Veterinary clinical emergency decision support method and system based on mapping knowledge domain

Abstract

The invention discloses a veterinary clinical first-aid decision support method and system based on a knowledge graph, and relates to the technical field of veterinary clinical first-aid decision support. Dynamic nature and accuracy of veterinary first-aid decision support are remarkably improved by constructing a static and dynamic combined knowledge graph and a two-way parallel reasoning mechanism; individual dynamic evolution of the knowledge graph is achieved, the actual effect feedback of field measures is fused into the graph through a state change weight algorithm, knowledge representation can adapt to the current case in real time, and the adaptability of the model is enhanced. In combination with forward diagnosis reasoning and backward traceability reasoning, treatment suggestions are deduced from symptoms, effectiveness of implemented measures can be evaluated, diagnosis hypotheses can be reversely corrected, a decision closed loop is formed, and reasoning reliability is improved; through conflict resolution and fusion of decision paths, a structured dynamic report is generated, the misdiagnosis risk is effectively reduced, and the disposal process is optimized.

Description

Technical Field

[0001] This application relates to the field of veterinary clinical emergency decision support technology, specifically to a veterinary clinical emergency decision support method and system based on knowledge graphs. Background Technology

[0002] In the field of veterinary emergency care, clinical decision-making relies heavily on the doctor's experience and the integration of real-time information. Traditional decision support systems are mostly based on static medical knowledge bases and cannot effectively integrate on-site temporal signs, performed operations, and their actual effects. Therefore, there is a need for an intelligent decision support method that can integrate multi-source information, realize dynamic knowledge evolution, and support bidirectional reasoning.

[0003] Existing technology, such as the invention application patent with publication number CN1994242A, discloses a veterinary remote automatic diagnosis and treatment system. The system is characterized by comprising an automatic diagnostic system and a safe medication automatic prescription system, which are connected via a network and auxiliary information channels. These auxiliary information channels include a video conferencing system and a telephone line. The automatic diagnostic system includes a color atlas of animal diseases and a database of veterinary experts with extensive clinical experience. With this invention, breeders or pet owners can input disease information of their sick animals into the system. The system automatically calculates a score and provides a diagnosis. Then, based on available medication resources and treatment habits, the breeder can select medication or prescriptions from the automatic prescription system to treat the sick animal.

[0004] Regarding the above-mentioned solutions, the inventors of this application have discovered that the above-mentioned technology has at least the following technical problems:

[0005] 1. Currently, there is a lack of integration of structured authoritative knowledge and unstructured clinical experience to construct an initial knowledge graph. There is a lack of initial weights, and qualitative knowledge is not transformed into a calculable and comparable quantitative form. This fails to provide a crucial data structure and starting point for subsequent dynamic adjustments based on actual effects. Furthermore, there is a lack of dynamic adjustment of the weights of relationships in the initial static knowledge graph based on implemented measures and observed immediate effects. There is no dynamic evolutionary knowledge graph generated for the current case, and real-time adjustments are not possible. This prevents personalization of the knowledge graph, and the reasoning basis does not align with the patient's unique current pathophysiological state.

[0006] 2. Currently, the system lacks a forward reasoning path starting from the current state, and does not perform a breadth-based search based on weights and priorities on a dynamic evolutionary knowledge graph, thus failing to generate suggested paths for diagnosis and treatment. It also lacks a backward reasoning path starting from implemented measures, making it impossible to verify their effectiveness or deduce diagnostic doubts or potential causes. Furthermore, it cannot ensure the system is based on the latest state, cannot comprehensively and systematically deduce all reasonable possibilities, and does not sort by confidence level, failing to assist doctors in broadening their thinking. It lacks the ability to simulate expert "debriefing" thinking, cannot proactively identify abnormal situations where measures are ineffective, and lacks the ability to trigger doubt and correction of the initial diagnosis, thus failing to prevent going further down the wrong path. Summary of the Invention

[0007] To address the aforementioned technical shortcomings, the purpose of this application is to provide a knowledge graph-based veterinary clinical emergency decision support method and system.

[0008] To solve the above-mentioned technical problems, this application adopts the following technical solution: In the first aspect, this application provides a veterinary clinical emergency decision support method based on knowledge graph, which includes the following steps: S1, constructing an initial static knowledge graph in the field of veterinary emergency care and obtaining case information of the current case.

[0009] S2. Based on case information, the initial static knowledge graph is dynamically evolved to generate a dynamically evolved knowledge graph for the current case.

[0010] S3. Based on the dynamic evolutionary knowledge graph and current state snapshot of the current case, the forward reasoning path set and backward reasoning conclusion set are derived through analysis.

[0011] S4. Conflict resolution and fusion are performed on the forward reasoning path set and the backward reasoning conclusion set to generate and output a dynamic decision support report.

[0012] Preferably, the case information includes time-series vital sign data, chief complaint text, and a sequence of implemented procedures.

[0013] Preferably, the construction of the initial static knowledge graph in the field of veterinary emergency care includes: acquiring structured knowledge sources and unstructured medical record texts in the field of veterinary emergency care to obtain multi-source knowledge data in the field of veterinary emergency care.

[0014] Based on the structured knowledge source, core entities are defined as species, diseases, symptoms, vital signs, examinations, drugs, and emergency procedures; and relationships are defined as causation, manifestation, treatment, contraindication, and monitoring relationships, to obtain the entity and relationship architecture of the veterinary emergency care knowledge graph.

