Main diagnostic rationality judgment method and system based on artificial intelligence

Abstract

The present invention relates to the technical field of medical information processing, and particularly to a method and system for judging the rationality of the main diagnosis based on artificial intelligence. The method includes the following steps: obtaining diagnostic request data, and using preset medical knowledge graph data and preset intelligent detection rule data to generate candidate main diagnoses for the diagnostic request data to obtain candidate main diagnosis data; generating diagnostic execution parameters according to the candidate main diagnosis data to obtain candidate main diagnosis parameter data; obtaining historical diagnostic request data and historical candidate main diagnosis data, and performing diagnostic linkage processing according to the historical diagnostic request data and the historical candidate main diagnosis data to obtain diagnostic linkage network data; performing node interaction processing on the candidate main diagnosis data according to the diagnostic linkage network data to obtain candidate main diagnosis weight data. The present invention improves the intelligent level of the judgment of the rationality of the diagnosis, the credibility of the scoring result and the adaptation ability of the clinical system.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing, and particularly to a method and system for judging the rationality of a primary diagnosis based on artificial intelligence. Background Art

[0002] With the continuous improvement of the level of medical informatization, electronic medical record systems, clinical pathway management systems, and intelligent auxiliary diagnosis systems have been gradually popularized in various medical institutions. Traditional auxiliary diagnosis methods mostly rely on rule bases, disease knowledge tables, or statistical models to recommend or verify primary diagnoses, and some systems have also introduced artificial intelligence algorithms to automatically discriminate diagnostic information. However, in the actual application process, the existing technologies generally have the problem that the process of generating diagnostic results lacks interpretability, and some models rely on black box algorithms and cannot provide clear logical bases for diagnostic suggestions, which is not conducive to medical staff's adoption. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a method and system for judging the rationality of a primary diagnosis based on artificial intelligence to solve at least one of the above technical problems.

[0004] The present application provides a method for judging the rationality of a primary diagnosis based on artificial intelligence, and the method includes:

[0005] S1. Obtain diagnostic request data, and use preset medical knowledge graph data and preset intelligent detection rule data to generate candidate primary diagnoses for the diagnostic request data to obtain candidate primary diagnosis data;

[0006] S2. Generate diagnostic execution parameters according to the candidate primary diagnosis data to obtain candidate primary diagnosis parameter data;

[0007] S3. Obtain historical diagnostic request data and historical candidate primary diagnosis data, and perform diagnostic linkage processing according to the historical diagnostic request data and the historical candidate primary diagnosis data to obtain diagnostic linkage network data;

[0008] S4. Perform node interaction processing on the candidate primary diagnosis data according to the diagnostic linkage network data to obtain candidate primary diagnosis weight data.

[0009] In the present invention, by constructing a medical knowledge graph and an intelligent detection rule engine, the intelligent generation of candidate primary diagnoses and the structured expression of context parameters are realized; further, historical diagnostic data is introduced to establish a diagnostic linkage network, and through graph structure modeling and node interaction scoring mechanism, multi-dimensional weight evaluation is performed on the candidate primary diagnosis, which improves the accuracy, interpretability, and automation level of the diagnostic rationality judgment, effectively enhances the intelligent auxiliary decision-making ability of the medical information system and the utilization efficiency of clinical data.

[0010] Optionally, the steps for constructing the preset medical knowledge graph data include the following steps:

[0011] Obtain structured electronic medical records, medical standard coding data, and unstructured medical literature data to construct a multi-source medical data set;

[0012] Perform entity recognition and entity relationship extraction on the multi-source medical data set to obtain medical entity data and initial association relationship data;

[0013] Construct semantic relationship edges between entities based on the medical entity data and the initial association relationship data to form medical relationship network data;

[0014] Perform semantic relationship enhancement processing on the medical entity data according to the preset medical knowledge base to obtain medical entity causal relationship, medical entity adaptation relationship, and medical entity concurrent association relationship data;

[0015] Calibrate the directionality of the edges of the medical relationship network data according to the medical entity causal relationship, medical entity adaptation relationship, and medical entity concurrent association relationship data to obtain medical knowledge graph data.

[0016] In the present invention, by integrating structured medical record data, standard medical coding, and unstructured literature, a multi-source heterogeneous medical data set is constructed. Combining entity recognition and relationship extraction technologies, a medical relationship network is generated, and further semantic relationship enhancement and edge directionality calibration are performed based on a preset medical knowledge base to construct a directed medical knowledge graph with causal relationships, adaptation relationships, and concurrent association relationships. The graph structure not only reflects the multi-layer semantic logic of medical knowledge but also has the capabilities of semantic reasoning and upstream and downstream dependency modeling. Compared with traditional static graphs or rule bases, it improves the systematicness of knowledge organization, the accuracy of relationship expression, and the interpretability of subsequent diagnostic reasoning, and has significant technical advantages and industry applicability.

[0017] Optionally, the preset intelligent detection rule data includes a primary diagnosis judgment rule, a surgery judgment rule, and a complication handling rule. The primary diagnosis judgment rule, the surgery judgment rule, and the complication handling rule include a rule identifier, a rule type, a trigger condition, parameter configuration, and priority information. The trigger condition includes a combination match of inspection indicators, chief complaint information, and medical record elements, and is used to realize the linkage judgment and intelligent trigger of diagnostic request data and rules.

[0018] In the present invention, by presetting multiple types of intelligent detection rules and defining rule identifiers, types, triggering conditions, parameter configurations, and priority information in the rules, a structured and configurable rule system is formed, which supports the linkage judgment and intelligent triggering of rules through the combined matching of inspection indicators, chief complaint information, and medical record elements. Compared with traditional fixed rule templates, this rule system has the characteristics of dynamic adaptation and refined management, can quickly match the most relevant rule paths in different clinical scenarios, improve the accuracy and controllability of primary diagnosis generation, and at the same time support the continuous optimization and maintenance of the rule library, with stronger scalability and engineering adaptability.

[0019] Optionally, S1 includes:

[0020] Obtain diagnostic request data, and perform entity matching on the diagnostic request data using preset medical knowledge graph data to obtain entity matching data;

[0021] Retrieve the initial diagnosis-related candidate set from the medical knowledge graph data based on the entity matching data to obtain initial diagnosis-related candidate set data;

[0022] Perform candidate primary diagnosis matching on the initial diagnosis-related candidate set data using preset intelligent detection rule data to obtain candidate primary diagnosis data.

[0023] In the present invention, by performing entity-level matching between the diagnostic request data and the medical knowledge graph, a semantic mapping relationship is constructed, and the initial diagnosis candidate set is quickly retrieved by combining the structural relevance in the knowledge graph; then, the candidate set is conditionally matched and filtered using preset intelligent detection rule data to obtain primary diagnosis candidates with high credibility. Compared with traditional diagnosis generation methods that rely on manual rule configuration or black box model reasoning, the present invention realizes the interpretability, rule-driven nature, and atlas structure optimization of the diagnostic reasoning process, effectively improving the accuracy, stability, and system scalability of candidate diagnosis generation.

[0024] Optionally, S2 includes:

[0025] Extract candidate diagnosis context from the diagnostic request data based on the candidate primary diagnosis data to obtain candidate diagnosis context data;

[0026] Initialize the rule parameter mapping for the candidate primary diagnosis data and the candidate diagnosis context data according to the preset intelligent detection rule data to obtain rule parameter mapping data;

[0027] Construct the candidate diagnosis parameter structure based on the rule parameter mapping data to obtain candidate primary diagnosis parameter data.

[0028] In the present invention, by extracting the context information of the candidate main diagnosis in the diagnosis request data and combining with the preset intelligent detection rule data for parameter mapping and structured construction, the structural expression of the candidate diagnosis in multi-dimensional parameter dimensions such as semantics, metrics, and rule conditions is realized. Compared with the traditional coarse-grained judgment method based on a rule table, this method can construct a candidate diagnosis parameter structure with context awareness ability and rule semantic alignment characteristics, providing clear, traceable, and configurable basic data support for subsequent diagnosis rationality analysis, and significantly improving the interpretation ability and parameterized management ability of the intelligent diagnosis system.

[0029] Optionally, S3 includes:

[0030] Obtain historical diagnosis request data and historical candidate main diagnosis data;

[0031] Perform diagnostic instance association extraction on the historical diagnosis request data and the historical candidate main diagnosis data to obtain diagnostic instance association data;

[0032] Mine the linkage relationship according to the diagnostic instance association data to obtain linkage relationship data;

[0033] Generate diagnostic linkage network data according to the diagnostic instance association data and the linkage relationship data to obtain diagnostic linkage network data.

[0034] In the present invention, by performing diagnostic instance-level association extraction on the historical diagnosis request data and the historical candidate main diagnosis data, and combining with the linkage relationship mining method, a diagnostic linkage network data containing diagnostic co-occurrence, sequence, and context dependence features is constructed to realize the semantic relationship modeling and structural expression between diagnostic nodes. Compared with the traditional diagnostic association methods based on static co-occurrence statistics or rule setting, the present invention can dynamically reflect the complex diagnostic linkage patterns in real medical data, significantly enhancing the ability to utilize historical experience and the basis of graph structure reasoning in the diagnostic reasoning process, and effectively improving the accuracy and adaptive ability of subsequent intelligent scoring and rationality judgment.

