Pathology knowledge base construction method and device, electronic equipment and storage medium

Abstract

The invention discloses a pathology knowledge base construction method and device, electronic equipment and a storage medium, the method and device are applied to the electronic equipment, specifically, a three-dimensional semantic mapping structure is constructed based on multiple information sources, and the three-dimensional semantic mapping structure comprises multiple types of core knowledge nodes; performing semantic riveting on the three-dimensional semantic mapping structure by utilizing a plurality of relationships to form a minimum closure sub-graph with complete three-dimensional semantics, and storing the minimum closure sub-graph into a graph database to obtain a pathology knowledge base; and optimizing the pathology knowledge base according to the feedback of the user. According to the description, the technical means of the pathology knowledge base with professional depth and interpretable semantic dimension can be systematically and standardly constructed and continuously improved, so that an authoritative, easy-to-understand and extensible knowledge basis can be provided for multi-scene intelligent application.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and more particularly, to a method, device, electronic device, and storage medium for constructing a pathological knowledge base. Background Art

[0002] In today's medical environment, tens of millions of pathological reports are generated every year. However, the vast majority of patients do not have sufficient medical background to interpret the professional terms and diagnostic logic in them. What they really want to know is the specific impact of this report on the severity of the disease, the risk of recurrence, and the treatment plan. However, the current paper or PDF reports in hospitals are mainly targeted at clinicians and are not user-friendly for patients; it is also difficult for doctors to fully explain to each patient during outpatient consultations, resulting in the continuous existence of the information gap and anxiety between doctors and patients. Therefore, there is an urgent need in clinical work for a technical tool that can interpret pathological reports for patients in a large-scale, authoritative, and personalized manner, so as to provide authoritative medical information interpretation services for ordinary non-professional users. Summary of the Invention

[0003] In view of this, this application provides a method, device, electronic device, and storage medium for constructing a pathological knowledge base, which is used to construct a knowledge base that can systematically organize pathological knowledge elements, taking into account both professional depth and semantic comprehensibility, so as to provide structured and extensible knowledge support for intelligent interpretation, assisted question answering, and multi-scenario medical applications.

[0004] To achieve the above object, the following solutions are proposed:

[0005] A method for constructing a pathological knowledge base, which is applied to an electronic device. The construction method includes the steps of:

[0006] Constructing a three-dimensional semantic mapping structure based on multiple information sources, where the three-dimensional semantic mapping structure includes multiple types of core knowledge nodes;

[0007] Using multiple relationships to perform semantic riveting on the three-dimensional semantic mapping structure to form a minimum closed subgraph with complete three-dimensional semantics, and storing the minimum closed subgraph in a graph database to obtain the pathological knowledge base;

[0008] Optimizing the pathological knowledge base according to the feedback of the user.

[0009] Optionally, the multiple information sources include authoritative medical guidelines and literature, expert interview records, and real patient consultation records;

[0010] The multiple types of core knowledge nodes include professional concept nodes, decision rule nodes, and candidate popular explanation nodes.

[0011] Optionally, the constructing a three-dimensional semantic mapping structure based on multiple information sources includes the steps of:

[0012] Extract from the information source using a named entity recognition model in the medical field to obtain the professional concept nodes;

[0013] Extract the premise-conclusion type logical structure from the information source by combining syntactic dependency analysis and rule template mining method to obtain the decision rule nodes;

[0014] Call a large model fine-tuned with pathological corpus to perform multi-layer semantic rewriting on the professional terms in the information source, generate multiple versions of popular explanations that can be understood by patients, and construct a term alignment mapping based on the popular explanations to obtain the candidate popular explanation nodes;

[0015] Optimize the professional concept nodes, the decision rule nodes, and / or the candidate popular explanation nodes based on structured expert knowledge or opinions.

[0016] Optionally, the multiple relationships include synonym link relationship, rule trigger relationship, and explanation binding relationship, where:

[0017] The rule trigger is used to establish a directed logical relationship between the concept nodes and the diagnosis and treatment rules they participate in, supporting many-to-many mapping of premises and results;

[0018] The explanation binding is used to bind the popular explanation nodes to the corresponding professional terms, and generate multi-layer expression versions by distinguishing the explanation methods through semantic tags.

[0019] Optionally, it further includes the steps of:

[0020] Respond to the user's query request, generate and feedback a final answer to the user for the patient message input by the user.

[0021] Optionally, responding to the user's query request, generating and feedbacking a final answer to the user for the patient message input by the user includes the steps of:

[0022] Perform entity recognition on the patient message based on the large model to obtain concept nodes;

[0023] Use the concept nodes as anchors to perform K-order semantic expansion or attribute matching in the graph database, and obtain the rules and explanation nodes associated with the concept nodes to form a closed subgraph;

[0024] Traverse all nodes in the closed subgraph to obtain the final answer.

