be matched in clinical text and the labels that the medical terms to be matched have.

[0081] In step S3060, the medical term to be matched can be vectorized. According to at least one embodiment of this disclosure, the BERT model can be used to vectorize the medical term to be matched to obtain a second semantic vector corresponding to the medical term to be matched.

[0082] In step S3070, the similarity between the representative semantic vectors of the medical term to be matched and the standard medical term clusters can be determined to identify candidate standard medical term clusters. According to at least one embodiment of this disclosure, multiple first similarities can be determined between the second semantic vector corresponding to the medical term to be matched and multiple representative semantic vectors corresponding to multiple standard medical term clusters. The first similarity may include, but is not limited to, cosine similarity. The standard medical term clusters corresponding to the representative semantic vectors with the highest first predetermined number of first similarities can be identified as candidate standard medical term clusters. For example, the standard medical term clusters can be sorted according to the first similarity from high to low, and the first predetermined number of standard medical term clusters ranked highest can be identified as candidate standard medical term clusters.

[0083] According to at least one embodiment of this disclosure, the labels of the medical term clusters to be matched can be used. As described above, when obtaining the medical terms to be matched from clinical text using the BERT model, the labels of the medical terms to be matched can be further obtained. For example, the medical terms to be matched may have a first label. The medical terms to be matched can be compared with representative semantic vectors of standard medical term clusters with the same first label from multiple standard medical term clusters to determine candidate medical term clusters. That is, it is not necessary to compare the medical terms to be matched with representative semantic vectors of standard medical term clusters with different, non-first labels from multiple standard medical term clusters. In this way, the computational load can be further reduced.

[0084] According to at least one embodiment of this disclosure, a first label can be determined for the medical terms to be matched. For example, matching can be performed only on medical terms with the first label and not on those with other labels. In this case, in step S3030, only the standard medical terms with the first label included in the standard medical term set can be clustered to determine multiple standard medical term clusters. In this way, the computational load when preparing standard medical term clusters in advance can be reduced.

[0085] In step S3080, standard medical terms, medical terms to be matched, and prompt words can be sent to the large language model. For example, standard medical terms, medical terms to be matched, and prompt words included in a candidate medical term cluster can be sent to the large language model. According to at least one embodiment of this disclosure, the prompt words can instruct the large language model to select candidate standard medical terms corresponding to the medical terms to be matched from the standard medical terms included in the candidate medical term cluster. Since a preliminary screening of the massive number of standard medical terms has already been performed in step S3070, the number of standard medical terms sent to the large language model will not be excessive. In this way, the computational load and communication bandwidth of the large language model can be reduced. More importantly, by reducing the number of standard medical terms sent, context explosion in the large language model can be prevented, thereby improving the accuracy of the inference results.

[0086] According to at least one embodiment of this disclosure, a large language model is trained based on prior clinical texts and a standard medical terminology set, wherein the medical terms to be matched included in the prior clinical texts are labeled with corresponding target standard medical terms. For example, the standard medical terminology set can be provided as a knowledge base, including the International Classification of Diseases (ICD) (e.g., the ICD Oncology Special Issue), and the US National Comprehensive Cancer Network Clinical Practice Guidelines for Oncology, etc., for the large language model to access. A large amount of prior clinical texts can be obtained in advance, and these prior clinical texts may include a large number of medical terms to be matched. The obtained prior clinical texts can be labeled, thereby labeling the medical terms to be matched included in the obtained prior clinical texts with corresponding target standard medical terms. The medical terms to be matched labeled with corresponding target standard medical terms can be sent to the large language model for learning. Further, prompt words can be designed to prompt the large language model to perform the medical terminology matching task. After training the large language model as described above, the large language model can be validated and deployed to perform subsequent terminology matching tasks.

[0087] According to at least one embodiment of this disclosure, multiple second similarities can be determined between the medical term to be matched and the standard medical terms included in the candidate medical term cluster. For example, for the candidate medical term cluster determined in step S3070, each standard medical term included in the candidate medical term cluster can be compared with the medical term to be matched to determine multiple second similarities. The second similarities may include cosine similarity, but this disclosure is not limited thereto. Additionally or alternatively, multiple second similarities can be sent to a large language model. In this way, the large language model can be assisted in making more accurate inferences.

