A Clinical Knowledge Retrieval and Suggestion Generation Method and System Based on Knowledge Graph
By splitting the records of patients who visited the same clinic, tracing the content formation path of existing decision data, identifying isolated and restored data, and forming a clinical fact subgraph for semantic retrieval, the problem of repeated participation in existing decisions and the easy exclusion of independent and objective results is solved, resulting in more independent and objective clinical knowledge retrieval results.
Patent Information
- Application Number
- CN202611343235.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-01
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, when patient records from the same visit are continuously updated, the repeated involvement of clinical decision-making affects the knowledge retrieval results, and subsequent independent and objective results are easily excluded, making it difficult to balance independence and integrity.
By collecting patient records from the same visit, data is divided into clinical event data and report field data according to its source. The content formation path of existing decision data is traced to identify isolated data and recovered data. A clinical fact subgraph is formed for semantic retrieval, excluding isolated data and retaining recovered data.
It improves the independence of clinical knowledge retrieval input and the objectivity of retrieval results, reduces the self-backfeeding effect of existing decisions on knowledge retrieval results, reduces the recall of duplicate or biased knowledge items, and improves the reference effectiveness of clinical knowledge suggestions.
Smart Images

Figure CN123115186A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semantic retrieval technology, and more specifically, to a method and system for clinical knowledge retrieval and suggestion generation based on knowledge graphs. Background Technology
[0002] With the development of electronic medical records, clinical decision support systems, and medical knowledge graphs, clinical knowledge retrieval has gradually shifted from simple keyword matching to medical entity recognition, standard medical concept mapping, and semantic relationship querying. These systems typically extract medical information such as symptoms, diseases, examinations, tests, and treatments from patient medical records or user queries, and utilize medical knowledge graphs to connect medical concepts, clinical knowledge, and knowledge sources, thereby providing medical personnel with clinical knowledge suggestions relevant to the current patient information. Since patient records from the same visit continuously increase with the process of consultation, examinations, tests, diagnosis, and treatment, how to form retrieval input suitable for the current clinical knowledge request from these constantly updated patient records directly affects the clinical reference value of the knowledge retrieval results.
[0003] In the prior art, Chinese invention patent application CN121565500A discloses a clinical decision-making interaction method and system based on structured evidence reasoning. It discloses the process of medical entity recognition, conflict resolution, and entity standardization of user input, querying a structured knowledge graph based on the standardized core entities and target intent, and generating answers based on the retrieved knowledge fragments. Its main focus is on entity recognition, intent judgment, and knowledge matching within the query content. When the search input comes from continuously updated patient records from the same visit, the patient's original symptoms, examination and test results, and existing diagnoses, requests, and treatment records may all participate in subsequent searches. Already formed clinical decisions may again become the basis for similar knowledge searches, causing the search results to be repeatedly affected by existing decision content. Furthermore, if the use of data added later during the visit is uniformly restricted, clinically valuable information may be missed, making it difficult to balance independence and completeness in the information upon which subsequent knowledge searches are based.
[0004] Therefore, it is necessary to design a clinical knowledge retrieval and suggestion generation method and system based on knowledge graphs to solve the problems existing in the current technology. Summary of the Invention
[0005] In view of this, the present invention proposes a clinical knowledge retrieval and suggestion generation method and system based on knowledge graph, which aims to solve the problem that existing clinical decisions and their derived data are repeatedly involved in knowledge retrieval during the same medical visit, while subsequent independent objective results are easily excluded.
[0006] This invention proposes a clinical knowledge retrieval and suggestion generation method based on knowledge graphs, including: Collect patient records from the same visit and receive clinical knowledge requests. Segment the patient records according to their origin to obtain clinical event data and report field data. Based on the clinical knowledge request, a target suggestion type is determined, existing decision data consistent with the target suggestion type is extracted from the clinical event data, and the existing decision data is traced to the content formation path of the clinical event data and the report field data; The existing decision data and the content formation path that includes copying, referencing, or rewriting the existing decision data are identified as isolated data; the report field data that is triggered by the existing decision data, whose content formation path does not include copying, referencing, or rewriting the existing decision data, and which is generated by an independent objective data source are identified as restored data. Excluding isolated data and decision-triggered event data, the recovery data and other clinical event data are mapped to standard medical concepts to form a clinical fact subgraph; Semantic retrieval of the clinical knowledge graph is performed based on the clinical fact subgraph to obtain clinical knowledge items, and clinical knowledge suggestions are generated based on the clinical knowledge items.
