Emergency obstetrics and gynecology pre-examination triage reasoning method and system based on knowledge graph

Abstract

The application discloses an emergency gynecological pre-examination triage reasoning method and system based on a knowledge graph, relates to the technical field of medical intelligent triage, and provides the following scheme, which comprises the following steps: acquiring chief complaint information of an object to be triaged, constructing a symptom algorithm knowledge graph, and constructing a symptom algorithm knowledge graph containing trigger relationships between symptom concepts and symptom algorithms, containing relationships between symptom algorithms and decision factors, and dependency relationships and influence criteria between decision factors and collection tasks. The application sets a preset safety level according to a triage level constraint, and selects decision factors that trigger a triage level not lower than the preset safety level when verification is established from decision factors in an unknown verification state to form a decision factor set to be checked, thereby solving the problems that pre-examination checking lacks risk priority criteria and is easily disturbed by low-value information to miss high-risk checking points, and realizing a check target focus and priority constraint of a check process in the safety level.

Description

Technical Field

[0001] This invention relates to the field of medical intelligent triage technology, and more specifically, this application relates to a reasoning method and system for emergency obstetrics and gynecology triage based on knowledge graphs. Background Technology

[0002] Emergency obstetrics and gynecology triage is typically a preliminary step before patients enter the emergency treatment process. Its purpose is to quickly identify potential critical risks based on the patient's chief complaint and preliminary physical examination, and to assign patients to the appropriate triage level and treatment channel to ensure timely treatment and rational resource allocation. In practice, triage personnel generally first obtain the patient's chief complaint and basic information, and, in conjunction with existing triage protocols or treatment guidelines, make a preliminary judgment on potentially relevant symptoms and risk clues. They then conduct necessary medical history taking, verification, and physical examinations according to the protocol requirements. Subsequently, they synthesize the collected observations to determine the triage level and generate corresponding records. However, due to the fast pace, fragmented information, and rapid changes in patient condition in the emergency setting, there are often significant differences in the order of collection, completeness, and expression of chief complaints and observations. This places higher demands on the triage process organization and information processing capabilities in terms of "which key points to verify, which information to collect first, and how to reach a consistent conclusion," leading to the specific problems exposed in the existing process regarding chief complaint mapping, verification organization, and cost constraints.

[0003] Current emergency obstetrics and gynecology triage systems largely rely on manual experience or static forms for item-by-item verification. When patients present their complaints in natural language, with diverse expressions, negative modifiers, and missing information, it is difficult to consistently map these complaints to an executable triage decision path and implement "known / unknown" state management for key decision factors. This leads to triage personnel either over-collecting information according to a fixed checklist or overlooking verification points that decisively affect the triage level, resulting in triage delays or level deviations. Furthermore, when faced with multiple decision points and resource constraints, existing processes lack a calculable task selection mechanism, failing to optimize the overall costs of time, manpower, and invasiveness while covering all key verification points. This results in a lack of priority and cost constraints in triage guidance, limiting execution efficiency and consistency. Therefore, this paper proposes a knowledge graph-based reasoning method and system for emergency obstetrics and gynecology triage to address this problem. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a knowledge graph-based triage and reasoning method and system for emergency obstetrics and gynecology, which solves the problems mentioned in the background section.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, this application provides a knowledge graph-based method for triage and reasoning in emergency obstetrics and gynecology departments, the method comprising:

[0007] Obtain the chief complaint information of patients to be triaged and construct a symptom algorithm knowledge graph;

[0008] The symptom algorithm knowledge graph includes the triggering relationship between symptom concepts and symptom algorithms, the inclusion relationship between symptom algorithms and determining factors, and the dependency relationship and influence criteria between determining factors and collection tasks. The determining factors are associated with triage level constraints, and the collection tasks have cost attributes.

[0009] The chief complaint information is converted into a set of symptom concepts through semantic mapping. Based on this, the corresponding set of symptom algorithms is retrieved and determined in the symptom algorithm knowledge graph, and the verification status of each determinant is initialized as valid, invalid, or unknown.