[0088] According to at least one embodiment of this disclosure, the context associated with the medical term to be matched can be determined. For example, the context associated with the medical term to be matched may include at least one of a first predetermined length of text in the clinical text preceding the medical term to be matched and a second predetermined length of text in the clinical text following the medical term to be matched. That is, in addition to the medical term to be matched, the context of the medical term to be matched in the clinical text can be further identified. Alternatively or additionally, the context associated with the medical term to be matched can be sent to a large language model. By introducing the context associated with the medical term to be matched as a verification criterion, combined with a multi-dimensional scoring framework, the large language model can not only rely on the term itself, but also combine the clinical context to judge the rationality of the mapping and matching process, thus overcoming the limitations of traditional shallow semantic matching.

[0089] The large language model can receive the medical term to be matched, the standard medical terms included in the candidate medical term cluster, and the context related to the medical term to be matched. In this way, guided by prompt words, the large language model can perform verification from multiple dimensions such as semantic consistency, clinical context compliance, terminology standardization, and confidence, thereby outputting the matching result. Furthermore, the matching result also includes the reasons why the large language model selected the candidate standard medical terms from the candidate medical term cluster, such as supplementing the contextual basis of the matching result. In this way, the matching accuracy of the matching result can be improved. According to at least one embodiment of this disclosure, the matching result can be output in JSON format.

[0090] According to at least one embodiment of this disclosure, the following can be sent to a large language model: standard medical terminology, medical terminology to be matched, second similarity, context related to the medical terminology to be matched, and prompt words, but this disclosure is not limited thereto:

[0091] # System Character Settings

[0092] You are a senior oncology expert and medical terminologist. Your task is to accurately verify the mapping relationship between a given [preceding context of the medical term to be matched], the [medical term to be matched], and the [standard medical terms in the candidate standard medical terminology set].

[0093] # Enter information

[0094] [Medical term to be matched]: {Original word}

[0095] [Contextual fragments of medical terms to be matched]: {Contextual fragments of the original term}

[0096] [Standard Medical Term 1 in the Candidate Standard Medical Terminology Set]: {Standard Term 1} (Similarity Score: {score1})

[0097] [Standard Medical Terminology 2 in the Candidate Standard Medical Terminology Set]: {Standard Word 2} (Similarity Score: {score2})

[0098] # Validation Dimensions

[0099] Please rate each candidate mapping on a scale of 1-5 (5 being the best) from the following four dimensions and provide a brief explanation:

[0100] 1. Semantic consistency: Are the two concepts fundamentally consistent in the medical context? Is there a relationship of generalization versus specificity, or inclusion versus being included?

[0101] 2. Clinical contextual compliance: Is this mapping reasonable and unambiguous in clinical oncology practice (such as diagnosis, treatment, and prognosis)?

[0102] 3. Terminology standardization: Are the candidate standard terms more standardized and in line with authoritative standards than the original terms?

[0103] 4. Mapping Confidence: Considering the above three points, how confident are you in the overall mapping relationship?

[0104] # Task Requirements

[0105] - Please score each candidate standard term according to the four dimensions mentioned above.

[0106] - Finally, please provide your overall recommendation: choose the one standard term you think is most appropriate, or judge it as "cannot be mapped".

[0107] - If "Unable to map" is recommended, please explain why (e.g., the original wording is vague, ambiguous, or not within the scope of the standard system).

[0108] # Output Format

[0109] Please strictly adhere to the following JSON format for output:

[0110] {

[0111] "candidate_reviews": [

[0112] {

[0113] "candidate_standard_term": "Standard medical term 1 in the candidate standard medical terminology set",

[0114] "semantic_consistency": {score},

[0115] "clinical_context": {score},

[0116] "term_standardization": {score},

[0117] "overall_confidence": {score},

[0118] "reason": "..."

[0119] },

[0120] ... / / Other candidates

[0121] ],

[0122] "final_recommendation": "The most appropriate standard medical term or 'unmapped'",

[0123] "final_reason": "Comprehensive reasons..."

[0124] }

[0125] # candidate_reviews (list of candidate terms for review)

[0126] # candidate_standard_term (candidate standard term)

[0127] # Semantic Consistency (Semantic Consistency Score)

[0128] # clinical_context (clinical context compliance score)

[0129] # term_standardization (term standardization score)

[0130] # Overall confidence score

[0131] # reason

[0132] # final_recommendation (Final Recommendation)

[0133] # final_reason (final recommendation reason)

[0134] The above information is provided for the purpose of facilitating understanding, and this disclosure is not limited thereto.

