General practice robot multi-disease collaborative diagnosis method and system based on knowledge graph

Abstract

This invention relates to the field of intelligent medical technology and proposes a method and system for collaborative diagnosis of multiple diseases using a general practice robot based on a knowledge graph. The method includes: jointly extracting element relationships from a pre-set medical knowledge base to generate medical triples; constructing a medical knowledge graph for the general practice robot based on the triples; converting multimodal clinical data into standardized clinical features according to the graph's rule framework; obtaining disease nodes by retrieving and matching the knowledge graph with clinical features, and generating candidate reasoning paths through multi-hop traversal; extracting path features to complete confidence scoring and screening, retaining valid reasoning paths, and thereby locking potential disease nodes and initial pathological evidence chains; performing temporal analysis on the diagnostic and treatment events within the evidence chain to sort out the pathological evolution sequence, and constructing a dynamic disease causal chain by combining the logic of adjacent events in the sequence; and conducting multi-disease association mining based on this causal chain to output a complete disease collaborative association structure. This invention can improve the efficiency of collaborative diagnosis of multiple diseases using a general practice robot.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and in particular to a method and system for collaborative diagnosis of multiple diseases using a general practitioner robot based on knowledge graphs. Background Technology

[0002] Existing knowledge graph-based intelligent diagnostic computing frameworks lack standardized multimodal data fusion processing procedures. It is difficult to achieve unified ontology alignment and semantic disambiguation for heterogeneous clinical data, such as test indicators and disease course texts, leading to feature bias when mapping data to knowledge graph entity nodes. Furthermore, traditional retrieval methods only support shallow single-node matching and lack multi-hop path traversal and quantitative confidence filtering mechanisms. This results in numerous meaningless reasoning branches during the computation process, making it difficult to quickly generate logically reliable links of pathological evidence.

[0003] Current intelligent diagnostic computing architectures lack the logical framework for analyzing the temporal sequence of diagnostic events. This prevents the orderly arrangement of disease progression events into an evolutionary sequence based on time, hinders the automatic derivation of dynamic causal relationships between different symptoms, and makes it difficult to batch-mine collaborative structures under multiple coexisting diseases. The output only presents isolated symptom entities, lacking complete temporal deduction and comorbidity linkage logic, resulting in insufficient overall computational accuracy and output completeness. Therefore, improving the computational accuracy and analytical completeness of multi-disease clinical data reasoning in general practice robots has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a knowledge graph-based method and system for collaborative diagnosis of multiple diseases using a general practitioner robot, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot, comprising:

[0006] D1. Perform joint extraction of element relationships on the preset medical knowledge base to obtain the medical triplet of the medical knowledge base, and construct the robot's general medical knowledge graph based on the medical triplet.

[0007] D2. Based on the rule framework of the general practice knowledge graph, perform semantic mapping on the patient's multimodal clinical data to obtain the patient's clinical characteristics;

[0008] D3. Based on the clinical characteristics, perform matching and retrieval on the general medical knowledge graph to obtain the patient's symptom nodes, and perform multi-hop traversal on the symptom nodes to obtain the candidate reasoning paths of the symptom nodes.

[0009] D4. Based on the path characteristics of the candidate inference path, the confidence of the candidate inference path is evaluated to obtain the effective inference path of the symptom node, and based on the effective inference path, the potential symptom node and the initial pathological evidence chain of the patient are determined.

[0010] D5. Perform time-series analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determine the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence.

[0011] D6. Based on the dynamic causal chain of the disease, perform multi-disease association mining on the potential disease nodes to obtain the disease collaborative association structure of the patient.

[0012] In a preferred embodiment, the step of performing joint extraction of element relationships from a preset medical knowledge base to obtain medical triples of the medical knowledge base, and constructing a general medical knowledge graph for the robot based on the medical triples, includes:

[0013] A preset medical knowledge base is obtained, and the structure of the medical knowledge base is parsed to obtain the semantic framework and entity type annotation information of the medical knowledge base;

[0014] Based on the semantic framework and the entity type annotation information, attribute anchoring is performed on the medical knowledge base to obtain the medical elements in the medical knowledge base.

[0015] Based on the semantic relationships of the medical elements, the medical knowledge base is merged by relation type to obtain the medical triples of the medical knowledge base;

[0016] The medical triples are stored in a structured serialization to obtain the robot's general medical knowledge graph.

[0017] In a preferred embodiment, the step of semantically mapping the patient's multimodal clinical data according to the rule framework of the general practice knowledge graph to obtain the patient's clinical characteristics includes:

[0018] Modality separation was performed on the patient's multimodal clinical data to obtain the patient's structured test index data and unstructured medical record text;

[0019] Based on the rule framework of the general practice knowledge graph, ontology alignment is performed on the structured test index data to obtain the standardized test indexes of the patient.

[0020] Key medical terms are extracted from the unstructured medical record text to obtain the text feature words of the patient;

[0021] Based on the rule framework of the general practice medical knowledge graph, semantic disambiguation is performed on the text feature words to obtain the standardized symptom description of the patient;

[0022] The standardized test indicators and the standardized symptom descriptions are fused to obtain the patient's clinical characteristics.

[0023] In a preferred embodiment, the step of matching and retrieving the general medicine knowledge graph based on the clinical features to obtain the patient's symptom nodes, and performing multi-hop traversal on the symptom nodes to obtain candidate reasoning paths for the symptom nodes, includes:

[0024] The clinical features are parsed to obtain the feature nodes of the clinical features;

[0025] The feature nodes are matched with the nodes in the general medicine knowledge graph to obtain the patient's symptom nodes;

[0026] Using the symptom node as the root node, the general medical knowledge graph is traversed hierarchically to obtain the full path association subgraph of the patient.

[0027] Based on a preset logical confidence threshold, the endpoint nodes of the full path association subgraph are filtered to obtain the candidate reasoning path of the patient.

[0028] In a preferred embodiment, the step of evaluating the confidence level of the candidate inference paths based on their path characteristics to obtain the effective inference paths for the symptom nodes, and determining the potential symptom nodes and initial pathological evidence chain for the patient based on the effective inference paths, includes:

[0029] The nodes and relation edges in the candidate reasoning path are parsed to obtain the semantic association strength and structural connectivity of the candidate reasoning path;

[0030] Based on the semantic association strength and the structural connectivity, the candidate inference paths are weighted and calculated to obtain a comprehensive confidence score for the candidate inference paths;

[0031] Based on the comparison results between the comprehensive confidence score and the preset screening threshold, the candidate inference paths are filtered to obtain the effective inference paths for the symptom nodes.

[0032] By utilizing the logical pointers of the nodes in the effective reasoning path, the potential symptom nodes of the patient are determined;

[0033] The nodes on the effective reasoning path are used as pathological elements, and the relational edge attributes on the effective reasoning path are used as the association logic.

[0034] Following the topological order of the effective reasoning path, the pathological elements and the associated logic are sequentially assembled to obtain the patient's initial pathological evidence chain.