[0010] From the decision factors whose verification status is unknown, select the decision factors whose triage level is not lower than the preset security level and form a set of decision factors to be verified;

[0011] Based on the collection tasks that depend on the decision factors to be verified, calculate the collection task set that can cover all elements in the set of decision factors to be verified and has the lowest total cost attribute, and output it as a pre-inspection guidance list.

[0012] Obtain the observations generated after execution according to the pre-inspection guidance checklist, compare them with the impact criteria to update the verification status of each element in the set of decision factors to be verified;

[0013] If the set of determinants to be verified contains determinants whose verification status is valid, a triage conclusion is generated based on the corresponding triage level constraint. If none are found, a triage conclusion is generated based on the minimum triage level constraint, and an explanatory evidence chain is constructed and output based on the associated symptom algorithm, determinants, and observation results.

[0014] Secondly, this application provides a knowledge graph-based emergency obstetrics and gynecology triage and reasoning system for implementing the knowledge graph-based emergency obstetrics and gynecology triage and reasoning method described in any of the above claims, including:

[0015] The knowledge graph construction module is used to obtain the chief complaint information of the object to be triaged and construct a symptom algorithm knowledge graph. The symptom algorithm knowledge graph includes the triggering relationship between symptom concepts and symptom algorithms, the inclusion relationship between symptom algorithms and determining factors, and the dependency relationship and influence criteria between determining factors and collection tasks. The determining factors are associated with triage level constraints, and the collection tasks have cost attributes.

[0016] The chief complaint semantic parsing module is used to convert chief complaint information into a set of symptom concepts through semantic mapping, retrieve and determine the corresponding set of symptom algorithms in the symptom algorithm knowledge graph, and initialize the verification status of each determinant in the set as valid, invalid, or unknown.

[0017] The triage factor screening module is used to screen out the relevant triage level that is not lower than the preset security level from the decision factors whose verification status is unknown, and form a set of decision factors to be verified.

[0018] The pre-inspection task optimization module is used to calculate the collection task set that can cover all elements in the set of decision factors to be verified and has the lowest total cost attribute based on the collection tasks dependent on the decision factors to be verified, and output it as the pre-inspection guidance list.

[0019] The status dynamic update module is used to obtain the observation results generated after execution according to the pre-inspection guidance list, compare them with the influence criteria, and update the verification status of each element in the set of decision factors to be verified.

[0020] The triage decision generation module is used to generate a triage conclusion based on the corresponding triage level constraint if there are valid determination factors in the set of determination factors to be verified. If there are no valid determination factors, a triage conclusion is generated based on the minimum triage level constraint. The module also constructs and outputs an explanatory evidence chain based on the associated symptom algorithm, determination factors, and observation results.

[0021] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described knowledge graph-based emergency obstetrics and gynecology triage reasoning method.

[0022] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described knowledge graph-based emergency obstetrics and gynecology triage reasoning method.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] This application constructs a symptom algorithm knowledge graph that includes symptom concepts, symptom algorithms, determinants, data collection tasks, influencing criteria, and triage level constraints. It performs semantic mapping, synonym unification, and affirmative or negative modification parsing on the chief complaint information to generate a set of symptom concepts and a set of symptom algorithms and initializes the verification state of determinants. This solves the problem that natural language chief complaints are difficult to stably fall into executable decision paths and that the lack of state management of the validity, invalidity, and unknown determinants leads to an uncertain starting point for reasoning. It realizes a structured reasoning entry point and reusable decision basis organization for pre-examination triage.

[0025] This application solves the problems of lack of risk priority criteria and easy interference from low-value information in pre-inspection verification, which leads to the omission of high-risk verification points. It sets a preset security level based on the triage level constraint and selects the factors that trigger the triage level to be no lower than the preset security level when the verification is successful from the factors with unknown verification status. This achieves the goal of security level-oriented verification to focus on the priority constraints of the verification process, thereby reducing triage delays and level deviations. Attached Figure Description

[0026] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Wherein:

[0027] Figure 1 This is a flowchart of the knowledge graph-based triage reasoning method for emergency obstetrics and gynecology proposed in this invention;

[0028] Figure 2 This is a structural block diagram of the knowledge graph-based emergency obstetrics and gynecology triage reasoning system proposed in this invention. Detailed Implementation

[0029] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0030] Reference Figure 1 As shown, this application proposes a knowledge graph-based pre-examination and triage reasoning method for emergency obstetrics and gynecology departments, characterized in that the method includes:

[0031] Obtain the chief complaint information of the patients to be triaged and construct a symptom algorithm knowledge graph; the chief complaint information can be text, speech transcription or structured fields;

[0032] The symptom algorithm knowledge graph includes the triggering relationship between symptom concepts and symptom algorithms, the inclusion relationship between symptom algorithms and determining factors, and the dependency relationship and influence criteria between determining factors and collection tasks. The determining factors are associated with triage level constraints, and the collection tasks have cost attributes.