[0135] In this way, the large language model is no longer a simple "matcher," but a "verifier" that acts as an "expert in oncology terminology." The scores provided by the large language model enhance the credibility and traceability of the results.

[0136] In step S3090, a matching result may be received. For example, the matching result may be received from a large language model. According to at least one embodiment of this disclosure, the matching result may include candidate standard medical terms corresponding to the medical term to be matched, determined by the large language model based on prompt words from the standard medical terms included in the candidate medical term cluster. Further, the matching result may also include the reasons why the large language model selected the candidate standard medical terms from the standard medical terms included in the candidate medical term cluster.

[0137] In step S3100, the target standard medical term corresponding to the medical term to be matched can be determined. According to at least one embodiment of this disclosure, the received matching results can be displayed to users such as reviewers and medical experts. Confirmation results for candidate standard medical terms can be received from the user. For example, after viewing the candidate standard medical terms and corresponding reasons included in the matching results, the user can perform one or more of the following operations: accept the candidate standard medical term, reject the candidate standard medical term, and modify the candidate standard medical term. Further, the candidate standard medical term can be determined as the target standard term based on the confirmation result. For example, in response to receiving a confirmation result indicating acceptance of the candidate standard medical term, the candidate standard medical term can be determined as the target standard medical term. For example, in response to receiving a confirmation result indicating modification of the candidate standard medical term, the modified candidate standard medical term can be determined as the target standard medical term. For example, in response to receiving a confirmation result indicating rejection of the candidate standard medical term, the large model can regenerate the matching results. In this way, the user only needs to perform "confirm," "modify," or "reject" operations on a small number of candidate standard terms, which is far more efficient than performing matching operations on a massive number of standard medical terms.

[0138] According to at least one embodiment of this disclosure, the BERT model and the large language model described above can be iteratively updated. For example, in response to a confirmation result indicating acceptance of a candidate standard medical term, the accepted candidate standard medical term (i.e., the target standard medical term) and the medical term to be matched can be stored as a "gold standard dataset" for subsequent iterative training. For example, the accepted candidate standard medical terms and the medical term to be matched in the "gold standard dataset" can be used to retrain at least one of the BERT model and the large language model. As another example, in response to receiving a second predetermined number of confirmation results indicating rejection of candidate standard medical terms and modification of at least one of the candidate standard medical terms for the medical term to be matched, the prompt words can be modified. For example, when a user repeatedly modifies or rejects the matching results returned by the large language model, it can be determined that the quality of the matching results returned by the large language model is poor. In this case, the reasons for the user's modification or rejection of the matching results can be analyzed, thereby further modifying the prompt words. That is, users can shift from cumbersome manual comparison to quality control of high-value cases, and user feedback data (including annotation results of context) can be used for further training of the BERT model and optimization of the prompt words of the large language model. In this way, the system can acquire context-aware self-evolution capabilities, enabling low-cost and high-efficiency dynamic maintenance of the terminology system. Through dynamic feedback and updates, the update cycle of the medical terminology system can be shortened from the monthly or yearly level to the daily or weekly level, becoming a "living" or dynamically updated system that can evolve with the development of medicine, fundamentally solving the technical problems of slow iteration and high cost of the traditional model.

[0139] According to at least one embodiment of this disclosure, the aforementioned models (e.g., including but not limited to BERT models and large language models) can be encapsulated as application programming interfaces (APIs) for user invocation. These APIs can be applied to electronic medical record structuring. For example, after personnel such as doctors enter medical records, the API can be automatically invoked to convert non-standard medical terms into standard medical terms. These APIs can be applied to research data integration. For example, during the data cleaning stage, the API can be invoked to normalize terminology in data from different hospitals. These APIs can be applied to AI model training. For example, the API can be invoked to provide clean, standardized input data for AI models in diagnosis, prognosis, etc. Furthermore, these APIs can also be applied to real-world research, etc., but this disclosure is not limited thereto.

[0140] Figure 4 This is a schematic diagram of a matching device for medical terminology according to an embodiment of the present disclosure.

[0141] like Figure 4 As shown, the matching device 400 for medical terms may include a text processing module 410, an acquisition module 420, a comparison module 430, a sending module 440, a receiving module 450, and a determination module 460.

