Method and system for constructing medical knowledge base based on medical examination and medium

Abstract

The invention belongs to the technical field of data processing, and discloses a construction method and system based on a medical examination medical knowledge base and a medium. Comprising the steps that a basic medical data set is collected, data cleaning is conducted, and a structured medical data set is output; constructing a multistage medical dictionary and establishing semantic mapping with the structured medical data set to generate a standard medical data set; carrying out medical related entity extraction on the standard medical data set and establishing a semantic relationship of each medical entity to obtain a medical entity set; performing semantic alignment on the medical entity set to obtain a semantic alignment entity set; constructing a medical examination knowledge rule base and constructing an initial medical examination knowledge graph in combination with the semantic alignment entity set; performing entity association extension on the initial medical examination knowledge graph, and outputting an enhanced medical examination knowledge graph; real-time clinical data are collected, and the medical examination knowledge graph is updated and enhanced based on the real-time clinical data to obtain the medical examination knowledge base, so that convenience is provided for medical decision making and intelligent reasoning diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, more particularly, the present application relates to a construction method and system of a medical examination medical knowledge base and a medium. BACKGROUND

[0002] With the continuous development of medical information systems, hospitals, physical examination centers and related research institutions face more and more patients every day, and the amount of medical data accumulated is also increasing, especially the related medical examination data from the clinical laboratory; Therefore, constructing a medical knowledge base as a carrier of multi-source medical data to provide convenience for medical staff when working has become a core task of smart medical care.

[0003] In the existing knowledge base construction field, it often depends on manual annotation, resulting in semantic split between the collected data and the actual semantic information of the data, and difficulty in knowledge association; Moreover, the existing technology is difficult to realize accurate semantic modeling of various fields in the data, there may be semantic ambiguity fields or polysemy fields interfering with semantic mapping effect, resulting in semantic deviation; At the same time, when performing entity extraction, it often depends on fixed window or simple syntax rules, which is easy to miss semantic association across sentences or paragraphs; In the traditional medical knowledge base, there is a lack of sufficient data analysis capability for the potential semantic relationship or semantic path, resulting in knowledge gaps; In addition, the traditional medical knowledge base is mostly statically deployed, lacks timeliness, and is difficult to support real-time data semantic evaluation, rule matching and dynamic knowledge update.

[0004] In view of this, the present application provides a construction method and system of a medical examination medical knowledge base and a medium to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a construction method of a medical examination medical knowledge base, comprising:

[0006] S1. Collecting medical examination data and patient diagnosis and treatment data of a medical data system to obtain a basic medical data set; performing data cleaning on the basic medical data set to output a structured medical data set;

[0007] S2. Constructing a multi-level medical dictionary, establishing semantic mapping between the structured medical data set and the multi-level medical dictionary, and generating a standard medical data set;

[0008] S3. Performing medical related entity extraction on the standard medical data set to obtain a medical entity set, and simultaneously constructing a language processing model to establish semantic relationships of each medical entity in the medical entity set;

[0009] S4. Performing semantic alignment on the medical entity set to obtain a semantic alignment entity set;

[0010] S5. Constructing a medical examination knowledge rule base based on the semantic alignment entity set, combining the semantic alignment entity set and the medical examination knowledge rule base to construct an initial medical examination knowledge graph;

[0011] S6. Extending the entity association of the initial medical examination knowledge graph, and outputting an enhanced medical examination knowledge graph;

[0012] S7. Connecting the enhanced medical examination knowledge graph with the actual clinical application scene, collecting real-time clinical data; based on the real-time clinical data, incrementally updating the enhanced medical examination knowledge graph to obtain a medical examination knowledge base, and sending the medical examination knowledge base to a medical data system for storage.

[0013] Further, the way of establishing semantic mapping between the structured medical data set and the multi-level medical dictionary comprises:

[0014] Obtaining the medical terminology field, the structure field and the association field of the medical database, and establishing different levels of medical dictionaries based on the medical terminology field, the structure field and the association field, and integrating to obtain a multi-level medical dictionary;

[0015] Classifying the structured medical data set by field, and performing semantic matching on different types of fields layer by layer using the multi-level medical dictionary to output a standard medical data set; when there are fuzzy fields in the structured medical data set, using a pre-trained word vector model to convert the fuzzy fields into fuzzy word vectors; based on the multi-level medical dictionary, performing preliminary fuzzy matching on the fuzzy word vectors to obtain candidate fields of the fuzzy word vectors; by comparison with the multi-level medical dictionary, calculating the vector similarity of each candidate field, screening and mapping fields based on the vector similarity, and using the mapping fields to realize semantic matching; when a single field is matched with multiple fields, performing semantic disambiguation on the single field, if the semantic disambiguation cannot be performed, the single field is determined as a fuzzy field and converted into a fuzzy word vector for semantic matching, otherwise the disambiguated field is directly used for semantic matching.

[0016] Further, the way of extracting medical related entities from the standard medical data set comprises:

[0017] Obtaining medical related entity types based on a medical database, and performing medical related entity annotation on a standard medical dataset based on the medical related entity types; constructing a dynamic sliding window, and performing content scanning on the annotated medical related entities in the standard medical dataset by using the dynamic sliding window; performing entity center field extraction in each dynamic sliding window to obtain an entity theme field; obtaining a neighboring field of the entity theme field based on the position of the entity theme field, and obtaining a medical content segment associated with the annotated medical related entity based on all the fields, i.e., a candidate entity segment of the annotated medical related entity; determining an entity boundary of each annotated medical related entity based on the candidate entity segment, and obtaining a candidate entity interval; adjusting the entity coverage of each annotated medical related entity based on the candidate entity interval, and comparing the annotated medical related entity with a multi-level medical dictionary to obtain a medical entity, and integrating all the medical entities to obtain a medical entity set.

[0018] Further, the manner in which the synchronous language processing model establishes semantic relationships of each medical entity in the medical entity set comprises:

[0019] Constructing a syntax tree structure of data in any one dynamic sliding window, and taking any two medical entities in each syntax tree structure as a medical entity pair; calculating a shortest syntax path of any one medical entity pair, and taking the shortest syntax path as a semantic path of the medical entity pair; extracting all fields and semantic features between the semantic paths, judging the dependency relationship between the fields in the semantic paths, and determining the medical terms corresponding to the semantic features by using a multi-level medical dictionary, and constructing a semantic feature set comprising all the fields, dependency relationships and corresponding medical terms;

[0020] Performing normalization on the semantic feature set to obtain a semantic judgment vector; performing semantic judgment on the medical entity pair by using the language processing model based on the semantic judgment vector, outputting a semantic relationship label of each medical entity pair, and associating the corresponding medical entity by using the semantic relationship label; at the same time, constructing a judgment accuracy function, calculating the accuracy score of the semantic relationship label of each medical entity pair by using the judgment accuracy function, setting an accuracy score threshold, and eliminating the ambiguous semantic relationship label based on the accuracy score and the accuracy score threshold.