[0025] A construction device for a pathological knowledge base, which is applied to an electronic device, and the construction device includes:

[0026] A three-piece building block configured to build a three-dimensional semantic mapping structure based on multiple information sources, the three-dimensional semantic mapping structure including multiple types of core knowledge nodes;

[0027] A knowledge base construction module configured to perform semantic riveting on the three-dimensional semantic mapping structure using multiple relationships to form a minimum closed subgraph, and store the minimum closed subgraph in a graph database to obtain the pathological knowledge base;

[0028] A model optimization module configured to perform optimization processing on the pathological knowledge base according to user feedback.

[0029] Optionally, it further includes:

[0030] An information query response module configured to respond to a user's query request, generate and feedback a final answer to the user for the patient message input by the user.

[0031] An electronic device, the electronic device includes at least one processor and a memory connected to the processor, wherein:

[0032] The memory is used to store computer programs or instructions;

[0033] The processor is used to execute the computer programs or instructions so that the electronic device implements the construction method as described above.

[0034] A computer-readable storage medium, applied to an electronic device, the storage medium carrying one or more computer programs, the one or more computer programs being executable by the electronic device so that the electronic device can implement the construction method as described above.

[0035] As can be seen from the above technical solutions, the present application discloses a method, device, electronic device and storage medium for constructing a pathological knowledge base. The method and device are applied to an electronic device. Specifically, a three-dimensional semantic mapping structure is constructed based on multiple information sources. The three-dimensional semantic mapping structure includes multiple types of core knowledge nodes; multiple relationships are used to perform semantic riveting on the three-dimensional semantic mapping structure to form a minimum closed subgraph, and the minimum closed subgraph is stored in a graph database to obtain the pathological knowledge base; the pathological knowledge base is optimized according to user feedback. As can be seen from the above description, this solution can systematically and standardly construct and continuously improve the technical means of a pathological knowledge base with both professional depth and interpretable semantic dimensions, so as to provide an authoritative, easy-to-understand and extensible knowledge base for multi-scenario intelligent applications. Description of the Drawings

[0036] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0037] Figure 1 It is a flowchart of a method for constructing a pathological knowledge base according to an embodiment of the present application;

[0038] Figure 2 It is a flowchart of another method for constructing a pathological knowledge base according to an embodiment of the present application;

[0039] Figure 3 It is a block diagram of a device for constructing a pathological knowledge base according to an embodiment of the present application;

[0040] Figure 4 It is a block diagram of another device for constructing a pathological knowledge base according to an embodiment of the present application;

[0041] Figure 5 It is a block diagram of an electronic device according to an embodiment of the present application. Detailed implementation manners

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0043] Figure 1 It is a flowchart of a method for constructing a pathological knowledge base according to an embodiment of the present application.

[0044] As Figure 1 shown, the construction method provided in this embodiment is applied to an electronic device, and is used to construct a pathological knowledge base for a user that can provide medical information interpretation services for the user. This electronic device can be understood as a computer, a server, a cloud platform, etc. with data computing capabilities and information processing capabilities. The construction method specifically includes the following steps:

[0045] S1. Construct a three-dimensional semantic mapping structure based on multiple information sources.

[0046] The three-dimensional semantic mapping structure includes multiple types of core knowledge nodes. This application synchronously collects corpus from three major information sources, and through natural language processing and expert quick proofreading, generates a "professional concept node, decision rule node, and candidate popular explanation node" set. That is, based on three types of core information sources, three types of core knowledge nodes are constructed to form a three-dimensional semantic mapping structure.

[0047] The three major information sources of this application include authoritative medical guidelines and literature, interviews with pathology experts, and real patient consultation records.

[0048] Authoritative medical guidelines and standard literature are used as sources of term definitions and grading standards to provide pathological classification, diagnostic criteria, and evidence-based rules; expert interview records can supplement unstandardized concepts or items with fuzzy ontology boundaries, and are used to supplement judgment logics that are common in practical experience but not covered by standard literature; real patient consultation records are used to identify fuzzy concept mappings under patient expressions, and capture term variants and common patient inquiry methods that appear in actual use; pathological report texts are used as verification corpus for the actual use of terms.

[0049] Professional concept nodes are used to identify core medical entities such as tumor types, grades, morphological manifestations, immunological markers, and molecular mutations from the information sources by using named entity recognition models in the medical field (BioBERT, MedNER, and joint extraction of professional concept nodes based on large model prompts to achieve multi-model collaborative extraction), and align with standard medical terminology databases (such as ICD-O, SNOMED, NCIt) to form uniquely identified professional concept nodes.

[0050] Decision rule nodes are used to extract premise-conclusion type logical structures from the information sources by combining syntactic dependency analysis and rule template mining methods for pathological explanations and subsequent diagnosis and treatment suggestions. For example, "Premise: IDH mutation --> Conclusion: Better prognosis". The rules are encoded as graph nodes in the logical structure <premise set, conclusion set> of the premise set and conclusion set, and can be attached with attributes such as applicable diseases, evidence levels, reference document numbers, etc., to support subsequent graph-based reasoning and case matching.