[0035] In a preferred embodiment, the formula for calculating the comprehensive confidence score is as follows:

[0036] ;

[0037] in, The overall confidence score is... For the first The semantic association strength of the relation edges, For the first The structural connectivity of the relation edges, The number of relation edges in the candidate reasoning path. This is the preset path attenuation coefficient.

[0038] In a preferred embodiment, the step of performing time-series analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determining the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence, includes:

[0039] The timing of the diagnosis and treatment events in the initial pathological evidence chain is obtained by parsing the timestamps of the events.

[0040] Based on the occurrence time, the diagnostic and treatment events are arranged chronologically to obtain the patient's pathological evolution sequence;

[0041] Based on the pathological evolution sequence, the logical relationship between adjacent events of the diagnosis and treatment events is determined to obtain the association logic of the diagnosis and treatment events;

[0042] Based on the aforementioned association logic, causal connections are made between the diagnostic and treatment events in the pathological evolution sequence to obtain the dynamic causal chain of the patient's symptoms.

[0043] In a preferred embodiment, the step of determining the logical relationship between adjacent events of the diagnostic and treatment events based on the pathological evolution sequence to obtain the association logic of the diagnostic and treatment events includes:

[0044] Event extraction is performed on adjacent diagnostic and treatment events in the pathological evolution sequence to obtain the preceding and subsequent events of the diagnostic and treatment events.

[0045] Substitute the preceding event and the following event into the general medicine knowledge graph and perform a relationship retrieval to obtain candidate relationship edges pointing from the preceding event to the following event;

[0046] Based on the relationship type of the candidate relationship edges, logical matching is performed on the preceding events and the following events to obtain the association logic of the diagnosis and treatment events.

[0047] In a preferred embodiment, the step of performing multi-disease association mining on the potential disease nodes based on the dynamic symptom causal chain to obtain the patient's symptom collaborative association structure includes:

[0048] Based on the type of the diagnosis and treatment event, the potential symptom nodes are filtered for related events to obtain the related symptom nodes of the dynamic symptom causal chain;

[0049] Based on the associated symptom nodes, subgraph extraction is performed on the general medicine knowledge graph to obtain the co-disease association paths between the associated symptom nodes;

[0050] Semantic logic analysis is performed on the co-disease association paths to obtain the co-disease association relationships between the associated symptom nodes;

[0051] With the associated symptom node as the center and the co-disease association relationship as the edge, construct the symptom collaborative association structure of the patient.

[0052] To address the aforementioned problems, this invention also provides a knowledge graph-based multi-disease collaborative diagnosis system for general practitioners, the system comprising:

[0053] The graph construction module is used to perform joint extraction of element relationships from a preset medical knowledge base to obtain medical triples of the medical knowledge base, and to construct a general medical knowledge graph of the robot based on the medical triples.

[0054] The clinical feature acquisition module is used to perform semantic mapping on the patient's multimodal clinical data according to the rule framework of the general practice medical knowledge graph to obtain the patient's clinical features;

[0055] The symptom path mining module is used to match and retrieve the general medical knowledge graph based on the clinical features to obtain the symptom nodes of the patient, and to perform multi-hop traversal of the symptom nodes to obtain the candidate reasoning paths of the symptom nodes.

[0056] The path confidence assessment module is used to assess the confidence of the candidate inference path based on the path characteristics of the candidate inference path, obtain the effective inference path of the symptom node, and determine the potential symptom node and the initial pathological evidence chain of the patient based on the effective inference path.

[0057] The temporal causal deduction module is used to perform temporal analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determine the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence.

[0058] The comorbidity association analysis module is used to perform multi-disease association mining on the potential disease nodes based on the dynamic disease causal chain to obtain the disease synergistic association structure of the patient.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. This invention generates standardized medical triples by jointly extracting elements from a medical knowledge base and constructs a general medical knowledge graph. Based on the graph rules, it completes the separation, standardization transformation, and feature fusion of multimodal clinical data. It obtains the full path association subgraph by combining multi-hop traversal and uses a weighted confidence formula that integrates semantic association strength, structural connectivity, and path decay coefficient to complete path filtering. This can efficiently filter invalid reasoning branches, quickly match symptom nodes, and generate a logically ordered initial pathological evidence chain, significantly improving the computational efficiency of clinical data parsing, graph retrieval, and path reasoning.

[0061] 2. This invention automatically determines the inherent logical relationship between adjacent events and constructs a dynamic causal chain of diseases by parsing the timestamps and sorting the time sequence of diagnostic and treatment events within the pathological evidence chain. Based on the causal chain, it mines the co-occurrence path of multiple diseases, builds a complete symptom synergistic association structure, and can simultaneously output the time sequence evolution of the disease course and the linkage relationship of multiple diseases, enriching the data dimensions of intelligent diagnosis output, ensuring the continuity and traceability of pathological deduction logic, and effectively improving the rigor and comprehensiveness of the results of joint analysis of multiple diseases. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating a knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot, provided in an embodiment of the present invention.

[0063] Figure 2 This is a functional module diagram of a knowledge graph-based general practitioner robot collaborative diagnosis system for multiple diseases, provided in an embodiment of the present invention.

[0064] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0065] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0066] This application provides a knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot. The execution entity of this knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0067] Reference Figure 1 The diagram shown is a flowchart illustrating a knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot, according to an embodiment of the present invention. In this embodiment, the knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot includes:

[0068] D1. Perform joint extraction of element relationships on the preset medical knowledge base to obtain the medical triplet of the medical knowledge base, and construct the robot's general medical knowledge graph based on the medical triplet.

[0069] In this embodiment of the invention, the step of performing joint extraction of element relationships from a preset medical knowledge base to obtain medical triples of the medical knowledge base, and constructing a general medical knowledge graph for the robot based on the medical triples, includes:

[0070] A preset medical knowledge base is obtained, and the structure of the medical knowledge base is parsed to obtain the semantic framework and entity type annotation information of the medical knowledge base;

[0071] Based on the semantic framework and the entity type annotation information, attribute anchoring is performed on the medical knowledge base to obtain the medical elements in the medical knowledge base.

[0072] Based on the semantic relationships of the medical elements, the medical knowledge base is merged by relation type to obtain the medical triples of the medical knowledge base;

[0073] The medical triples are stored in a structured serialization to obtain the robot's general medical knowledge graph.

[0074] This process involves pre-integrating authoritative disease diagnosis and treatment guidelines, national pharmacopoeias, clinical pathway standards, medical textbooks, and publicly available clinical case databases covering all medical specialties, including internal medicine, surgery, gynecology, and pediatrics. A complete pre-defined medical knowledge base is then formed by summarizing these comprehensive medical knowledge entries, including clinical diagnosis and treatment guidelines, drug usage instructions, human physiology and pathology records, and standardized medical treatment items. The entire text segment within this pre-defined medical knowledge base is then broken down into paragraphs, and the accompanying text and images are analyzed to complete the structural analysis. A semantic framework is generated by sorting and differentiating the fixed semantic expression formats within the pre-defined medical knowledge base. Simultaneously, the text is labeled with all disease, drug, symptom, examination instrument, and human tissue categories, encompassing seven core entities: disease, symptoms, drugs, laboratory indicators, examination methods, human anatomy, and complications. Entity annotation employs the BIO sequence annotation system, utilizing an entity-relationship joint extraction network based on a pre-trained medical BERT model for automatic annotation. This is supplemented by manual verification by at least two licensed physicians to ensure an entity annotation accuracy of no less than 95%. The generated entity type annotation information and semantic framework are then stored as the direct output of this operation.