[0035] Optionally, S4 includes:

[0036] Construct a candidate main diagnosis subgraph for the diagnostic linkage network data according to the candidate main diagnosis data to obtain candidate main diagnosis subgraph data;

[0037] Extract node linkage features according to the candidate main diagnosis subgraph data and the candidate main diagnosis data to obtain node linkage feature data;

[0038] Perform interactive scoring according to the candidate main diagnosis data and the node linkage feature data to obtain preliminary scoring data;

[0039] Perform simplified graph attention mechanism processing based on the candidate primary diagnosis data and the preliminary scoring data to obtain candidate primary diagnosis weight data.

[0040] In the present invention, by constructing a subgraph associated with the candidate primary diagnosis in the diagnostic linkage network, extracting the linkage features of candidate diagnosis nodes in the graph structure, and performing interactive scoring in combination with context information, and further dynamically adjusting the weights between nodes using the simplified graph attention mechanism, candidate primary diagnosis weight data with structural interpretability is generated. Compared with traditional static scoring or rule-based methods, the present invention introduces a graph structure semantic propagation and multi-source feature fusion mechanism, which can adaptively evaluate the rationality priority of the primary diagnosis according to the diagnostic context and the relationship of the historical graph, effectively improving the accuracy, semantic consistency and system intelligence level of the diagnostic weight judgment.

[0041] Optionally, the interactive scoring includes:

[0042] Extract context semantic features from the candidate primary diagnosis data to obtain context semantic feature data;

[0043] Perform bilinear attention calculation based on the node linkage feature data and the context semantic feature data to obtain matching weight data;

[0044] Perform feature fusion on the node linkage feature data and the context semantic feature data according to the matching weight data to obtain preliminary scoring data.

[0045] In the present invention, during the interactive scoring process, first extract the context semantic features corresponding to the candidate primary diagnosis, and combine the node linkage feature data. Based on the bilinear attention mechanism, calculate the matching weight between the candidate diagnosis and the graph structure, and then generate a preliminary scoring result through weight-driven feature fusion. Compared with traditional weighted average or rule-based scoring methods, the bilinear attention mechanism introduced in the present invention can more accurately model the deep coupling relationship between context semantics and the graph structure, realize the adaptive dynamic adjustment of diagnostic scoring and the enhancement of semantic perception, and significantly improve the discrimination, reliability and clinical semantic consistency of the scoring results.

[0046] Optionally, the simplified graph attention mechanism processing includes:

[0047] Perform adjacent node similarity calculation based on the candidate primary diagnosis data and the preliminary scoring data to obtain adjacent node edge weight data;

[0048] Perform simplified attention coefficient calculation based on the adjacent node edge weight data and the preliminary scoring data to obtain simplified attention coefficient data;

[0049] Perform multi-order aggregation based on the simplified attention coefficient data to obtain multi-order neighbor aggregation data;

[0050] Perform residual connection based on multi - order neighbor aggregation data and preliminary scoring data to obtain candidate main diagnosis weight data.

[0051] In the present invention, by calculating the similarity of adjacent nodes for the candidate main diagnosis node and its preliminary scoring result, edge weight data is generated, and further combining the edge weight and node score to calculate a simplified attention coefficient, an efficient graph - structure attention distribution is realized; through a multi - order neighbor aggregation mechanism, the global semantic perception ability is enhanced, and under the residual connection, the original score and graph aggregation information are fused to generate a diagnosis weight result. Compared with traditional static propagation or full - graph attention methods, the present invention greatly reduces the computational complexity while ensuring the aggregation accuracy, improves the diagnostic inference efficiency and edge - deployment adaptability, and retains the original diagnostic semantics, effectively enhancing the robustness, flexibility and interpretability of the diagnostic scoring process. Traditional attention calculations often require multi - layer neural network calculations, with a large number of feature transformations, concatenations and activation functions, high computational complexity, and strong dependence on computing resources and training samples during training / deployment, which is not conducive to lightweight embedded deployment or rule - collaborative logic expression in clinical systems. In the attention mechanism design of the present invention, a linear combination calculation of structural edge weights and preliminary scoring data is used to replace the traditional attention score method based on deep - neural - network parameter learning, constructing a simplified attention coefficient calculation mechanism that is untrained, easy to deploy and interpretable. This not only reduces the system's dependence on computing resources but also improves the transparency and traceability of the diagnostic scoring process, especially suitable for scenarios where medical assistance systems require high reliability and lightweight deployment capabilities.

[0052] Optionally, the present application also provides an artificial - intelligence - based main diagnosis rationality judgment system for performing the artificial - intelligence - based main diagnosis rationality judgment method as described above. The artificial - intelligence - based main diagnosis rationality judgment system includes:

[0053] A candidate diagnosis generation module, configured to obtain diagnosis request data, and use preset medical knowledge graph data and preset intelligent detection rule data to generate candidate main diagnoses for the diagnosis request data, obtaining candidate main diagnosis data;

[0054] A diagnosis parameter extraction module, configured to generate diagnosis execution parameters based on the candidate main diagnosis data, obtaining candidate main diagnosis parameter data;

[0055] A diagnosis linkage graph - building module, configured to obtain historical diagnosis request data and historical candidate main diagnosis data, and perform diagnosis linkage processing based on the historical diagnosis request data and historical candidate main diagnosis data to obtain diagnosis linkage network data;

[0056] A diagnosis weight evaluation module, configured to perform node interaction processing on the candidate main diagnosis data based on the diagnosis linkage network data to obtain candidate main diagnosis weight data.

[0057] The object of the present invention is that through constructing a medical knowledge graph with causal association, adaptation relationship and concurrent logic, the present invention establishes a multi-level structure expression model between diagnostic entities and semantic relationship edges, realizes the unified semantic mapping of unstructured symptom descriptions, structured examination indicators and chief complaint elements in diagnostic requests; based on the semantic paths between graph nodes and the upstream and downstream dependency relationships of entities, combined with a configurable intelligent detection rule engine, semantic triggering and rule screening are performed on the initially matched candidate primary diagnosis set, so as to realize the intelligent generation and structural traceability of candidate primary diagnoses.

[0058] On this basis, the present invention further introduces historical diagnostic request data and candidate primary diagnosis instances. By constructing a "diagnosis-context-candidate" triple graph, the co-occurrence frequency, path dependency and sequence ranking relationship between diagnoses are extracted to form diagnostic linkage network data with directionality and context condition constraints, providing multi-dimensional historical association support in the real world for candidate primary diagnoses. This graph structure not only has the topological propagation ability between entities, but also can be used as a path network for the propagation of scoring information between diagnostic candidate nodes.

[0059] For the task of judging the rationality of diagnosis, the present invention proposes a bidirectional scoring mechanism that combines node linkage features and semantic context. By introducing a bilinear attention structure, the graph structure features of candidate nodes and the diagnostic context vectors are aligned and calculated to dynamically generate semantic matching weights, realizing a refined expression of the node-level scoring results. Subsequently, a lightweight graph attention aggregation mechanism is introduced. Based on multi-order neighbor information, semantic strengthening and structural enhancement of node weights are completed, and the residual connection mechanism is combined to maintain the initial semantic stability, and a credible weight score for diagnostic decision-making is output.

[0060] Compared with the traditional method for generating primary diagnoses based on rule tables or model outputs, the present invention has significant technical advantages such as clear knowledge modeling structure, traceable scoring process, strong context fusion ability, and high graph structure reasoning efficiency, providing structured, intelligent and interpretable underlying technical support for the task of judging the rationality of diagnosis, and greatly improving the integration and utilization efficiency of multi-source data and the automated diagnosis management ability of medical auxiliary decision-making systems. Brief Description of the Drawings

[0061] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objects and advantages of the present application will become more obvious:

[0062] Figure 1 Shows a flowchart of the steps of a method for judging the rationality of a primary diagnosis based on artificial intelligence in an embodiment;

[0063] Figure 2 Shows a flowchart of the steps of a method for generating candidate diagnoses in an embodiment;

[0064] Figure 3 The flowchart of the steps of a diagnostic parameter extraction method according to an embodiment is shown;

[0065] Figure 4 The flowchart of the steps of a diagnostic linkage mapping method according to an embodiment is shown;

[0066] Figure 5 The flowchart of the steps of a diagnostic weight evaluation method according to an embodiment is shown;

[0067] The realization of the object of the present invention, functional features and advantages will be further described with reference to the embodiments and the accompanying drawings. Detailed Embodiments

[0068] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0069] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0070] It should be understood that although terms such as "first", "second", etc. may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly, the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0071] This product is an AI medical inference engine with artificial intelligence and medical knowledge graph as the core. It has the capabilities of medical large text analysis, clinical diagnosis and treatment inference, and efficient computing. It can accurately and quickly extract key medical descriptions from medical records for judging the accuracy of diagnosis, surgical coding, and the reasonableness of charges. Through artificial intelligence + medical knowledge graph + scenario rules, it completes the rationality analysis of the use of medical insurance funds in different scenarios, and fully realizes the artificial intelligence scanning of the data of the diagnosis and treatment process, which can deeply empower the supervision of medical insurance funds and the intelligent audit business of hospitals.

[0072] Product function description includes:

[0073] Knowledge graph management covers multi-dimensional data such as disease conditions, causes, and drug information, facilitating accurate query and application of knowledge by medical staff. It has a knowledge association function that can connect different knowledge points according to medical logic, such as the association between diseases and corresponding diagnostic criteria and treatment drugs, to assist in intelligent diagnostic reasoning.

[0074] Intelligent inspection rule management allows professionals to configure and view intelligent detection rules. It has a built-in rule template library that covers main diagnosis judgment rules, surgical judgment rules, etc. It is convenient for on-site personnel to quickly understand the judgment logic of the rules, reducing the learning cost. At the same time, it also supports on-site configuration of rule parameters to meet on-site personalized needs.