[0033] It should be noted that triggering relationships are used to locate symptom algorithms from symptom concepts; inclusion relationships are used to locate the determinants that need to be verified within the algorithm; dependency relationships are used to deduce executable data collection tasks from the determinants; influence criteria are used to map observation results to whether the determinants are true or false; triage level constraints are used to solidify the influence of the determinants on the triage level into computable constraints; and cost attributes are used for combinatorial optimization when multiple tasks are verifiable.

[0034] The chief complaint information is converted into a set of symptom concepts through semantic mapping. Based on this, the corresponding set of symptom algorithms is retrieved and determined in the symptom algorithm knowledge graph, and the verification status of each determinant is initialized as valid, invalid, or unknown.

[0035] From the decision factors whose verification status is unknown, select the decision factors whose triage level is not lower than the preset security level and form a set of decision factors to be verified;

[0036] For example, a preset security level is used to filter unknown factors that, if verified, will trigger a decision factor that is no lower than the security level. The basis for setting this level is that emergency triage should prioritize covering checkpoints that pose a risk of pushing up the triage level in order to reduce the risk of missing critical cases. In the four-level triage (critical, acute, sub-acute, non-acute), the value range is {acute, sub-acute}, and it is usually set to acute.

[0037] Based on the collection tasks that depend on the decision factors to be verified, calculate the collection task set that can cover all elements in the set of decision factors to be verified and has the lowest total cost attribute, and output it as a pre-inspection guidance list.

[0038] Obtain the observations generated after execution according to the pre-inspection guidance checklist, compare them with the impact criteria to update the verification status of each element in the set of decision factors to be verified;

[0039] If there are determinants in the set of determinants to be verified that are valid, a triage conclusion is generated based on the corresponding triage level constraint. If there are no determinants, a triage conclusion is generated based on the minimum triage level constraint, and an explanatory evidence chain is constructed and output based on the associated symptom algorithm, determinants and observation results.

[0040] Through the above technical solution, this embodiment establishes a computable association between the chief complaint, determinants and collection tasks through a symptom algorithm knowledge graph, drives the screening of verification targets with a preset security level, generates a pre-examination guidance list with minimum cost coverage, and updates the status of determinants in combination with the influence criteria, thereby achieving risk-priority, cost-controllable and interpretable pre-examination triage reasoning under resource constraints.

[0041] In an optional embodiment, constructing the symptom algorithm knowledge graph specifically includes:

[0042] The symptom descriptions in the emergency obstetrics and gynecology pre-examination criteria are broken down into atomic symptom concepts, and logical relationships between symptom concepts are established based on clinical concurrency logic.

[0043] It should be noted that the complex expressions in the pre-screening criteria are broken down into reusable minimum symptom units and connected by logical relationships such as concurrency / accompaniment / exclusion, which facilitates cross-algorithm sharing and retrieval.

[0044] The symptom algorithm is a judgment logic flow triggered by a specific symptom concept, and multiple determining factors and their logical operation rules are preset in the symptom algorithm. The logical operation rules are the influence criteria. Clinical rules are solidified into an executable logic flow, determining factors are key judgment nodes, and influence criteria are the determining factor status judgment rules.

[0045] Each determining factor is configured with a corresponding triage level constraint, which includes four levels: critical, acute, sub-acute, and non-acute.

[0046] The data collection task is defined as an operation instruction to obtain patient vital signs information or medical history details, and each data collection task is labeled with a comprehensive weight including time consumption, manpower consumption, and equipment invasiveness as a cost attribute; among them, operations such as consultation / vital signs / examination are abstracted into tasks, and an optimizable cost metric is formed by the comprehensive weight of time, manpower, and invasiveness.