[0142] The text processing module 410 can be configured to process clinical text to obtain the medical terms to be matched included in the clinical text.

[0143] The acquisition module 420 can be configured to acquire multiple standard medical term clusters, wherein the multiple standard medical term clusters are determined by clustering multiple standard medical terms included in the standard medical term set, and each of the multiple standard medical term clusters has a representative semantic vector.

[0144] The comparison module 430 can be configured to compare the medical term to be matched with the representative semantic vectors of each of the plurality of standard medical term clusters, so as to determine a first predetermined number of standard medical term clusters from the plurality of standard medical term clusters as candidate medical term clusters based on the highest similarity between the medical term to be matched and the plurality of representative semantic vectors of the plurality of standard medical term clusters.

[0145] The sending module 440 can be configured to send the standard medical terms included in the candidate medical term cluster, the medical term to be matched, and a prompt word to the large language model, wherein the prompt word instructs the large language model to select a candidate standard medical term corresponding to the medical term to be matched from the standard medical terms included in the candidate medical term cluster.

[0146] The receiving module 450 can be configured to receive from the large language model matching results including candidate standard medical terms corresponding to the medical term to be matched.

[0147] The determination module 460 can be configured to determine, based on the matching results, the target standard medical term corresponding to the medical term to be matched.

[0148] According to at least one embodiment of the present disclosure, a medical terminology matching apparatus is provided, wherein the text processing module 410 may be specifically configured to process clinical text using a transformer-based bidirectional encoder representation model to obtain medical terms to be matched included in the clinical text.

[0149] A medical terminology matching apparatus according to at least one embodiment of the present disclosure, wherein a transformer-based bidirectional encoder representation model is trained based on prior clinical text and a standard medical terminology set, and wherein the medical terms to be matched included in the prior clinical text are labeled with corresponding target standard medical terms.

[0150] A matching apparatus for medical terms according to at least one embodiment of the present disclosure, wherein the medical term to be matched has a first label among a plurality of labels; and wherein clustering a plurality of standard medical terms included in a set of standard medical terms includes: clustering the standard medical terms included in the set of standard medical terms that have the first label to determine a plurality of standard medical term clusters.

[0151] According to at least one embodiment of the present disclosure, a matching apparatus for medical terms includes clustering a plurality of standard medical terms included in a set of standard medical terms, which comprises: vectorizing the plurality of standard medical terms included in the set of standard medical terms to obtain a plurality of first semantic vectors; and clustering the plurality of first semantic vectors to obtain a plurality of standard medical term clusters.

[0152] According to at least one embodiment of the medical term matching apparatus of the present disclosure, the comparison module 430 may be specifically configured to vectorize the medical term to be matched to obtain a second semantic vector; determine a plurality of representative semantic vectors corresponding to a plurality of standard medical term clusters based on a plurality of first semantic vectors; determine a plurality of first similarities between the second semantic vector and the plurality of representative semantic vectors; and determine the standard medical term cluster corresponding to the representative semantic vector having the highest first predetermined number of first similarities as a candidate medical term cluster.

[0153] According to at least one embodiment of the present disclosure, a matching device for medical terms is provided, wherein the representative semantic vector is the centroid semantic vector of a standard medical term cluster corresponding to the representative semantic vector.

[0154] According to at least one embodiment of the present disclosure, a matching apparatus for medical terms is provided, wherein vectorization of a plurality of standard medical terms included in a standard medical terminology set and vectorization of the medical terms to be matched are performed by a transformer-based bidirectional encoder representation model.

[0155] According to at least one embodiment of the present disclosure, a matching apparatus for medical terms is used, wherein clustering of a plurality of first semantic vectors is performed using the K-means method.

[0156] A matching apparatus for medical terminology according to at least one embodiment of the present disclosure, wherein a large language model is trained based on prior clinical text and a standard medical terminology set, and wherein the medical terms to be matched included in the prior clinical text are labeled with corresponding target standard medical terms.

[0157] According to at least one embodiment of the present disclosure, a matching apparatus for medical terms may be specifically configured to determine a plurality of second similarities between the medical term to be matched and standard medical terms included in a candidate medical term cluster; and to send the plurality of second similarities to a large language model.