[0075] Intelligent inspection execution statistics comprehensively statistics the interface call situation with Yuzhi Medical, including the number of executions, interface response duration, the rule with the highest trigger frequency, etc.

[0076] Service log tracking details every operation log of the operation of Yuzhi Medical, covering key behaviors such as user login, data query, and intelligent inspection task startup, ensuring that operations are traceable. For abnormal situations such as system errors and detection interruptions, it accurately locates the fault point, records the error details and occurrence time, assisting technical personnel to quickly troubleshoot and repair. It supports log filtering and retrieval, and can query according to multiple dimensions such as users, time, and operation types, facilitating operation and maintenance personnel to focus on the root cause of problems and ensuring the stable operation of the system.

[0077] Rule-driven matrix constructs a visual rule-driven relationship matrix, presenting the linkage, priority, and mutual influence relationships between intelligent inspection rules in a graphical way to assist in rule optimization decisions. It supports interactive operations on matrix nodes. Clicking on a single rule node can view its detailed configuration, associated rules, and trigger conditions, facilitating in-depth analysis of the logic of the rule system. According to clinical feedback and intelligent inspection statistical data, dynamically adjust the matrix layout and rule weights to continuously improve the accuracy and reliability of the system's intelligent diagnosis.

[0078] In one embodiment, the main idea of the rationality judgment of the main diagnosis (rotavirus enteritis) in the above case.

[0079] The first step: Judge whether the main diagnosis (rotavirus enteritis) is reasonable. First, it is necessary to judge whether the patient has rotavirus enteritis. The diagnostic medical knowledge graph will contain the judgment basis for rotavirus enteritis, such as a positive rotavirus antigen test. The AI scans the full-text data of the electronic medical record, performs precise semantic recognition and analysis of the context, searches for relevant descriptions of "positive rotavirus antigen test" in the medical record document, and judges whether the patient has rotavirus enteritis based on this.

[0080] Step 2: Determine whether the primary diagnosis (rotavirus enteritis) consumes the most medical resources. Combining the monitoring model and scenario rules set by artificial intelligence, analyze the detailed cost data of the patient's entire hospitalization process. Based on the medical knowledge graph, classify the costs according to the diagnosis. If rotavirus enteritis is the diagnosis that consumes the most medical resources, directly determine that the selection of the primary diagnosis (rotavirus enteritis) is reasonable; otherwise, proceed to the next judgment.

[0081] Step 3: Determine whether the hospitalization purpose is consistent with the primary diagnosis (rotavirus enteritis). If the resource consumption of the primary diagnosis (rotavirus enteritis) is not the highest but is close to that of the diagnosis with the highest medical resource consumption, further analyze in combination with the patient's chief complaint. The content of the chief complaint is: chest tightness, shortness of breath, and cough for 3 months, which is quite different from the clinical manifestations of rotavirus enteritis (patients with rotavirus enteritis generally have symptoms such as watery diarrhea, vomiting, and low fever). Therefore, it can be determined that the hospitalization purpose is not consistent with the primary diagnosis (rotavirus enteritis), and the selection of the primary diagnosis is unreasonable.

[0082] Please refer to Figure 1 , this application provides a method for judging the rationality of the primary diagnosis based on artificial intelligence, and the method includes:

[0083] S1. Obtain diagnostic request data, and use the preset medical knowledge graph data and preset intelligent detection rule data to generate candidate primary diagnoses for the diagnostic request data, obtaining candidate primary diagnosis data;

[0084] In one embodiment, obtain a diagnostic request data of an outpatient patient from the hospital HIS system, including: the patient's chief complaint (such as "cough for 2 weeks, accompanied by low fever"), examination results (such as blood routine, chest CT, etc.). Input the above data into the natural language processing module, extract key medical entities (such as symptoms, examinations, past medical history, etc.), and perform entity alignment with the disease nodes in the medical knowledge graph. The knowledge graph contains the causal relationship paths between structured diseases - symptoms - examinations. Through graph search (such as based on the path Rank algorithm), identify several diseases with relatively high matching degrees (such as pulmonary tuberculosis, bronchitis, etc.) as candidate primary diagnoses. At the same time, the system scores and ranks the candidate diagnoses according to the intelligent detection rules (such as "if the CT result shows nodules + chronic cough, then prompt to screen for pulmonary tuberculosis"), obtaining candidate primary diagnosis data.

[0085] S2. Generate diagnostic execution parameters based on the candidate primary diagnosis data, obtaining candidate primary diagnosis parameter data;

[0086] In one embodiment, for the candidate primary diagnosis of "pulmonary tuberculosis", the system automatically matches the standard diagnosis and treatment parameters to be executed, such as recommended further examinations (sputum acid-fast staining, tuberculin test), the ICD code of the disease (such as A15.0), the recommended treatment cycle (6-month anti-tuberculosis drug treatment), and the medical insurance review risk level (medium to high). These parameters are automatically extracted through a knowledge base template generation system or an AI model (such as an ICD encoder fine-tuned by BERT and a semantic editing model with similar functions) to form structured candidate primary diagnosis parameter data.

[0087] S3. Obtain historical diagnosis request data and historical candidate primary diagnosis data, and perform diagnostic linkage processing based on the historical diagnosis request data and the historical candidate primary diagnosis data to obtain diagnostic linkage network data.

[0088] In one embodiment, the system retrieves the diagnosis requests and candidate primary diagnosis records of similar medical records (matched through a similarity model such as SimCSE) in the past year from the data platforms of local or consortium hospitals. A linkage network diagram among "diagnosis - symptom - parameter" is constructed. The nodes in the diagram include diseases, symptoms, examinations, treatment paths, etc., and the weights of the edges represent co-occurrence frequencies or temporal relationships. For example, if the proportion of "cough + abnormal chest CT" diagnosed as "bronchitis" is 60% and "pulmonary tuberculosis" is 30% in the past 100 cases, then this part of the historical linkage information forms the diagnostic linkage network data.

[0089] In one embodiment, the diagnostic linkage network data is constructed based on historical diagnosis request data and historical candidate primary diagnosis data. Considering that this process depends on a large number of historical records and involves complex relationship extraction and graph structure generation operations, to improve the system response efficiency, the construction process of the diagnostic linkage network is designed as a periodic offline preprocessing task, which can be run regularly by day, week, or month, and the constructed network structure is stored in a graph database for calling. In the actual diagnosis process, the system does not generate a diagnostic linkage graph in real time for the received current diagnosis request, but directly queries the existing linkage network structure, and locates its subgraph area and adjacent structure in the graph according to the candidate primary diagnosis for subsequent node interaction modeling and weight evaluation.

[0090] In the offline stage, the system periodically retrieves historical diagnosis request data and corresponding candidate primary diagnosis records from the data platforms of local or consortium medical institutions, and filters similar instances based on a semantic similarity model (such as SimCSE); for these historical data, the system extracts the co-occurrence relationships and temporal transformation trends among entities such as "diagnosis - symptom - examination - treatment path" to construct a linkage graph structure; the nodes in the graph include medical entities such as diseases, symptoms, examination results, treatment paths, etc., the edges represent semantic or temporal relationships, and the edge weights represent co-occurrence frequencies, transformation probabilities, or clinical weights; the constructed diagnostic linkage network is stored in the form of a graph database.

[0091] The system only selects the diagnostic request records and diagnostic result data within a preset time range (such as the past three years) to build the basis of the atlas, avoiding the introduction of outdated information caused by factors such as clinical pathway updates and diagnostic standard adjustments. This time range can be flexibly adjusted according to the actual deployment scenario (such as the hospital database update frequency). Perform field integrity checks on the diagnostic records in the historical data, and eliminate abnormal records that lack keywords such as the chief complaint, examination results, confirmed diagnosis information, and diagnostic codes. Preferably, only samples with a higher degree of structurality and available for entity recognition and standardization are retained. The system performs coding legality verification on diagnostic and operation fields, and preferentially retains diagnostic items that conform to national standard terms (such as ICD-10, SNOMED CT) or the hospital's standard coding system, and eliminates spelling mistakes, unknown categories, or abnormally manually entered data. For diagnostic nodes that appear extremely rarely but are not rare disease labels, conduct investigations in combination with the corresponding departments and confirmed diagnosis processes. If it is found that they are likely to originate from diagnostic errors or unreasonable paths, they can be marked as "low-confidence samples" or eliminated. Introduce basic statistical indicators (such as case gender, age group, department source, etc.) to evaluate the balance of sample distribution, and ensure that the coverage of different diseases and their characteristic nodes in the diagnostic linkage map is not affected by extreme biases of a certain group.

[0092] S4. Perform node interaction processing on the candidate primary diagnosis data according to the diagnostic linkage network data to obtain candidate primary diagnosis weight data.

[0093] In one embodiment, a graph neural network (such as GraphSAGE, GAT, or a graph neural network with similar functions) is used to aggregate and update the candidate primary diagnosis nodes in the diagnostic linkage network. The features of the nodes include the historical confirmation rate of the candidate diagnosis, the matching degree with the symptoms in the current request, the historical medical insurance return rate, etc. The graph neural network calculates the weight value of each candidate diagnosis node to form candidate primary diagnosis weight data. For example: tuberculosis 0.62, bronchitis 0.35, and the weights of the remaining diagnoses are less than 0.1. The system will recommend "tuberculosis" as the primary diagnosis for the current diagnostic request and visualize the sorting of the weights of the remaining primary diagnoses.