[0047] Establish dependencies between data collection tasks and decision factors, ensuring that each decision factor points to at least one data collection task that can verify its state; guarantee that each decision factor can be verified by at least one task to avoid inference chain breaks.

[0048] Through the above technical solution, this embodiment constructs an executable knowledge graph by using atomic symptom concepts and symptom algorithm logic flow, and configures triage level constraints for determining factors and cost attributes and dependencies for collection tasks, thereby providing a unified, computable and uninterrupted basic structure for subsequent risk priority screening and cost minimum coverage optimization.

[0049] In an optional embodiment, the chief complaint information is converted into a set of symptom concepts through semantic mapping, specifically including:

[0050] The chief complaint information is processed by phrase segmentation, medical entity recognition and synonym normalization to obtain a set of symptom keywords; the chief complaint is converted from natural language into structured keywords to reduce the difference in expression;

[0051] The symptom keyword set is semantically similar to the symptom concepts in the symptom algorithm knowledge graph, and symptom concepts with similarity higher than a preset similarity threshold are selected to form a symptom concept set.

[0052] For example, a preset similarity threshold is used for semantic similarity matching between symptom keywords and symptom concepts; the setting is based on suppressing synonymous noise and mismatches of non-symptom words while ensuring recall; the value range is 0.75-0.92, and the typical value is 0.85 (taking cosine similarity as an example).

[0053] Based on the symptom concept set, the corresponding symptom algorithm set is retrieved and determined in the symptom algorithm knowledge graph. The determinants contained in the symptom algorithm set are extracted. Based on the affirmative or negative modifiers in the chief complaint information, the verification status of the corresponding determinants is initialized to true or false, and the remaining related determinants are initialized to unknown. In this way, based on the affirmative / negative modifiers, some determinants are directly set to true or false, and the rest are unknown, thereby reducing invalid checks before data collection.

[0054] Through the above technical solution, this embodiment maps the chief complaint to a set of symptom concepts and retrieves the set of symptom algorithms by setting a similarity threshold T_sim. At the same time, it uses affirmative / negative modifiers to perform three-value initialization on the verification status of the determinant, thereby reducing false triggering caused by differences in chief complaint statements and reducing unnecessary subsequent data collection tasks at the inference starting point.

[0055] In an optional embodiment, the set of collection tasks that covers all elements in the set of decision factors to be verified and has the lowest total cost attribute is calculated, specifically including:

[0056] Extract the decision factors that are associated with the highest triage level from the decision factors to be verified and mark them as high-priority verification targets. The purpose of setting high-priority verification targets is to first identify the verification targets that are most likely to lead to the highest level of triage, so as to ensure risk control.

[0057] Based on the dependencies in the symptom algorithm knowledge graph, candidate collection tasks that can cover high-priority verification targets are retrieved, and the intersection operation of candidate collection tasks is performed in combination with the coverage of other decision factors to be verified. First, high-priority targets are covered, and then tasks that can also cover other decision factors are included in the candidates and the combination space is compressed.

[0058] Under the constraint of covering all decision factors to be verified, the branch and bound search algorithm is used to traverse the combination space of candidate collection tasks and calculate the cumulative sum of the cost attributes of collection tasks in different combinations.

[0059] The combination with the smallest cumulative sum is selected as the collection task set. The collection tasks in this set are sorted according to the priority of preset medical operation criteria to generate a structured pre-examination guidance list. The set of minimum costs is transformed into an executable operation sequence.

[0060] For example, the priority of the preset medical operation guidelines is: used to sort the collection task list; the basis for setting is to obtain low-risk, low-cost evidence that can quickly rule out high-risk evidence first; typical guidelines are: consultation before examination, bedside before transport, non-invasive before invasive, fast before time-consuming, low resource consumption before high resource consumption.

[0061] Through the above technical solution, this embodiment obtains the collection task set with the lowest total cost attribute by taking the highest triage level determinant as the high-priority verification target and using branch and bound search under coverage constraints. Then, it outputs a structured list by combining the preset medical operation criteria priority, thereby achieving risk-first, full-coverage, low-cost computable pre-examination guidance under resource constraints.