[0007] Furthermore, when extracting existing decision data, this includes: Based on the type mapping relationship, determine the target decision type that is consistent with the target suggestion type; The target decision type data generated during the same medical visit at a time earlier than the time the clinical knowledge request was received and which is in a valid state is identified as the existing decision data.
[0008] Furthermore, when classifying the report field data, the following are included: Identify examination or testing requests from the clinical event data to obtain request event data; The report field data brought in by the application event data is determined as the application inheritance field, the report field data with the original result identifier and generated by the independent objective data source is determined as the objective result field, and the report field data that simultaneously references the application inheritance field and the objective result field is determined as the comprehensive conclusion field.
[0009] Furthermore, when establishing the content formation path, it includes: Based on the generation records in the application event data, an acquisition trigger relationship is established from the existing decision data to the application event data; based on the application form identifier, an acquisition trigger relationship is established from the application event data to the report field data; based on the source field identifier, a content derivation relationship is established from the referenced data to the report field data; based on the original result identifier, a result generation relationship is established from the independent objective data source to the objective result field; the acquisition trigger relationship, the content derivation relationship, and the result generation relationship are connected to obtain the content formation path.
[0010] Furthermore, determining the isolated data includes: for the clinical event data or the report field data, if any content formation path from the existing decision data to the clinical event data or the report field data contains the content derivation relationship, the clinical event data or the report field data is determined as the isolated data.
[0011] Furthermore, when tracing the indirect content derivation relationships of the reported field data, this includes: When the source field identifier points to another report field data, the data is traced level by level along the source field identifiers. When any tracing branch reaches the existing decision data or the application inheritance field, the report field data of the tracing starting point is determined as the isolated data. When none of the tracing branches reach the existing decision data or the application inheritance field, the report field data of the tracing starting point is determined according to the original result identifier to see if it meets the recovery data determination condition.
[0012] Furthermore, when determining the recovered data, the following criteria are included: the entire content formation path of the report field data does not contain the content derivation relationship; the report field data is associated with the application event data through the acquisition trigger relationship and with the independent objective data source through the result generation relationship; and the report field data has the original result identifier. In these cases, the report field data is determined as the recovered data.
[0013] Furthermore, when processing the report field data in the same inspection report or test report, the following steps are taken: when the same inspection report or test report contains the application inheritance field, the objective result field, and the comprehensive conclusion field, the application inheritance field that can be traced back to the existing decision data and the comprehensive conclusion field that references the application inheritance field are determined as the isolated data, and the objective result field that meets the data recovery determination conditions is determined as the recovered data.
[0014] Furthermore, when forming a clinical fact subgraph, the following is included: If the same standard medical concept is semantically mapped to both the isolated data and the recovered data, only the semantic mapping between the recovered data and the standard medical concept is retained. Standard medical concepts that are semantically mapped only to the isolated data are not included in the clinical fact subgraph.
[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: By splitting patient records from the same visit according to their origin, the original clinical events, examination and test report fields, and decision data generated during the visit are processed separately. Based on the clinical knowledge request, the target suggestion type for this round of retrieval is determined, thus targeting only existing decision data that may affect similar knowledge retrieval. By tracing the content formation path from existing decision data to clinical event data and report field data, existing decision data and data formed by copying, referencing, or rewriting existing decision content are identified as isolated data. This prevents the repetitive reinforcement formed by existing decision content such as diagnoses, applications, or treatments re-entering similar semantic retrieval through medical record write-back, field inheritance, or report conclusions. Simultaneously, the existing... Report field data obtained through decision-triggered events but whose content is not affected by existing decisions and generated from independent objective data sources are identified as recovered data. This avoids the loss of information with practical retrieval value, such as examination values, test results, and independent measurement results, due to the uniform exclusion of subsequent records. After excluding isolated data and decision-triggered event data, the recovered data and other clinical event data are used to form a clinical fact subgraph. This ensures that the input for semantic retrieval does not contain content derived from existing decisions, while retaining objective facts generated independently later. This reduces the self-backfeeding effect of existing decisions on knowledge retrieval results, reduces the recall of duplicate or biased knowledge items, and improves the independence of clinical knowledge retrieval input, the relevance of retrieval results to objective clinical facts, and the reference effectiveness of clinical knowledge suggestions in scenarios of continuously updated patient records.