[0094] Optionally, the steps for constructing the preset medical knowledge graph data include the following steps:

[0095] Obtain structured electronic medical records, medical standard coding data, and unstructured medical literature data to construct a multi-source medical data set;

[0096] In one embodiment, the structured electronic medical record data refers to the electronic medical data with a standard field structure collected through medical informatization platforms such as the hospital information system (HIS), laboratory information system (LIS), and picture archiving and communication system (PACS). The structured electronic medical record data includes, but is not limited to, the following fields: patient_id: the coding information used to uniquely identify the patient's identity; visit_date: the date-type data indicating the patient's visit time; chief_complaint: the patient's chief complaint information used to describe the main symptom manifestations during the visit; diagnosis: the preliminary or specific diagnosis information of the patient's current disease by the doctor; prescription: the list of drug prescriptions issued for this visit; check_result: the inspection or test result data related to this visit, including laboratory test indicators, imaging descriptions, and other contents.

[0097] The medical standard coding data refers to the medical information coding system used to standardize the identification of medical entities and their relationships, including, but not limited to, the following categories: ICD-10 is used to represent the classification standard of diseases and related health problems and is widely used in clinical diagnosis and medical insurance settlement scenarios; SNOMED-CT is used to describe clinical terms such as diseases, symptoms, signs, procedures, and devices and has strong semantic expression capabilities; LOINC is mainly used for the standardized coding of laboratory test items and clinical observation items. The above coding systems are used to assign unified structured codes to medical entities such as diagnosis names, inspection indicators, surgical operations, and medication behaviors, facilitating entity alignment and semantic fusion in the atlas modeling process.

[0098] The unstructured medical literature data refers to the literature resources in the medical field that exist in natural language form without being structured. Its sources include, but are not limited to, publicly available papers in medical literature databases such as PubMed and Embase; various authoritative medical guidelines and clinical pathway specification documents; Chinese core medical journals, specialty papers, disease knowledge base texts, etc. This type of data is crawled and downloaded through web crawler tools or open interfaces, and after subsequent steps such as PDF document to text conversion, HTML structure parsing, and text cleaning of the main body, the pure text corpus of medical knowledge content is extracted to provide the original semantic materials for entity recognition and relationship extraction.

[0099] Entity recognition and entity relationship extraction are performed based on the multi-source medical data set to obtain medical entity data and initial association relationship data;

[0100] In one embodiment, in the medical entity recognition process, the predefined entity categories include, but are not limited to, types such as Disease, Symptom, Test, Drug, Procedure, Body Part, etc. This classification system is used to guide the subsequent entity annotation and model training processes.

[0101] In the entity recognition stage, a deep learning-based named entity recognition model is used to model medical texts. The model can be a Bidirectional Long Short-Term Memory Network combined with Conditional Random Field (BiLSTM-CRF) structure, or a pre-trained language model BERT combined with a CRF output layer structure, or a semantic editing model with similar functions. The training corpus is selected from the Chinese Clinical Medicine Entity Annotation Corpus (such as the CMeEE dataset). This corpus covers more than six types of medical entity tags and has been manually verified to ensure semantic accuracy and label consistency.

[0102] Taking the original text sample "The patient has fever and cough. The CT shows a shadow in the lungs, and ceftriaxone is given for treatment" as an example, the following medical entities can be extracted through the entity recognition model: "fever" and "cough" are recognized as Symptom entity types; "shadow in the lungs" is recognized as a Check Finding entity type; "ceftriaxone" is recognized as a Drug entity type.

[0103] On the basis of completing entity recognition, a relation extraction model is further used to identify the semantic relations between entity pairs. The model can be a pre-trained structure based on sentence pair modeling (such as BERT-PAIR or a semantic editing model with similar functions), or a model based on a Convolutional Neural Network (CNN) combined with an attention mechanism. The output form of relation extraction is a triple structure, i.e., (entity 1, relation type, entity 2), such as ("shadow in the lungs", indicates, "pneumonia"), ("cough", symptom, "pneumonia"), ("ceftriaxone", treats, "pneumonia").

[0104] Based on the medical entity data and the initial association relation data, semantic relation edges between entities are constructed to form medical relation network data;

[0105] In one embodiment, the medical knowledge graph is a graph data structure with structured semantic expression capabilities, wherein each graph node represents a standardized medical entity, which may be a disease, symptom, drug, examination item, examination result or surgery, etc.; each graph edge represents a semantic association relationship between entities, and the relationships include but are not limited to the following initial relationship types: symptom-manifestation-disease: indicating that a symptom is a typical manifestation of a disease; drug-treatment-disease: indicating that a drug is suitable for the treatment of a disease; examination result-support-disease: indicating that a certain examination finding can be used to support the diagnosis of a disease.

[0106] Each edge not only contains the relationship type, but also comes with multiple edge attribute information, including co-occurrence frequency, which indicates the frequency of the relationship appearing in historical medical records or documents; text source identifier, which records the original text number or data source of the source of the relationship; semantic similarity score, which calculates the contextual semantic similarity between entities based on a pre-trained language model (such as BERT or a semantic editing model with similar functions), and the value range is 0 to 1.

[0107] To quantify the importance of edges, a numerical weight is assigned to each edge during the composition process. The weight value is calculated according to the following formula: .in, and is the weighting coefficient, set to =0.6, =0.4; the co-occurrence frequency is normalized by Min-Max, and the semantic similarity is calculated by the BERT model or a semantic editing model with similar functions to calculate the sentence vector cosine value. The constructed knowledge graph is stored and managed in the form of a graph database, preferably using a graph database system that supports attribute graph modeling and efficient graph query, such as Neo4j or OrientDB.

[0108] According to the preset medical knowledge base, the medical entity data is processed with semantic relationship enhancement to obtain the medical entity causal relationship, medical entity adaptation relationship and medical entity concurrent association relationship data;

[0109] In one embodiment, to avoid redundancy or semantic conflicts with the initial entity relationships extracted by the foregoing natural language processing method, when the system performs semantic enhancement of the knowledge base, it preferentially uses the existing entity relationships in the knowledge graph as the main structure and only introduces enhancement edges in the following two cases: 1. When the relationship type does not exist in the graph, that is, when there are no semantic edges such as causality, adaptation, and concurrency between two entities, if there is a clearly defined relationship in the knowledge base, it is newly added and supplemented; 2. The confidence of the existing relationship is insufficient. For the low-confidence edges automatically extracted from the graph (such as weak expressions from unstructured text), if there is a strong semantic relationship between the same entity pair in the knowledge base, the knowledge base edge is used to replace the original edge or the edge weight is increased. Before the system performs enhancement, it will perform deduplication and consistency checks on the candidate edges to ensure that only a single semantic edge is retained for the same entity pair, avoiding relationship conflicts or reasoning ambiguities. Therefore, the role of the preset knowledge base is to supplement, correct, and enhance the automatically extracted relationships.

[0110] The medical knowledge base for enhancement processing includes, but is not limited to, the following data resources, such as national standard medical terminology libraries such as SNOMED CT and ICD coding systems; official clinical pathway guidelines including various disease clinical pathway documents issued by the National Health Commission; drug-related knowledge bases such as national pharmacopoeias, drug instructions, and drug adverse reaction databases; medical insurance rule libraries including medical insurance restricted drug use and reasonable diagnosis and treatment determination rules issued by local medical insurance bureaus; drug indication instructions are derived from drug use guidelines filed or publicly released by the drug regulatory department and are used to identify the indication relationships between drugs and diseases.

[0111] The above knowledge base may contain structured data (such as coded tables, databases) and semi-structured data (such as PDF format guidelines, rule descriptions in the form of HTML web pages). During the actual construction process, the following parsing methods are adopted to extract semantic relationships for enhancing the knowledge graph. The entries in the knowledge base are mapped to the medical entity nodes in the knowledge graph. Methods such as exact name matching, medical term synonym expansion, ICD code alignment, and calculating semantic similarity based on word vector models (such as BERT or semantic editing models with similar functions) are used for entity alignment and standardization processing. For structured knowledge tables, the field relationships are directly identified. For example, the field "disease → recommended drug" is parsed as the relationship "drug - indication - disease". For semi-structured text data, rule templates or natural language-based relationship extraction models are used to identify the semantic patterns between entity pairs, such as expression patterns like "suitable for treating...", "often complicated with...", "mainly caused by...". By parsing the structured or semi-structured information in the above knowledge base, semantic enhancement relationships for supplementing the knowledge graph are extracted, mainly including causal relationships, indicating that a medical entity is the cause or inducing factor of another entity, such as "inhaled pathogen" → "pneumonia"; indication relationships, indicating that a certain treatment method is applicable to a specific disease, such as "ceftriaxone" → "bacterial pneumonia"; complication relationships, indicating that two diseases have a relatively high probability of co-occurrence clinically, such as "hypertension" associated with "diabetes".

[0112] On the basis of completing the construction of the knowledge graph, the following semantic enhancement strategies are further introduced. For entity pairs with clear relationship definitions in the knowledge base, a new graph edge is automatically added as an "enhanced edge"; the enhanced edge has an independent relationship type identifier and is given a relatively high confidence weight. Preferably, a weight value of 0.3 is increased to improve its priority in graph reasoning. Before edge construction, text-level similarity is calculated using edit distance (Levenshtein Distance), Jaccard similarity, etc. If the similarity > threshold (such as 0.85), it is directly matched as the same node. Or, the knowledge base entity and the graph node name are encoded as vector representations (semantic editing models with similar functions such as BERT, SimCSE, ESimCSE, etc. can be used), and the cosine similarity of the two entity vectors is calculated. If the semantic similarity > 0.9, it is mapped to the same graph node; if there is a unique high-confidence match (such as the highest semantic similarity and exceeding the threshold), the knowledge base entity is bound to this node in the graph; if there are multiple candidate nodes, the top item with the highest similarity is retained and marked as "partial match"; if there is no match, the original entity is retained for manual review or use during heterogeneous graph expansion, so as to achieve a one-to-one correspondence between the knowledge base entity and the graph node; if there are entity aliases, abbreviations, or language differences, the matching result with a word vector similarity greater than the set threshold (such as 0.85) is preferentially used as the alignment basis.