[0062] In an optional embodiment, the observations generated after execution according to the pre-inspection guidance checklist are obtained and compared with the influence criteria to update the verification status of each element in the set of decision factors to be verified, specifically including:

[0063] Extract the observation results fed back after the execution of the pre-examination guidance checklist, and semantically map them with the influence criteria corresponding to each element in the set of decision factors to be verified. The influence criteria are pre-encapsulated as multivariate conditional expressions containing numerical threshold ranges, clinical category labels and Boolean logic terms. The judgment conditions of each decision factor are standardized into executable expressions, supporting joint judgment of numerical thresholds, category labels and Boolean logic.

[0064] The conditional expressions corresponding to each influencing criterion are called one by one to perform logical judgment on the observation results. When the measured value of the observation result falls into the numerical threshold range, matches the corresponding clinical category label, or satisfies the Boolean logic term, the verification status of the determinant is updated from unknown to established. The establishment update is triggered by the observation result to form a positive evidence closed loop.

[0065] Iterate through all the associated influence criteria within the set of determinants to be verified. If all influence criteria are determined to be inconsistent with the observations and there is no execution bias, update the verification status of the determinant to invalid. Give a negative conclusion when the evidence is sufficient and no valid condition is met, thus avoiding long-term uncertainty.

[0066] Calculate the execution completeness of the collection task set on which the set of decision factors to be verified depends. When it is determined that there are collection tasks that have not returned observation results, keep the verification status of the corresponding decision factor as unknown and extract the missing task identifier. After receiving the subsequent supplementary observation results, trigger the comparison process of the above-mentioned influence criteria again. Explicitly retain the missing results as unknown and support secondary judgment after supplementation.

[0067] It should be noted that the execution of multivariate conditional expressions involves establishing three types of sub-conditions for each influencing criterion: numerical threshold range judgment, category label matching, and Boolean logic term judgment. During execution, the observation result fields are normalized and then judged item by item. If any condition is met, the determining factor is set to true. If all sub-conditions are not met and the data collection process is complete and unbiased, the condition is set to false. If there are missing tasks, they are kept unknown and the missing task identifier is recorded.

[0068] Through the above technical solution, this embodiment encapsulates the influencing criteria into an executable multivariate conditional expression and combines it with the execution completeness mechanism to determine and update the status of the observation results. This enables traceable updates of the determination factor verification status in common pre-inspection scenarios involving missing or supplementary information, reducing fluctuations in triage conclusions caused by missing or misjudgment.

[0069] In an optional embodiment, after generating the triage conclusion, a feedback optimization step is also included:

[0070] After obtaining the triage conclusion, the clinician enters the final diagnosis result, and the triage conclusion is compared with the final diagnosis result for consistency; the triage conclusion is compared with the doctor's final diagnosis to evaluate the correctness of the reasoning and to find rule deviations.

[0071] When the consistency comparison results are inconsistent, extract the verification status and observation results of each determinant in the chain of evidence; use the chain of evidence to locate which determinants / observations drove the erroneous conclusion.

[0072] Based on the discrepancy between the observed results and the final diagnosis, locate the symptom algorithm nodes or influencing criterion parameters that lead to the reasoning error; attribute the error to graph nodes or parameters such as thresholds / labels to form correctable objects;

[0073] The trigger weights of corresponding nodes in the symptom algorithm knowledge graph are adjusted through backpropagation, or the triage level constraints of the determinant association are corrected, and the symptom algorithm knowledge graph is updated accordingly.

[0074] It should be noted that the backpropagation mechanism is used to trigger weight adjustments: the trigger edges and determinant nodes activated in a single inference step are considered as computational graph paths; when the triage conclusion is inconsistent with the final diagnosis, the error is assigned to the trigger weights on the path and updated according to the learning rate, so that the trigger weights strongly correlated with the incorrect conclusion are reduced and the trigger weights correlated with the correct conclusion are increased; at the same time, discrete corrections to the triage level constraints are allowed to eliminate systematic biases.

[0075] Through the above technical solution, this embodiment introduces error attribution driven by final diagnostic feedback and trigger weight / triage level constraint updates, enabling the symptom algorithm knowledge graph to have iterative optimization capabilities, thereby continuously reducing the systematic bias caused by static rule solidification and improving triage consistency.