[0015] Based on the multi-source knowledge data and entity and relation architecture, an initial static knowledge graph in the field of veterinary emergency care is constructed through knowledge fusion, and initial global confidence weights are assigned to each treatment and relief relation in the initial static knowledge graph.

[0016] Preferably, the step of dynamically evolving the initial static knowledge graph based on case information to generate a dynamically evolved knowledge graph for the current case includes: performing standardization processing and trend feature extraction operations based on the time-series vital sign data to obtain a vital sign feature vector.

[0017] Based on the stated textual complaint, entity recognition and linking operations are performed to map the identified entities to the corresponding entities in the initial static knowledge graph, thus forming an initial symptom entity set.

[0018] Based on the sequence of implemented operational measures, the trend of vital sign changes after the execution of each implemented operational measure is extracted from the time-series vital sign data to obtain the effect trend sequence.

[0019] Based on the effect trend sequence and the initial static knowledge graph, the weight adjustment amount of each implemented operation measure on its expected target state node in the initial static knowledge graph is calculated by the state change weight algorithm to generate a dynamic evolutionary knowledge graph for the current case; the vital sign feature vector and the initial symptom entity set are combined to form a snapshot of the current state of the current case.

[0020] Preferably, the calculation of the weight adjustment of each implemented action on its expected target state node in the initial static knowledge graph to generate a dynamic evolutionary knowledge graph for the current case includes: calculating the weight adjustment using the formula... The result is the The weight adjustment amount of each implemented operational measure. This indicates the number corresponding to the operation that has been performed. , This represents the total number of operations performed. This is represented as the adjustment factor coefficient, which is a preset constant. The function representing the trend of the effect. Indicated as the trend of effect, This represents the initial global confidence weight.

[0021] Through calculation formula The result is the Updated weight values ​​for each implemented operational measure This updates the initial static knowledge graph. The weights of implemented operational measures are assigned to generate a dynamic evolutionary knowledge graph for the current case.

[0022] Preferably, the analysis yields a forward reasoning path set, including: based on the vital sign feature vector in the current state snapshot of the case and the initial symptom entity set, performing a restricted breadth-first search on the dynamic evolution knowledge graph of the current case to obtain a candidate relation edge set.

[0023] Based on the relation types and updated weight values ​​in the candidate relation edge set, the confidence of each inference chain formed by connecting the initial state node to the recommended operation node via the candidate relation edge is calculated to obtain a set of candidate inference paths with confidence.

[0024] The candidate inference path set with confidence is sorted in descending order based on its path confidence, and the top few paths are selected to obtain the forward inference path set, which includes several diagnostic hypotheses, recommended measures and path confidence.

[0025] Preferably, the backward reasoning conclusion set is analyzed as follows: Based on the sequence of implemented operational measures, for each implemented operational measure in the sequence, the treatment relationship starting from the implemented operational measure is traversed backward in the dynamic evolution knowledge graph of the current case to locate the set of target symptom entity nodes and vital sign entity nodes that are expected to be improved.

[0026] Based on the current state snapshot of the case, the actual state data corresponding to each node in the set of target symptom entity nodes and vital sign entity nodes is extracted to obtain the actual state dataset of the target node.

[0027] For each implemented operational measure, the actual state dataset of the corresponding target node is compared with the preset expected state improvement threshold of the measure. The effectiveness evaluation value of the measure is calculated by the measure effectiveness evaluation algorithm to obtain the measure effectiveness evaluation sequence, and then the confidence level of the diagnostic correction hypothesis is obtained. A backward inference conclusion set is obtained, which includes several measure effectiveness evaluations, diagnostic correction hypotheses and confidence levels of diagnostic correction hypotheses.

[0028] Preferably, the step of deriving the confidence level of the diagnostic correction hypothesis includes: based on the effectiveness evaluation sequence of the measures and the dynamic evolutionary knowledge graph of the current case, for implemented operational measures whose effectiveness evaluation values ​​are lower than a preset ineffectiveness threshold, searching in the dynamic evolutionary knowledge graph for alternative disease entity nodes that can simultaneously connect to the initial symptom entity set and explain the ineffectiveness of the measures, to generate a diagnostic correction hypothesis; and calculating the confidence level of the diagnostic correction hypothesis based on the connection strength between the diagnostic correction hypothesis and the relevant entities in the dynamic evolutionary knowledge graph and the degree of ineffectiveness of the measures.

[0029] Preferably, the step of resolving and fusing the conflict between the forward inference path set and the backward inference conclusion set to generate and output a dynamic decision support report includes: performing conflict detection and confidence adjustment operations on the forward inference path set based on the measure effectiveness evaluation sequence in the backward inference conclusion set to obtain the resolved forward inference path set.

[0030] Based on the resolved forward inference path set and the backward inference conclusion set, a path fusion and comprehensive confidence ranking operation is performed to obtain a ranked candidate decision path set.

[0031] Based on the sorted candidate decision path set and the backward reasoning conclusion set, a report synthesis operation is performed according to a preset report template to generate and output a dynamic decision support report.

[0032] In its second aspect, this application provides a system for a knowledge graph-based veterinary clinical emergency decision support method, comprising: preferably, a data construction and acquisition module for constructing an initial static knowledge graph in the field of veterinary emergency care and acquiring case information of the current case.

[0033] The dynamic evolutionary knowledge graph generation module dynamically evolves the initial static knowledge graph based on case information to generate a dynamic evolutionary knowledge graph for the current case.