[0075] Based on the semantic framework generated by the previous operation, each fixed semantic expression position in the preset medical knowledge base is locked. Relying on the entity type annotation information, the text area corresponding to all entity content inside the text is located. The attribute anchoring operation is completed for the fixed entity binding corresponding inherent description content one by one. All entities that have completed the binding operation and their accompanying descriptions are uniformly organized and summarized to form medical elements. The medical elements fully inherit all the annotation content of the semantic framework and entity type annotation information.

[0076] The semantic relationships between medical elements are identified by extracting the internal text of each medical element and identifying the mutual descriptions contained within the text. Semantic relationships expressing the same type of interaction logic are uniformly classified into the same category to complete the relationship type merging operation. After merging, six core relationship types are formed: "Induction", "Clinical Manifestations", "Auxiliary Diagnosis", "Drug Treatment", "Complications", and "Pathophysiological Changes". Each relationship type is further subdivided into two secondary relationships. A single medical element is used as the head entity to correspond to the merged relationship type, and another medical element is used as the tail entity to complete the combination construction. All the combined entity relationship pairings are uniformly organized to generate medical triples. Medical triples fully contain all medical elements and the various semantic relationships after merging.

[0077] All generated medical triples are sequentially split into individual entries according to a unified and fixed arrangement order. They are then stored sequentially according to the fixed arrangement order of head entity, relation type, and tail entity. Each complete medical triple is stored in an independent storage unit. The Neo4j graph database is used for structured storage, and an inverted index of entity name and an index of relation type are built simultaneously. Millisecond-level node retrieval and multi-hop path traversal are supported. All storage units are integrated and summarized to form a complete general medical knowledge graph that can be called and read by robots. The general medical knowledge graph fully contains all the medical triples that have been serialized and stored.

[0078] The beneficial effects are as follows: relying on a fully-established pre-built medical knowledge base, semantic and entity information is decomposed layer by layer, and element extraction, relationship integration, and graph structured storage are completed in layers. The entire process relies on standardized semantic annotation to anchor medical entity elements, and various medical semantic relationships are uniformly organized to generate standardized medical triples. Then, through ordered serialized storage, a general medical knowledge graph is built. This can completely preserve all entities and related logic within the original medical knowledge base, eliminate the problem of scattered, messy, and unrelated medical knowledge content, and generate a general medical knowledge graph with a well-organized and logically clear structure. It can be directly adapted to the robot's subsequent multimodal clinical data semantic mapping, disease matching retrieval, multi-hop path reasoning, and other general diagnosis and treatment related operations, thereby improving the completeness and efficiency of the robot's reading and parsing of medical knowledge content.

[0079] D2. Based on the rule framework of the general practice knowledge graph, perform semantic mapping on the patient's multimodal clinical data to obtain the patient's clinical characteristics;

[0080] In this embodiment of the invention, the step of semantically mapping the patient's multimodal clinical data according to the rule framework of the general practice knowledge graph to obtain the patient's clinical characteristics includes:

[0081] Modality separation was performed on the patient's multimodal clinical data to obtain the patient's structured test index data and unstructured medical record text;

[0082] Based on the rule framework of the general practice knowledge graph, ontology alignment is performed on the structured test index data to obtain the standardized test indexes of the patient.

[0083] Key medical terms are extracted from the unstructured medical record text to obtain the text feature words of the patient;

[0084] Based on the rule framework of the general practice medical knowledge graph, semantic disambiguation is performed on the text feature words to obtain the standardized symptom description of the patient;

[0085] The standardized test indicators and the standardized symptom descriptions are fused to obtain the patient's clinical characteristics.

[0086] The patient's multimodal clinical data was completely disassembled, containing various information carriers including images, text, and numerical records. The multimodal clinical data covers three core clinical data categories: laboratory test reports, electronic medical record progress notes, and textual conclusions of imaging examinations. Based on the storage format of the information carriers, two independent information contents were divided: information in digital table format that can be directly read from the numerical range boundaries was separately collected to form structured test indicator data; and diagnostic and treatment text records written in free paragraphs without a fixed table format were separately collected to form unstructured progress notes. Textual conclusions of imaging examinations were simultaneously included in the category of unstructured progress notes for unified processing. The structured test indicator data and unstructured progress notes were the direct output of this operation, and all original clinical information was completely preserved.

[0087] The system retrieves the classification standards of the test indicators recorded within the rule framework of the pre-built general practice knowledge graph. It reads the original indicator name corresponding to each set of values ​​in the structured test indicator data one by one, matches each original indicator name with the unified standard indicator ontology built into the rule framework of the general practice knowledge graph, replaces the original custom indicator name with the unified standard name of the graph, and simultaneously matches the corresponding indicator normal value range classification definition to complete the alignment process. All indicator content after unified standardization conversion is summarized to form standardized test indicators. The standardized test indicators are completely consistent with the rule framework ontology definition specification of the general practice knowledge graph.

[0088] The entire text of the unstructured medical record is read through sentence by sentence. The text is compared word by word with the complete list of medical terms recorded within the framework of the general medical knowledge graph. The text fragments that record disease symptoms, body parts, examination methods and medication names are accurately located. All successfully matched text fragments are completely extracted without deletion or modification. All extracted text fragments are compiled and organized into text feature words. The text feature words are completely preserved, containing all core text content with medical reference value in the medical record.

[0089] The semantic distinction standards for terms recorded within the rule framework of the general practice medical knowledge graph are retrieved. All contextual disease-related text content corresponding to each individual text feature word is read one by one. Combined with the polysemous distinction logic of terms recorded within the rule framework, the unique medical definition corresponding to each text feature word in the current context is determined. Ambiguous interpretations of text feature words that do not match the current diagnosis and treatment scenario are removed. Only the standard interpretation content suitable for the current patient's diagnosis and treatment scenario is retained to complete the semantic disambiguation operation. The text content after unified and standardized expression through interpretation filtering is summarized to form a standardized symptom description. The standardized symptom description uniformly follows the semantic definitions of the rule framework of the general practice medical knowledge graph and does not contain any polysemous or confusing content.

[0090] The system collects two independent and standardized medical information categories: standardized test indicators and standardized symptom descriptions. It integrates these information according to the rules and framework of the general medicine knowledge graph and the pre-set feature combination logic. It establishes a binding association between a single standardized test indicator and its corresponding standardized symptom description. It integrates all the two types of standardized information that have been bound together to form a complete and unified data unit. All the integrated data units are summarized to form the patient's clinical characteristics. The clinical characteristics simultaneously carry two types of complete patient diagnosis and treatment content: quantitative test data and textual symptom information.