[0016] On the other hand, this application also provides a knowledge graph-based clinical knowledge retrieval and suggestion generation system for performing the aforementioned knowledge graph-based clinical knowledge retrieval and suggestion generation method, including: The data acquisition module is used to acquire patient records and clinical knowledge requests; The path processing module is used to extract existing decision data, establish content formation paths, and determine isolated and restored data. The subgraph forming module is used to exclude the isolated data and the clinical event data triggered by the existing decision data, and to form a clinical fact subgraph based on the recovery data and the remaining clinical event data. The suggestion generation module is used to retrieve clinical knowledge graphs based on the clinical fact subgraph and generate clinical knowledge suggestions.
[0017] It is understandable that the aforementioned knowledge graph-based clinical knowledge retrieval and suggestion generation methods and systems have the same beneficial effects, and will not be elaborated further here. Attached Figure Description
[0018] The accompanying drawings are used to illustrate the technical solutions of the present invention and do not constitute a limitation on the scope of protection of the present invention. The process nodes, regional relationships, and functional block diagrams in the drawings are used to illustrate the implementation methods of the embodiments of the present invention. In the drawings: Figure 1 A flowchart illustrating a knowledge graph-based clinical knowledge retrieval and suggestion generation method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process of determining isolated data and recovery data as provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the content formation path provided in the embodiments of the present invention; Figure 4 This is a functional block diagram of a knowledge graph-based clinical knowledge retrieval and suggestion generation system provided in an embodiment of the present invention. Detailed Implementation
[0019] The specific embodiments of the present invention will now be described with reference to the accompanying drawings. These embodiments are used to illustrate the technical solutions of the present invention and do not limit the scope of protection of the present invention. Unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0020] In some embodiments of this application, see Figure 1-3 As shown, this application proposes a clinical knowledge retrieval and suggestion generation method based on knowledge graphs, including: S100: Collect patient records from the same visit and receive clinical knowledge requests. Segment the patient records according to their origin to obtain clinical event data and report field data.
[0021] S200: Determine the target recommendation type based on the clinical knowledge request, extract existing decision data consistent with the target recommendation type from the clinical event data, and trace the content formation path of the existing decision data to the clinical event data and report field data.
[0022] S300: Existing decision data and data whose content creation path includes copying, referencing, or rewriting existing decision data are identified as isolated data. Report field data that is triggered by existing decision data, whose content creation path does not include copying, referencing, or rewriting existing decision data, and which is generated from an independent objective data source are identified as restored data.
[0023] S400: Exclude isolated data and decision-triggered event data, map recovery data and other clinical event data to standard medical concepts and form a clinical fact subgraph.
[0024] S500: Perform semantic retrieval on the clinical knowledge graph based on the clinical fact subgraph to obtain clinical knowledge items, and generate clinical knowledge suggestions based on the clinical knowledge items.
[0025] Specifically, this embodiment uses the example of initiating a diagnostic clinical knowledge request again during an outpatient visit to illustrate the method of clinical knowledge retrieval and suggestion generation based on knowledge graphs. The same visit is limited by a visit identifier. Chief complaint, present illness history, physical examination, diagnosis, examination requests, laboratory requests, and treatment records generated under the same visit identifier are considered clinical event data. The independent fields in the examination and laboratory reports are considered report field data. Each clinical event data record at least includes the event identifier, event type, generation time, data status, and generation record. Each report field data record at least includes the report identifier, field identifier, field type, field value, source field identifier, and original result identifier. The source field identifier is used to locate upstream data that has been copied, referenced, or modified. The original result identifier is used to locate the original result record generated by the examination equipment, laboratory equipment, or independent measurement terminal. If the source field identifier cannot locate the upstream data, or the original result identifier cannot locate the original result record, the corresponding report field data is not determined as recovered data.
[0026] After collecting patient records from the same visit and receiving clinical knowledge requests, the patient records are broken down according to their data origin. Records formed by patient statements, medical staff observations, or clinical procedures are identified as clinical event data. Independent fields carried by examination or laboratory reports are identified as report field data. Fields in reports such as clinical diagnoses, examination purposes, and medical history summaries, which are brought in by examination or laboratory requests, are identified as request-inherited fields. Fields with original result identifiers that can locate independent objective data sources are identified as objective result fields. Examination impressions, laboratory conclusions, or comprehensive descriptions formed by referencing both request-inherited fields and objective result fields are identified as comprehensive conclusion fields. Examination requests, laboratory requests, or treatment records directly triggered by existing decision data are identified as decision-triggered event data. Decision-triggered event data is used to establish content formation paths and is not considered as clinical facts in this round of semantic retrieval.