[0158] According to at least one embodiment of the present disclosure, a matching apparatus for medical terms, wherein a sending module 440 may be specifically configured to determine a context associated with a medical term to be matched, wherein the context associated with the medical term to be matched includes at least one of a first predetermined length of clinical text preceding the medical term to be matched and a second predetermined length of clinical text following the medical term to be matched; and to send the context associated with the medical term to be matched to a large language model.

[0159] A matching apparatus for medical terminology according to at least one embodiment of the present disclosure, wherein the matching result further includes the rationale for selecting standard medical terms from a candidate medical terminology cluster by a large language model.

[0160] According to at least one embodiment of the present disclosure, a matching apparatus for medical terminology, wherein a determining module 460 may be specifically configured to receive a confirmation result for a candidate standard medical term, wherein the confirmation result indicates at least one of accepting the candidate standard medical term, rejecting the candidate standard medical term, and modifying the candidate standard medical term; and to determine a target standard term based on the confirmation result and the candidate standard medical term.

[0161] According to at least one embodiment of the present disclosure, a medical term matching apparatus is provided, wherein the text processing module 410 may be specifically configured to obtain medical terms to be matched included in clinical text through a transformer-based bidirectional encoder representation model, and the medical term matching apparatus further includes an update module, which may be configured to accept candidate standard medical terms in response to a confirmation result indication, and retrain the transformer-based bidirectional encoder representation model using the accepted candidate standard medical terms and the medical terms to be matched.

[0162] According to at least one embodiment of the present disclosure, the matching apparatus for medical terms may include an update module specifically configured to modify a prompt word in response to receiving a confirmation result of receiving a second predetermined number of indications to reject candidate standard medical terms and modifying at least one of the candidate standard medical terms for the medical term to be matched.

[0163] According to another aspect of this disclosure, a matching device for medical terminology is also provided. Figure 5 A schematic diagram of a matching device for medical terminology according to an embodiment of the present disclosure is shown.

[0164] like Figure 5 As shown, the medical term matching device 500 may include one or more processors 510 and one or more memories 520. The memories 520 store computer-readable code that, when executed by the one or more processors 510, can perform the medical term matching method as described above.

[0165] The processor in the embodiments of this disclosure can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x84 architecture or an ARM architecture.

[0166] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0167] Figure 6 A schematic diagram of the architecture of an exemplary computing device according to an embodiment of the present disclosure is shown.

[0168] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 6 The architecture of the computing device 600 shown is used for implementation. For example... Figure 6 As shown, the computing device 600 may include a bus 610, one or more CPUs 620, a read-only memory (ROM) 630, a random access memory (RAM) 640, a communication port 650 connected to a network, an input / output component 660, a hard disk 670, etc. The storage devices in the computing device 600, such as the ROM 670 or the hard disk 670, may store various data or files used for processing and / or communication of the medical terminology matching method provided in this disclosure, as well as program instructions executed by the CPU. The computing device 600 may also include a user interface 680. Of course, Figure 6 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 6 One or more components in the computing device shown.

[0169] According to another aspect of this disclosure, a computer-readable storage medium is also provided. The computer storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, a matching method for medical terms according to embodiments of this disclosure, as described with reference to the above-drawn figures, can be performed. The computer-readable storage medium in the embodiments of this disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), 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), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0170] Embodiments of this disclosure also provide a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a matching method for medical terminology according to embodiments of this disclosure.

[0171] Embodiments of this disclosure provide methods, apparatus, devices, storage media, and computer program products for matching medical terms.

[0172] The methods, apparatus, devices, storage media, and computer program products for matching medical terms provided in the embodiments of this disclosure can improve the processing efficiency of medical terms, enhance the accuracy and coverage of medical term matching, and improve the sustainability and adaptability of medical term matching. The methods, apparatus, devices, storage media, and computer program products for matching medical terms provided in the embodiments of this disclosure combine the advantages of both "clustering initial screening" and "large language model verification," achieving simultaneous improvement in efficiency and accuracy, and reducing invalid verification of the large language model.

[0173] Specifically, in terms of improving the processing efficiency of medical terminology, this disclosure effectively overcomes the efficiency bottlenecks of traditional manual mapping or simple rule matching methods, which suffer from slow processing speed and poor real-time performance when dealing with massive, multi-center medical terminology data. The automated matching mechanism constructed through this disclosure enables high-speed processing of large-scale clinical text data, thereby meeting the real-time and high-throughput requirements for terminology standardization or normalization in clinical and research settings.