[0021] Further, the manner in which the synchronous language processing model establishes semantic relationships of each medical entity in the medical entity set comprises:

[0022] The context content of each medical entity in the medical entity set is scanned to obtain coexistence fields and dependency fields in the context content of each medical entity; a context semantic network is constructed based on the medical entities and the corresponding coexistence fields and dependency fields; a structure vector corresponding to each medical entity in the context semantic network is extracted, and a semantic relationship vector and a field vector of each medical entity are obtained, and the structure vector, the semantic relationship vector and the field vector of each medical entity are fused to obtain a multi-dimensional medical entity fusion vector;

[0023] The attention unit is deployed to perform semantic enhancement on each multi-dimensional medical entity fusion vector to obtain an enhanced medical entity fusion vector; an entity similarity matrix is constructed by calculating the similarity of any two enhanced medical entity fusion vectors; semantic clustering is performed on all entity similarity matrices to obtain a semantic clustering cluster; the medical entities corresponding to all entity similarity matrices in each semantic clustering cluster are detected, and the medical entity with the highest similarity to other medical entities is taken as a standard entity, and all medical entities in the same clustering cluster are standardized based on the standard entity to obtain a standardized clustering cluster; and all standardized clustering clusters are integrated to obtain a semantic alignment entity set.

[0024] Further, the manner of constructing a medical examination knowledge rule library based on the semantic alignment entity set comprises:

[0025] Each semantic alignment entity in the semantic alignment entity set is compared with a multi-level medical dictionary based on semantics, and a hierarchical constraint rule is generated based on the multi-level medical dictionary; a standard medical examination knowledge rule for each semantic alignment entity in the corresponding semantic alignment entity set is designed based on a medical database, and a medical examination knowledge rule library is obtained by integrating the standard medical examination knowledge rule and the hierarchical constraint rule;

[0026] The manner of constructing an initial medical examination knowledge graph in combination with the semantic alignment entity set and the medical examination knowledge rule library comprises:

[0027] The semantic alignment entities in the semantic alignment entity set are matched with the medical examination knowledge rule library, and if an entity combination formed by the semantic alignment entities meets a rule triggering condition set by the medical examination knowledge rule library, an entity multi-tuple is output; all entity multi-tuples are structured and arranged based on semantic relationship labels, an entity triple is output, and a graph structure relationship graph is constructed based on the entity triple; the semantic alignment entities are taken as nodes of the graph structure relationship graph, a rule connection relationship of each entity triple is extracted, and the rule connection relationship is taken as an edge between the corresponding nodes; the graph structure relationship graph is processed for repeated relationships to obtain an initial medical examination knowledge graph, and the medical examination knowledge rule library is taken as a semantic constraint basis of the initial medical examination knowledge graph.

[0028] Further, the manner of extending the initial medical examination knowledge graph includes:

[0029] Performing structural analysis on the initial medical examination knowledge graph, identifying isolated entity nodes and weak connection relationship paths in the initial medical examination knowledge graph, integrating all nodes on the isolated entity nodes and weak connection relationship paths to obtain a set of associated extension candidate nodes, using the medical database as a basis to fill logical entities for the associated extension candidate nodes, and constructing a set of completed nodes based on all nodes related to the filled logical entities, calculating the matching confidence of the completed nodes in the set of completed nodes and the existing semantically aligned entities, setting a confidence threshold, filtering the supplementable nodes based on the confidence threshold and the matching confidence, and adding the supplementable nodes to the initial medical examination knowledge graph to obtain a supplemented medical examination knowledge graph.

[0030] Identifying structurally sparse areas in the supplemented medical examination knowledge graph, and using a pre-trained semantic prediction model to mine semantic relationships in the structurally sparse areas to obtain a set of candidate potential paths, calculating the spatial angle of the candidate potential paths in the vector space, and if the angle is less than a threshold angle, the candidate potential path is considered as a normal relationship path, adding the normal relationship path and all nodes therein to the supplemented medical knowledge graph to obtain an enhanced medical examination knowledge graph.

[0031] Further, the manner of incrementally updating the enhanced medical examination knowledge graph based on real-time clinical data includes:

[0032] Performing medical-related entity extraction and semantic alignment on real-time clinical data to obtain semantically aligned clinical data, while extracting new medical entities in the semantically aligned clinical data during processing, performing rule matching on the new medical entities and the medical examination knowledge rule library, if the new medical entities cannot be matched, the new medical entities are determined as candidate extension entities, otherwise the new medical entities are directly added to the enhanced medical examination knowledge graph and the connection relationship is generated simultaneously, calculating the matching confidence of the new medical entities and the existing semantically aligned entities, constructing an update evaluation model, using the update evaluation model to evaluate the matching effect of the new medical entities, and outputting a matching score, calculating an execution fusion score based on the matching confidence and the matching score, setting a fusion score threshold, and adding the new medical entities with an execution fusion score greater than the fusion score threshold to the enhanced medical examination knowledge graph, obtaining the rule connection relationship of the new medical entities based on the medical database, and generating a medical examination knowledge base.

[0033] A construction system of a medical examination medical knowledge base is used to implement a construction method of a medical examination medical knowledge base, and is characterized by comprising:

[0034] The data acquisition module is used to collect medical test data and patient diagnosis and treatment data from the medical data system to obtain a basic medical data set; the basic medical data set is cleaned and output as a structured medical data set;

[0035] Semantic mapping module, used to build a multi-level medical dictionary, establish semantic mapping between structured medical datasets and multi-level medical dictionaries, and generate standard medical datasets;

[0036] The entity extraction module is used to extract medical-related entities from the standard medical dataset to obtain a medical entity set, and simultaneously build a language processing model to establish the semantic relationship between each medical entity in the medical entity set;

[0037] A semantic alignment module is used to semantically align the medical entity set to obtain a semantically aligned entity set;

[0038] A graph construction module is used to construct a medical laboratory knowledge rule base based on a semantically aligned entity set, and to construct an initial medical laboratory knowledge graph by combining the semantically aligned entity set and the medical laboratory knowledge rule base;

[0039] The association extension module is used to perform entity association extension on the initial medical examination knowledge graph and output an enhanced medical examination knowledge graph;

[0040] The graph optimization module is used to connect the enhanced medical test knowledge graph with actual clinical application scenarios and collect real-time clinical data; incrementally update the enhanced medical test knowledge graph based on real-time clinical data to obtain a medical test knowledge base, and send the medical test knowledge base to the medical data system for storage; each module is connected by wired and / or wireless means.