[0027] The decision-triggered event data referred to in this application refers to clinical event data that is directly triggered by existing decision data and contains event identifiers stored in the generation record. Decision-triggered event data is used to connect existing decision data with subsequent report field data and does not participate in standard medical concept mapping. The remaining clinical event data referred to in this application refers to clinical event data other than existing decision data, isolated data, and decision-triggered event data.
[0028] The target suggestion type is determined based on the clinical knowledge request. A pre-defined type mapping relationship is established between the target suggestion type and the target decision type; for example, diagnostic clinical knowledge requests are mapped to diagnostic types, examination clinical knowledge requests are mapped to examination request types, and treatment clinical knowledge requests are mapped to treatment types. Existing decision data is obtained by filtering clinical event data whose generation time is earlier than the time the clinical knowledge request was received, whose type matches the target decision type, and whose status is valid. A valid status means that the data has not been withdrawn, invalidated, or replaced by a subsequent decision of the same type when the clinical knowledge request was received. Withdrawn or replaced data is not used as the starting point for this content formation path.
[0029] When multiple target decision type data points meet the aforementioned conditions exist during the same medical visit, each target decision type data point is identified as existing decision data and used as the starting point for tracing the content formation path. If the content formation path from any existing decision data point to the corresponding clinical event data or report field data meets the criteria for isolated data, the corresponding clinical event data or report field data is identified as isolated data.
[0030] When establishing a content formation path, firstly, based on the generation records in the application event data, determine whether the application event data was triggered by existing decision data. If the generation record stores the event identifier of existing decision data, then an acquisition trigger relationship is established between the existing decision data and the application event data. Secondly, based on the application form identifier, associate the application event data with the subsequent inspection report or test report, and establish an acquisition trigger relationship between the application event data and the report field data. When the source field identifier points to existing decision data, application event data, or another report field data, a content derivation relationship is established between the pointed-to data and the report field data. When the original result identifier can locate the original result generated by an independent objective data source, a result generation relationship is established between the independent objective data source and the objective result field. An acquisition trigger relationship only indicates that an existing decision causes subsequent data acquisition behavior, and does not indicate that the existing decision determines the specific content of the subsequent objective result. A content derivation relationship indicates that the semantic content of downstream data originates from upstream data. A result generation relationship indicates that the field value directly originates from the objective measurement or inspection result.
[0031] When report field data has multiple source field identifiers, a separate tracing branch is established for each source field identifier, and tracing proceeds level by level along each source field identifier. When a tracing branch reaches existing decision data or a requested inheritance field, the content derivation relationship of the corresponding tracing branch is determined. When a tracing branch reaches an objective result field, cannot continue locating upstream data, or reaches data with an already recorded access identifier again, the current tracing branch is terminated. When any tracing branch reaches existing decision data or a requested inheritance field, the report field data at the tracing starting point is identified as isolated data. If none of the tracing branches reach existing decision data or requested inheritance fields, then processing is performed according to the data recovery determination criteria.
[0032] For example, if the source field identifier of the composite conclusion field points to the clinical data summary field, and the source field identifier of the clinical data summary field continues to point to the application inheritance field, the application inheritance field can be reached along the two levels of source field identifiers. Therefore, the composite conclusion field is identified as isolated data. If the composite conclusion field also has another source field identifier pointing to the objective result field, it does not affect the aforementioned isolation result.
[0033] Starting from existing decision data, clinical event data and report field data are traced along triggering and content derivation relationships. During the tracing process, an access identifier is recorded for each accessed data point to prevent circular references between source fields. For the same clinical event data or report field data, if any content formation path contains a content derivation relationship, the corresponding clinical event data or report field data is identified as isolated data. This determination uses a path existence rule, without setting similarity or credibility thresholds. Even if the same report field data has both result generation relationships, as long as there is a path indicating that the report field content copies, references, or rewrites existing decision data, the report field data is still identified as isolated data.