[0034] The reasoning conclusion set generation module analyzes and derives the forward reasoning path set and the backward reasoning conclusion set based on the dynamic evolutionary knowledge graph of the current case and the current state snapshot.

[0035] The dynamic decision support report output module is used to resolve and merge the conflict between the forward reasoning path set and the backward reasoning conclusion set in order to generate and output a dynamic decision support report.

[0036] The beneficial effects of this application are as follows: 1. The veterinary clinical emergency decision support method and system based on knowledge graph provided in this application significantly improves the dynamism and accuracy of veterinary emergency decision support by constructing a knowledge graph that combines static and dynamic elements and a dual-path parallel reasoning mechanism; it realizes the case-specific dynamic evolution of the knowledge graph, and integrates the actual effect feedback of on-site measures into the graph through a state change weighting algorithm, enabling the knowledge representation to adapt to the current case in real time and enhancing the model's adaptability. Combining forward diagnostic reasoning and backward tracing reasoning, it not only derives treatment suggestions from symptoms, but also evaluates the effectiveness of implemented measures and reverses diagnostic hypotheses, forming a decision-making closed loop and improving the reliability of reasoning; through the conflict resolution and fusion of decision paths, it generates a structured dynamic report, effectively reducing the risk of misdiagnosis and optimizing the treatment process.

[0037] 2. This application integrates structured authoritative knowledge with unstructured clinical experience to construct an initial knowledge graph, which represents knowledge more comprehensively and is closer to clinical practice. The introduction of initial weights transforms qualitative knowledge into a calculable and comparable quantitative form, providing a key data structure and starting point for subsequent dynamic adjustments based on actual results.

[0038] 3. Based on the implemented measures and the observed immediate effects, this application dynamically adjusts the weights of relationships in the initial static knowledge graph using a state change weighting algorithm to generate a dynamic evolutionary knowledge graph for the current case. This real-time adjustment personalizes the knowledge graph and makes the reasoning basis more closely aligned with the patient's current unique pathophysiological state.

[0039] 4. The forward reasoning path in this application starts from the current state and performs a breadth-first search based on weighted priorities on a dynamic evolutionary knowledge graph to generate suggested diagnostic and treatment paths. The backward tracing reasoning path starts from the implemented measures, verifies their effectiveness in reverse, and infers diagnostic doubts or potential causes. Forward reasoning ensures that the system can comprehensively and systematically deduce all reasonable possibilities based on the latest state and sort them by confidence level to help doctors broaden their thinking. Backward reasoning simulates the "debriefing" thinking of experts, which can proactively discover abnormal situations where measures are ineffective and trigger doubts and corrections to the initial diagnosis, effectively dealing with clinical uncertainties and preventing going further down the wrong path. Both reasoning paths rely on dynamically updated graphs and real-time states, making the reasoning conclusions not static predictions but dynamic judgments based on the latest evidence, thus increasing their credibility. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the implementation steps of the method described in this application.

[0042] Figure 2 This is a schematic diagram of the system structure connection of this application.

[0043] Figure 3 This is a graph data example of the entity and relationship architecture of this application. Detailed Implementation

[0044] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] Please see Figure 1 As shown, this application provides a knowledge graph-based veterinary clinical emergency decision support method in the first aspect, including: S1, constructing an initial static knowledge graph in the field of veterinary emergency care and obtaining case information of the current case.

[0046] In one specific instance, the case information includes time-series vital signs data, chief complaint text, and a sequence of implemented procedures.

[0047] It should be noted that the emergency case flow is obtained in real time through the system interface, and the case information of the current case is extracted from the emergency case flow.

[0048] It should be noted that the aforementioned time-series vital sign data refers to a sequence of continuous or periodic physiological parameter measurements from vital sign monitoring equipment that changes over time. This data stream includes dynamic changes in key physiological indicators such as heart rate, blood pressure, and blood oxygen saturation, used to reflect the physiological state of the sick animal (case) and its response to treatment measures in real time.

[0049] It should be noted that the aforementioned text complaint refers to a short natural language text entered by emergency personnel during the admission process that describes the core issues of the case; its content usually includes information such as species, age, and main abnormal manifestations, and is a key qualitative input for quickly understanding the initial situation of the case, used to trigger initial disease or symptom associations.

[0050] It should be noted that the implemented sequence of procedures refers to a list of medical measures that have been performed in the current case emergency response, recorded in chronological order; each item in the implemented sequence of procedures represents a specific emergency response procedure, examination, or medication action; the implemented sequence of procedures records the history of interventions taken and is the basis for assessing the current state, differences from the initial state, and analyzing the effectiveness of the measures.

[0051] In a specific example, the construction of the initial static knowledge graph in the field of veterinary emergency care includes: acquiring structured knowledge sources and unstructured medical record texts in the field of veterinary emergency care to obtain multi-source knowledge data in the field of veterinary emergency care.

[0052] Based on the structured knowledge source, core entities are defined as species, diseases, symptoms, vital signs, examinations, drugs, and emergency procedures; and relationships are defined as causation, manifestation, treatment, contraindication, and monitoring relationships, to obtain the entity and relationship architecture of the veterinary emergency care knowledge graph.

[0053] Based on the multi-source knowledge data and entity and relation architecture, an initial static knowledge graph in the field of veterinary emergency care is constructed through knowledge fusion, and initial global confidence weights are assigned to each treatment and relief relation in the initial static knowledge graph.