[0051] Candidate popular explanation nodes are used to call large models (such as GLM-4, DeepSeek) fine-tuned with pathological corpus to generate multiple versions of semantic rewrite explanations for professional terms, covering different semantic levels (such as definitions, metaphors, question-and-answer expressions). All explanation versions are classified and managed through semantic labels (such as "definitional explanation", "analogical / figurative explanation", "question-and-answer explanation"), and a term alignment mapping is established with the corresponding professional concept nodes. For example, one of the popular explanations corresponding to the term "glioblastoma (WHO grade IV)" is "This is a brain tumor with a relatively high degree of malignancy and requires prompt treatment."

[0052] After the above - mentioned all nodes are structurally encoded, they are stored in the knowledge graph after being reviewed by experts to ensure professional accuracy and expressibility. Standardized semantic units are established to lay a foundation for semantic connection and reasoning in the knowledge graph.

[0053] In addition, this embodiment further includes optimizing the above - mentioned professional concept nodes, decision - rule nodes, and candidate popular - explanation nodes based on a structured expert review process to make all nodes more scientific and reasonable.

[0054] S2. Perform semantic riveting on the three - dimensional semantic mapping structure to obtain a pathological knowledge base.

[0055] Specifically, use multiple relationships to perform semantic riveting on the three - dimensional semantic mapping structure, form a minimum closed sub - graph, and store the minimum closed sub - graph in a graph database to obtain a pathological knowledge base.

[0056] The three types of nodes form a complete knowledge structure through three types of semantic edges. The semantic connection relationships include synonym - link relationships, rule - trigger relationships, and explanation - binding relationships.

[0057] The synonym - link relationship is used to connect multiple expression forms of the same medical concept in different contexts, supporting two - way or multi - way unification and normalization among terms, abbreviations, and common sayings. For example, the nodes "glioblastoma multiforme", "GBM", and "the most malignant type of brain tumor" are interconnected through the synonym - link relationship to form a semantic - equivalence cluster, facilitating concept generalization and matching.

[0058] The rule - trigger relationship is used to establish a directed logical relationship between a concept node and the diagnosis and treatment rules it participates in, supporting the structural modeling of multiple premises and multiple conclusions. For example: The nodes "IDH wild - type" and "tumor grade IV" are respectively connected to the rule node "require concurrent radiotherapy and chemotherapy", forming a many - to - many mapping relationship where the premise set points to the conclusion set. Each rule - trigger relationship can include edge - weight attributes, such as trigger - condition types, reference evidence (such as the evidence - based level recommended by guidelines).

[0059] The explanation - binding relationship is used to establish a binding relationship between a candidate popular - explanation node and the corresponding professional - concept node, and different explanation methods (definitions, metaphors, analogies, etc.) are identified through semantic tags, which are used to generate multi - level expression versions to adapt to different user semantic requirements. For example: The node "temozolomide" is bound to the explanation: "This is an oral chemotherapy drug commonly used for brain tumors, with relatively mild side effects".

[0060] The system constructs a minimum closed subgraph using graph traversal algorithms (such as depth - first matching or weighted semantic expansion). That is, taking any concept node as an anchor point, it automatically collects the decision rules and explanatory nodes directly connected to it to form a subgraph that satisfies three - dimensional semantic completeness. This subgraph is regarded as an "answerable unit" and is stored in a graph database (such as Neo4j) as the smallest semantically callable unit in the knowledge base. This structure facilitates subsequent execution of structured queries, semantic expansion, or GraphRAG retrieval calls in embedded question - answering tasks. For example, when the user queries "My mother was diagnosed with glioblastoma multiforme. How should she be treated?"

[0061] The solution in this embodiment retrieves the rules and explanations associated with the "GBM" concept node as an anchor point through a graph traversal algorithm;

[0062] The obtained minimum closed subgraph contains:

[0063] • Concept nodes: high - grade glioma, IDH status, MGMT status, radiotherapy, temozolomide;

[0064] • Rule nodes: concurrent radiotherapy + TMZ strategy;

[0065] • Explanation nodes: popularized explanations corresponding to each term (multiple - version candidates with semantic label annotations).

[0066] This subgraph provides structured semantic input for subsequent multi - modal question - answering, automatic reasoning, and auxiliary interpretation.

[0067] S3. Optimize the pathological knowledge base according to the user's feedback.

[0068] This solution continuously collects the user's "comprehension rate" feedback. Explanations with too low scores will automatically enter the pool to be optimized, keeping the knowledge base fresh and accurate. Specifically, it is to optimize the popular explanation nodes.

[0069] The front - end application interface continuously receives the user's feedback information on the explanation content, especially the "comprehension score" or "understandability level" annotation on the popular expression of terms. For popular explanation nodes that are repeatedly marked as "difficult to understand" or have scores lower than the set threshold, the system automatically includes them in the "pool to be optimized".