[0174] To enhance the accuracy and coverage of medical terminology matching, this disclosure overcomes the key shortcomings of traditional methods that rely on literal matching and struggle to handle semantically similar terms, neologisms, and complex expressions. By introducing large-scale models and contextual analysis, it significantly improves the accuracy of medical terminology matching and its generalization coverage of diverse medical texts.

[0175] In terms of improving the sustainability and adaptability of medical terminology matching, this disclosure fundamentally changes the static maintenance model of traditional terminology systems that relies on periodic manual review and input. By establishing a dynamic update mechanism with self-learning capabilities, the system can continuously absorb newly emerging terminology knowledge and automatically optimize the matching model, thereby forming a medical terminology engine that can intelligently adapt to the rapid evolution of clinical practice and significantly reduce long-term maintenance costs.

[0176] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing at least one executable instruction for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0177] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0178] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations of these embodiments or their features can be made without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure.

Claims

1. A matching method for medical terms, comprising: The clinical text is processed to obtain the medical terms to be matched included in the clinical text; Multiple standard medical term clusters are obtained, wherein the multiple standard medical term clusters are determined by clustering multiple standard medical terms included in the standard medical term set, and each of the multiple standard medical term clusters has a representative semantic vector, wherein the representative semantic vector is the centroid semantic vector of the standard medical term cluster corresponding to the representative semantic vector, and the centroid semantic vector is obtained by: vectorizing multiple standard medical terms included in the standard medical term set to obtain multiple first semantic vectors, and determining the semantic vector having the average value of the multiple first semantic vectors corresponding to the multiple standard medical terms included in the corresponding standard medical term cluster as the representative semantic vector; The medical term to be matched is compared with the representative semantic vectors of each of the plurality of standard medical term clusters, but not with the standard medical terms included in the plurality of standard medical term clusters, so as to determine a first predetermined number of standard medical term clusters from the plurality of standard medical term clusters whose similarity to the plurality of representative semantic vectors of the medical term to be matched is the highest as candidate medical term clusters. The large language model is sent with the standard medical terms included in the candidate medical term cluster, the medical term to be matched, and a prompt word, wherein the prompt word instructs the large language model to select a candidate standard medical term corresponding to the medical term to be matched from the standard medical terms included in the candidate medical term cluster, wherein the number of standard medical terms included in the candidate medical term cluster is less than the number of standard medical terms included in the plurality of standard medical term clusters; Receive from the large language model matching results including candidate standard medical terms corresponding to the medical term to be matched; and Based on the matching results, the target standard medical term corresponding to the medical term to be matched is determined.

2. The matching method according to claim 1, wherein, The process of processing the clinical text to obtain the medical terms to be matched included in the clinical text: The clinical text is processed using a transformer-based bidirectional encoder representation model to obtain the medical terms to be matched included in the clinical text.

3. The matching method according to claim 2, wherein, The transformer-based bidirectional encoder representation model is trained on prior clinical text and the standard medical terminology set, wherein the medical terms to be matched included in the prior clinical text are labeled with corresponding tags.

4. The matching method according to claim 3, wherein, The medical term to be matched has a first label among multiple labels; and The clustering of multiple standard medical terms included in the standard medical terminology set includes: Clustering is performed on the standard medical terms with a first label included in the standard medical terminology set to determine the multiple standard medical terminology clusters.

5. The matching method according to claim 1, wherein, The clustering of multiple standard medical terms included in the standard medical terminology set includes: Cluster the plurality of first semantic vectors to obtain the plurality of standard medical term clusters.

6. The matching method according to claim 5, wherein, The step of comparing the medical term to be matched with the representative semantic vectors of the plurality of standard medical term clusters to determine candidate medical term clusters from the plurality of standard medical term clusters includes: The medical terms to be matched are vectorized to obtain a second semantic vector; Based on the multiple first semantic vectors, multiple representative semantic vectors corresponding to the multiple standard medical term clusters are determined respectively; Determine multiple first similarities between the second semantic vector and the plurality of representative semantic vectors; and The standard medical term clusters corresponding to the representative semantic vectors with the highest first predetermined number of first similarities are determined as the candidate medical term clusters.