[0034] For report field data, if none of the content generation paths contain content derivation relationships, and the report field data is associated with the application event data through trigger relationships and with an independent objective data source through result generation relationships, and the original result identifier can locate a unique original result, then the report field data is identified as recovered data. If the original result identifier cannot locate the original result, the same original result identifier locates multiple inconsistent results, or the source field identifier contains unresolved references, then the report field data is not identified as recovered data. The above processing does not delete the original content in the patient records; it only controls whether the corresponding data participates in this round of clinical knowledge graph retrieval.
[0035] The criteria for determining recoverable data as used in this application refer to the following: the entire content formation path of the report field data does not contain content derivation relationships; the report field data is associated with the application event data through triggering relationships and with an independent objective data source through result generation relationships; and the original result identifier can uniquely locate the original result. If any of these conditions are not met, the report field data will not be determined as recoverable data.
[0036] For the same report field data, isolation data determination is performed first, followed by recovery data determination. If any content formation path contains a content derivation relationship, the report field data is identified as isolated data, and recovery data determination is stopped. If none of the content formation paths contain a content derivation relationship, the determination of whether the report field data meets the recovery data determination criteria continues. If the recovery data determination criteria are met, the report field data is identified as recovery data. If the recovery data determination criteria are not met, the report field data is not added to this round of semantic search. The processing only changes whether the data is included in this round of search; it does not delete or modify the original data in the patient records.
[0037] Specifically, in the patient record identified as "Outpatient 2025072801," the patient's chief complaint of "thirst and excessive drinking" was recorded at 09:00, a temporary diagnosis of "abnormal glucose metabolism" was recorded at 09:08, a fasting blood glucose test request was generated based on the temporary diagnosis at 09:12, and the test report was generated at 09:38. The test report includes a clinical diagnosis field, a fasting blood glucose result field, and a test conclusion field. The source field identifier of the clinical diagnosis field points to "abnormal glucose metabolism" in the test request; therefore, the clinical diagnosis field is an inherited field. The fasting blood glucose result field has a value of 9.6 mmol / L, and the original result identifier points to the result record generated by the biochemical analyzer; therefore, the fasting blood glucose result field is an objective result field. The test conclusion field states "elevated fasting blood glucose, clinical assessment should be conducted in conjunction with abnormal glucose metabolism," and references both the clinical diagnosis field and the fasting blood glucose result field; therefore, the test conclusion field is a comprehensive conclusion field.
[0038] The patient initiated another diagnostic clinical knowledge request at 09:45. The type of the temporary diagnostic record is consistent with the diagnostic clinical knowledge request, its generation time is earlier than the time the clinical knowledge request was received, and it has not been withdrawn or replaced; therefore, it is determined to be existing decision data. An acquisition trigger relationship is established between the temporary diagnostic record and the test request. Based on the request form identifier, acquisition trigger relationships are established between the test request and the clinical diagnosis field, the fasting blood glucose result field, and the test conclusion field, respectively. A content derivation relationship is established between the clinical diagnosis field and the diagnostic content in the test request. Based on the original result identifier, a result generation relationship is established between the biochemical analyzer and the fasting blood glucose result field, and the result record generated by the biochemical analyzer is located by the original result identifier. A content derivation relationship is established between the test conclusion field and the clinical diagnosis field and the fasting blood glucose result field, respectively.
[0039] Because the content formation path of the clinical diagnosis field includes content derivation relationships, the clinical diagnosis field is identified as isolated data. Because the test conclusion field has at least one content formation path that passes through the clinical diagnosis field, it is also identified as isolated data and will not be recovered simply because it also references the fasting blood glucose result field. The fasting blood glucose result field has an acquisition trigger relationship with the test request and a result generation relationship with the biochemical analyzer. The original result identifier can locate the unique result record generated by the biochemical analyzer, and the content formation path from existing decision data to the fasting blood glucose result field does not contain content derivation relationships; therefore, the fasting blood glucose result field is identified as recovered data. Provisional diagnosis records belong to existing decision data, and test requests belong to decision-triggered event data; neither participates in this round of semantic retrieval. The patient's chief complaint of "thirst and polydipsia" and the fasting blood glucose result field participate in this round of semantic retrieval.
[0040] When mapping recovered data and other clinical event data to standard medical concepts, precise matching is performed sequentially based on medical codes, laboratory test codes, and a standardized thesaurus. If any matching method uniquely identifies a standard medical concept, a candidate semantic mapping is generated. If multiple standard medical concept candidates are obtained, the corresponding data is not added to the clinical fact subgraph. The fasting blood glucose result field is mapped to the standard medical concept of "fasting blood glucose" based on the laboratory test code, retaining the field value of 9.6 mmol / L. "Thirst" and "excessive thirst" are mapped to their respective symptom concepts.