[0070] For the nodes to be optimized, the system can automatically trigger the following iterative optimization process:

[0071] 1) Regenerate candidate expression versions: Call a large model fine - tuned with pathological corpus to generate multiple popular explanation versions for the original professional terms, covering different expression styles and semantic granularities. For example:

[0072] "This means that there is a change in a certain gene, which may make taking medicine more effective."

[0073] "If the test is positive, it means you may be more sensitive to that oral chemotherapy drug."

[0074] 2) Popularity assessment and version screening: The system can use A / B testing, user click preference analysis, or expert review mechanisms to compare the understandability of the generated candidate versions and select the optimal expression version.

[0075] 3) Node replacement and version control: The system replaces the optimal explanatory version as the new version of the current popular explanatory node, preserves the original node version in the archive, and attaches metadata such as timestamps, scoring records, and generation model configurations to support subsequent traceability management and explanatory content comparison.

[0076] 4) Medical consistency guarantee: All regenerations and replacements of popular explanations are limited to be executed within the semantic boundaries of the original knowledge graph to ensure that the explanatory content does not deviate from the clinical meaning and diagnostic logic carried by the original professional terms. When necessary, a medical review session can be set as a pre-update verification.

[0077] Through the node evolution of the above user feedback, the dynamic adaptive optimization of the explanatory dimension in the pathological knowledge base is realized. On the basis of maintaining professional accuracy, the popularity and user-friendliness of semantic expression are continuously improved.

[0078] Specific optimization processes include manual verification and revision, including:

[0079] 1. The review of professional concept nodes includes the inspection of term standardization and the confirmation of mapping to the standard thesaurus;

[0080] 2. The review of decision rule nodes includes the logical consistency review between the preconditions and the conclusions, and the manual confirmation of their corresponding evidence-based levels;

[0081] 3. The review of candidate popular explanatory nodes includes the evaluation of language understandability, medical accuracy, and expression style. Experts can score, rank, or select the best expression for multiple versions of explanations;

[0082] In addition, the nodes that pass the review can be written into the graph database and bound with the reviewer ID, review timestamp, and version number for subsequent optimization traceability and knowledge update control.

[0083] Suppose the candidate popular explanatory node "Positive MGMT methylation indicates sensitivity to temozolomide" is marked as "unintelligible" by users multiple times. The system executes the optimization process as follows:

[0084] 1) The system detects a low feedback score --> automatically adds this explanatory node to the "pool to be optimized" 2) Calls the large model to regenerate multiple expression versions, such as:

[0085] - "This means there is a change in a certain gene, which may make taking medicine more effective."

[0086] - "If the test is positive, it means you may be more sensitive to that oral chemotherapy drug."

[0087] 3) Expert review or AB test user preferences to select a clearer and more understandable version.

[0088] 4) Replace the original explanation node with the best version, and retain the old version to ensure semantic controllability and quality traceability.

[0089] All optimizations are based on the scope of control of the atlas content to ensure that medical explanations do not deviate from the original diagnostic basis.

[0090] As can be seen from the above technical solution, this embodiment provides a method for constructing a pathological knowledge base, which is applied to an electronic device. Specifically, a three-dimensional semantic mapping structure is constructed based on multiple information sources. The three-dimensional semantic mapping structure includes multiple types of core knowledge nodes; multiple relationships are used to perform semantic riveting on the three-dimensional semantic mapping structure to form a minimum closed subgraph, and the minimum closed subgraph is stored in a graph database to obtain a pathological knowledge base; the pathological knowledge base is optimized according to user feedback. As can be seen from the above description, this solution can systematically and standardly construct and continuously improve a technical means for a pathological knowledge base with both professional depth and interpretable semantic dimensions, so as to provide an authoritative, easy-to-understand and extensible knowledge base for multi-scenario intelligent applications.

[0091] In addition, as Figure 2 shown, in a specific implementation manner of the present application, the following steps are further included:

[0092] S4. Provide a final answer to the user according to the user's query request.

[0093] The atlas layer of the present application exposes a unified structured query interface externally, supporting the input of natural language pathological messages. After the GraphRAG service layer receives user input (such as a patient's question or a pathological report paragraph), it starts the query response process. This pathological knowledge base also supports dynamic information invocation and answer generation based on the user's query request.

[0094] When processing the user's query request, the process of processing a description of a pathological report is as follows:

[0095] 1) Entity recognition and concept node extraction:

[0096] Use a structured extraction module fine-tuned based on large models (such as GLM, DeepSeek) to perform entity recognition on the input text, extract the core pathological entities or expression features therein, and generate corresponding concept nodes. For example, for the input: "IDH1 negative, Ki-67 15%, ATRX retained". Extraction result: IDH status = wild type; Ki-67 index = 15%; ATRX status = retained.

[0097] This step is particularly suitable for extracting implicit or semantically complex expressions in reports, such as: "TP53 suspiciously positive", "Enhanced GFAP staining in some areas".