[0091] The beneficial effects are that by splitting multimodal clinical data into modalities, clinical information in different formats can be processed separately and in a standardized manner. Based on the rule framework of the general practice medical knowledge graph, the alignment of test indicators and semantic disambiguation of symptom terms are completed separately, unifying the expression of all clinical information. Then, quantitative indicators and textual symptoms are integrated to generate complete clinical features, avoiding the problem of information not matching the graph due to the messy format and ambiguous terminology of multimodal clinical data. The standardized clinical features can be directly connected to the general practice medical knowledge graph to carry out disease matching and retrieval operations, ensuring that the patient's clinical information and the medical entities in the graph are accurately matched in the subsequent diagnosis and treatment reasoning process, and improving the data accuracy of subsequent disease reasoning operations.

[0092] D3. Based on the clinical characteristics, perform matching and retrieval on the general medical knowledge graph to obtain the patient's symptom nodes, and perform multi-hop traversal on the symptom nodes to obtain the candidate reasoning paths of the symptom nodes.

[0093] In this embodiment of the invention, the step of matching and retrieving the general medical knowledge graph based on the clinical features to obtain the patient's symptom nodes, and performing multi-hop traversal on the symptom nodes to obtain candidate reasoning paths for the symptom nodes, includes:

[0094] The clinical features are parsed to obtain the feature nodes of the clinical features;

[0095] The feature nodes are matched with the nodes in the general medicine knowledge graph to obtain the patient's symptom nodes;

[0096] Using the symptom node as the root node, the general medical knowledge graph is traversed hierarchically to obtain the full path association subgraph of the patient.

[0097] Based on a preset logical confidence threshold, the endpoint nodes of the full path association subgraph are filtered to obtain the candidate reasoning path of the patient.

[0098] The clinical features are broken down into two categories of information: standardized test indicators and standardized symptom descriptions. The numerical value, classification, and abnormality judgment of each standardized test indicator are extracted. At the same time, the onset site, symptom manifestation, and duration of each standardized symptom description are extracted. All the extracted inherent information is organized and formed according to the unified node construction standard of the general medicine knowledge graph. The completed independent information unit is the feature node. The feature node completely replicates all the patient diagnosis and treatment information carried by the clinical feature.

[0099] All entity nodes stored within the general medical knowledge graph are retrieved. The feature nodes generated in the previous step are compared one by one with the textual definitions and attribute information carried by each type of node in the graph. The comparison adopts the attribute weighted similarity calculation rule, with symptom manifestation attribute weight accounting for 50%, onset site weight accounting for 30%, and the combined weight of onset duration and abnormality of indicators accounting for 20%. If the overall similarity is higher than 0.8, it is considered a successful match. The graph node is then locked as the symptom node that fits the patient's actual diagnosis and treatment situation. The symptom node is a standard medical entity node in the general medical knowledge graph that completely matches the patient's clinical information and completely corresponds to all the abnormal symptoms and abnormal test results currently appearing in the patient.

[0100] The selected symptom nodes are set as the fixed starting root nodes for the hierarchical expansion traversal operation. Following the pre-stored entity relationships between nodes within the general medical knowledge graph, the operation extends outward layer by layer to read connected nodes. After reading the first-level adjacent nodes directly associated with the root node, the operation continues to read the second-level adjacent nodes associated with the first-level adjacent nodes. This process continues uninterruptedly, extending outward to read multiple layers of connected nodes and the connections between nodes. The traversal uses a breadth-first search algorithm with a maximum traversal jump of 5 hops. During the traversal, duplicate nodes and backtracking paths are removed. At each hop, the top 20 adjacent nodes with the strongest semantic association are retained to complete pruning and avoid path explosion. All nodes and connections generated during the traversal are fully recorded. All recorded content is integrated to form a complete full-path association subgraph that covers multiple layers of association relationships. The full-path association subgraph contains all hierarchical association links generated by extending outward from the symptom nodes.

[0101] A fixed logical confidence threshold is pre-defined based on the general medical logic of general practice diagnosis and treatment. The logical confidence threshold corresponds to the basic judgment criteria that the links meet the conventional pathological development logic. The logical confidence threshold is set as follows: the semantic similarity between the first and last nodes of the path is not less than 0.6, and the path contains at least one disease-type entity node. The endpoint node corresponding to the end of each link in the full path association subgraph is read one by one. The association logic between the nodes of each link is checked one by one to see if it meets the judgment criteria corresponding to the logical confidence threshold. Only complete links whose overall association logic meets the fixed criteria are retained, and invalid links whose link logic does not meet the criteria are eliminated. All the retained complete and valid links are uniformly summarized to form candidate inference paths. The candidate inference paths are all multi-layer association links in the full path association subgraph that meet the basic medical logic criteria.

[0102] The beneficial effects are as follows: First, clinical features are decomposed to generate feature nodes that can be connected to the atlas. Then, the symptom nodes corresponding to the patient's condition are accurately located through full-dimensional attribute matching. Next, the atlas is traversed in multiple layers starting from the symptom nodes to obtain complete related subgraphs. Based on fixed medical logic standards, compliant candidate reasoning paths are selected. This can fully explore the multi-layered medical relationships of the symptoms, avoid missing potential related pathological links in single-level retrieval, filter invalid related links that obviously violate basic pathological common sense, unify and standardize the selection criteria for reasoning links, and the generated candidate reasoning paths can be directly used for subsequent confidence assessment and pathological causal deduction, improving the comprehensiveness and basic rationality of the discovery of diagnosis and treatment reasoning links.

[0103] D4. Based on the path characteristics of the candidate inference path, the confidence of the candidate inference path is evaluated to obtain the effective inference path of the symptom node, and based on the effective inference path, the potential symptom node and the initial pathological evidence chain of the patient are determined.

[0104] In this embodiment of the invention, the step of evaluating the confidence level of the candidate inference paths based on their path characteristics to obtain the effective inference paths for the symptom nodes, and determining the potential symptom nodes and initial pathological evidence chain for the patient based on the effective inference paths, includes:

[0105] The nodes and relation edges in the candidate reasoning path are parsed to obtain the semantic association strength and structural connectivity of the candidate reasoning path;

[0106] Based on the semantic association strength and the structural connectivity, the candidate inference paths are weighted and calculated to obtain a comprehensive confidence score for the candidate inference paths;

[0107] Based on the comparison results between the comprehensive confidence score and the preset screening threshold, the candidate inference paths are filtered to obtain the effective inference paths for the symptom nodes.

[0108] By utilizing the logical pointers of the nodes in the effective reasoning path, the potential symptom nodes of the patient are determined;

[0109] The nodes on the effective reasoning path are used as pathological elements, and the relational edge attributes on the effective reasoning path are used as the association logic.

[0110] Following the topological order of the effective reasoning path, the pathological elements and the associated logic are sequentially assembled to obtain the patient's initial pathological evidence chain.

[0111] The formula for calculating the comprehensive confidence score is as follows:

[0112] ;

[0113] in, The overall confidence score is... For the first The semantic association strength of the relation edges, For the first The structural connectivity of the relation edges, The number of relation edges in the candidate reasoning path. This is the preset path attenuation coefficient.