[0054] It should be noted that the acquisition of structured knowledge sources and unstructured medical record texts in the field of veterinary emergency care involves accessing pre-stored or online knowledge bases and medical record databases, simultaneously extracting knowledge data in both formats. Structured knowledge sources are knowledge carriers with well-organized and clearly defined structures, such as veterinary emergency care textbooks, clinical operation guidelines, and drug manuals, whose content is organized by chapters, entries, and tables. Unstructured medical record texts are historical medical records recorded in natural language and not arranged in a fixed format, such as past electronic medical record descriptions and consultation records. By simultaneously acquiring structured knowledge sources and unstructured medical record texts in the field of veterinary emergency care, authoritative and standardized knowledge and real case data reflecting the diversity of clinical practice are provided for the construction of the knowledge graph, thereby ensuring that the knowledge graph possesses both theoretical correctness and practical relevance.

[0055] It should be noted that the core entities defined include species, diseases, symptoms, vital signs, examinations, drugs, and emergency procedures; and the relationships defined include causal, manifestation, treatment, contraindication, and monitoring relationships. This refers to the abstraction of objects (i.e., entities) representing key concepts in the field of veterinary emergency care, as well as the semantic connections (i.e., relationships) between these objects, from the structured knowledge source based on its logical structure and semantics, through automated rules. Species entities refer to the animal category receiving emergency care, such as dogs, cats, and horses; disease entities refer to specific pathological states, such as gastric volvulus and toxic shock; symptom entities refer to abnormal phenomena exhibited by the animal, such as vomiting and abdominal distension; vital sign entities refer to quantifiable physiological indicators, such as heart rate and blood pressure; examination entities refer to medical procedures used for diagnosis, such as X-ray examinations and blood biochemistry tests; drug entities refer to drugs used for treatment, such as dopamine and antibiotics; and emergency procedure entities refer to emergency measures taken to stabilize the condition, such as endotracheal intubation and intravenous infusion. The "cause" relation describes how a disease or pathological state may trigger another disease or symptom; the "manifestation" relation describes how a disease or symptom typically manifests as a specific change in vital signs or a combination of symptoms; the "treatment" relation describes the emergency procedures, examinations, or medications that should be taken for a specific disease or symptom; the "contraindication" relation describes specific procedures or medications that should not be taken in certain diseases or states; and the "monitoring" relation describes the subsequent vital signs or examination indicators that need to be closely monitored after performing a specific procedure or using a specific medication. Defining the architecture of each entity and relation provides a standardized data model for the subsequent construction of the knowledge graph.

[0056] It should be noted that the construction of an initial static knowledge graph in the field of veterinary emergency care through knowledge fusion refers to the extraction of information, entity alignment, and relationship linking of the multi-source knowledge data based on the entity and relationship architecture, forming a graph-structured data model composed of nodes (entities) and edges (relationships). This initial static knowledge graph reflects a relatively fixed network of veterinary emergency care professional knowledge summarized and extracted from authoritative knowledge sources, serving as the foundational knowledge base for case-specific reasoning.

[0057] It should be noted that assigning an initial global confidence weight (quantifying the expected strength or effectiveness of the treatment plan in general circumstances, used as the initial basis for prioritizing decisions in subsequent reasoning) to each treatment and relief relationship in the initial static knowledge graph means assigning an initial confidence value to the edge connecting the disease or symptom entity and the emergency operation or drug entity in the knowledge graph. This initial global confidence weight is set based on the authority level of the knowledge source (e.g., guideline recommendation level), the evidence level of clinical research, or expert consensus.

[0058] This application integrates structured authoritative knowledge with unstructured clinical experience to construct an initial knowledge graph, which represents knowledge more comprehensively and is closer to clinical practice. The introduction of initial weights transforms qualitative knowledge into a calculable and comparable quantitative form, providing a key data structure and starting point for subsequent dynamic adjustments based on actual results.

[0059] S2. Based on case information, the initial static knowledge graph is dynamically evolved to generate a dynamically evolved knowledge graph for the current case.

[0060] In a specific instance, the step of dynamically evolving the initial static knowledge graph based on case information to generate a dynamic evolutionary knowledge graph for the current case includes: performing standardization processing and trend feature extraction operations based on the time-series vital sign data to obtain a vital sign feature vector.

[0061] Based on the stated textual complaint, entity recognition and linking operations are performed to map the identified entities to the corresponding entities in the initial static knowledge graph, thus forming an initial symptom entity set.

[0062] Based on the sequence of implemented operational measures, the trend of vital sign changes after the execution of each implemented operational measure is extracted from the time-series vital sign data to obtain the effect trend sequence.

[0063] Based on the effect trend sequence and the initial static knowledge graph, the weight adjustment amount of each implemented operation measure on its expected target state node in the initial static knowledge graph is calculated by the state change weight algorithm to generate a dynamic evolutionary knowledge graph for the current case; the vital sign feature vector and the initial symptom entity set are combined to form a snapshot of the current state of the current case.

[0064] It should be noted that standardization and trend feature extraction are data preprocessing and feature engineering operations performed on the time-series vital sign data. Standardization refers to converting vital sign data (such as heart rate and blood pressure) from different sources and with different dimensions (such as heart rate and blood pressure) into a standard numerical sequence limited to a specific interval (such as [0,1]) using the Z-score standardization method to eliminate the influence of dimensions. Trend feature extraction refers to calculating statistics (such as mean, slope, and variance) within a sliding window on the standardized time series, thereby condensing the original high-dimensional time-series data into a feature vector that can characterize its short-term and long-term change patterns. The vital sign feature vector is a numerical vector, and each dimension represents the core statistical characteristics or change patterns of a certain vital sign or a group of vital signs of the diseased animal within a specific time period after processing. The vital sign feature vector is a quantitative snapshot of the physiological state of the case and serves as a numerical input for assessing the severity of the current state, matching historical patterns, or predicting future changes in subsequent reasoning processes.