[0098] 2) Subgraph Retrieval and Structure Invocation (GraphRAG):

[0099] Use the identified concept nodes as anchors to perform K-order semantic expansion or attribute matching in the graph database, retrieve the associated rule and explanation nodes, and form the smallest closed subgraph that conforms to the current query context. This subgraph is the "callable semantic unit" for subsequent reasoning and answer generation.

[0100] 3) Rule Reasoning and Diagnostic Recommendation Generation:

[0101] Traverse all rule nodes in the subgraph, match the conditions to trigger the rules, and combine the graph reasoning engine (graph attention network) to judge the diagnostic conclusions, recommended paths or risk indicators that meet the condition combinations. For example: "IDH wild type + Ki-67>10%" → Reasoning result: "High-risk glioblastoma, recommend evaluating the MGMT methylation status to guide TMZ treatment". All reasoning paths record the reasoning chain for visualization and auditing.

[0102] For example, taking brain tumors as an example:

[0103] Input combination: "IDH1 negative, Ki-67 15%, ATRX retained"

[0104] 1) Entity Recognition and Extraction: IDH status = wild type; Ki-67 index = 15%; ATRX status = retained;

[0105] 2) Subgraph Retrieval Results: Find the relevant diagnostic paths, molecular feature combinations and popular explanations of "primary GBM";

[0106] 3) Reasoning Output: "High-risk primary glioblastoma, it is recommended to evaluate the MGMT methylation status to guide temozolomide treatment"

[0107] All reasoning paths can record the generated reasoning chain to support visual display.

[0108] According to the target users (medical staff vs patients) and interaction scenarios, the system adopts two types of response methods to support the generation of personalized answers.

[0109] 1) Template filling method;

[0110] Based on the structured template, fill in the professional concepts and reasoning results to form a standardized text answer, which is applicable to electronic medical record annotations or doctor decision-making assistance systems. For example, it is interpreted as IDH wild-type glioblastoma with a relatively high Ki-67, indicating active tumor proliferation. It is recommended to supplement the MGMT methylation test to evaluate the sensitivity of temozolomide treatment.

[0111] 2) Large model polishing method:

[0112] Use the closed subgraph as a semantic prompt (prompt) and inject it into the general large model to generate a natural, fluent, and easy-to-understand answer text. Through term alignment and traceability path verification, ensure that the output semantics is accurate and professionally controllable. Example output:

[0113] “This is a highly malignant brain tumor called glioblastoma. There are several genetic tests suggesting that it may grow and progress rapidly. The doctor may recommend further tests to see if taking medicine can be more effective.”

[0114] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes may occur in a different order than that marked in the accompanying drawings. For example, two consecutive boxes shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0115] Although the operations are depicted in a particular order, this should not be construed as requiring that the operations be performed in the particular order shown or in a sequential order. In certain circumstances, multitasking and parallel processing may be advantageous.

[0116] It should be understood that the various steps described in the method embodiments of the present disclosure may be executed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this regard.

[0117] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages or combinations thereof. The above-mentioned programming languages include, but are not limited to, object-oriented programming languages - such as Java, Smalltalk, C++; and also include conventional procedural programming languages - such as the C language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN) - or may be connected to an external computer.

[0118] Figure 3 It is a block diagram of a device for constructing a pathological knowledge base according to an embodiment of the present application.

[0119] As Figure 3 shown, the construction device provided in this embodiment is applied to an electronic device and is used to construct a pathological knowledge base for a user that can provide medical information interpretation services for the user. The electronic device can be understood as a computer, a server, a cloud platform, etc. with data computing capabilities and information processing capabilities. The construction device specifically includes a three-piece set construction module 10, a knowledge base construction module 20, and a model optimization module 30.

[0120] The three-piece set construction module is used to construct a three-dimensional semantic mapping structure based on multiple information sources.

[0121] The three-dimensional semantic mapping structure includes multiple types of core knowledge nodes. In this application, corpora are synchronously collected from three major information sources, and through natural language processing and expert quick proofreading, a "professional concept node, decision rule node, and candidate popular explanation node" three-piece set is generated. That is, based on three types of core information sources, three types of core knowledge nodes are constructed to form a three-dimensional semantic mapping structure.

[0122] The three major information sources of this application include authoritative medical guidelines and literature, pathological expert interviews, and real patient consultation records.

[0123] Professional concept nodes are used to identify core medical entities such as tumor types, grades, morphological manifestations, immune markers, and molecular mutations from the information source through the collaborative extraction of multiple models using named entity recognition models in the medical field (BioBERT, MedNER, and joint extraction of professional concept nodes based on large model prompts), and align them with standard medical terminology libraries (such as ICD-O, SNOMED, NCIt) to form uniquely identified professional concept nodes.

[0124] Decision rule nodes are used to extract conditional-conclusion judgment logic from standardized guidelines and expert knowledge through syntactic parsing and knowledge extraction methods for pathological interpretation and subsequent diagnosis and treatment recommendations. Such as "IDH mutation --> better prognosis". The rules are encoded into the logical structure of the premise set and the conclusion set for subsequent graph reasoning.