[0113] Calibrate the directionality of the edges in the medical relationship network data based on the causal relationship, adaptation relationship, and concurrent association relationship data of medical entities to obtain medical knowledge graph data.

[0114] In one embodiment, to ensure the logical consistency and reasoning effectiveness of the medical knowledge graph relationship expression, the directionality and type of the graph edges are defined and encoded in a standardized manner. According to the causal logic or co-occurrence characteristics of different semantic relationship types, the directionality of the graph edges is set. For example, a causal edge is used to represent that a certain medical entity is the pathogenic factor or antecedent event of another entity, and the direction of the edge points from the cause entity to the result entity. For example: "Bacterial infection" → "Pneumonia"; an adaptation edge is used to represent that a certain treatment method or drug is applicable to the treatment of a specific disease, and the direction of the edge points from the treatment plan entity to the target disease entity. For example: "Ceftriaxone" → "Pneumonia"; a concurrent edge is used to represent the relationship that two diseases or medical states have a high probability of comorbidity or occur simultaneously in clinical practice. This type of edge is a bidirectional edge without a causal sequence. For example: "Hypertension" "Coronary heart disease". To achieve the manageability of the graph structure and computer recognizability, a unique identifier code is configured for each type of semantic relationship type. Causal relationship: encoded as R1; adaptation relationship: encoded as R2; concurrent relationship: encoded as R3; the encoding can be used as an attribute field of the edge in the graph database for relationship screening, graph path calculation, or type recognition during the training of graph neural networks.

[0115] Optionally, the preset intelligent detection rule data includes a main diagnosis judgment rule, a surgery judgment rule, and a complication handling rule. The main diagnosis judgment rule, surgery judgment rule, and complication handling rule include a rule identifier, a rule type, a trigger condition, parameter configuration, and priority information. The trigger condition includes a combined match of inspection indicators, chief complaint information, and medical record elements, and is used to achieve the linkage judgment and intelligent trigger between the diagnostic request data and the rule.

[0116] In one embodiment, the preset intelligent detection rule data includes a primary diagnosis judgment rule, a surgical judgment rule, and a complication handling rule. All three types of rules include a rule identifier, a rule type, a trigger condition, parameter configuration, and priority information. The primary diagnosis judgment rule is used to judge candidate primary diagnoses based on fields such as chief complaint information, examination indicators, and medical imaging results. The trigger condition includes multi-field combination matching, and after being triggered, it outputs a candidate diagnosis name and a confidence score. The surgical judgment rule is used to judge and recommend a surgical plan based on imaging features and laboratory indicators in a specific diagnosis scenario, and outputs execution parameters such as a surgical code and an anesthesia method. The complication handling rule is used to intelligently prompt the risk of concurrent diseases when there is a high-risk primary diagnosis or an abnormal indicator combination, and recommend corresponding screening examinations. The trigger conditions of the rules are combined and matched using Boolean logic expressions. After the system receives the diagnosis request data, it automatically traverses the rule library, performs sequential matching and intelligent triggering based on the rule priorities, and realizes the logical linkage and decision explanation of the diagnostic reasoning process. For example, if the chief complaint contains "cough", the WBC index in the blood routine is greater than 10, and the chest CT shows lung shadows, then the primary diagnosis judgment rule "D001" is triggered, and the candidate diagnosis "bacterial pneumonia" is output, and a confidence score of 0.85 is assigned. The parameter configuration fields include information such as ICD codes (such as J15.9), recommended drugs (such as ceftriaxone), and priorities (such as P1). The system uses a rule engine method to execute rule traversal and condition judgment, supports multiple rules to be matched in parallel, and outputs candidate results according to the rule priorities for the diagnostic reasoning and recommendation processes to call.

[0117] Optionally, please refer to Figure 2 , S1 includes:

[0118] S11. Obtain the diagnosis request data, and perform entity matching on the diagnosis request data using the preset medical knowledge graph data to obtain entity matching data;

[0119] In one embodiment, for text fields in diagnostic request data, such as chief complaint descriptions, imaging examination report texts, etc., a pre-trained language model trained on medical domain corpora (such as Chinese-Clinical-BERT or a semantic editing model with similar functions) is used for named entity recognition to extract medical entities such as diseases, symptoms, examination items, drugs, etc. from them. This model can automatically recognize non-standard terms in clinical expressions and output structured entity label results. For the recognized text entities, the system performs entity standardization processing by comparing medical term dictionaries (such as ICD term phrase dictionaries, SNOMED-CT term libraries) to achieve semantic alignment of multi-source texts in the medical knowledge graph. For example, non-standard expressions such as "there is a shadow in the lungs" and "CT reveals patchy density shadows" are uniformly mapped to the entity node of "lung shadow". For structured numerical fields in diagnostic request data (such as blood routine results, imaging conclusion parameters, etc.), the system uses a rule matching mechanism for classification. For example, if the WBC value is greater than 10.0, it can be matched to the entity of "elevated white blood cells" in the knowledge graph according to the preset rules; if the imaging report contains the keyword "patchy shadow", it is matched to the entity of "lung shadow".

[0120] S12. Retrieve the initial diagnosis-related candidate set data from the medical knowledge graph data according to the entity matching data to obtain the initial diagnosis-related candidate set data;

[0121] In one embodiment, to retrieve the candidate diseases most relevant to the current diagnostic request from the knowledge graph, the system takes the "disease" node as the target node and performs a reverse graph retrieval operation based on the semantic association relationships between the entities matched with the current diagnostic request in the graph to construct the initial diagnosis candidate set. Specifically, the system first looks up the disease nodes in the knowledge graph that have semantic relationships such as "manifested as" and "supported by" with the structured medical entities (such as "cough", "fever", "lung shadow", "elevated white blood cells", etc.) extracted from the current diagnostic request data. Each disease node is regarded as an alternative object for the candidate main diagnosis, and a relevance score is calculated based on the hit situation of the associated entities. The scoring process uses a weighted co-occurrence scoring model, denoted as: , where is the matching score of the candidate disease node, and different types of entities are set with different weight coefficients to reflect their importance in diagnostic reasoning. For example, the weight of symptom entities is set to 1.0, the weight of imaging entities is set to 1.5, and the weight of laboratory index entities is set to 1.2. The system sorts according to the matching scores of each candidate disease node and selects the top N high-score nodes (such as Top10) as the initial diagnosis-related candidate set corresponding to the current diagnostic request.

[0122] S13. Use the preset intelligent detection rule data to perform matching of candidate main diagnosis items on the initial diagnosis-related candidate set data, and obtain candidate main diagnosis data.

[0123] In one embodiment, according to the preset intelligent detection rule data, trigger condition matching judgment is performed on each candidate diagnosis in the initial diagnosis-related candidate set, and the trigger conditions include the combination relationship of the inspection indicators, the chief complaint information, and the graph-matching entities in the diagnosis request; if the rule is triggered, the candidate diagnosis is determined as a candidate main diagnosis to form candidate main diagnosis data.

[0124] Optionally, please refer to Figure 3 , S2 includes:

[0125] S21. Extract candidate diagnosis context from the diagnosis request data according to the candidate main diagnosis data to obtain candidate diagnosis context data;

[0126] In one embodiment, according to the candidate main diagnosis data, context information semantically related to each candidate diagnosis is extracted from the diagnosis request data. The context information includes symptom descriptions, laboratory indicators, imaging features, etc. Candidate diagnosis context data is obtained through the reverse matching and semantic association analysis method of the knowledge graph. Based on the established entity relationship path in the medical knowledge graph, with the candidate diagnosis as the central node, the system retrieves its adjacent typical symptoms, inspection indicators, and other supporting entities in reverse, and judges whether these entities exist in the current diagnosis request data. If there is a match, it is recognized as the context feature of the candidate diagnosis. For example, when the candidate diagnosis is "bacterial pneumonia", its adjacent nodes in the graph include "fever", "cough", "lung shadow", "elevated white blood cells", etc.; if the diagnosis request contains the above information, it can be extracted as the context data of "bacterial pneumonia". For the unstructured text information in the diagnosis request (such as the chief complaint description, imaging report, etc.), the system uses a natural language processing module based on a pre-trained language model to extract keyword vectors, and calculates the similarity with the semantic vectors of the candidate diagnosis entities. If the similarity is higher than the set threshold (such as 0.75), the field corresponding to the keyword is extracted as the context supplementary feature of the candidate diagnosis. This method can be used to identify information matching of implicit expressions or synonymous expressions.

[0127] S22. Initialize the rule parameter mapping for the candidate main diagnosis data and the candidate diagnosis context data according to the preset intelligent detection rule data to obtain rule parameter mapping data;

[0128] In one embodiment, the preset intelligent detection rule data is called to perform rule matching on the candidate primary diagnosis data and its context data. If the diagnosis context meets the trigger condition of the corresponding rule, the standardized diagnosis parameters corresponding to the rule are extracted, including ICD codes, medical insurance risk levels, recommended treatment path IDs, recommended drugs, etc., to form rule parameter mapping data.