[0065] It should be noted that entity recognition and linking operations involve natural language processing of the textual complaint. Entity recognition refers to automatically identifying entity mentions (such as dogs, vomiting, and abdominal pain) related to the veterinary emergency field from the textual complaint using a named entity recognition model or a rule-based extractor. Entity linking refers to mapping the identified entity mentions to standardized entity nodes with unique identifiers in the knowledge graph by querying the entity library in the initial static knowledge graph and performing precise semantic matching. For example, linking "vomiting" to the "vomiting" symptom entity with ID "S002" in the knowledge graph. The initial symptom entity set is a collection of standard entity node identifiers in the knowledge graph. It is a structured list of symptoms extracted from the user-inputted complaint text that can be understood and processed by the knowledge graph, providing a clear initial description of the case state based on knowledge graph semantics for subsequent reasoning.

[0066] It should be noted that extracting the trend of vital sign changes after the implementation of each operational measure from the time-series vital sign data refers to locating the execution time point of each measure in the sequence of implemented operational measures and extracting a segment of time-series vital sign data within a preset observation window after that execution time point. Trend extraction refers to analyzing this data segment to determine the overall direction of change of specific key vital signs (such as blood pressure and blood oxygen related to the expected goals of the measures) within the observation window. Further, this can be achieved by calculating the slope of a linear regression of the data within the window and comparing it with a preset threshold (this involves quantifying and classifying continuous vital sign time-series data). First, linear regression refines complex fluctuations into a slope value representing the overall direction. Then, by introducing a clinical threshold, a binary decision boundary is established: when the slope value is less than the clinical threshold, it is recorded as "no significant clinical change, basically unchanged"; when the slope value is greater than or equal to the clinical threshold, it is "significant clinical change, and it is increasing"; when the slope value is less than or equal to a negative clinical threshold, it is "significant clinical change, and it is decreasing". This process continuously simulates physiological signals, thereby quantifying the trend as "increasing", "decreasing", or "basically unchanged". The effect trend sequence is an ordered list, where each element corresponds to a measure in the sequence of implemented operational measures, recording the actual direction of change of the expected monitored vital signs after the implementation of the measure (it is an objective physiological response record of the case to the historical intervention measures, providing a basis for subsequent quantitative evaluation of the actual effect of the measures).

[0067] It should be noted that combining the current state snapshot representing the current case refers to creating a structured data object containing two core components: the vital sign feature vector and the initial symptom entity set. The state snapshot is a digitized case state descriptor that integrates quantified physiological indicators and structured symptom descriptions. Its role is to serve as a complete and machine-readable representation of the case's state at a specific moment (usually the current or most recent moment), used as input into subsequent inference modules to interact with a dynamically evolving knowledge graph to generate decision recommendations.

[0068] In a specific instance, calculating the weight adjustment of each implemented action on its expected target state node in the initial static knowledge graph to generate a dynamic evolutionary knowledge graph for the current case includes: calculating a formula... The result is the The weight adjustment amount of each implemented operational measure. This indicates the number corresponding to the operation that has been performed. , This represents the total number of operations performed. This is represented as the adjustment factor coefficient, which is a preset constant. The function representing the trend of the effect. Indicated as the trend of effect, This represents the initial global confidence weight.

[0069] Through calculation formula The result is the Updated weight values ​​for each implemented operational measure This updates the initial static knowledge graph. The weights of implemented operational measures are assigned to generate a dynamic evolutionary knowledge graph for the current case.

[0070] It should be noted that the state change weighting algorithm is a core mathematical method for calculating weight adjustments based on the effect trend sequence and the initial static knowledge graph. This algorithm calculates a weight adjustment for each implemented action (belonging to the sequence of implemented actions), which is used to adjust the weights of the relation edges in the knowledge graph pointing from that action to its expected target state node (e.g., the target disease node "dehydration" connected to the action "intravenous infusion" in the knowledge graph via the "treatment" relation).

[0071] Furthermore, the weight adjustment amount for implemented operational measures is a scalar value (the amount by which the global confidence weight of the relevant relation edges in the knowledge graph needs to be increased or decreased based on the actual effect of the measure). The adjustment factor coefficient controls the overall adjustment magnitude to prevent excessively large abrupt changes in weights due to a single case feedback, ensuring the stability of the knowledge graph's evolution. The input to the effect trend function is the first... The output is a symbolic numerical value representing the trend of the effects of each implemented action (e.g., "rising", "falling", "remaining basically unchanged").

[0072] For example, it can be defined as: if the trend is "upward" (indicating that the expected improvement in vital signs has actually improved), then If the trend is "declining" (indicating that the expected improvement in vital signs has actually worsened), then If the trend is "basically unchanged", then This transforms qualitative trend descriptions into quantifiable impact factors that can be used for calculation.