[0125] Candidate plain explanation nodes are used to call large models (such as ChatGLM) to simplify the semantics of professional terms, generate plain explanations understandable to patients, and construct term alignment mappings. For example, "Glioblastoma (WHO Grade IV)" is transformed into "Glioblastoma is a type of brain tumor with a relatively fast progression and requires prompt treatment".

[0126] All of the above nodes are structured and encoded, reviewed by experts, and then stored in the knowledge graph to establish standardized semantic units, laying a foundation for subsequent relationship establishment and reasoning in the knowledge graph.

[0127] In addition, this embodiment also includes optimizing the above-mentioned professional concept nodes, decision rule nodes, and candidate plain explanation nodes based on expert knowledge or suggestions to make all nodes more scientific and reasonable.

[0128] The knowledge base construction module is used to perform semantic riveting on the three-dimensional semantic mapping structure to obtain a pathological knowledge base;

[0129] Specifically, multiple relationships are used to perform semantic riveting on the three-dimensional semantic mapping structure to form a minimum closed subgraph, and the minimum closed subgraph is stored in the graph database to obtain a pathological knowledge base.

[0130] The three types of nodes form a complete knowledge structure through three semantic edges, and the semantic connection relationships include synonym link relationships, rule trigger relationships, and explanation binding relationships.

[0131] Synonym link relationships are used to connect different expressions of the same medical concept (such as "Glioblastoma multiforme" / "GBM" / "the most malignant type of brain tumor"), supporting the multi-directional unification of common sayings, terms, and abbreviations.

[0132] The rule triggering relationship is used to establish a directed logical relationship between concept nodes and the diagnosis and treatment rules they participate in, supporting many-to-many mappings between premises and results. For example, the nodes "IDH wild type" and "tumor grade IV" are linked to the rule: "Synchronous radiotherapy and chemotherapy are required." Edge weight attributes include triggering logic and reference evidence (such as the evidence-based level recommended by the guidelines).

[0133] The interpretation binding relationship is used to bind candidate plain language interpretation nodes to corresponding professional terms, distinguishing interpretation methods (definitions, metaphors, analogies, etc.) through semantic tags, and is used to generate multi-level expression versions. For example, the node "temozolomide" is bound to the interpretation: "This is an oral chemotherapy drug commonly used for brain tumors with relatively mild side effects."

[0134] The system uses a graph traversal algorithm (such as depth-first matching) to construct a minimum closed subgraph, that is, with a concept as the core, automatically including the semantic full graph of all its decision rules and interpretation nodes. This subgraph is regarded as an "answerable unit", which is the smallest semantic unit that can be called in the knowledge graph and is stored in a graph database (such as Neo4j) for subsequent RAG processing. For example, the patient's family member inputs "My mother was diagnosed with glioblastoma multiforme. How should she be treated?"

[0135] The solution adopted in this embodiment retrieves the rules and interpretations associated with "GBM" through a graph traversal algorithm;

[0136] The obtained subgraph includes:

[0137] • Nodes: high-grade glioma, IDH status, MGMT status, radiotherapy, temozolomide;

[0138] • Rules: synchronous radiotherapy + TMZ strategy;

[0139] • Plain language interpretation: Each term has an aligned interpretation.

[0140] The model optimization module is used to optimize the pathological knowledge base according to the user's feedback.

[0141] This solution continuously collects the user's "comprehension rate" feedback. Explanations with too low scores will automatically enter the pool to be optimized, keeping the knowledge base fresh and accurate.

[0142] Suppose the candidate plain language interpretation node "MGMT methylation positive indicates sensitivity to temozolomide" is marked as "not understood" by the user multiple times. The system executes the optimization process as follows:

[0143] 1) The system detects a low feedback score --> automatically adds this interpretation node to the "pool to be optimized" 2) Calls the large model to regenerate multiple expression versions, such as:

[0144] - "This means that there is a change in a certain gene, which may make the medication more effective."

[0145] - "If the test is positive, it means you may be more sensitive to that oral chemotherapy drug."

[0146] 3) Expert review or AB test user preferences to select a clearer and more understandable version.

[0147] 4) Replace the original explanation node with the best version, and keep the old version to ensure semantic controllability and quality traceability.

[0148] All optimizations are based on the scope of control of the atlas content to ensure that medical explanations do not deviate from the original diagnostic basis.

[0149] As can be seen from the above technical solution, this embodiment provides a device for constructing a pathology knowledge base. The device is applied to an electronic device, specifically to construct a three-dimensional semantic mapping structure based on multiple information sources. The three-dimensional semantic mapping structure includes multiple types of core knowledge nodes; use multiple relationships to perform semantic riveting on the three-dimensional semantic mapping structure to form a minimum closed subgraph, and store the minimum closed subgraph in a graph database to obtain a pathology knowledge base; optimize the pathology knowledge base according to the user's feedback. As can be seen from the above description, this solution can systematically and standardly construct and continuously improve a technical means for a pathology knowledge base with both professional depth and interpretable semantic dimensions, so as to provide an authoritative, easy-to-understand and extensible knowledge basis for multi-scenario intelligent applications.