[0114] Each candidate reasoning path is broken down into its own inherent attributes, including all entity nodes and the relationships between them. The semantic association strength is formed by summarizing the medical semantic fit information of each pair of adjacent nodes based on their corresponding relationships. The structural connectivity is formed by summarizing the complete connection information of the nodes in the entire candidate reasoning path without interruption or branch breakage. The semantic association strength and structural connectivity fully carry the two core path features of the candidate reasoning path: semantic matching degree and link integrity.

[0115] After attribute parsing, all relation edges within a single candidate reasoning path generate semantic association strength and structural connectivity. The semantic association strength of a single relation edge is assigned based on the level of evidence-based medicine evidence corresponding to the relation, with a value range of [0, 1]. The values ​​are directly assigned according to the general evidence-based medicine grading standards: Level I, high-quality evidence-based evidence such as large-sample randomized controlled trials and meta-analyses, corresponds to a value of 1.0; Level II, prospective cohort studies, corresponds to a value of 0.8; Level III, retrospective case-control studies, corresponds to a value of 0.6; Level IV, case series and case reports, corresponds to a value of 0.4; and evidence without clinical trial support, but based on expert consensus reached by multiple physicians, corresponds to a value of 0.2. During the graph construction phase, each relation edge is bound with an original literature source tag, allowing the system to automatically identify the corresponding evidence level and complete the score matching without additional complex calculations, thus uniformly quantifying the medical logical credibility of a single relation edge. The structural connectivity of a single relation edge depends on the completeness of the connection between the nodes at both ends of the relation edge without any breaks. The edge is directly generated, with a value range of [0, 1]. The structural connectivity of the edge is the normalized ratio of the total frequency of the current relation edge in the entire general medical knowledge graph to the highest frequency of all relation edges under the same major category. The total number of relation edges within a single candidate reasoning path is counted to obtain the number of relation edges. The path decay coefficient is calibrated using a large number of general medical case reasoning samples, grouped according to different link lengths, and the matching degree of the scores after substituting multiple sets of constants with the credibility of physician annotations is compared. The appropriate value of 0.2 is locked. The value of 0.2 is chosen because when the coefficient is greater than 0.3, it will over-compress the scores of long paths and mistakenly screen out compliant comorbidity links, while less than 0.1 will be difficult to suppress excessively long invalid paths. The value of 0.2 allows the score to decay smoothly as the link length increases, taking into account the clinical evaluation criteria of both long and short reasoning paths. The summation operation traverses each independent relation edge in the candidate reasoning path to complete the item-by-item value selection. All parameters are derived from the path features output by the candidate reasoning path attribute parsing stage and the pre-defined medical reasoning constraint standards.

[0116] The overall calculation process consists of two layers of computational logic. The first layer multiplies the semantic association strength and structural connectivity of the same relation edge to quantify and integrate the reliability of the matching of a single link. The second layer performs a step-by-step accumulation operation on the reliability values ​​of all relation edges in the path to obtain the basic reliability total value of the entire path. The third layer uses exponential decay operation to uniformly correct the accumulated basic reliability total value. The correction operation is used to offset the problem of decreased credibility of pathological logic caused by the increase in the number of hops in the inference path. The entire calculation finally outputs a comprehensive confidence score. The comprehensive confidence score fully quantifies the overall reliability of a single candidate inference path in conforming to clinical pathological rules, and provides a unified and standardized judgment basis for subsequent comparison with preset screening thresholds to filter invalid inference links.

[0117] When the number of relation edges within a candidate inference path remains constant, the higher the corresponding values ​​of semantic association strength and structural connectivity of a single relation edge, the higher the reliability value obtained by multiplying a single link segment. The total base reliability value after accumulating all relation edges also increases synchronously, and the final output comprehensive confidence score increases synchronously. When the semantic association strength and structural connectivity of all relation edges within a candidate inference path remain constant, the continuous increase in the total number of relation edges contained in the path will amplify the numerical compression effect brought about by exponential decay, and the final output comprehensive confidence score will continuously decrease. When the path decay coefficient remains constant, the decay compression magnitude is only determined by the total number of relation edges in the path. The comprehensive confidence score of the entire path decreases monotonically with the increase of path length and increases monotonically with the increase of the reliability of matching a single relation edge segment.

[0118] A fixed preset screening threshold is pre-defined based on general practice clinical pathology and diagnosis knowledge. This threshold is calibrated by simulation testing using a dataset of real general practice clinical cases. During the test, all values ​​from 0 to 1 are traversed, and the diagnostic miss rate and misdiagnosis rate are statistically analyzed at each threshold: when the threshold is below 0.3, a large number of invalid paths with extremely low credibility are retained, and the misdiagnosis rate exceeds 40%; when the threshold is above 0.7, long inference paths for multiple diseases are all eliminated, and the miss rate for patients with multiple coexisting diseases exceeds 35%; a threshold of 0.5 is the optimal balance point for balancing misdiagnosis and miss diagnosis, and is used as the default benchmark value; a threshold of [0.3, 0.5] is suitable for primary screening at the grassroots level, and a threshold of [0.5, 0.5] is suitable for primary screening. [7] Applicable to precise review of specialist cases, the default preset screening threshold is set to 0.5, with a value range of [0.3, 0.7]. The comprehensive confidence score corresponding to each candidate reasoning path is retrieved one by one, and each comprehensive confidence score is compared with the preset screening threshold. All candidate reasoning paths with comprehensive confidence scores higher than the preset screening threshold are retained, and all candidate reasoning paths with comprehensive confidence scores lower than the preset screening threshold are removed. All complete reasoning links retained after the judgment are uniformly collected to form the effective reasoning path corresponding to the symptom node. All effective reasoning paths meet the basic medical logic credibility standard.

[0119] Read the fixed logical pointing relationships formed by the relation edges between all nodes within a single valid reasoning path one by one. Taking the symptom node as the root node of the traversal as the logical starting point, locate all disease-type entity nodes at the end of the path along the unidirectional logical extension direction of the relation edges within the path. Accumulate the confidence of the same disease node pointed to by multiple paths, and select the top 10 disease nodes with the highest accumulated scores to be included in the candidate range as potential symptom nodes. Potential symptom nodes represent various suspicious related disease entities that can be deduced from the current patient's clinical characteristics.

[0120] All entity nodes on a single valid reasoning path are uniformly classified as pathological elements. Simultaneously, all attribute contents of the relation edges of each connecting node on a single valid reasoning path are extracted. The relation edge attributes completely record the medical association logic at the pathological, symptom, and examination levels between adjacent pathological elements. The pathological elements and association logic completely cover all medical entities and inter-entity interactions contained in the entire valid reasoning path.

[0121] Following the fixed topological arrangement of the effective reasoning path from the initial symptom node to the potential symptom node, the pathological elements and their associated logics at the corresponding positions are retrieved sequentially to complete the orderly splicing operation. The previous pathological element is paired with the corresponding associated logic to connect the next pathological element, thus completely splicing all the elements and logical content of the entire path. The complete and orderly medical logic link after splicing is the patient's initial pathological evidence chain. The initial pathological evidence chain completely restores the complete reasoning basis from the existing clinical characteristics to the suspected potential symptom.