[0076] In an optional embodiment, constructing and outputting the explanatory chain of evidence specifically includes:

[0077] Using the generated triage conclusion as the root node, the determinant of the state that triggers the conclusion is located by tracing back the triage level constraint, and the data collection task and the observation results obtained on which the determinant depends are associated, thus extracting the complete directed association path from the conclusion to the original observation results; the determinant is deduced from the conclusion to the determinant, and the data collection task and the observation results are associated again to ensure that the interpretation can be traced back to the original evidence from the conclusion.

[0078] Multiple directed association paths extracted are merged and redundancy is removed. The hierarchical relationship between triage conclusions, determinants, data collection tasks and observation results is organized using a directed acyclic graph structure to form an interpretive evidence topology. Duplicate nodes and edges are merged to form an acyclic structure for display and auditing.

[0079] To explain the verification status and node type label of each node in the evidence topology, a structured report is generated by integrating the reasoning path, observation summary, and triage level as the output of the explanation evidence chain; the reasoning link, observation summary, and triage level are solidified into a structured output to facilitate clinical communication and record keeping;

[0080] It should be noted that the explanation of the evidence topology construction is as follows: starting from the triage conclusion, the reverse dependency relationship along the triage level constraint → determining factors → data collection task → observation results is traversed to generate multiple directed paths; the paths are deduplicated and edges are merged to ensure that the topology is a directed acyclic graph; the node type and verification status are added to the topology and a hierarchical structure is output to ensure that the explanation is both readable and auditable.

[0081] Through the above technical solution, this embodiment constructs and outputs an explanatory evidence chain of triage conclusions, determining factors, collection tasks, and observation results by reverse tracing based on triage conclusions and DAG deduplication, thereby achieving structured presentation and auditable traceability of the basis for conclusions and improving clinical interpretability and quality control capabilities.

[0082] See Figure 2 As shown, this solution proposes a knowledge graph-based emergency obstetrics and gynecology triage and reasoning system to implement the aforementioned knowledge graph-based emergency obstetrics and gynecology triage and reasoning method, including:

[0083] The knowledge graph construction module is used to obtain the chief complaint information of the object to be triaged and construct a symptom algorithm knowledge graph. The symptom algorithm knowledge graph includes the triggering relationship between symptom concepts and symptom algorithms, the inclusion relationship between symptom algorithms and determining factors, and the dependency relationship and influence criteria between determining factors and collection tasks. The determining factors are associated with triage level constraints, and the collection tasks have cost attributes.

[0084] The chief complaint semantic parsing module is used to convert chief complaint information into a set of symptom concepts through semantic mapping, retrieve and determine the corresponding set of symptom algorithms in the symptom algorithm knowledge graph, and initialize the verification status of each determinant in the set as valid, invalid, or unknown.

[0085] The triage factor screening module is used to screen out the relevant triage level that is not lower than the preset security level from the decision factors whose verification status is unknown, and form a set of decision factors to be verified.

[0086] The pre-inspection task optimization module is used to calculate the collection task set that can cover all elements in the set of decision factors to be verified and has the lowest total cost attribute based on the collection tasks dependent on the decision factors to be verified, and output it as the pre-inspection guidance list.

[0087] The status dynamic update module is used to obtain the observation results generated after execution according to the pre-inspection guidance list, compare them with the influence criteria, and update the verification status of each element in the set of decision factors to be verified.

[0088] The triage decision generation module is used to generate a triage conclusion based on the corresponding triage level constraint if there are valid determination factors in the set of determination factors to be verified. If there are no valid determination factors, a triage conclusion is generated based on the minimum triage level constraint. The module also constructs and outputs an explanatory evidence chain based on the associated symptom algorithm, determination factors, and observation results.

[0089] In another embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above embodiments.

[0090] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps described above.

[0091] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps described above.