[0041] A computer-readable storage medium is characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, a method for constructing a medical knowledge base based on medical laboratory testing is implemented.

[0042] The technical effects and advantages of the method, system and medium for constructing a medical knowledge base based on medical laboratory testing of the present invention are as follows:

[0043] Through collecting and processing the data of medical examination scene, the processes of establishing semantic mapping, entity extraction, semantic alignment and knowledge graph are carried out in sequence, and the medical examination knowledge base is constructed by collecting real-time data for updating; not only the analysis accuracy of the collected data is improved, but also the standardization and structuralization of the data are realized, and the readability of the data is improved; the accuracy of medical entity relationship recognition is enhanced, and the accuracy of entity edge connection in the process of knowledge graph construction is significantly improved; at the same time, the standardization of the entity is completed in the process, and the problem of inconsistent entity naming under heterogeneous data sources is solved; through the correlation extension of the knowledge graph, the structure space of the knowledge graph is expanded, more potential semantic paths are mined, and the wide coverage of knowledge representation is realized; finally, the generated medical examination knowledge base has semantic consistency and evolvability, and is convenient for subsequent continuous utilization of real-time data for updating, and provides convenience for medical decision and intelligent reasoning diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 FIG. 1 is a schematic diagram of a medical examination medical knowledge base construction method according to the present application;

[0045] Figure 2 FIG. 2 is a schematic diagram of a medical examination medical knowledge base construction system according to the present application. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0047] Embodiment 1

[0048] Please refer to Figure 1 The medical examination medical knowledge base construction method described in this embodiment includes:

[0049] S1. Collecting medical examination data and patient diagnosis and treatment data of a medical data system to obtain a basic medical data set; performing data cleaning on the basic medical data set to output a structured medical data set;

[0050] S2. Constructing a multi-level medical dictionary, establishing semantic mapping of the structured medical data set and the multi-level medical dictionary, and generating a standard medical data set;

[0051] S3. Performing medical related entity extraction on the standard medical data set to obtain a medical entity set, and simultaneously constructing a language processing model to establish the semantic relationship of each medical entity in the medical entity set;

[0052] S4. performing semantic alignment on the medical entity set to obtain a semantic alignment entity set;

[0053] S5. constructing a medical examination knowledge rule base based on the semantic alignment entity set, combining the semantic alignment entity set and the medical examination knowledge rule base to construct an initial medical examination knowledge graph;

[0054] S6. performing entity association extension on the initial medical examination knowledge graph to output an enhanced medical examination knowledge graph;

[0055] S7. connecting the enhanced medical examination knowledge graph with an actual clinical application scenario, collecting real-time clinical data; based on the real-time clinical data, incrementally updating the enhanced medical examination knowledge graph to obtain a medical examination knowledge base, and sending the medical examination knowledge base to a medical data system for storage.

[0056] Obtain medical examination data and patient diagnosis and treatment data by connecting the hospital laboratory data system; wherein the medical examination data includes common examination type data such as biochemical examination data, blood routine data, blood gas analysis data and hormone determination; the patient diagnosis and treatment data includes text type record data such as patient symptoms, doctor's diagnosis record of disease based on patient's examination data, treatment plan and patient's electronic medical record, etc. Fill in the missing values of the basic medical data set, modify the field misplacement and denoise the data to obtain a more accurate structured medical data set.

[0057] The semantic mapping method for the structured medical data set and the multi-level medical dictionary includes:

[0058] Obtain the medical terminology field, structure field and association field of the medical database, and respectively establish different levels of medical dictionaries based on the medical terminology field, structure field and association field, and integrate to obtain a multi-level medical dictionary. In this embodiment, the medical database is constructed based on existing medical materials and related field expert experience, wherein the medical terminology field includes examination field terminology and disease name terminology, such as "blood sugar content", "ALT" and "glucose content"; the structure field is used to represent the structured attributes of the data, such as the numerical range, unit and reference value of certain examination data; the association field is used to describe the logical relationship between fields, such as the content change of certain examination data indicating the abnormality of certain function or disease; the field that connects multiple fields and outputs a certain conclusion is the association field; based on the medical terminology field, a medical terminology dictionary is constructed, based on the structure field, a structure dictionary is constructed, and based on the association field, an association dictionary is constructed, and a multi-level medical dictionary across levels is integrated.

[0059] The structured medical data set is field classified, different types of fields are matched layer by layer using a multi-level medical dictionary, and a standard medical data set is output. The field classification includes test item fields, disease name fields, unit fields, and numerical value fields, and the like. The multi-level medical dictionary is used for semantic comparison with different types of fields, and the fields are matched layer by layer using the corresponding level of the dictionary according to the level of the field, so that the matching is ordered and efficient, and the matching is not chaotic, which can cause low efficiency. The medical term field is matched first, then the associated field is matched, and finally the structure field is matched. When the structured medical data set has a fuzzy field, the fuzzy field is converted into a fuzzy word vector using a pre-trained word vector model. The fuzzy field represents inaccurate information embodied by the field, such as " ", which mixes medical terms and mark characters together and cannot clearly express the specific meaning of the field. The word vector model in this embodiment is a model trained based on medical corpus, such as a Word2Vec model. The word vector model is used to encode the fuzzy field to capture the context relationship in the medical context, thereby mapping the fuzzy word vector. The fuzzy word vector is preliminarily matched based on the multi-level medical dictionary to obtain the candidate field of the fuzzy word vector. The fuzzy word vector is compared with the vector representation of each field in the multi-level medical dictionary, and the nearest neighbor is selected based on the Euclidean distance between the fuzzy word vector and any one vector. The field with a distance less than a predetermined threshold is determined as the candidate field. The vector similarity of each candidate field is calculated by comparing with the multi-level medical dictionary, and the mapping field is selected based on the vector similarity. The mapping field is used for semantic matching, and the vector similarity is calculated using the cosine similarity calculation method, which is widely used to determine the similarity between two vectors. Based on the preset vector similarity threshold, the candidate field greater than or equal to the threshold is selected as the mapping field. The mapping field is obtained by first performing coarse-grained matching on the fuzzy field and then performing fine-grained selection, which solves the problem of fuzzy fields in establishing semantic mapping and makes the semantic matching more accurate. When a single field is matched with multiple fields, the semantic disambiguation of the single field is performed. If the semantic disambiguation cannot be performed, the single field is determined as a fuzzy field and converted into a fuzzy word vector for semantic matching. Otherwise, the disambiguated field is directly used for semantic matching. There may be ambiguous fields in the semantic matching process, such as "RA" which can represent rheumatoid arthritis and right atrium. At this time, the field needs to be contextually disambiguated to prevent semantic matching errors. In this embodiment, the NLP natural language processing algorithm is used for semantic disambiguation. In the prior art, the NLP algorithm is widely used for processing language type data.