[0073] It should be noted that, based on the aforementioned weight adjustment, the weights of the corresponding treatment or relief relationships in the initial static knowledge graph are updated to generate a dynamic evolutionary knowledge graph for the current case. Updating the weights of the corresponding treatment or relief relationships in the initial static knowledge graph means that for each implemented action, the edge between the entity node of that action and its expected target state node (determined by the "treatment" relationship in the knowledge graph) is found in the initial static knowledge graph. The initial global confidence weight of this edge is then added to (or subtracted from) by the weight adjustment amount to obtain the updated weight value. After performing this operation on the relationship edges corresponding to all actions in the sequence, the weight values ​​of the relevant edges in the initial static knowledge graph are updated, thereby generating a dynamic evolutionary knowledge graph for the current case whose weight parameters have been calibrated to the actual effect of the current case. This dynamic evolutionary knowledge graph is physically a graph data structure, with its node and edge structure consistent with the initial graph, but the weight values ​​of some relationship edges have changed. This reflects the immediate fine-tuning of the confidence in the effectiveness of specific treatment measures based on feedback from the current case, providing a more realistic knowledge base for subsequent accurate reasoning for the current case.

[0074] Based on the implemented measures and the observed immediate effects, this application dynamically adjusts the weights of relationships in the initial static knowledge graph using a state change weighting algorithm to generate a dynamic evolutionary knowledge graph for the current case. This real-time adjustment personalizes the knowledge graph, making the reasoning basis more closely aligned with the patient's current unique pathophysiological state.

[0075] S3. Based on the dynamic evolutionary knowledge graph and current state snapshot of the current case, the forward reasoning path set and backward reasoning conclusion set are derived through analysis.

[0076] In a specific instance, the analysis yields a set of forward reasoning paths, including: based on the vital sign feature vectors in the current state snapshot of the case and the initial symptom entity set, performing a restricted breadth-first search on the dynamic evolutionary knowledge graph of the current case to obtain a set of candidate relation edges.

[0077] Based on the relation types and updated weight values ​​in the candidate relation edge set, the confidence of each inference chain formed by connecting the initial state node to the recommended operation node via the candidate relation edge is calculated to obtain a set of candidate inference paths with confidence.

[0078] The candidate inference path set with confidence is sorted in descending order based on its path confidence, and the top few paths are selected to obtain the forward inference path set, which includes several diagnostic hypotheses, recommended measures and path confidence.

[0079] It should be noted that the restricted breadth-first search uses the vital sign feature vector and entities in the initial symptom entity set as the starting nodes for the search, traversing the graph structure of the dynamic evolutionary knowledge graph of the current case. The search follows a pre-defined logical chain rule of "symptom entity, possible disease entity, recommended examination or treatment entity" for path expansion. That is, when the search reaches a symptom node, the next step only expands the "manifestation" relation edges pointing to disease entities; when the search reaches a disease node, the next step only expands the "treatment" relation edges pointing to examination, medication, or emergency operation entities. The "restriction" of the search is reflected in the search depth and branch selection. The depth is typically limited to 2 to 3 steps to meet the timeliness requirements of emergency decision-making, and the branch selection is based on priority ranking according to the updated weight values ​​on the relation edges.

[0080] It's important to note that the updated weight values ​​are used as part of the transfer cost in prioritizing search paths. In the breadth-first search queue, when multiple logically linked edges are extended from the current node, the weight value of each edge is converted into a cost value. A lower weight value indicates a higher likelihood that the treatment plan is evaluated as having poor short-term efficacy or being contraindicated in the current case's dynamic evolutionary knowledge graph; therefore, the priority of continuing the search along that edge is lower. Specifically, each edge from node one to node two can be assigned a selection priority score, where the selection priority score is the updated weight value of that edge. When expanding nodes, the search is prioritized to continue from the node pointed to by the edge with the higher priority score.

[0081] It should be noted that the candidate relation edge set is the set of all relation edges that conform to the "symptom, disease, treatment" logic and can be extended from the current search frontier node under the constraints of the current search depth and logical chain. Each edge in this set represents a possible pathological inference or treatment suggestion direction starting from the current state of the case, and is the basic unit that constitutes a complete reasoning path.

[0082] It should be noted that the path confidence calculation algorithm is used to quantify the reliability of each inference path from the initial symptom / sign node to the final recommended operation node. The confidence of a path is determined by the weights of all relation edges traversed by that path. Furthermore, the calculation formula of the path confidence calculation algorithm... derive path confidence Its value is between 0 and 1; Represented as the first on the path The updated weight values ​​of the edges. This represents the index of the relation edge traversed by the path. This represents the total number of relation edges traversed by the path. This represents a series of multiplication operations. Represented as the path length penalty factor, it is the path length. (i.e., number of sides) The path confidence calculation algorithm treats a path as a chain reaction of related events, where the overall confidence is equal to the product of the confidence of each link. The path length penalty factor is introduced to prevent excessively long paths, even with acceptable weights for each link, from achieving excessively high confidence due to accumulated uncertainty. Typically, the path length penalty factor is set to... form.

[0083] It should be noted that the set of candidate inference paths with confidence scores includes all complete inference chains discovered during the constrained search process and assigned a confidence score by the path confidence calculation algorithm. Each path is a structured data object that contains at least: a list of starting nodes (initial symptoms or vital signs), a sequence of intermediate nodes (such as disease entities), a termination node (recommended action entity), and the calculated path confidence score.

[0084] In a specific instance, the backward reasoning conclusion set is analyzed as follows: Based on the sequence of implemented operational measures, for each implemented operational measure in the sequence, the treatment relationship starting from the implemented operational measure is traversed backward in the dynamic evolutionary knowledge graph of the current case to locate the set of target symptom entity nodes and vital sign entity nodes that are expected to be improved.