[0129] In one embodiment, the function of the rule parameter mapping is as follows: mapping diagnosis-related information to multiple standardized fields, including but not limited to ICD codes (used for disease identification and classification), medical insurance review grades (used for rationality assessment), preoperative examination suggestions (used for preoperative safety assessment), recommended treatment path IDs (used for clinical path execution docking), etc. The system realizes the call and trigger of parameter rules through a matching mechanism. For example, for rule retrieval condition matching, each rule sets a combination of applicable diagnosis names and trigger conditions. The system makes a matching judgment based on the current candidate primary diagnosis name, the extracted context keywords, examination index results, etc. The trigger condition supports Boolean logic combination expressions, such as: "cough + WBC > 10 and there is a'shadow' in the image", etc. Example of rule structure: For example, for the candidate diagnosis of "bacterial pneumonia", if the system detects that the WBC value in the laboratory index is greater than 10 and there is a description of "lung shadow" in the imaging report, the following parameter rules can be triggered: Rule_ID: P001, ICD10 code: J15.9, medical insurance risk level: medium risk, recommended path ID: PATH_03, treatment suggestion: cefotaxime is the first choice for antibiotics. If a candidate diagnosis matches multiple parameter rules at the same time, the system will select the rule with the highest priority as the basis for parameter output according to the rule priority field; if some parameters in the rule are missing or the context information does not meet the complete trigger condition, the parameter item will be marked as "to be manually confirmed" and its context source will be retained for manual review or secondary inference of the system.

[0130] S23. Construct the candidate diagnosis parameter structure according to the rule parameter mapping data to obtain the candidate primary diagnosis parameter data.

[0131] In one embodiment, according to the rule parameter mapping data, construct the structured parameter information of the candidate primary diagnosis. The parameter structure includes diagnosis name, coding information, confidence score, clinical path ID, recommended drug information, and whether review is required field, and generate the candidate primary diagnosis parameter data.

[0132] Optionally, please refer to Figure 4 , S3 includes:

[0133] S31. Obtain the historical diagnosis request data and the historical candidate primary diagnosis data;

[0134] In one embodiment, historical diagnostic request data and its corresponding historical candidate primary diagnostic data are obtained. The data comes from the historical medical records of medical institutions and the output records of the candidate diagnostic system, and undergoes integrity screening and structured cleaning processing.

[0135] S32. Extract the diagnostic instance associations for the historical diagnostic request data and the historical candidate primary diagnostic data to obtain diagnostic instance association data;

[0136] In one embodiment, to construct a diagnostic linkage structure based on historical data, the system performs instance-level association analysis on historical diagnostic request records and their corresponding candidate primary diagnostic information. Specifically, the system determines whether there is semantic commonality, similar diagnostic structure, or disease evolution trend between two historical diagnostic records to identify diagnostic instance pairs with significant linkage relationships. The judgment of the diagnostic instance association is based on the following three strategies: Symptom or feature semantic similarity analysis. The system vectorizes and encodes the text fields such as the chief complaint, symptom description, and examination information in each diagnostic request data, and uses a sentence vector modeling method (such as SimCSE, BERT-CLS, or a semantic editing model with similar functions) to generate corresponding semantic vectors, and calculates the semantic similarity between historical instances. When the similarity is higher than a set threshold (such as 0.85), it is considered that the two records have high semantic consistency in subjective symptoms or objective descriptions, and is marked as "semantically similar". Candidate diagnosis overlap determination. The system compares the coincidence of the candidate primary diagnosis sets in two historical instances. If the ratio of the number of intersections to the number of unions is greater than or equal to 0.5 (that is, the candidate diagnosis overlap reaches 50%), it is considered that they have candidate neighbor features at the structural level and is marked as "diagnostic candidate similar". Similarity evaluation of the evolution path after diagnosis. For cases that have been diagnosed and have follow-up records, the system determines whether there is consistency in the complications that occur within 7 days after diagnosis or the clinical paths entered. If two cases show similar complications or receive the same treatment path in the short term, they are regarded as having a "clinically evolving similar" relationship. The results of the above-mentioned association analysis are output in a structured form. Each historical diagnostic record is associated with a set of similar diagnostic instance identifiers (Instance_ID), and is accompanied by a similar type label (such as semantically similar, diagnostic candidate similar, evolution path similar), the corresponding similarity score, and time difference information (such as within 3 days, within 7 days, etc.).

[0137] S33. Mine the linkage relationships based on the diagnostic instance association data to obtain linkage relationship data;

[0138] In one embodiment, the system performs linkage relationship recognition and modeling operations based on historical instance association data and the knowledge graph structure. The linkage relationship refers to the co-occurrence, causal, evolutionary, or structural connection between a candidate diagnosis and other diagnoses, symptoms, or examination features. Specifically, the mining of the linkage relationship includes the following three methods: co-occurrence relationship statistical analysis, in which the system traverses all historical instances and their identified context entities, counts the frequency of simultaneous occurrence of candidate diagnosis A and other entity B (which can be a disease, symptom, or examination index) in the same instance, and calculates their co-occurrence rate. If the co-occurrence frequency exceeds 10%, it is considered that there is a weak linkage relationship between the two; if it exceeds 30%, it is defined as a strong linkage relationship, which is used to construct a co-occurrence edge with high confidence. Temporal transformation relationship recognition, in which the system performs sequence analysis on historical cases with time tags to determine whether diagnosis A significantly precedes diagnosis or symptom B in time. If there is a stable sequence, record its average transformation time interval and directionality (such as A→B), and construct an evolutionary relationship edge accordingly, indicating the disease development path or the change trend of the clinical state. Graph path relationship enhancement, in combination with the medical knowledge graph structure, the system retrieves whether there is a direct path connection or a shared upper semantic node (i.e., a common parent node) between candidate diagnosis A and target entity B. If there is a co-occurrence path, causal chain, or co-disease lineage relationship between A and B, the linkage credibility is improved, which is used to enhance the structural consistency and reasoning reliability in the graph. The linkage relationship is represented in the form of a structured triple, with the format: 〈entity A, linkage type, entity B〉, where the linkage type includes type identifiers such as "co-occurrence", "transformation", "evolution", "graph structure", etc.; each relationship edge is attached with multiple attribute fields, including the linkage weight value, the number of supporting instances, whether it is a temporal relationship, the average time interval, etc.

[0139] S34. Generate diagnostic linkage network data according to the diagnostic instance association data and the linkage relationship data, and obtain the diagnostic linkage network data.

[0140] In one embodiment, according to the diagnostic instance association data and the linkage relationship data, a diagnostic linkage network data including multiple relationship types, multi-edge weight features, and time directionality is constructed with candidate diagnoses, symptoms, examination items, etc. as nodes.

[0141] Optionally, please refer to Figure 5 , S4 includes:

[0142] S41. Construct a candidate main diagnosis subgraph for the diagnostic linkage network data according to the candidate main diagnosis data, and obtain the candidate main diagnosis subgraph data;

[0143] In one embodiment, the system takes each candidate primary diagnosis node as the center and extracts the subgraph within its K-hop range from the diagnostic linkage graph. Preferably, it is set as the 2-hop adjacency range to cover the direct features (1 hop) of the diagnosis and its context-related nodes (2 hops), forming a diagnostic context subgraph with a complete local structure. The nodes in the subgraph include, but are not limited to, symptom entities directly related to the candidate diagnosis (such as "cough", "fever"); examination or test indicators (such as "elevated white blood cells", "pulmonary shadow"); concurrent diseases or sequential evolution diagnoses (such as "lung abscess", "chronic bronchitis"), etc. The edges in the subgraph retain the relationship types and their weight information in the original graph, and the relationship types include "manifest as", "support", "concurrent with", "co-occur in", "evolve into", etc. When there are multiple paths between the candidate diagnosis and a certain associated node, the system preferentially retains the path with the largest edge weight value to ensure that only the most representative semantic path is retained in the subgraph structure; to avoid overfitting of features caused by an overly large graph structure, an upper limit N is set for the number of subgraph nodes, preferably set to within 30 nodes; edges with relatively low edge weights or insufficient semantic confidence can be eliminated using a threshold filtering strategy (such as setting the threshold to 0.85). Through the above subgraph construction method, the system generates a context semantic subgraph with a complete structure and clear relationships for each candidate primary diagnosis.

[0144] S42. Extract node linkage features based on the candidate primary diagnosis subgraph data and the candidate primary diagnosis data to obtain node linkage feature data;

[0145] In one embodiment, for each node in the candidate diagnostic sub-graph, the system extracts multi-class features in the following dimensions. For the structural feature dimension, the node degree is included, which represents the number of edges directly connected to the current node in the graph and is used to measure the connection strength of the entity in the local structure; the hop distance represents the number of hops of the shortest path between this node and the candidate diagnostic central node (such as 1 hop, 2 hops), reflecting its association tightness; the occurrence frequency is the number of times or frequency that this node appears in the historical instance graph and is used to measure its global representativeness or importance. For the semantic feature dimension, the semantic similarity is included, which is the cosine similarity between the node word vector generated based on a pre-trained language model (such as BERT or a semantic editing model with similar functions) and the current diagnostic request chief complaint vector, and is used to measure the degree of fit between the entity expression and the chief complaint semantics; the node type is represented in a One-hot manner to indicate the category to which the node belongs, such as symptoms, examination items, diagnostic entities, etc.; the edge type and edge weight features represent the type of connection relationship (such as "manifested as", "co-occurred in") between the current node and the candidate diagnostic node and the weight value of the edge, and are used to convey relationship semantics and structural importance. For the context matching feature dimension, the context hit status is included, which is a boolean field indicating whether the medical entity corresponding to this node exists in the current diagnostic request data (for example, if the chief complaint contains "fever", then the value of the "fever" node is True), and is used to reflect the degree of association between the entity and the current request context.