[0060] The medical related entity extraction method for the standard medical data set includes:

[0061] ​The medical database is used to obtain medical entity types, and the standard medical data set is marked with medical entities based on the medical entity types. The medical entity types include disease entities, test item entities, diagnosis record entities, etc. The fields in the standard medical data set are scanned, and appropriate medical entity types are selected for matching. The entity types are used for labeling. A dynamic sliding window is constructed, and the content of the labeled medical entities in the standard medical data set is scanned using the dynamic sliding window. The size of the dynamic sliding window is dynamically adjusted to ensure that each window can include complete text content and will not cause the text to be incomplete due to the presence of only part of a sentence. The text content between the labeled medical entities is scanned by the dynamic sliding window. In this embodiment, each dynamic sliding window includes at least two labeled medical entities. The entity center field is extracted in each dynamic sliding window to obtain the entity theme field. The adjacent field of the entity theme field is obtained based on the position of the entity theme field, and the medical content segment associated with the labeled medical entity is obtained based on all fields, which is the candidate entity segment of the labeled medical entity. Each dynamic sliding window includes the local context content of several labeled medical entities. The entity theme field represents the core semantic part of the labeled medical entity, i.e., a field that can represent a certain medical concept, such as the name of a disease entity. The position of the entity theme field is the row, column, or category position of the field in the original data structure. The sentence or paragraph containing the entity theme field and the adjacent field is taken as the candidate entity segment. The entity boundaries of each labeled medical entity are determined based on the candidate entity segment to obtain the candidate entity interval. The syntax structure, punctuation separator, and unit of each candidate entity segment are detected. The entity start point and the entity end point are marked word by word or sentence by sentence from left to right for each row or the whole, ensuring that the content is complete and the syntax is logically correct. The part between the entity start point and the entity end point is taken as the candidate entity interval. The entity coverage of each labeled medical entity is adjusted based on the candidate entity interval, and the labeled medical entity is compared with the multi-level medical dictionary to obtain the medical entity. The medical entity set is obtained by integrating all medical entities. The content of the candidate entity interval is compared with the terms, associated relationships, and entity corresponding values in the multi-level medical dictionary to determine whether the labeled medical entity to be expressed is complete in the candidate entity interval. The content in the candidate entity interval is processed based on the standard medical concepts in the dictionary to remove redundant content, and the adjusted medical entity is output. The medical entity set is obtained by integrating all adjusted medical entities.

[0062] The way to establish the semantic relationship of each medical entity in the medical entity set by synchronously constructing a language processing model includes:

[0063] Construct a syntactic tree structure of data in any dynamic sliding window, and take any two medical entities in each syntactic tree structure as a medical entity pair. In this embodiment, dependency syntactic analysis is performed on the content in each dynamic sliding window, a syntactic dependency tree corresponding to the content in the window is constructed, and the grammatical structure and modification relationship in the content corresponding to the syntactic dependency tree are identified; each pair of medical entities is extracted as a medical entity pair, which is used as a basic unit for subsequent construction of semantic relationships; the shortest grammatical path of any medical entity pair is calculated, and the shortest grammatical path is used as the semantic path of the medical entity pair. All grammatical paths connecting two medical entities in any medical entity pair are searched in the syntactic dependency tree, and the shortest grammatical path is taken as the semantic path; wherein the grammatical path refers to the word structure that connects the two entities in a sentence or paragraph composed of the two medical entities in each medical entity pair. The path formed by the sentence is constructed; for example, elevated ALT indicates that there may be liver function problems, in which the two entities are ALT and liver function, and the other words are all part of the grammatical path; all fields and semantic features between the semantic paths are extracted, the dependency relationships between the fields in the semantic path are determined, and the medical terms corresponding to the semantic features are determined using a multi-level medical dictionary. A semantic feature set including all fields, dependency relationships and corresponding medical terms is constructed, where the semantic features represent the semantic expression information of each semantic path, including contextual relationships, meaning tendencies and semantic connection types. For example, in the sentence "elevated ALT indicates that there may be liver function problems", "elevated" is a state term and "explanation" is an inference term. The two entities are ALT and liver function. This sentence describes the connection between elevated ALT content and liver function problems; dependency relationships represent grammatical relationships such as subject-predicate, verb-object and attributive-adverbial in the semantic path.

[0064] Normalize the semantic feature set to obtain a semantic judgment vector, encode each semantic feature and perform type conversion to convert it into a vector representation; concatenate all vectors and reduce the dimension to obtain a semantic judgment vector; based on the semantic judgment vector, use the language processing model to perform semantic judgment on the medical entity pair, output the semantic relationship label of each medical entity pair, and use the semantic relationship label to associate the corresponding medical entity. In this embodiment, As the basic structure of the language processing model, the model forward propagates the semantic judgment vector, identifies whether there is a semantic relationship between the two medical entities in the medical entity pair, and outputs the corresponding semantic relationship label. For example, if the two medical entities are a certain test indicator and a certain disease, and the test indicator meets a certain disease, then the semantic relationship label is in the form of "indicator-symptom". At the same time, a judgment accuracy function is constructed, and the judgment accuracy function is used to calculate the accuracy score of the semantic relationship label of each medical entity pair. The accuracy score threshold is set, and the unknown semantic relationship labels are eliminated based on the accuracy score and the accuracy score threshold. The calculation formula of the judgment accuracy function is: ;in, denotes an accuracy score; denotes a matching degree of two medical entities in a semantic relation label, and the matching degree is quantified as a numerical value by querying a medical database; denotes a matching degree of a semantic relation label and a syntactic logic of original content corresponding to each medical entity pair, and the matching degree of the label and the syntactic logic is quantified by a pre-trained BERT model; a semantic relation label with an accuracy score less than a preset accuracy score threshold is determined as an ambiguous semantic relation label.