7. The matching method according to claim 6, wherein, Vectorization of multiple standard medical terms included in the standard medical terminology set and vectorization of the medical terms to be matched are performed using a transformer-based bidirectional encoder representation model.

8. The matching method according to claim 5, wherein, The clustering of the multiple first semantic vectors was performed using the K-means method.

9. The matching method according to claim 1, wherein, The large language model is trained based on previous clinical texts and the standard medical terminology set, wherein the medical terms to be matched included in the previous clinical texts are labeled with corresponding target standard medical terms.

10. The matching method according to claim 1, further comprising: Determine multiple second similarities between the medical term to be matched and the standard medical terms included in the candidate medical term cluster; as well as The multiple second similarities are sent to the large language model.

11. The matching method according to claim 1, further comprising: Determine the context associated with the medical term to be matched, wherein the context associated with the medical term to be matched includes at least one of the clinical text of a first predetermined length preceding the medical term to be matched and the clinical text of a second predetermined length following the medical term to be matched; and Send the context associated with the medical term to be matched to the large language model.

12. The matching method according to claim 1, wherein, The matching results also include the reasons why the large language model selects the candidate standard medical terms from the standard medical terms included in the candidate medical term cluster.

13. The matching method according to claim 1, wherein, The determination of the target standard medical terminology based on the matching results includes: Receive confirmation results for the candidate standard medical terminology, wherein the confirmation results indicate at least one of accepting the candidate standard medical terminology, rejecting the candidate standard medical terminology, and modifying the candidate standard medical terminology; and Based on the confirmation results and the candidate standard medical terms, the target standard terminology is determined.

14. The matching method according to claim 13, wherein, The acquisition of the medical terms to be matched included in the clinical text is performed using a transformer-based bidirectional encoder representation model, and the method further includes: In response to the confirmation result indicating acceptance of the candidate standard medical term, the transformer-based bidirectional encoder representation model is retrained using the accepted candidate standard medical term and the medical term to be matched.

15. The matching method according to claim 13, further comprising: In response to the confirmation result of receiving a second predetermined number of indications to reject the candidate standard medical terms and modifying at least one of the candidate standard medical terms, the prompt word is modified.

16. A matching device for medical terminology, comprising: A text processing module is configured to process clinical text to obtain medical terms to be matched included in the clinical text; The acquisition module is configured to acquire multiple standard medical term clusters, wherein the multiple standard medical term clusters are determined by clustering multiple standard medical terms included in the standard medical term set, and each of the multiple standard medical term clusters has a representative semantic vector, wherein the representative semantic vector is the centroid semantic vector of the standard medical term cluster corresponding to the representative semantic vector, and the centroid semantic vector is obtained by: vectorizing multiple standard medical terms included in the standard medical term set to obtain multiple first semantic vectors, and determining the semantic vector having the average value of the multiple first semantic vectors corresponding to the multiple standard medical terms included in the corresponding standard medical term cluster as the representative semantic vector; The comparison module is configured to compare the medical term to be matched with the representative semantic vectors of each of the plurality of standard medical term clusters, but not with the standard medical terms included in the plurality of standard medical term clusters, so as to determine a first predetermined number of standard medical term clusters from the plurality of standard medical term clusters as candidate medical term clusters, based on the highest similarity between the medical term to be matched and the plurality of representative semantic vectors of the plurality of standard medical term clusters. The sending module is configured to send the standard medical terms included in the candidate medical term cluster, the medical term to be matched, and a prompt word to the large language model, wherein the prompt word instructs the large language model to select a candidate standard medical term corresponding to the medical term to be matched from the standard medical terms included in the candidate medical term cluster, wherein the number of standard medical terms included in the candidate medical term cluster is less than the number of standard medical terms included in the plurality of standard medical term clusters; The receiving module is configured to receive from the large language model matching results including candidate standard medical terms corresponding to the medical term to be matched; and The determination module is configured to determine the target standard medical term corresponding to the medical term to be matched based on the matching result.

17. A matching device for medical terminology, comprising: One or more processors; as well as One or more memories storing a computer-executable program, which, when executed by the processor, performs the method of any one of claims 1-15.

18. A computer-readable storage medium having stored thereon computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-15.

19. A computer program product comprising a computer program or instructions, wherein, The computer program or instructions, when executed by a processor, implement the method as described in any one of claims 1-15.

Images