[0092] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A knowledge graph-based reasoning method for triage and diagnosis in emergency obstetrics and gynecology departments, characterized in that, The method includes: Obtain the chief complaint information of patients to be triaged and construct a symptom algorithm knowledge graph; The symptom algorithm knowledge graph includes the triggering relationship between symptom concepts and symptom algorithms, the inclusion relationship between symptom algorithms and determining factors, and the dependency relationship and influence criteria between determining factors and collection tasks. The determining factors are associated with triage level constraints, and the collection tasks have cost attributes. The chief complaint information is converted into a set of symptom concepts through semantic mapping. Based on this, the corresponding set of symptom algorithms is retrieved and determined in the symptom algorithm knowledge graph, and the verification status of each determinant is initialized as valid, invalid, or unknown. From the decision factors whose verification status is unknown, select the decision factors whose triage level is not lower than the preset security level and form a set of decision factors to be verified; Based on the collection tasks that depend on the decision factors to be verified, calculate the collection task set that can cover all elements in the set of decision factors to be verified and has the lowest total cost attribute, and output it as a pre-inspection guidance list. Obtain the observations generated after execution according to the pre-inspection guidance checklist, compare them with the impact criteria to update the verification status of each element in the set of decision factors to be verified; If the set of determinants to be verified contains determinants whose verification status is valid, a triage conclusion is generated based on the corresponding triage level constraint. If none are found, a triage conclusion is generated based on the minimum triage level constraint, and an explanatory evidence chain is constructed and output based on the associated symptom algorithm, determinants, and observation results.

2. The method according to claim 1, characterized in that, The construction of a symptom algorithm knowledge graph specifically includes: The symptom descriptions in the emergency obstetrics and gynecology pre-examination criteria are broken down into atomic symptom concepts, and logical relationships between symptom concepts are established based on clinical concurrency logic. The symptom algorithm is a judgment logic flow triggered by a specific symptom concept, and multiple determining factors and their logical operation rules are preset in the symptom algorithm. The logical operation rules are the influence criteria. Each determining factor is configured with a corresponding triage level constraint, which includes four levels: critical, acute, sub-acute, and non-acute. The collection task is defined as an operation instruction to obtain patient vital signs information or medical history details, and each collection task is labeled with a comprehensive weight including time consumption, manpower consumption and equipment invasiveness as a cost attribute; Establish dependencies between data collection tasks and determinants, ensuring that each determinant points to at least one data collection task that can verify its status.

3. The method according to claim 1, characterized in that, The chief complaint information is converted into a set of symptom concepts through semantic mapping, specifically including: The chief complaint information is processed by phrase segmentation, medical entity recognition and synonym normalization to obtain a set of symptom keywords; The symptom keyword set is semantically similar to the symptom concepts in the symptom algorithm knowledge graph, and symptom concepts with similarity higher than a preset similarity threshold are selected to form a symptom concept set. Based on the symptom concept set, the corresponding symptom algorithm set is retrieved and determined in the symptom algorithm knowledge graph. The determinants contained in the symptom algorithm set are extracted. Based on the affirmative or negative modifiers present in the chief complaint information, the verification status of the corresponding determinant is initialized to true or false, and the remaining related determinants are initialized to unknown.

4. The method according to claim 1, characterized in that, The calculation identifies the collection task set that covers all elements in the set of decision factors to be verified and has the lowest total cost attribute, specifically including: Extract the decision factors with the highest triage level from the set of decision factors to be verified, and mark them as high-priority verification targets; Based on the dependencies in the symptom algorithm knowledge graph, candidate collection tasks that can cover high-priority verification targets are retrieved, and the intersection operation of the candidate collection tasks is performed in combination with the coverage of other decision factors to be verified. Under the constraint of covering all decision factors to be verified, the branch and bound search algorithm is used to traverse the combination space of candidate collection tasks and calculate the cumulative sum of the cost attributes of collection tasks in different combinations. The combination with the smallest cumulative sum is selected as the collection task set. The collection tasks in this set are sorted according to the priority of preset medical operation guidelines to generate a structured pre-examination guidance list.