[0065] The manner of performing semantic alignment on the medical entity set comprises:

[0066] The context content of each medical entity in the medical entity set is scanned to obtain coexisting fields and dependency fields in the context content of each medical entity, and the context content is limited to content within a dynamic sliding window in which each entity is located in the semantic alignment process; the coexisting fields represent fields associated with the medical entity and in the same paragraph as the medical entity, and the dependency fields represent fields capable of reflecting dependency relations, such as modifier words, attribute words, and judgment words; a context semantic network is constructed based on the medical entity and the corresponding coexisting fields and dependency fields, the medical entity and the coexisting fields are taken as nodes of the context semantic network, and the dependency fields and the syntactic dependency relations between the fields are taken as edges of the context semantic network; the context semantic network reflects the semantic adjacency state and the expression syntactic structure of the medical entity in the corpus space; a structure vector corresponding to each medical entity in the context semantic network is extracted, and a semantic relation vector and a field vector of each medical entity are obtained, feature fusion is performed on the structure vector, the semantic relation vector, and the field vector of each medical entity, and a multi-dimensional medical entity fusion vector is obtained, wherein the structure vector represents a structural feature corresponding to a topological position of the medical entity in the context semantic network; the semantic relation vector represents a semantic dependency relation between any one medical entity and other medical entities, such as a logical relation and a causal relation; and the field vector represents attribute information of the field itself, such as a numerical value, a unit, and a label; the multi-dimensional medical entity fusion vector is obtained by vector splicing and fusion coding.

[0067] The deployment attention unit performs semantic enhancement on each multi-dimensional medical entity fusion vector to obtain an enhanced medical entity fusion vector, wherein the algorithm basis of the attention unit is a multi-head attention mechanism. The unit is used to model the global context attention of each multi-dimensional medical entity fusion vector, adjust the semantic expression ability of different feature dimensions through attention weight, and output a context semantic sensitive enhanced medical entity fusion vector. An entity similarity matrix is constructed by calculating the similarity of any two enhanced medical entity fusion vectors. Each element in the matrix is the cosine similarity of any two elements in the two vectors. Semantic clustering is performed on all entity similarity matrices to obtain semantic clustering clusters. In this embodiment, spectral clustering is used as the semantic clustering method. A Laplacian matrix of the graph is constructed based on the entity similarity matrix, and dimensionality reduction is performed on the matrix. The nodes are projected into a new vector space, and then the DBSCAN clustering algorithm is used in the vector space to obtain a plurality of semantic clustering clusters. The medical entities corresponding to all entity similarity matrices in each semantic clustering cluster are detected. The medical entity with the highest similarity to other medical entities is taken as a standard entity. Based on the standard entity, all medical entities in the same cluster are standardized to obtain a standardized cluster. Since medical entities with different names may have the same semantics, entity naming unification is needed. This process is different from the semantic disambiguation in step S2, which is only for a single field. Entity naming unification is to ensure the semantic consistency of entities. In this embodiment, the role of clustering is to cluster medical entities with similar semantics and determine a standard entity, which is used as a unique identifier assigned to all entities in the same cluster, to realize one-to-one aligned semantic mapping. Integrating all standardized clusters obtains a semantic aligned entity set.

[0068] The method for constructing a medical test knowledge rule base based on the semantic aligned entity set includes:

[0069] The semantic alignment entity set is compared with the multi-level medical dictionary, and hierarchical constraint rules are generated based on the multi-level medical dictionary. The different fields in the multi-level medical dictionary have a hierarchical relationship. For example, liver function belongs to biochemical examination in the test data, and ALT belongs to one of the six liver function tests. Therefore, ALT is subordinate to liver function. Three levels of hierarchical constraint rules are designed based on the multi-level medical dictionary, including upper level, sub-level and parallel level. The standard medical test knowledge rules for each semantic alignment entity in the corresponding semantic alignment entity set are designed based on the medical database. The standard medical test knowledge rules and the hierarchical constraint rules are integrated to obtain a medical test knowledge rule library. The medical database includes general medical standard rules, so the standard medical test knowledge rules corresponding to the semantic alignment entity can be directly obtained. The integrated medical test knowledge rule library includes hierarchical relationships, rule trigger conditions and multi-entity association rules, etc. For example, if creatinine is elevated and urine output is reduced, it may be kidney dysfunction. If the blood sugar content is greater than 7.0 mmol / L, it is suspected to be diabetes, etc.

[0070] The initial medical test knowledge graph is constructed by combining the semantic alignment entity set and the medical test knowledge rule library. The construction method includes:

[0071] ​The semantic alignment entities in the semantic alignment entity set are matched with the medical test knowledge rule base, if the entity combination formed by the semantic alignment entities meets the rule trigger condition set by the medical test knowledge rule base, the entity tuple is output, if the entity combination formed by several semantic alignment entities meets the rule trigger condition of the corresponding rule in the rule base, the combination is taken as the entity tuple, for example, creatinine elevation and urine volume reduction may be kidney dysfunction, creatinine elevation, urine volume reduction and kidney dysfunction form an entity tuple; all entity tuples are structured and sorted based on semantic relationship labels, entity triples are output, and a graph structure relationship graph is constructed based on the entity triples, any two entities with semantic relationship in each entity tuple are extracted to construct an entity triple, and the semantic relationship label is taken as the connection condition of the entity triple, the form is entity-semantic relationship label-entity; the semantic alignment entity is taken as the node of the graph structure relationship graph, the rule connection relationship of each entity triple is extracted, and the rule connection relationship is taken as the edge between the corresponding nodes, wherein each edge is used to connect two semantic alignment entities, and the rule connection relationship is embedded at the same time; in this embodiment, the rule connection relationship represents the logical, semantic relationship and rule trigger condition between entities; the graph structure relationship graph is processed for repeated relationship to obtain an initial medical test knowledge graph, and the medical test knowledge rule base is taken as the semantic constraint basis of the initial medical test knowledge graph, the medical test knowledge rule base is taken as the semantic constraint basis to ensure the logical consistency, semantic correctness and medical rationality of the connection between nodes in the graph generation process; the redundant repeated edges in the graph structure relationship graph are removed to ensure the simplicity and rationality of the graph structure.