[0122] The beneficial effects are as follows: Path features of candidate inference paths are extracted from two dimensions, path semantics and link structure, to complete multi-dimensional confidence assessment. Invalid inference links with insufficient credibility are screened out by relying on comprehensive confidence scores and fixed screening thresholds. The retained valid inference paths are used to accurately locate the potential disease nodes corresponding to the patient. Then, pathological elements and related logic are integrated in an orderly manner based on the path topology to generate a complete initial pathological evidence chain. The entire process relies on the inherent attributes of the graph to carry out standardized assessment and evidence construction, avoiding the inference bias caused by single-dimensional evaluation of the path. The generated initial pathological evidence chain is logically coherent and has complete derivation basis, which can be directly used for subsequent diagnosis and treatment event time sequence analysis and pathological evolution derivation, greatly improving the traceability and medical rigor of the intelligent diagnosis derivation process.

[0123] D5. Perform time-series analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determine the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence.

[0124] In this embodiment of the invention, the step of performing time-series analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determining the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence, includes:

[0125] The timing of the diagnosis and treatment events in the initial pathological evidence chain is obtained by parsing the timestamps of the events.

[0126] Based on the occurrence time, the diagnostic and treatment events are arranged chronologically to obtain the patient's pathological evolution sequence;

[0127] Based on the pathological evolution sequence, the logical relationship between adjacent events of the diagnosis and treatment events is determined to obtain the association logic of the diagnosis and treatment events;

[0128] Based on the aforementioned association logic, causal connections are made between the diagnostic and treatment events in the pathological evolution sequence to obtain the dynamic causal chain of the patient's symptoms.

[0129] The step of determining the logical relationship between adjacent events of the diagnostic and treatment events based on the pathological evolution sequence to obtain the association logic of the diagnostic and treatment events includes:

[0130] Event extraction is performed on adjacent diagnostic and treatment events in the pathological evolution sequence to obtain the preceding and subsequent events of the diagnostic and treatment events.

[0131] Substitute the preceding event and the following event into the general medicine knowledge graph and perform a relationship retrieval to obtain candidate relationship edges pointing from the preceding event to the following event;

[0132] Based on the relationship type of the candidate relationship edges, logical matching is performed on the preceding events and the following events to obtain the association logic of the diagnosis and treatment events.

[0133] We analyze all the diagnostic and treatment behaviors recorded in the initial pathological evidence chain, identify the clinical time record information corresponding to each diagnostic and treatment behavior, accurately identify the specific time node where the diagnostic and treatment behavior occurred, perform complete analysis and verification of all time record information, clarify the precise occurrence time corresponding to each diagnostic and treatment event, determine the unique occurrence time corresponding to each diagnostic and treatment event, and completely retain the time dimension information of all diagnostic and treatment events.

[0134] Based on the actual occurrence time of all diagnostic and treatment events and following the objective rule of the unidirectional progression of clinical disease development, all diagnostic and treatment events are systematically arranged and sorted in order from earliest to latest time. This process links scattered and independent individual diagnostic and treatment events into a coherent disease development timeline. Parallel diagnostic and treatment events occurring at the same time point are prioritized according to the order of symptom record, test results, and diagnostic and treatment procedures. After overall organization and summarization, a complete pathological evolution sequence that records the timeline of the patient's disease development is formed. This sequence completely replicates the temporal process of the patient's entire disease progression and treatment.

[0135] The diagnostic and treatment events arranged in close succession in the pathological evolution sequence are split into pairs to distinguish between the diagnostic and treatment content that comes first and the diagnostic and treatment content that comes later. The grouping and splitting of all adjacent diagnostic and treatment events is completed, and the preceding and following events in each pair are accurately divided. All medical information, such as symptoms, examination results, and diagnostic and treatment procedures, contained in the two pairs of adjacent events are completely preserved.

[0136] Based on the complete entity and relationship system of the general practice medical knowledge graph, the medical entity information corresponding to the split antecedent and consequent events is completely imported into the graph retrieval system. The system traverses all entity relationships in the graph, matches all the relationship links between the two types of event entities one by one, and locks all relationship links that can realize the logical transmission from the antecedent event to the consequent event. The retrieval scope is limited to three types of causal relationships: "inducing", "aggravating", and "secondary". Parallel descriptions and non-causal relationships such as treatment interventions are excluded. All successfully matched relationship links are summarized to form candidate relationship edges from the antecedent event to the consequent event, covering all compliant medical relationship forms between the two types of events.

[0137] By combining the standardized relationship definitions built into the general medicine knowledge graph, the exclusive medical relationship type corresponding to each candidate relationship edge is checked. Combined with the patient's real clinical diagnosis and treatment scenarios and event content, the relationship logic of each candidate relationship edge and the adaptability of the preceding and following events are verified one by one. When there are multiple candidate relationship edges, they are sorted from high to low according to the semantic association strength, and the relationship edge ranked first is taken as the final association logic. Finally, the real inherent medical connection between two sets of adjacent diagnosis and treatment events is determined, forming the association logic of the precise corresponding diagnosis and treatment events.

[0138] By combining the specific correlation logic corresponding to each group of adjacent diagnosis and treatment events, and taking the chronological arrangement of the pathological evolution sequence as the overall framework, the preceding events in the chronological order are unidirectionally connected to the subsequent events in the chronological order through the corresponding medical correlation logic. The causal connection of all adjacent diagnosis and treatment events is completed group by group, and a progressive, chronologically coherent, and logically closed-loop pathological transmission link is constructed. All the complete pathological links after being connected and integrated are summarized to form the dynamic causal chain of the patient's disease, which fully reflects the core law of the dynamic evolution of the patient's disease and the mutual induction of symptoms.

[0139] The beneficial effects include: accurately analyzing the time information of diagnosis and treatment events to complete temporal reconstruction, objectively restoring the true temporal development of the patient's condition; by splitting adjacent diagnosis and treatment events and combining them with the standardized relationship system of the general medical knowledge graph to match the event association logic, accurately mining the inherent causal relationship between different diagnosis and treatment events and symptom changes, realizing the transformation from static pathological evidence chain to dynamic causal link, effectively making up for the shortcomings of traditional pathological analysis that can only present static diagnosis and treatment information and cannot reflect the dynamic evolution of the disease. The constructed dynamic symptom causal chain is time-sensitive and the causal logic conforms to the laws of clinical medicine, which can provide rigorous dynamic data support for subsequent multi-symptom association mining and accurate disease reasoning, significantly improving the objectivity and accuracy of intelligent diagnosis and treatment reasoning.

[0140] D6. Based on the dynamic causal chain of the disease, perform multi-disease association mining on the potential disease nodes to obtain the disease collaborative association structure of the patient.