[0041] When forming the clinical fact subgraph, candidate semantic mappings are preserved. If candidate semantic mappings for isolated and recovered data point to the same standard medical concept, only the candidate semantic mappings for the recovered data are retained. For example, if both the test conclusion field and the fasting blood glucose result field can be mapped to the standard medical concept of "fasting blood glucose," the test conclusion field belongs to isolated data, and the fasting blood glucose result field belongs to recovered data; therefore, only the semantic mapping between the fasting blood glucose result field and the standard medical concept of "fasting blood glucose" is retained. Standard medical concepts derived solely from the candidate semantic mappings of isolated data are not added to the clinical fact subgraph. The final clinical fact subgraph includes the symptom concept corresponding to the patient's complaint, the standard medical concept of fasting blood glucose, and field values, but excludes temporary diagnostic records, test requests, request inheritance fields, and comprehensive conclusion fields. Excluding isolated data means not retaining isolated data and the candidate semantic mappings formed by isolated data in the final clinical fact subgraph.
[0042] A pre-built clinical knowledge graph includes at least standard medical concept nodes, clinical knowledge item nodes, and knowledge source nodes. Standard medical concept nodes are connected to clinical knowledge item nodes through knowledge association relationships, and clinical knowledge item nodes are connected to knowledge source nodes through source relationships. Each clinical knowledge item node records at least the knowledge item identifier, suggestion type, knowledge title, applicable conditions, and knowledge content.
[0043] During semantic retrieval, standard medical concept identifiers in the clinical fact subgraph are read, and clinical knowledge items with consistent suggestion types are obtained by connecting them through knowledge association relationships, limited by the target suggestion type. When the same clinical knowledge item is matched by multiple standard medical concepts, they are merged into one item according to the knowledge item identifier, and the matched standard medical concepts are retained as supporting information. In this embodiment, clinical knowledge items associated with "thirst," "excessive drinking," and "fasting blood glucose" are obtained, and clinical knowledge suggestions are generated according to the knowledge title, applicable conditions, knowledge content, and knowledge source. When no matching clinical knowledge item is found, the search result of "no matching clinical knowledge item" is output, and no disease diagnosis conclusion or treatment measure is directly formed.
[0044] This embodiment distinguishes between two different relationships: "data obtained by triggering existing decisions" and "field content determined by existing decisions." In the same test report generated after a provisional diagnosis, only the clinical diagnosis field and the comprehensive conclusion field, which inherit the content of the existing decision, are isolated. Simultaneously, the fasting blood glucose result field, independently generated by the biochemical analyzer, is restored. Compared to uniformly excluding all data after a provisional diagnosis, this embodiment preserves the objective clinical facts obtained subsequently. Compared to restoring the entire test report, this embodiment avoids the existing decision from re-entering the same knowledge retrieval process through the inherited field or comprehensive conclusion field. Therefore, this embodiment can block the indirect reinjection of existing decision content without reducing the utilization rate of subsequent objective results, ensuring that clinical knowledge recommendations are primarily supported by the patient's original symptoms and independent objective results.
[0045] Based on another preferred embodiment described above, see [link to preferred embodiment]. Figure 4 As shown, this embodiment provides a knowledge graph-based clinical knowledge retrieval and suggestion generation system for performing the aforementioned knowledge graph-based clinical knowledge retrieval and suggestion generation method, including: The data acquisition module is used to acquire patient records and clinical knowledge requests; The path processing module is used to extract existing decision data, establish content formation paths, and determine isolated and restored data. The subgraph formation module is used to exclude isolated data and clinical event data triggered by existing decision data, and to form a clinical fact subgraph based on the recovery data and the remaining clinical event data. The suggestion generation module is used to retrieve clinical knowledge graphs based on clinical fact subgraphs and generate clinical knowledge suggestions.
[0046] When the above system executes the aforementioned method, it can achieve the same technical effect as the aforementioned method.