[0072] The entity association extension mode of the initial medical test knowledge graph includes:

[0073] The initial medical examination knowledge graph is analyzed in structure, isolated entity nodes and weak connection relationship paths in the initial medical examination knowledge graph are identified, in the embodiment, the method for analyzing the structure of the initial medical examination knowledge graph is a node degree analysis method, which is widely used in understanding network graph structure; based on the node degree analysis method, the nodes in the graph that do not form effective connections with other nodes are identified as isolated entity nodes; based on the node degree analysis method, the absolute value of the difference between the degrees of any two nodes is calculated, if the value is greater than the preset degree difference threshold, the edge between the two nodes is determined as a single-sided dependent edge, that is, a weak connection relationship path; all nodes on the isolated entity nodes and the weak connection relationship paths are integrated to obtain a set of associated extension candidate nodes; the medical database is used as a basis to fill the logical entities of the associated extension candidate nodes, and based on all nodes related to the filled logical entities, a set of completed nodes is constructed, wherein since the medical entities corresponding to the associated extension candidate nodes may not exist in the graph, the medical entities that the associated extension candidate nodes may exist in the graph are inferred based on the hierarchical relationship of the entities in the medical database, the corresponding standard medical examination knowledge rules and entity semantic information, that is, the logical entities; the matching confidence of the completed nodes in the completed node set and the existing semantic alignment entities is calculated, a confidence threshold is set, and based on the confidence threshold and the matching confidence, the supplementable nodes are screened and added to the initial medical examination knowledge graph to obtain a supplemented medical examination knowledge graph, wherein the calculation formula of the matching confidence is: ; wherein, represents the matching confidence of any one completed node and the existing semantic alignment entity. represents the cosine similarity between any one completed node and any one existing semantic alignment entity. represents the field similarity between any one completed node and any one existing semantic alignment entity, which is output by a pre-trained BERT model. represents the co-occurrence frequency of any one completed node and any one existing semantic alignment entity, if the co-occurrence frequency is high, it means that the association is close, otherwise it means that the association is weak. , and are weights of , and , respectively, in the embodiment , , .

[0074] The structural sparse area in the supplementary medical examination knowledge graph is identified, and a pre-trained semantic prediction model is used to mine the semantic relationship of the structural sparse area to obtain a candidate potential path set. A node quantity threshold is set, and the area with a node quantity less than the node quantity threshold is regarded as a structural sparse area. A BERT model trained based on a medical corpus is used as the basic structure of the semantic prediction model. A path that is potential but does not constitute an edge is extracted through semantic relationship mining, that is, a candidate potential path. The spatial angle of the candidate potential path in the vector space is calculated. If the spatial angle is less than a preset angle threshold, the candidate potential path is regarded as a normal relationship path. A graph embedding representation method is used to map each candidate potential path to a vector space. The spatial angle of each candidate potential path with an existing path in the graph is calculated. The semantic direction consistency is judged based on the spatial angle. The smaller the spatial angle, the higher the similarity of the semantic direction with the existing path. The normal relationship path and all nodes in the normal relationship path are added to the supplementary medical knowledge graph to obtain an enhanced medical examination knowledge graph. The edges of the normal relationship path and all nodes involved in the normal relationship path are added to the supplementary medical knowledge graph to expand the entity node set and the edge structure of the graph.

[0075] The incremental updating mode of the enhanced medical examination knowledge graph based on real-time clinical data comprises:

[0076] The medical related entity extraction and semantic alignment of real-time clinical data are performed to obtain semantic alignment clinical data, and new medical entities in the semantic alignment clinical data are extracted in the processing. The real-time clinical data represent data extracted in real time based on an actual clinical application scenario, including the latest patient medical test data and diagnosis and treatment data, and the new medical entities are obtained through preprocessing and entity extraction and semantic alignment. The new medical entities are matched with a medical test knowledge rule base. If the new medical entities cannot be matched, the new medical entities are determined as candidate extended entities. Otherwise, the new medical entities are directly added to the enhanced medical test knowledge graph and a connection relationship is generated synchronously. The fact that the new medical entities cannot be matched with the medical test knowledge rule base indicates that the rules related to the new medical entities are not in the existing medical test knowledge rule base, and the new medical entities are helpful for subsequent graph expansion. If the new medical entities can be matched, it indicates that the new medical entities have a connection relationship with some entities in the existing graph, and therefore the new medical entities can be directly added to the graph. The connection relationship between the new medical entities and other existing entities is generated based on the existing medical test knowledge rule base. The matching confidence of the new medical entities and the existing semantic alignment entities is calculated. The calculation method and steps are the same as those in step S6 when the logical entities are added, but the purposes of the two are different. In step S6, the existing entities are filled in to enhance the structural integrity of the graph. In the present operation, the new knowledge is determined whether to be absorbed to improve the expansibility and timeliness of the graph. An update evaluation model is constructed, and the update evaluation model is used to evaluate the matching effect of the new medical entities. The matching score is output. The update evaluation model is a machine learning model, and a random forest supervised learning model is used as the basic structure of the update evaluation model. The model is widely used in data classification in the prior art, and is used in the present operation to identify the new medical entities that can be matched and output the corresponding probability value. The value is the matching score. The execution fusion score is calculated based on the matching confidence and the matching score. The new medical entities with an execution fusion score greater than a fusion score threshold are added to the enhanced medical test knowledge graph. The execution fusion score is obtained by weighted summation of the matching confidence and the matching score. In the present embodiment, the weight of the matching confidence is , and the weight of the matching score is . The rule connection relationship of the new medical entities is obtained based on a medical database, a medical test knowledge base is generated, and the related rules corresponding to the new medical entities are obtained based on the medical database after the new medical entities are added to the enhanced medical test knowledge graph. The rule connection relationship is constructed to obtain an updatable enhanced medical test knowledge graph. The nodes, edges and connection relationships in the updatable enhanced medical test knowledge graph are converted into manageable metadata items through metadata modeling of the updatable enhanced medical test knowledge graph. The metadata items are structured and output to form the medical test knowledge base.

[0077] The embodiment collects and processes data of a medical examination scene, and sequentially performs the processes of establishing semantic mapping, entity extraction, semantic alignment and knowledge graph establishment, and constructs a medical examination knowledge base by collecting real-time data for updating; not only improves the analysis accuracy of the collected data, but also realizes the standardization and structuralization of the data, improves the readability of the data, enhances the accuracy of medical entity relationship recognition, significantly improves the accuracy of entity edge connection in the knowledge graph construction process, and realizes the wide coverage of knowledge representation; and finally generates a medical examination knowledge base with semantic consistency and evolvability, which is convenient for subsequent continuous use of real-time data for updating, and provides convenience for medical decision and intelligent reasoning diagnosis.