[0141] In this embodiment of the invention, the step of performing multi-disease association mining on the potential disease nodes based on the dynamic disease causal chain to obtain the patient's disease collaborative association structure includes:

[0142] Based on the type of the diagnosis and treatment event, the potential symptom nodes are filtered for related events to obtain the related symptom nodes of the dynamic symptom causal chain;

[0143] Based on the associated symptom nodes, subgraph extraction is performed on the general medicine knowledge graph to obtain the co-disease association paths between the associated symptom nodes;

[0144] Semantic logic analysis is performed on the co-disease association paths to obtain the co-disease association relationships between the associated symptom nodes;

[0145] With the associated symptom node as the center and the co-disease association relationship as the edge, construct the symptom collaborative association structure of the patient.

[0146] The system distinguishes the category of each recorded content within the dynamic causal chain of symptoms, based on fixed classification standards in general practice. Different categories correspond to four fixed areas: symptom records, test results, medication interventions, and confirmed diagnosis records. Each potential symptom node is linked to multiple corresponding causal events within the causal chain. Irrelevant events that lack symptom linkage reference value are eliminated by comparing the event category, retaining only causal events that can corroborate the simultaneous occurrence and mutual influence of multiple diseases. Based on the retained effective causal events, matching potential symptom nodes are identified. All the nodes that are selected and retained are uniformly used as associated symptom nodes in the dynamic causal chain of symptoms. All associated symptom nodes are supported by clinical evidence of concurrent and mutual influence.

[0147] Using all the filtered related disease nodes as the retrieval benchmark anchor points, the system fully traverses all the connections between entities within the general medical knowledge graph. The retrieval scope covers the multi-level connection links extending outward from each anchor point, and fully includes complete and continuous links that enable bidirectional communication between different related disease nodes. All interconnected links are uniformly collected and organized to form comorbidity association paths between related disease nodes. The comorbidity association paths fully carry the complete link information related to various concurrent, induced, and combined onset of different diseases recorded in the graph.

[0148] The medical semantic content carried by the entities and connections in each comorbidity path is analyzed one by one. Combined with the general medical rules for determining comorbidity in general practice, the pathological effects transmitted in the entire link are judged. Different links are distinguished into four fixed logical forms: mutual induction of symptoms, synchronous coexistence, one symptom aggravating another, and the derivation of complications. Semantic logic in the link that cannot be matched with all the current patient's diagnosis and treatment information is removed. Only the logical definitions that are completely consistent with the evolution of the patient's dynamic causal chain of symptoms are retained. The standardized medical connection content determined after screening is the comorbidity relationship between the related symptom nodes. The comorbidity relationship clearly defines the specific pathological pattern of the interaction between the two symptoms.

[0149] All the screened related symptom nodes are used as independent core vertices within the structure. The co-pathological relationships between each pair of vertices are matched with the co-pathological relationships analyzed in the previous step as connecting edges between vertices. The unidirectional or bidirectional direction of each connecting edge is determined according to the pathological action direction corresponding to the co-pathological relationship. All vertices and their matching connecting edges are integrated and arranged into a complete visual logical network. The integrated overall network framework is the patient's symptom co-pathological relationship structure. The symptom co-pathological relationship structure fully displays all the internal pathological connections between the synchronous onset and mutual influence of various suspected symptoms of the patient.

[0150] The beneficial effects are as follows: relying on the dynamic causal chain of diseases to complete the screening of related disease nodes through various diagnosis and treatment events, using the general medical knowledge graph to extract complete comorbidity links and analyze standardized comorbidity association logic, and finally building a collaborative association network structure for multiple diseases. This breaks through the limitations of independent analysis of single diseases, and fully restores the intrinsic pathological connections between multiple potential diseases that induce each other and co-occur. The entire process relies on standardized medical graph rules to complete the mining, avoiding the problem of subjective bias in manual judgment of multiple disease associations. The generated disease collaborative association structure is logically intuitive and the association basis is complete and traceable. It can provide data support for the joint judgment of multiple diseases for general intelligent diagnostic equipment, and effectively improve the comprehensiveness and clinical fit of intelligent diagnosis in the scenario of multiple coexisting diseases.

[0151] like Figure 2 The diagram shown is a functional block diagram of a knowledge graph-based general practitioner robot collaborative diagnosis system for multiple diseases, provided in an embodiment of the present invention.

[0152] The knowledge graph-based general practice robot multi-disease collaborative diagnosis system 100 of this invention can be installed in an electronic device. Depending on the functions implemented, the knowledge graph-based general practice robot multi-disease collaborative diagnosis system 100 may include a knowledge graph construction module 101, a clinical feature acquisition module 102, a disease path mining module 103, a path confidence assessment module 104, a temporal causal inference module 105, and a comorbidity association analysis module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0153] In this embodiment, the functions of each module / unit are as follows:

[0154] The graph construction module 101 is used to perform joint extraction of element relationships from a preset medical knowledge base to obtain medical triples of the medical knowledge base, and to construct a general medical knowledge graph of the robot based on the medical triples.

[0155] The clinical feature acquisition module 102 is used to perform semantic mapping on the patient's multimodal clinical data according to the rule framework of the general practice knowledge graph to obtain the patient's clinical features.

[0156] The symptom path mining module 103 is used to match and retrieve the general medicine knowledge graph based on the clinical features to obtain the symptom nodes of the patient, and to perform multi-hop traversal of the symptom nodes to obtain the candidate reasoning paths of the symptom nodes.

[0157] The path confidence assessment module 104 is used to assess the confidence of the candidate inference path based on the path characteristics of the candidate inference path, obtain the effective inference path of the symptom node, and determine the potential symptom node and the initial pathological evidence chain of the patient based on the effective inference path.

[0158] The temporal causal deduction module 105 is used to perform temporal analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determine the patient's dynamic disease causal chain based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence.

[0159] The comorbidity association analysis module 106 is used to perform multi-disease association mining on the potential disease nodes based on the dynamic disease causal chain to obtain the disease synergistic association structure of the patient.

[0160] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0161] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0162] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0163] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0164] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A general robot multi-disease collaborative diagnosis method based on a knowledge graph, characterized in that, The method includes: D1. Perform joint extraction of element relationships on the preset medical knowledge base to obtain the medical triplet of the medical knowledge base, and construct the robot's general medical knowledge graph based on the medical triplet. D2. Based on the rule framework of the general practice knowledge graph, perform semantic mapping on the patient's multimodal clinical data to obtain the patient's clinical characteristics; D3. Based on the clinical characteristics, perform matching and retrieval on the general medical knowledge graph to obtain the patient's symptom nodes, and perform multi-hop traversal on the symptom nodes to obtain the candidate reasoning paths of the symptom nodes. D4. Based on the path characteristics of the candidate inference paths, evaluate the confidence level of the candidate inference paths to obtain the effective inference paths of the symptom nodes, and determine the potential symptom nodes and initial pathological evidence chain of the patient based on the effective inference paths, including: The nodes and relation edges in the candidate reasoning path are parsed to obtain the semantic association strength and structural connectivity of the candidate reasoning path; Based on the semantic association strength and the structural connectivity, the candidate inference paths are weighted and calculated to obtain a comprehensive confidence score for each candidate inference path. The formula for calculating the comprehensive confidence score is as follows: ； wherein, is the comprehensive confidence score, is the semantic association strength of the i-th relationship edge, is the structural connectivity of the i-th relationship edge, is the number of relationship edges in the candidate reasoning path, is the structural connectivity of the i-th relationship edge, is the number of relationship edges in the candidate reasoning path, is a preset path attenuation coefficient; Based on the comparison results between the comprehensive confidence score and the preset screening threshold, the candidate inference paths are filtered to obtain the effective inference paths for the symptom nodes. By utilizing the logical pointers of the nodes in the effective reasoning path, the potential symptom nodes of the patient are determined; The nodes on the effective reasoning path are used as pathological elements, and the relational edge attributes on the effective reasoning path are used as the association logic. Following the topological order of the effective reasoning path, the pathological elements and the associated logic are sequentially assembled to obtain the patient's initial pathological evidence chain. D5. Perform time-series analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determine the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence. D6. Based on the dynamic causal chain of the disease, perform multi-disease association mining on the potential disease nodes to obtain the disease collaborative association structure of the patient.