[0146] S43. Perform an interactive scoring based on the candidate main diagnosis data and the node linkage feature data to obtain preliminary scoring data;

[0147] In one embodiment, based on a predefined feature importance rule, for each adjacent node in the candidate diagnostic sub-graph, the method calculates its degree of support for the main diagnosis. The specific calculation formula is as follows:

[0148] (Node edge weight × Tone similarity between the node and the chief complaint × Context hit flag), where is the preliminary scoring data. The edge weight represents the strength of the structural relationship between the diagnostic node and the adjacent node. The semantic similarity is calculated based on the cosine value between the entity word vector and the chief complaint semantic vector. The context hit flag is a boolean value indicating whether this node is mentioned in the current diagnostic request. The contribution degrees of all supportive nodes are accumulated through a weighted summation method to form the context confidence score of the candidate diagnosis.

[0149] In one embodiment, the system may select a shallow graph neural network model (such as GCN, GraphSAGE, or a graph neural network with similar functions) to perform node feature aggregation on the diagnostic subgraph structure. In specific operations, taking the candidate diagnostic node as the target node, receiving the embedding representations of its adjacent nodes in the structural and semantic feature spaces, completing multi-hop neighbor feature fusion through graph convolution operations, outputting the representation vector of the diagnostic node, and generating a score value through a linear mapping layer as the confidence score of the diagnosis in the current context. For example, the score of a certain candidate diagnosis after aggregation in its context graph structure is 0.78.

[0150] S44. Perform simplified graph attention mechanism processing based on the candidate main diagnosis data and the preliminary score data to obtain candidate main diagnosis weight data.

[0151] In one embodiment, for each candidate main diagnosis node, the system traverses all adjacent nodes (such as symptoms, examinations, concurrent diagnoses, etc.) in its subgraph, and comprehensively calculates the attention weights based on the following factors: node importance, including the occurrence frequency of the node in the global graph structure, type (such as typical symptoms vs. secondary symptoms), and structural connectivity with the main diagnosis (such as whether it is a 1-hop adjacent node); semantic similarity, the semantic matching degree between the word vector corresponding to the node and the diagnosis chief complaint vector, measuring the degree of fit between the entity expression and the current context; structural edge weight, the edge weight value between the candidate diagnosis node and the adjacent node in the original graph, reflecting its structural association strength. After the above factors are normalized and fused, the attention weight of each adjacent node is calculated and used to weight the feature vector of the node. Subsequently, the system aggregates the weighted features of all adjacent nodes into the representation of the candidate diagnosis node and outputs the diagnosis confidence weight value; this mechanism can achieve the strengthening of key information and the suppression of interference information in the diagnostic subgraph. For example, when typical symptoms such as "fever" and "cough" exist in the subgraph corresponding to the candidate diagnosis "bacterial pneumonia", due to the high importance and semantic similarity of these nodes, they will be given higher attention weights; while for non-specific symptoms such as "slight fatigue", the attention is lower and the contribution to the weight is smaller. The system realizes the fusion modeling of structural information and semantic information through the adaptive attention mechanism and outputs the confidence score of the candidate diagnosis node as the decision basis.

[0152] Optionally, the interaction score includes:

[0153] Extract context semantic features based on the candidate main diagnosis data to obtain context semantic feature data;

[0154] In one embodiment, text such as the chief complaint, symptoms, and examination fields is input into a pre-trained language model dedicated to the medical field (such as Chinese-Clinical-BERT, and a semantic editing model with similar functions); the sentence vector representation of the entire context is extracted. The typical symptoms and examination items of the candidate primary diagnosis in the knowledge graph are extracted; it is determined whether these elements are hit in the current request, and a set of boolean flags or numerical enhancement dimensions are set. A context vector schematic dimension is obtained, such as 128 dimensions, which contains sub-fields such as text semantic dimension, hit enhancement dimension, disease type identifier, and feature missing identifier.

[0155] Bilinear attention calculation is performed based on the node linkage feature data and the context semantic feature data to obtain the matching weight data;

[0156] In one embodiment, when the bilinear attention mechanism calculates the matching weight between the context and the graph node, a bidirectional interaction representation of the context node features is constructed, and the matching degree in its multi-dimensional semantic space is measured by a learnable weight matrix or a static weight function. For example, when the diagnostic request context contains "CT shows a shadow in the lungs", the bilinear interaction score between the "lung shadow" node in the graph and the context is relatively high, and a higher attention weight is assigned; conversely, if a certain node (such as "headache") does not appear or has no semantic association in the current context, its attention score will be suppressed.

[0157] The system extracts the semantic representation of the diagnostic request text, denoted as the context vector , and the linkage feature vectors of each node in the candidate diagnostic sub-graph in the graph , and constructs the following bilinear interaction scoring function: , is the original attention score, the matching score between the context and the node , is the transposed representation of the context vector, the semantic representation of the current request (chief complaint, medical history, etc.), is the bilinear weight matrix, which is used to capture the interaction relationship between the context and the graph node (it is a learnable parameter matrix, that is, it is automatically updated through gradient backpropagation during the model training process, or a diagonal matrix specifying the dimension weights), is the node linkage feature data, which represents the multi-dimensional semantic and structural features of the adjacent node in the sub-graph; Softmax calculation is performed on the set of all adjacent nodes of the candidate diagnostic node to obtain the attention weight of each adjacent node, is the attention weight of the node , is the original attention score, is the exponential function, For the node sequence number thinking, Is the candidate diagnosis node Of the first-order neighbor set, Is the node ; Use the attention weight to weighted aggregate the adjacent node features to form an enhanced representation of the candidate diagnosis node in the current context: , Is the matching weight data, Is the node sequence item, Is the candidate diagnosis node Of the first-order neighbor set, Is the node Of the attention weight, Is the node linkage feature data, indicating the multi-dimensional semantics and structural features of the adjacent node In the subgraph.

[0158] According to the matching weight data, perform feature fusion on the node linkage feature data and the context semantic feature data to obtain the preliminary scoring data.

[0159] In one embodiment, aggregate the linkage features of all graph nodes according to the attention weight, fuse them with the context semantic vector, and form a preliminary confidence score for the diagnosis candidate. Perform weighted average on the feature vectors of all graph nodes according to the corresponding attention weights to obtain a graph feature aggregation vector. Concatenate, non-linearly map, or take the vector inner product of the context semantic vector and the graph aggregation vector to generate a scalar score, which is the matching weight data; this score is the semantic rationality score of the candidate diagnosis in the current request context.

[0160] Optionally, the simplified graph attention mechanism processing includes:

[0161] According to the candidate main diagnosis data and the preliminary scoring data, perform adjacent node similarity calculation to obtain the adjacent node edge weight data;

[0162] In one embodiment, for each candidate main diagnosis node, all first-order adjacent nodes (such as symptoms, examinations, concurrent diagnoses, etc.) in its corresponding subgraph, the system extracts the multi-dimensional feature representation of the node, including but not limited to structural linkage features (such as degree value, hop distance, node type), context relevance features (such as hit status, semantic similarity score), etc., to form a unified node feature vector representation. Calculate the cosine similarity between the candidate diagnosis node feature vector and the adjacent node feature vector to measure the basic similarity between the two in the structural and semantic spaces; introduce the context score (from the previous module) of the adjacent node in the current diagnosis request context as a semantic importance weight enhancement factor; , Is the adjacent node edge weight data, is the weight of similarity, is similarity, is the weight of preliminary scoring data, is preliminary scoring data, and is an adjustable weighting coefficient, preferably set to = 0.6, = 0.4, to ensure a weight distribution strategy that emphasizes structural similarity and supplements semantic signals.

[0163] Calculate the simplified attention coefficient based on the adjacent node edge weight data and the preliminary scoring data to obtain the simplified attention coefficient data;

[0164] In one embodiment, for each candidate main diagnosis node, all its first-order adjacent edges have obtained edge weights through weighted calculation of structural similarity and context semantic scores . The system performs a Softmax operation on all adjacent edges of the node to generate a simplified attention coefficient. The calculation method is as follows: , is the node 's attention score for the candidate diagnosis node , is the original edge weight, which is the original edge weight between the candidate main diagnosis node and the adjacent node , obtained through calculation of structural similarity and context scoring, is the exponential function, is the adjacent node index, is the candidate diagnosis node 's set of all adjacent nodes, is the node and the th adjacent node's edge weight. The simplified attention mechanism has the following characteristics: it does not use trainable parameters, avoiding introducing additional model burdens; it does not introduce a multi-head mechanism and non-linear activation, only retaining static attention scores based on graph structure and semantic feature distributions; the attention scores reflect the relative contribution degrees of adjacent nodes and are used for weighted aggregation processing of node features. Through this simplified attention normalization method, the system can achieve an efficient and controllable feature propagation mechanism while maintaining the integrity of the graph structure.