[0078] Embodiment 2

[0079] Please refer to Figure 2 The embodiment does not describe some parts in detail, which are described in embodiment 1, and provides a construction system of a medical examination medical knowledge base, which comprises:

[0080] A data collection module is configured to collect medical examination data and patient diagnosis and treatment data of a medical data system to obtain a basic medical data set, and perform data cleaning on the basic medical data set to output a structured medical data set;

[0081] A semantic mapping module is configured to construct a multi-level medical dictionary, establish semantic mapping between the structured medical data set and the multi-level medical dictionary, and generate a standard medical data set;

[0082] An entity extraction module is configured to extract medical related entities from the standard medical data set to obtain a medical entity set, and simultaneously construct a language processing model to establish semantic relationships of each medical entity in the medical entity set;

[0083] A semantic alignment module is configured to perform semantic alignment on the medical entity set to obtain a semantic alignment entity set;

[0084] A graph construction module is configured to construct a medical examination knowledge rule base based on the semantic alignment entity set, and construct an initial medical examination knowledge graph by combining the semantic alignment entity set and the medical examination knowledge rule base;

[0085] An association extension module is configured to perform entity association extension on the initial medical examination knowledge graph to output an enhanced medical examination knowledge graph;

[0086] A graph optimization module is configured to connect the enhanced medical examination knowledge graph with actual clinical application scenarios, collect real-time clinical data, incrementally update the enhanced medical examination knowledge graph based on the real-time clinical data, obtain a medical examination knowledge base, and send the medical examination knowledge base to a medical data system for storage.

[0087] The above merely provides preferred embodiments of the present application but should not be used to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, the technical solutions recorded in the foregoing embodiments can be modified or some technical features can be replaced by equivalent features by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0088] It should be noted that, in this document, the terms "comprise", "comprising", or any other variant thereof are intended to cover non-exclusive inclusions, so that processes, methods, articles, or devices that comprise a list of elements not only include those elements, but also include other elements that are not expressly listed, or other elements inherent to such processes, methods, articles, or devices. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element.

[0089] In the description of the present application, it should be understood that the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0090] In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0091] In the description of the present application, the meaning of "several" is one or more, and the meaning of "a large number" is two or more.

[0092] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0093] The formula of the present specification is a value calculated by de-dimensioning, the formula is obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and preset parameters and threshold values in the formula are set by a person skilled in the art according to actual conditions.

[0094] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the claims and their equivalents.

Claims

1. A method for constructing a medical laboratory knowledge base, characterized in that: include: S1. Collect medical test data and patient diagnosis and treatment data from the medical data system to obtain a basic medical data set; Perform data cleaning on basic medical datasets to output structured medical datasets; S2. Build a multi-level medical dictionary, establish a semantic mapping between the structured medical dataset and the multi-level medical dictionary, and generate a standard medical dataset; S3. Extract medical-related entities from a standard medical dataset to obtain a medical entity set. Simultaneously, a language processing model is constructed to establish the semantic relationships between the medical entities in the medical entity set. S4. semantically align the medical entity set to obtain a semantically aligned entity set; S5. Build a medical laboratory knowledge rule base based on the semantically aligned entity set, and combine the semantically aligned entity set and the medical laboratory knowledge rule base to build an initial medical laboratory knowledge graph; S6. Extend the initial medical examination knowledge graph by entity association and output an enhanced medical examination knowledge graph; S7. Connect the enhanced medical laboratory knowledge graph with actual clinical application scenarios to collect real-time clinical data; The enhanced medical examination knowledge graph is incrementally updated based on real-time clinical data to obtain a medical examination knowledge base, which is then sent to a medical data system for storage.

2. A method for constructing a medical laboratory medicine knowledge base according to claim 1, characterized in that: The method of establishing semantic mapping between structured medical data sets and multi-level medical dictionaries includes: Obtain medical terminology fields, structure fields, and association fields of a medical database, and establish medical dictionaries of different levels based on the medical terminology fields, structure fields, and association fields, and integrate them to obtain a multi-level medical dictionary; The fields of structured medical datasets are classified, and multi-level medical dictionaries are used to perform semantic matching of different types of fields layer by layer to output a standard medical dataset. When fuzzy fields exist in the structured medical dataset, the fuzzy fields are converted into fuzzy word vectors using a pre-trained word vector model. Preliminary fuzzy matching is performed on the fuzzy word vectors based on the multi-level medical dictionary to obtain candidate fields for the fuzzy word vectors. By comparing with the multi-level medical dictionary, the vector similarity of each candidate field is calculated, and the mapping fields are screened based on the vector similarity, and semantic matching is achieved using the mapping fields. When a single field is semantically matched and corresponds to multiple fields at the same time, semantic disambiguation is performed on the single field. If semantic disambiguation is not possible, the single field is judged as a fuzzy field and converted into a fuzzy word vector for semantic matching. Otherwise, the disambiguated field is directly semantically matched.

3. The method for constructing a medical laboratory medicine knowledge base according to claim 2, characterized in that: The method of extracting medical-related entities from a standard medical dataset includes: Medical-related entity types are obtained based on the medical database, and medical-related entities are annotated in the standard medical dataset based on the medical-related entity types; a dynamic sliding window is constructed, and the annotated medical-related entities in the standard medical dataset are scanned for content using the dynamic sliding window; entity center fields are extracted within each dynamic sliding window to obtain entity subject fields; neighboring fields of the entity subject field are obtained based on the position of the field, and medical content fragments associated with the annotated medical-related entity are obtained based on all fields, which are candidate entity fragments of the annotated medical-related entity; the entity boundary of each annotated medical-related entity is determined based on the candidate entity fragments to obtain the candidate entity interval; the entity coverage of each annotated medical-related entity is adjusted based on the candidate entity interval, and a multi-level medical dictionary is used to compare with the annotated medical-related entity to obtain the medical entity, and all medical entities are integrated to obtain a medical entity set.

4. The method for constructing a medical laboratory medicine knowledge base according to claim 3, characterized in that: The method of synchronously constructing a language processing model to establish semantic relationships between medical entities in the medical entity set includes: Construct a syntactic tree structure for the data within any dynamic sliding window, and treat any two medical entities in each syntactic tree structure as a medical entity pair; calculate the shortest syntactic path for any medical entity pair, and use this shortest syntactic path as the semantic path of the medical entity pair; extract all fields and semantic features between the semantic paths, determine the dependencies between the fields in the semantic paths, and use a multi-level medical dictionary to determine the medical terms corresponding to the semantic features, and construct a semantic feature set that includes all fields, dependencies, and corresponding medical terms; The semantic feature set is normalized to obtain a semantic judgment vector; based on the semantic judgment vector, a language processing model is used to perform semantic judgment on medical entity pairs, and the semantic relationship labels of each medical entity pair are output. The corresponding medical entities are associated with the semantic relationship labels. At the same time, a judgment accuracy function is constructed, and the accuracy score of the semantic relationship label of each medical entity pair is calculated using the judgment accuracy function. An accuracy score threshold is set, and unclear semantic relationship labels are eliminated based on the accuracy score and the accuracy score threshold.