[0085] Based on the current state snapshot of the case, the actual state data corresponding to each node in the set of target symptom entity nodes and vital sign entity nodes is extracted to obtain the actual state dataset of the target node.

[0086] For each implemented operational measure, the actual state dataset of the corresponding target node is compared with the preset expected state improvement threshold of the measure. The effectiveness evaluation value of the measure is calculated by the measure effectiveness evaluation algorithm to obtain the measure effectiveness evaluation sequence, and then the confidence level of the diagnostic correction hypothesis is obtained. A backward inference conclusion set is obtained, which includes several measure effectiveness evaluations, diagnostic correction hypotheses and confidence levels of diagnostic correction hypotheses.

[0087] It should be noted that reverse traversal of treatment relationships starting from the implemented action measures refers to, in the dynamic evolutionary knowledge graph of the current case, starting from a certain implemented action measure entity node, searching backwards along the edges connected to that node with the relationship type of treatment. This process aims to locate the expected treatment target of the action measure in the knowledge graph, that is, to find which disease or symptom nodes are "treated by this measure".

[0088] It should be noted that the set of target symptom entity nodes and vital sign entity nodes comprises all disease and symptom entity nodes found through the aforementioned reverse traversal operation. For disease nodes, specific vital sign nodes may be further associated with "manifestations as" relationships. This set clarifies the specific clinical indicators that need to be considered when evaluating the effectiveness of this operational measure.

[0089] It should be noted that the actual state dataset of the target node is the quantified or descriptive state value corresponding to the target node extracted from the current state snapshot. For vital sign entity nodes, their actual state data is the numerical value of the corresponding dimension in the vital sign feature vector or its feature statistics. For symptom entity nodes, their actual state data may be the existence identifier from the initial symptom entity set, or a scalar value converted from the severity description extracted from the latest medical record text. This dataset serves as the factual basis for evaluating whether the measures have produced the expected results.

[0090] It should be noted that the measure effectiveness evaluation algorithm calculates a quantitative effectiveness evaluation value by comparing the actual state of the target node after the implementation of the measure with the expected state. The calculation formula for the measure effectiveness evaluation algorithm is as follows: Determine the effectiveness assessment value of the measures. It is a dimensionless scalar; This represents the corresponding number of the target node (symptom or vital sign entity) that needs to be evaluated in the current case dynamic evolution knowledge graph. , This represents the total number of target nodes (symptom or vital sign entities) that need to be evaluated in the current case dynamic evolution knowledge graph corresponding to this measure; Indicates the first The actual state quantification value of each target node in the current state snapshot of the case has been standardized and is comparable to the expected threshold; Indicates that for the first For each target node, the pre-set threshold or target value for expected state improvement after implementing this measure is also a standardized value.

[0091] Furthermore, the algorithm for evaluating the effectiveness of a measure calculates the average relative improvement rate of a medical measure on its multiple expected treatment goals (defined by the "treatment" relationships in a knowledge graph). For a single goal, the ratio... The ratio of the actual observed state change to the expected change was calculated. If the actual improvement meets or exceeds the expectation (e.g., the actual increase in blood pressure is equal to or greater than the expected threshold), the ratio is greater than or equal to 1; if the improvement does not meet the expectation, the ratio is less than 1 but greater than 0; if the state worsens after the measures are implemented (the actual value changes in a worse direction), the actual state quantification value may be negative or much smaller than the baseline value, causing the ratio to become a very small positive number, zero, or negative number.

[0092] It should be noted that the measure effectiveness assessment sequence is an ordered list, with each element corresponding one-to-one with the implemented operational measure sequence. Each element stores the effectiveness assessment value of the corresponding measure. This sequence systematically records the current case's actual response to all implemented measures and is a key input for backward diagnostic tracing.

[0093] It should be noted that the backward reasoning conclusion set includes diagnostic correction hypotheses generated for one or more ineffective measures and their confidence levels, as well as the effectiveness assessment values ​​for each implemented operational measure. Each conclusion is a structured inference that identifies the problems with the interventions taken and proposes new possibilities, providing clinicians with decision support for "review" and "correction".

[0094] In a specific instance, the process of deriving the confidence level of the diagnostic correction hypothesis includes: based on the effectiveness assessment sequence of the measures and the dynamic evolutionary knowledge graph of the current case, for implemented operational measures whose effectiveness assessment values ​​are lower than a preset ineffectiveness threshold, searching in the dynamic evolutionary knowledge graph for alternative disease entity nodes that can simultaneously connect to the initial symptom entity set and explain the ineffectiveness of the measures, to generate a diagnostic correction hypothesis; and calculating the confidence level of the diagnostic correction hypothesis based on the connection strength between the diagnostic correction hypothesis and the relevant entities in the dynamic evolutionary knowledge graph and the degree of ineffectiveness of the measures.

[0095] It should be noted that searching for alternative disease entity nodes in the dynamic evolutionary knowledge graph that can simultaneously connect to the initial symptom entity set and explain the ineffectiveness of the measure is a graph-based hypothesis generation process. Specifically, taking a measure evaluated as ineffective (E below the ineffectiveness threshold) as an example, the dynamic evolutionary knowledge graph searches for disease entity nodes D_alt that meet the following conditions: (1) the disease entity node is connected to most of the initial symptom entities through the "behaves as" relationship; (2) the weight value of the "treatment" relationship between the disease entity node and the ineffective measure is low, or the disease entity node is associated with the measure through other relationships (such as "contraindication"), thereby explaining from a knowledge level why the measure is ineffective for the disease entity node. The disease entity node is a potential diagnostic correction hypothesis that was not initially considered.