[0165] Perform multi-order aggregation based on the simplified attention coefficient data to obtain multi-order neighbor aggregation data;

[0166] In one embodiment, information is transmitted to nodes using attention coefficients to generate an enhanced representation of the candidate primary diagnosis node, which represents not only aggregating first-order neighbors but also including multi-order neighbor information (such as second-order neighbors) to increase semantic coverage. The features of each first-order neighbor are weighted and summed using attention coefficients to form a direct context aggregation vector for the candidate diagnosis, that is, first-order aggregation. Repeat the above aggregation method for the neighbors of the first-order neighbors (i.e., second-order neighbors); allow information to affect the primary diagnosis score through an indirect path (such as "cough" → "bronchitis" → "bacterial pneumonia"). For each additional hop, the weight decay coefficient is multiplied by (such as 0.5); avoid the over-amplification effect of distant neighbor nodes. The multi-order aggregation result is a new diagnostic feature vector that fuses structural + semantic information.

[0167] Perform a residual connection based on the multi-order neighbor aggregation data and the preliminary scoring data to obtain the candidate primary diagnosis weight data.

[0168] In one embodiment, for a candidate diagnosis with a scalar form score, its score is defined as: where is the candidate primary diagnosis weight data, is the fusion weight coefficient, preferably set to 0.6, which is used to maintain the dominant position of the semantic score while introducing a certain proportion of structural enhancement compensation, is the preliminary scoring data, is the multi-order neighbor aggregation data. When there are high-confidence adjacent nodes in the graph structure, even if the preliminary score is insufficient, the structural enhancement part can effectively improve the candidate diagnosis score, reflecting its potential contextual reasonableness; when the graph structure signal is weak and there is more noise in the aggregation path, the residual mechanism can retain the preliminary scoring result and prevent the uncertainty introduced by the structure from misleading the diagnosis score.

[0169] Optionally, the present application also provides an artificial intelligence-based main diagnosis reasonableness judgment system for performing the artificial intelligence-based main diagnosis reasonableness judgment method as described above. The artificial intelligence-based main diagnosis reasonableness judgment system includes:

[0170] A candidate diagnosis generation module, configured to obtain diagnosis request data, and use preset medical knowledge graph data and preset intelligent detection rule data to generate candidate primary diagnoses for the diagnosis request data, obtaining candidate primary diagnosis data;

[0171] A diagnosis parameter extraction module, configured to generate diagnosis execution parameters based on the candidate primary diagnosis data, obtaining candidate primary diagnosis parameter data;

[0172] The diagnostic linkage mapping module is used to obtain historical diagnostic request data and historical candidate primary diagnosis data, and perform diagnostic linkage processing based on the historical diagnostic request data and the historical candidate primary diagnosis data to obtain diagnostic linkage network data;

[0173] The diagnostic weight evaluation module is used to perform node interaction processing on the candidate primary diagnosis data according to the diagnostic linkage network data to obtain candidate primary diagnosis weight data.

[0174] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended application documents rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0175] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A main diagnostic rationality judgment method based on artificial intelligence, characterized in that, The method includes: S1. Obtain diagnostic request data, and generate candidate primary diagnoses for the diagnostic request data by using preset medical knowledge graph data and preset intelligent detection rule data, to obtain candidate primary diagnosis data; S2. Generate diagnostic execution parameters according to the candidate primary diagnosis data, to obtain candidate primary diagnosis parameter data, where the candidate primary diagnosis parameter data includes a diagnosis name, coding information, confidence score, clinical pathway ID, recommended drug information, and a field indicating whether a review is required; S3. Obtain historical diagnostic request data and historical candidate primary diagnosis data, and perform diagnostic linkage processing according to the historical diagnostic request data and the historical candidate primary diagnosis data, to obtain diagnostic linkage network data, where the diagnostic linkage processing is a process in which the system extracts diagnostic association information corresponding to diseases, symptoms, examinations, and treatment paths based on the historical diagnostic request data and the historical candidate primary diagnosis data, constructs a diagnosis-symptom-parameter ternary graph through matching with a preset similarity model, and calculates the co-occurrence frequency between each node as the edge weight; S4. Construct a candidate primary diagnosis subgraph for the diagnostic linkage network data according to the candidate primary diagnosis parameter data, to obtain candidate primary diagnosis subgraph data; extract node linkage feature data according to the candidate primary diagnosis subgraph data and the candidate primary diagnosis parameter data; perform an interaction score according to the candidate primary diagnosis parameter data and the node linkage feature data, to obtain preliminary score data; perform a simplified graph attention mechanism process according to the candidate primary diagnosis parameter data and the preliminary score data, to obtain candidate primary diagnosis weight data, and perform a main diagnosis recommendation operation on the diagnostic request data according to the candidate primary diagnosis weight data, and output the recommended main diagnosis result.

2. The method according to claim 1, wherein The construction steps of the preset medical knowledge graph data include the following steps: Obtain structured electronic medical records, medical standard coding data, and unstructured medical literature data to construct a multi-source medical data set; Perform entity recognition and entity relationship extraction according to the multi-source medical data set to obtain medical entity data and initial association relationship data; Construct semantic relationship edges between entities according to the medical entity data and the initial association relationship data to form medical relationship network data; Perform semantic relationship enhancement processing on the medical entity data according to a preset medical knowledge base to obtain medical entity causal relationship, medical entity adaptation relationship, and medical entity concurrent association relationship data; Calibrate the directionality of the edges of the medical relationship network data according to the medical entity causal relationship, medical entity adaptation relationship, and medical entity concurrent association relationship data to obtain medical knowledge graph data.

3. The method according to claim 1, wherein The preset intelligent detection rule data includes a primary diagnosis judgment rule, a surgery judgment rule, and a complication handling rule. The primary diagnosis judgment rule, the surgery judgment rule, and the complication handling rule include a rule identifier, a rule type, a trigger condition, parameter configuration, and priority information. The trigger condition includes a combined match of examination indicators, chief complaint information, and medical record elements, and is used to implement the linkage judgment and intelligent trigger between the diagnostic request data and the rules.

4. The method according to claim 1, wherein S1 includes: Obtain diagnostic request data, and perform entity matching on the diagnostic request data using the preset medical knowledge graph data to obtain entity matching data; Retrieve the initial diagnosis-related candidate set from the medical knowledge graph data according to the entity matching data to obtain the initial diagnosis-related candidate set data; Perform main diagnosis candidate item matching on the initial diagnosis-related candidate set data using the preset intelligent detection rule data to obtain candidate main diagnosis data.

5. The method according to claim 1, characterized in that S2 Including: Extract candidate diagnosis context from the diagnostic request data according to the candidate main diagnosis data to obtain candidate diagnosis context data; Initialize the rule parameter mapping for the candidate main diagnosis data and the candidate diagnosis context data according to the preset intelligent detection rule data to obtain rule parameter mapping data; Construct the candidate diagnosis parameter structure according to the rule parameter mapping data to obtain candidate main diagnosis parameter data.

6. The method according to claim 1, wherein S3 Including: Obtain historical diagnostic request data and historical candidate main diagnosis data; Extract the diagnostic instance association from the historical diagnostic request data and the historical candidate main diagnosis data to obtain diagnostic instance association data; Mine the linkage relationship according to the diagnostic instance association data to obtain linkage relationship data; Generate diagnostic linkage network data according to the diagnostic instance association data and the linkage relationship data.

7. The method according to claim 1, wherein The interaction score includes: Extract context semantic features from the candidate main diagnosis parameter data to obtain context semantic feature data; Perform bilinear attention calculation according to the node linkage feature data and the context semantic feature data to obtain matching weight data; Fuse the features of the node linkage feature data and the context semantic feature data according to the matching weight data to obtain preliminary score data.

8. The method according to claim 1, characterized in that, The simplified graph attention mechanism processing includes: Calculate the similarity of adjacent nodes according to the candidate main diagnosis parameter data and the preliminary score data to obtain adjacent node edge weight data; Calculate the simplified attention coefficient according to the adjacent node edge weight data and the preliminary score data to obtain simplified attention coefficient data; Perform multi-order aggregation according to the simplified attention coefficient data to obtain multi-order neighbor aggregation data; Perform residual connection according to the multi-order neighbor aggregation data and the preliminary score data to obtain candidate main diagnosis weight data.

9. An artificial intelligence-based main diagnosis rationality judgment system, characterized in that, For executing the artificial intelligence-based main diagnosis rationality judgment method as described in claim 1, the artificial intelligence-based main diagnosis rationality judgment system includes: A candidate diagnosis generation module, configured to obtain diagnostic request data, and generate candidate main diagnosis for the diagnostic request data by using the preset medical knowledge graph data and the preset intelligent detection rule data to obtain candidate main diagnosis data; A diagnostic parameter extraction module, configured to generate diagnostic execution parameters according to the candidate main diagnosis data to obtain candidate main diagnosis parameter data; A diagnostic linkage graph construction module, configured to obtain historical diagnostic request data and historical candidate main diagnosis data, and perform diagnostic linkage processing according to the historical diagnostic request data and the historical candidate main diagnosis data to obtain diagnostic linkage network data; A diagnostic weight evaluation module, which is used to construct a candidate primary diagnosis subgraph for the diagnostic linkage network data based on the candidate primary diagnosis parameter data to obtain candidate primary diagnosis subgraph data; extract node linkage features based on the candidate primary diagnosis subgraph data and the candidate primary diagnosis parameter data to obtain node linkage feature data; perform an interaction score based on the candidate primary diagnosis parameter data and the node linkage feature data to obtain preliminary score data; perform a simplified graph attention mechanism process based on the candidate primary diagnosis parameter data and the preliminary score data to obtain candidate primary diagnosis weight data, and perform a main diagnosis recommendation operation on the diagnostic request data according to the candidate primary diagnosis weight data, and output the recommended main diagnosis result.

Images