[0150] In addition, as Figure 4 shown, in a specific embodiment of the present application, it further includes an information query response module 40.

[0151] The information query response module is used to feedback a final answer to the user according to the user's query request.

[0152] The atlas layer of the present application exposes a standardized query interface externally. The GraphRAG service layer completes entity recognition, subgraph retrieval, and rule reasoning after receiving the patient message, and generates a final answer through template filling or large model enhancement.

[0153] When processing the user's query request, the process of processing a section of pathological report description is as follows:

[0154] 1) Entity recognition:

[0155] Use a structure extraction model fine-tuned based on a large model (such as GLM) to perform entity recognition on unstructured pathological text. It is especially suitable for extracting implicit or semantically complex expressions in the report, such as: "ATRX not lost", "Enhanced GFAP staining in some areas".

[0156] 2) Subgraph retrieval (GraphRAG):

[0157] Using the identified conceptual nodes as anchors, perform K-order semantic expansion or property matching in the graph database to obtain the associated rule and explanation nodes, and form a closed subgraph required for the response.

[0158] 3) Rule reasoning:

[0159] Traverse all rule nodes in the subgraph, match the conditions to trigger the rules, and combine with the graph reasoning engine (graph attention network) to judge the possible diagnostic tendencies or auxiliary suggestions. For example: "IDH wild-type + Ki-67>10%" → Reasoning result: "High-risk glioblastoma, it is recommended to evaluate the MGMT methylation status to guide TMZ treatment". All reasoning paths record the reasoning chain, which is available for visualization and auditing.

[0160] For example, taking brain tumors as an example:

[0161] Input: "IDH1 negative, Ki-67 15%, ATRX retained"

[0162] 1) Entity recognition and extraction: IDH status = wild-type; Ki-67 index = 15%; ATRX status = retained;

[0163] 2) Subgraph retrieval result: Find the relevant diagnostic paths, molecular feature combinations and popular explanations of "primary GBM";

[0164] 3) Reasoning output: "The result suggests primary glioblastoma, and the molecular markers support a higher malignancy. It is recommended to detect the MGMT status to evaluate the chemotherapy sensitivity."

[0165] According to the target users (medical staff vs patients) and interaction scenarios, the system adopts two types of response methods.

[0166] 1) Template filling;

[0167] Fill in professional concepts and reasoning results based on the structured template to form a standardized text answer, which is applicable to electronic medical record annotations or doctor decision-making assistance systems.

[0168] 2) Large model polishing and generation:

[0169] Use the closed subgraph as the input (prompt) and inject it into the large model to generate a natural, fluent and easy-to-understand answer text, and ensure the accurate semantics and professional controllability of the output through term alignment and traceability path verification. Example output:

[0170] "This is a brain tumor with a relatively high malignancy, called glioblastoma. There are several gene tests suggesting that it may grow and progress rapidly. Doctors may recommend further tests to see if taking medicine can be more effective." For example:

[0171] The units involved in the embodiments of the present disclosure can be implemented in software or in hardware. Among them, the name of the unit does not constitute a limitation to the unit itself in some cases. For example, the first acquisition unit can also be described as "the unit for acquiring at least two Internet protocol addresses".

[0172] The functions described above in this article can be performed, at least in part, by one or more hardware logic components. For example, without limitation, the exemplary types of hardware logic components that can be used include: field programmable gate array (FPGA), application specific integrated circuit (ASIC), application specific standard product (ASSP), system on chip (SOC), complex programmable logic device (CPLD), and so on.

[0173] Figure 5 It is a block diagram of an electronic device according to an embodiment of the present application.

[0174] The following refers to Figure 5 , which shows a schematic structural diagram suitable for implementing the electronic device in the embodiments of the present disclosure. The terminal device in the embodiments of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. This electronic device is only an example and should not bring any limitation to the functions and usage scope of the embodiments of the present disclosure.

[0175] The electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to the program stored in the read-only memory ROM 502 or the program loaded from the input device 506 into the random access memory RAM 503. In the RAM, various programs and data required for the operation of the electronic device are also stored. The processing device, the ROM, and the RAM are connected to each other through a bus 504. The input / output (I / O) interface 505 is also connected to the bus 504.

[0176] Generally, the following devices can be connected to the I / O interface: input devices including, for example, touch screens, touch pads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. The communication device 509 can allow the electronic device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows an electronic device with various devices, it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices can be implemented or had.

[0177] The present application also provides an embodiment of a computer-readable storage medium.