5. The method for constructing a medical laboratory medicine knowledge base according to claim 4, characterized in that: The method of semantically aligning the medical entity set includes: The context content of each medical entity in the medical entity set is scanned to obtain the co-existing fields and dependent fields in the context content of each medical entity; a context semantic network is constructed based on the medical entities and the corresponding co-existing fields and dependent fields; the structural vector corresponding to each medical entity in the context semantic network is extracted, and the semantic relationship vector and field vector of each medical entity are obtained at the same time. The structural vector, semantic relationship vector and field vector of each medical entity are feature fused to obtain a multidimensional medical entity fusion vector; Attention units are deployed to perform semantic enhancement on each multidimensional medical entity fusion vector to obtain an enhanced medical entity fusion vector; an entity similarity matrix is ​​constructed by calculating the similarity between any two enhanced medical entity fusion vectors; all entity similarity matrices are semantically clustered to obtain semantic clustering clusters; the medical entities corresponding to all entity similarity matrices in each semantic clustering cluster are detected, and the medical entity with the highest similarity to other medical entities is used as the standard entity. Based on the standard entity, all medical entities in the same cluster are entity standardized to obtain standardized clustering clusters; all standardized clustering clusters are integrated to obtain a set of semantically aligned entities.

6. The method for constructing a medical laboratory knowledge base according to claim 5, characterized in that: The method of constructing a medical examination knowledge rule base based on a semantically aligned entity set includes: Each semantically aligned entity in the semantically aligned entity set is semantically compared with a multi-level medical dictionary, and hierarchical constraint rules are generated based on the multi-level medical dictionary. Based on the medical database, a standard medical laboratory knowledge rule corresponding to each semantically aligned entity in the semantically aligned entity set is designed, and the standard medical laboratory knowledge rule and the hierarchical constraint rules are integrated to obtain a medical laboratory knowledge rule base. The methods of combining the semantically aligned entity set and the medical examination knowledge rule base to construct the initial medical examination knowledge graph include: The semantically aligned entities in the semantically aligned entity set are matched with the medical laboratory test knowledge rule base. If the entity combination composed of the semantically aligned entities meets the rule triggering conditions set by the medical laboratory test knowledge rule base, the entity tuple is output; all entity tuples are structured and organized based on the semantic relationship labels, and entity triples are output, and a graph structure relationship map is constructed based on the entity triples; the semantically aligned entities are used as nodes of the graph structure relationship map, the rule connection relationship of each entity triple is extracted, and the rule connection relationship is used as the edge between the corresponding nodes; the graph structure relationship map is subjected to repeated relationship processing to obtain the initial medical laboratory test knowledge map, and the medical laboratory test knowledge rule base is used as the semantic constraint basis of the initial medical laboratory test knowledge map.

7. The method for constructing a medical laboratory knowledge base according to claim 6, characterized in that: The method of extending entity association of the initial medical examination knowledge graph includes: Perform structural analysis on the initial medical test knowledge graph to identify isolated entity nodes and weakly connected relationship paths in the initial medical test knowledge graph; integrate all nodes on the isolated entity nodes and weakly connected relationship paths to obtain a set of associated extension candidate nodes; use the medical database as a basis for judgment, perform logical entity filling on the associated extension candidate nodes, and construct a complete node set based on all nodes related to the filled logical entities; calculate the matching confidence between the complete nodes in the complete node set and the existing semantically aligned entities, set a confidence threshold, and screen the nodes that can be supplemented based on the confidence threshold and the matching confidence and add them to the initial medical test knowledge graph to obtain a supplementary medical test knowledge graph; Identify the structural sparse areas in the supplementary medical examination knowledge graph, and use the pre-trained semantic prediction model to perform semantic relationship mining on the structural sparse areas to obtain a set of candidate potential paths; calculate the spatial angle of the candidate potential path in the vector space, and if it is less than the angle threshold, the candidate potential path is regarded as a normal relationship path; add the normal relationship path and all the nodes therein to the supplementary medical examination knowledge graph to obtain the enhanced medical examination knowledge graph.

8. The method for constructing a medical laboratory medicine knowledge base according to claim 7, characterized in that: The method of incrementally updating the enhanced medical test knowledge graph based on real-time clinical data includes: Medical-related entities are extracted and semantically aligned on real-time clinical data to obtain semantically aligned clinical data. At the same time, new medical entities in the semantically aligned clinical data are extracted during the processing process; the new medical entities are matched with the medical laboratory knowledge rule base. If no match is found, the new medical entities are judged as candidate extension entities. Otherwise, they are directly added to the enhanced medical laboratory knowledge graph and a connection relationship is generated simultaneously; the matching confidence of the new medical entities and the existing semantically aligned entities is calculated; an update evaluation model is constructed, and the matching effect of the new medical entities is evaluated using the update evaluation model, and a matching score is output; the execution fusion score is calculated based on the matching confidence and matching score, and a fusion score threshold is set. The new medical entities with an execution fusion score greater than the fusion score threshold are added to the enhanced medical laboratory knowledge graph; the rule connection relationship of the new medical entities is obtained based on the medical database to generate a medical laboratory knowledge base.

9. A construction system based on a medical laboratory medicine knowledge base, which is used to implement a construction method based on a medical laboratory medicine knowledge base according to any one of claims 1 to 8, characterized in that: include: The data acquisition module is used to collect medical test data and patient diagnosis and treatment data from the medical data system to obtain basic medical data sets; Perform data cleaning on basic medical datasets to output structured medical datasets; Semantic mapping module, used to build a multi-level medical dictionary, establish semantic mapping between structured medical datasets and multi-level medical dictionaries, and generate standard medical datasets; The entity extraction module is used to extract medical-related entities from the standard medical dataset to obtain a medical entity set, and simultaneously build a language processing model to establish the semantic relationship between each medical entity in the medical entity set; A semantic alignment module is used to semantically align the medical entity set to obtain a semantically aligned entity set; A graph construction module is used to construct a medical laboratory knowledge rule base based on a semantically aligned entity set, and to construct an initial medical laboratory knowledge graph by combining the semantically aligned entity set and the medical laboratory knowledge rule base; The association extension module is used to perform entity association extension on the initial medical examination knowledge graph and output an enhanced medical examination knowledge graph; The graph optimization module is used to connect the enhanced medical test knowledge graph with actual clinical application scenarios and collect real-time clinical data; The enhanced medical examination knowledge graph is incrementally updated based on real-time clinical data to obtain a medical examination knowledge base, which is then sent to a medical data system for storage; each module is connected via wired and / or wireless means.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a medical knowledge base based on medical laboratory testing according to any one of claims 1 to 8 is implemented.

Images