2. The knowledge graph-based general practice robot multi-disease collaborative diagnosis method according to claim 1, wherein, The step of performing joint extraction of element relationships from a preset medical knowledge base to obtain medical triples of the medical knowledge base, and constructing a general medical knowledge graph for the robot based on the medical triples, includes: A preset medical knowledge base is obtained, and the structure of the medical knowledge base is parsed to obtain the semantic framework and entity type annotation information of the medical knowledge base; Based on the semantic framework and the entity type annotation information, attribute anchoring is performed on the medical knowledge base to obtain the medical elements in the medical knowledge base. Based on the semantic relationships of the medical elements, the medical knowledge base is merged by relation type to obtain the medical triples of the medical knowledge base; The medical triples are stored in a structured serialization to obtain the robot's general medical knowledge graph.

3. The knowledge graph-based general practice robot multi-disease collaborative diagnosis method of claim 1, wherein, The step involves semantic mapping of the patient's multimodal clinical data based on the rule framework of the general practice knowledge graph to obtain the patient's clinical characteristics, including: Modality separation was performed on the patient's multimodal clinical data to obtain the patient's structured test index data and unstructured medical record text; Based on the rule framework of the general practice knowledge graph, ontology alignment is performed on the structured test index data to obtain the standardized test indexes of the patient. Key medical terms are extracted from the unstructured medical record text to obtain the text feature words of the patient; Based on the rule framework of the general practice medical knowledge graph, semantic disambiguation is performed on the text feature words to obtain the standardized symptom description of the patient; The standardized test indicators and the standardized symptom descriptions are fused to obtain the patient's clinical characteristics.

4. The knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot as described in claim 1, characterized in that, The process of matching and retrieving the general medical knowledge graph based on the clinical features to obtain the patient's symptom nodes, and performing multi-hop traversal on the symptom nodes to obtain candidate reasoning paths for the symptom nodes, includes: The clinical features are parsed to obtain the feature nodes of the clinical features; The feature nodes are matched with the nodes in the general medicine knowledge graph to obtain the patient's symptom nodes; Using the symptom node as the root node, the general medical knowledge graph is traversed hierarchically to obtain the full path association subgraph of the patient. Based on a preset logical confidence threshold, the endpoint nodes of the full path association subgraph are filtered to obtain the candidate reasoning path of the patient.

5. The knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot as described in claim 1, characterized in that, The step of performing a time-series analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determining the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence, includes: The timing of the diagnosis and treatment events in the initial pathological evidence chain is obtained by parsing the timestamps of the events. Based on the occurrence time, the diagnostic and treatment events are arranged chronologically to obtain the patient's pathological evolution sequence; Based on the pathological evolution sequence, the logical relationship between adjacent events of the diagnosis and treatment events is determined to obtain the association logic of the diagnosis and treatment events; Based on the aforementioned association logic, causal connections are made between the diagnostic and treatment events in the pathological evolution sequence to obtain the dynamic causal chain of the patient's symptoms.

6. The knowledge graph-based multi-disease collaborative diagnosis method for general practitioners using robots as described in claim 5, characterized in that, The step of determining the logical relationship between adjacent events of the diagnostic and treatment events based on the pathological evolution sequence to obtain the association logic of the diagnostic and treatment events includes: Event extraction is performed on adjacent diagnostic and treatment events in the pathological evolution sequence to obtain the preceding and subsequent events of the diagnostic and treatment events. Substitute the preceding event and the following event into the general medicine knowledge graph and perform a relationship retrieval to obtain candidate relationship edges pointing from the preceding event to the following event; Based on the relationship type of the candidate relationship edges, logical matching is performed on the preceding events and the following events to obtain the association logic of the diagnosis and treatment events.

7. The knowledge graph-based collaborative diagnosis method for multiple diseases using a general practitioner robot as described in claim 1, characterized in that, The process of mining multiple disease associations for the potential disease nodes based on the dynamic causal chain of the disease to obtain the symptom-related association structure of the patient includes: Based on the type of the diagnosis and treatment event, the potential symptom nodes are filtered for related events to obtain the related symptom nodes of the dynamic symptom causal chain; Based on the associated symptom nodes, subgraph extraction is performed on the general medicine knowledge graph to obtain the co-disease association paths between the associated symptom nodes; Semantic logic analysis is performed on the co-disease association paths to obtain the co-disease association relationships between the associated symptom nodes; With the associated symptom node as the center and the co-disease association relationship as the edge, construct the symptom collaborative association structure of the patient.

8. A knowledge graph-based general practice robot-based collaborative diagnosis system for multiple diseases, characterized in that, The system for implementing the knowledge graph-based multi-disease collaborative diagnosis method for general practitioners using a robot as described in claim 1 includes: The graph construction module is used to perform joint extraction of element relationships from a preset medical knowledge base to obtain medical triples of the medical knowledge base, and to construct a general medical knowledge graph of the robot based on the medical triples. The clinical feature acquisition module is used to perform semantic mapping on the patient's multimodal clinical data according to the rule framework of the general practice medical knowledge graph to obtain the patient's clinical features; The symptom path mining module is used to match and retrieve the general medical knowledge graph based on the clinical features to obtain the symptom nodes of the patient, and to perform multi-hop traversal of the symptom nodes to obtain the candidate reasoning paths of the symptom nodes. The path confidence assessment module is used to assess the confidence of the candidate inference path based on the path characteristics of the candidate inference path, obtain the effective inference path of the symptom node, and determine the potential symptom node and the initial pathological evidence chain of the patient based on the effective inference path. The temporal causal deduction module is used to perform temporal analysis on the diagnostic and treatment events in the initial pathological evidence chain to obtain the patient's pathological evolution sequence, and determine the patient's dynamic causal chain of symptoms based on the correlation logic of adjacent diagnostic and treatment events in the pathological evolution sequence. The comorbidity association analysis module is used to perform multi-disease association mining on the potential disease nodes based on the dynamic disease causal chain to obtain the disease synergistic association structure of the patient.

Images