[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for clinical knowledge retrieval and suggestion generation based on knowledge graphs, characterized in that, include: Collect patient records from the same visit and receive clinical knowledge requests. Segment the patient records according to their origin to obtain clinical event data and report field data. Based on the clinical knowledge request, a target suggestion type is determined, existing decision data consistent with the target suggestion type is extracted from the clinical event data, and the existing decision data is traced to the content formation path of the clinical event data and the report field data; The existing decision data and the content formation path that includes copying, referencing, or rewriting the existing decision data are identified as isolated data; the report field data that is triggered by the existing decision data, whose content formation path does not include copying, referencing, or rewriting the existing decision data, and which is generated by an independent objective data source are identified as restored data. Excluding isolated data and decision-triggered event data, the recovery data and other clinical event data are mapped to standard medical concepts to form a clinical fact subgraph; Semantic retrieval of the clinical knowledge graph is performed based on the clinical fact subgraph to obtain clinical knowledge items, and clinical knowledge suggestions are generated based on the clinical knowledge items.
2. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 1, characterized in that, When extracting existing decision data, the following should be included: Based on the type mapping relationship, determine the target decision type that is consistent with the target suggestion type; The target decision type data generated during the same medical visit at a time earlier than the time the clinical knowledge request was received and which is in a valid state is identified as the existing decision data.
3. The clinical knowledge retrieval and suggestion generation method based on knowledge graphs according to claim 1, characterized in that, When classifying the report field data, the following are included: Identify examination or testing requests from the clinical event data to obtain request event data; The report field data brought in by the application event data is determined as the application inheritance field, the report field data with the original result identifier and generated by the independent objective data source is determined as the objective result field, and the report field data that simultaneously references the application inheritance field and the objective result field is determined as the comprehensive conclusion field.
4. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 3, characterized in that, When establishing the content formation path, the following is included: Based on the generation records in the application event data, establish the acquisition trigger relationship between the existing decision data and the application event data; based on the application form identifier, establish the acquisition trigger relationship between the application event data and the report field data. Establish a content derivation relationship between the referenced data and the data in the report field based on the source field identifier; Based on the original result identifier, establish the result generation relationship between the independent objective data source and the objective result field; By connecting the trigger relationship, the content derivation relationship, and the result generation relationship, the content formation path is obtained.
5. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 4, characterized in that, Determining the isolated data includes: For the clinical event data or the report field data, if any content formation path from the existing decision data to the clinical event data or the report field data contains the content derivation relationship, the clinical event data or the report field data is identified as the isolated data.
6. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 4, characterized in that, When tracing the indirect content derivation relationships of the reported field data, this includes: When the source field identifier points to another report field data, trace along the source field identifiers at each level; When any tracking branch reaches the existing decision data or the application inheritance field, the report field data of the tracking starting point will be determined as the isolated data; If none of the tracking branches reach the existing decision data or the application inheritance field, determine whether the report field data of the tracking starting point meets the recovery data determination condition according to the original result identifier.
7. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 4, characterized in that, Determining the recovered data includes: The entire content formation path of the report field data does not contain the content derivation relationship. The report field data is associated with the application event data through the acquisition trigger relationship and with the independent objective data source through the result generation relationship. When the report field data has the original result identifier, the report field data is determined as the restored data.
8. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 7, characterized in that, When processing report field data in the same inspection report or test report, the following steps are taken: when the same inspection report or test report contains the application inheritance field, the objective result field, and the comprehensive conclusion field, the application inheritance field that can be traced back to the existing decision data and the comprehensive conclusion field that references the application inheritance field are determined as the isolated data, and the objective result field that meets the data recovery determination conditions is determined as the recovered data.
9. The clinical knowledge retrieval and suggestion generation method based on knowledge graph according to claim 1, characterized in that, When forming a clinical fact subgraph, the following are included: If the same standard medical concept is semantically mapped to both the isolated data and the recovered data, only the semantic mapping between the recovered data and the standard medical concept is retained. Standard medical concepts that are semantically mapped only to the isolated data are not included in the clinical fact subgraph.
10. A knowledge graph-based clinical knowledge retrieval and suggestion generation system, used to execute the knowledge graph-based clinical knowledge retrieval and suggestion generation method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire patient records and clinical knowledge requests; The path processing module is used to extract existing decision data, establish content formation paths, and determine isolated and restored data. The subgraph forming module is used to exclude the isolated data and the clinical event data triggered by the existing decision data, and to form a clinical fact subgraph based on the recovery data and the remaining clinical event data. The suggestion generation module is used to retrieve clinical knowledge graphs based on the clinical fact subgraph and generate clinical knowledge suggestions.
Citation Information
Patent Citations
Clinical decision interaction method and system based on structured evidence reasoning
CN121565500A