5. The method according to claim 1, characterized in that, Obtain the observations generated after execution according to the pre-inspection guidance checklist, compare them with the impact criteria to update the verification status of each element in the set of decision factors to be verified, specifically including: Extract the observation results fed back after the execution of the pre-examination guidance list, and semantically map them with the influence criteria corresponding to each element in the set of decision factors to be verified. The influence criteria are pre-encapsulated as multivariate conditional expressions containing numerical threshold ranges, clinical category labels and Boolean logic terms. The conditional expressions corresponding to each influencing criterion are called one by one to perform logical judgment on the observation results. When the measured value of the observation result falls within the numerical threshold range, matches the corresponding clinical category label, or satisfies the Boolean logic term, the verification status of the determinant is updated from unknown to valid. Iterate through all the associated influence criteria in the set of determinants to be verified. If all influence criteria are determined to be inconsistent with the observation results and there is no execution bias, update the verification status of the determinant to invalid. Calculate the execution completeness of the collection task set on which the set of decision factors to be verified depends. When it is determined that there are collection tasks that have not returned observation results, keep the verification status of the corresponding decision factor as unknown and extract the missing task identifier. After receiving the subsequent supplementary observation results, trigger the comparison process of the above-mentioned influence criteria again.

6. The method according to claim 1, characterized in that, After generating the triage conclusion, a feedback optimization step is also included: After obtaining the triage conclusion, the clinician enters the final diagnosis result, and the triage conclusion is compared with the final diagnosis result for consistency. When the consistency comparison results are inconsistent, extract the verification status and observation results of each determinant in the chain of evidence. Based on the discrepancy between the observed results and the final diagnosis, locate the symptom algorithm nodes or influencing criterion parameters that lead to the reasoning error; The trigger weights of corresponding nodes in the symptom algorithm knowledge graph are adjusted through backpropagation, or the triage level constraints associated with determining factors are corrected, and the symptom algorithm knowledge graph is updated accordingly.

7. The method according to claim 1, characterized in that, Constructing and outputting the explanatory chain of evidence specifically includes: Using the generated triage conclusion as the root node, the factors that trigger the establishment of the conclusion are located by tracing the triage level constraints in reverse, and the data collection tasks and observation results obtained on which the factors depend are associated, thus extracting the complete directed association path from the conclusion to the original observation results. Multiple directed association paths extracted are merged and redundancy is removed. The hierarchical relationship between triage conclusions, determinants, data collection tasks and observation results is organized using a directed acyclic graph structure to form an interpretive evidence topology. To explain the verification status and node type label of each node in the evidence topology, a structured report is generated by integrating the reasoning path, observation result summary and triage level, which serves as the output of the explanation evidence chain.

8. A knowledge graph-based emergency obstetrics and gynecology triage and reasoning system, characterized in that, The method for implementing the knowledge graph-based emergency obstetrics and gynecology triage reasoning method as described in any one of claims 1-7 includes: The knowledge graph construction module is used to obtain the chief complaint information of the object to be triaged and construct a symptom algorithm knowledge graph. The symptom algorithm knowledge graph includes the triggering relationship between symptom concepts and symptom algorithms, the inclusion relationship between symptom algorithms and determining factors, and the dependency relationship and influence criteria between determining factors and collection tasks. The determining factors are associated with triage level constraints, and the collection tasks have cost attributes. The chief complaint semantic parsing module is used to convert chief complaint information into a set of symptom concepts through semantic mapping, retrieve and determine the corresponding set of symptom algorithms in the symptom algorithm knowledge graph, and initialize the verification status of each determinant in the set as valid, invalid, or unknown. The triage factor screening module is used to screen out the relevant triage level that is not lower than the preset security level from the decision factors whose verification status is unknown, and form a set of decision factors to be verified. The pre-inspection task optimization module is used to calculate the collection task set that can cover all elements in the set of decision factors to be verified and has the lowest total cost attribute based on the collection tasks dependent on the decision factors to be verified, and output it as the pre-inspection guidance list. The status dynamic update module is used to obtain the observation results generated after execution according to the pre-inspection guidance list, compare them with the influence criteria, and update the verification status of each element in the set of decision factors to be verified. The triage decision generation module is used to generate a triage conclusion based on the corresponding triage level constraint if there are valid determination factors in the set of determination factors to be verified. If there are no valid determination factors, a triage conclusion is generated based on the minimum triage level constraint. The module also constructs and outputs an explanatory evidence chain based on the associated symptom algorithm, determination factors, and observation results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images