[0178] The above computer-readable storage medium is applied to an electronic device and carries one or more computer programs. When the above one or more computer programs are executed by the electronic device, the electronic device constructs a three-dimensional semantic mapping structure based on multiple information sources. The three-dimensional semantic mapping structure includes multiple types of core knowledge nodes; uses multiple relationships to perform semantic riveting on the three-dimensional semantic mapping structure to form a minimum closed subgraph, and stores the minimum closed subgraph in a graph database to obtain a pathological knowledge base; responds to a user's query request, generates and feeds back a final answer to the user for the patient message input by the user. It can be seen from the above description that this solution can systematically and standardly construct and continuously improve the technical means of a pathological knowledge base with both professional depth and interpretable semantic dimensions, so as to provide an authoritative, easy-to-understand and extensible knowledge base for multi-scenario intelligent applications.

[0179] It should be noted that the computer-readable medium in the present disclosure may be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.

[0180] In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable signal medium may send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0181] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other.

[0182] Although the preferred embodiments of the embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

[0183] Finally, it should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or terminal device comprising the element.

[0184] The technical solutions provided by the present invention have been introduced in detail above. Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, there will be changes in the specific implementation manners and application scopes according to the idea of the present invention. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for constructing a pathological knowledge base, applied to an electronic device, characterized in that The construction method includes the steps of: Constructing a three-dimensional semantic mapping structure based on multiple information sources, where the three-dimensional semantic mapping structure includes multiple types of core knowledge nodes; Semantically riveting the three-dimensional semantic mapping structure using multiple relationships to form a minimally closed subgraph with complete three-dimensional semantics, and storing the minimally closed subgraph in a graph database to obtain the pathological knowledge base; Optimizing the pathological knowledge base according to user feedback.

2. The construction method according to claim 1, characterized in that, The multiple information sources include authoritative medical guidelines and literature, expert interview records, and real patient consultation records; The multiple types of core knowledge nodes include professional concept nodes, decision rule nodes, and candidate popular explanation nodes.

3. The construction method according to claim 2, characterized in that, The constructing of the three-dimensional semantic mapping structure based on multiple information sources includes the steps of: Using a named entity recognition model in the medical field to extract from the information sources to obtain the professional concept nodes; Extracting a premise-conclusion type logical structure from the information sources by combining syntactic dependency analysis and rule template mining methods to obtain the decision rule nodes; Invoking a large model fine-tuned with pathological corpus to perform multi-layer semantic rewriting on the professional terms in the information sources, generating multiple versions of popular explanations understandable by patients, and constructing a term alignment mapping based on the popular explanations to obtain the candidate popular explanation nodes; Optimizing the professional concept nodes, the decision rule nodes, and / or the candidate popular explanation nodes based on structured expert knowledge or opinions.

4. The construction method according to claim 1, characterized in that The multiple relationships include synonym link relationships, rule trigger relationships, and explanation binding relationships, where: The rule trigger relationship is used to establish a directed logical relationship between a concept node and the diagnosis and treatment rules it participates in, supporting many-to-many mapping of premises and results; The explanation binding relationship is used to bind a popular explanation node to the corresponding professional term, generating multi-layer expression versions by distinguishing explanation methods through semantic tags.

5. The construction method according to any one of claims 1 to 4, characterized in that It also includes the steps of: Responding to a user's query request, generating and feeding back a final answer to the user for the patient message input by the user.

6. The construction method according to claim 5, characterized in that Responding to a user's query request, generating and feeding back a final answer to the user for the patient message input by the user, including the steps of: Performing entity recognition on the patient message based on a large model to obtain concept nodes; Using the concept nodes as anchors to perform K-order semantic expansion or attribute matching in the graph database, obtaining rules and explanation nodes associated with the concept nodes to form a closed subgraph; Traversing all nodes in the closed subgraph to obtain the final answer.

7. A construction device for a pathological knowledge base, applied to an electronic device, characterized in that The construction device includes: A three-piece construction module configured to construct a three-dimensional semantic mapping structure based on multiple information sources, where the three-dimensional semantic mapping structure includes multiple types of core knowledge nodes; A knowledge base construction module configured to semantically rivet the three-dimensional semantic mapping structure using multiple relationships to form a minimally closed subgraph, and storing the minimally closed subgraph in a graph database to obtain the pathological knowledge base; A model optimization module configured to optimize the pathological knowledge base according to user feedback.

8. The constructing device according to claim 7, characterized in that It also includes: An information query response module configured to respond to a user's query request, generating and feeding back a final answer to the user for the patient message input by the user.

9. An electronic device, characterized in that, The electronic device includes at least one processor and a memory connected to the processor, wherein: The memory is used for storing computer programs or instructions; The processor is used for executing the computer programs or instructions, so that the electronic device can implement the construction method described in any one of claims 1 to 6.

10. A computer-readable storage medium, applied to an electronic device, characterized in that, The storage medium carries one or more computer programs, and the one or more computer programs can be executed by the electronic device, so that the electronic device can implement the construction method described in any one of claims 1 to 6.

Images