[0096] It should be noted that the confidence level of the diagnostic correction hypothesis is calculated from two parts: first, the average connection weight between the surrogate disease node and the initial symptom entity set (the ability to explain the original symptoms); and second, the effectiveness assessment value of ineffective measures (reflecting the severity of the original diagnostic problem). Further, this is achieved through the calculation formula... Determine the confidence level of the diagnostic correction hypothesis. ,in It is the absolute value of the effectiveness assessment value (a negative effectiveness assessment value indicates deterioration and a more serious problem). It is the average weight of the "manifestation" relation edges from disease entity nodes to the initial symptom set. The higher the confidence level of the diagnostic correction hypothesis, the better the correction hypothesis explains the initial symptoms and the higher the agreement with observed intervention failures.

[0097] The forward reasoning path in this application starts from the current state and performs a breadth-first search based on weighted priorities on a dynamically evolving knowledge graph to generate suggested diagnostic and treatment paths. The backward tracing reasoning path starts from the implemented measures, verifies their effectiveness in reverse, and infers diagnostic doubts or potential causes. Forward reasoning ensures that the system can comprehensively and systematically deduce all reasonable possibilities based on the latest state and sort them by confidence level to help doctors broaden their thinking. Backward reasoning simulates the "debriefing" thinking of experts, which can proactively discover abnormal situations where measures are ineffective and trigger doubts and corrections to the initial diagnosis, effectively dealing with clinical uncertainties and preventing going further down the wrong path. Both reasoning paths rely on dynamically updated graphs and real-time states, making the reasoning conclusions not static predictions but dynamic judgments based on the latest evidence, thus increasing their credibility.

[0098] S4. Conflict resolution and fusion are performed on the forward reasoning path set and the backward reasoning conclusion set to generate and output a dynamic decision support report.

[0099] In a specific instance, the conflict resolution and fusion of the forward inference path set and the backward inference conclusion set to generate and output a dynamic decision support report includes: performing conflict detection and confidence adjustment operations on the forward inference path set based on the measure effectiveness evaluation sequence in the backward inference conclusion set to obtain the resolved forward inference path set.

[0100] Based on the resolved forward inference path set and the backward inference conclusion set, a path fusion and comprehensive confidence ranking operation is performed to obtain a ranked candidate decision path set.

[0101] Based on the sorted candidate decision path set and the backward reasoning conclusion set, a report synthesis operation is performed according to a preset report template to generate and output a dynamic decision support report.

[0102] It should be noted that the conflict detection and confidence adjustment operation includes two parallel data processing sub-processes: measure conflict resolution and diagnostic confidence correction. Measure conflict resolution refers to traversing the measure effectiveness evaluation sequence in the backward reasoning conclusion set and identifying measures evaluated as "ineffective" (i.e., the effectiveness evaluation value is lower than a preset ineffectiveness threshold). In the forward reasoning path set, the paths in the recommended measure list that contain the above-mentioned "ineffective" measures are found, which are the paths with logical conflicts. For each conflicting path, its path confidence needs to be significantly reduced.

[0103] Furthermore, the significant reduction in path confidence is achieved through a conflict penalty calculation, the specific calculation process of which is as follows: in, Represented as the first The original path confidence of the forward inference path. This is represented by the number corresponding to the forward reasoning path. This represents the path confidence after conflict penalty adjustment. It is represented as the conflict penalty coefficient, which is a preset constant between 0 and 1, used to control the magnitude of the confidence reduction; It is an indicator function, when the first A path has a value of 1 if it contains at least one recommended measure that is deemed "invalid" in the backward evaluation, and 0 otherwise. In the conflict penalty algorithm, when a measure recommended by a forward inference path is confirmed as invalid by the backward evaluation, the overall credibility of that path should be penalized by a deduction, the severity of which is determined by a preset conflict penalty coefficient.

[0104] It should be noted that diagnostic confidence correction refers to adopting the diagnostic correction hypothesis and its confidence level proposed in the backward reasoning conclusion set to enhance or weaken the local confidence level of the corresponding diagnostic node in the resolved forward reasoning path set. Specifically, in the resolved forward reasoning path set, diagnostic entity nodes related to the diagnostic correction hypothesis are located; if the correction hypothesis proposed by backward reasoning points to a different disease entity (i.e., an alternative diagnosis), and this alternative diagnosis has a competitive or mutually exclusive relationship with the original diagnosis in the knowledge graph, then the local confidence contribution of the original diagnostic node in this path is reduced proportionally according to the confidence level of the correction hypothesis; if the correction hypothesis is a further confirmation or refinement of the original diagnosis, then its local confidence contribution is increased proportionally.

[0105] It should be noted that the resolved forward inference path set is a structured set of paths, in which the overall confidence of each path and the local confidence of the diagnostic nodes within the path have been dynamically adjusted based on backward feedback information. This set reflects a revised and more reliable inference result on the future treatment plan for the current case after incorporating historical evaluations of the effectiveness of measures.

[0106] It should be noted that the path fusion and comprehensive confidence ranking operation first uses the resolved forward inference path set as the report backbone, and embeds the effectiveness assessment conclusions of each implemented or related measure from the backward inference conclusion set as "clinical notes" information into the corresponding measure node of the corresponding path. Then, a comprehensive confidence score is calculated for each path.

[0107] Furthermore, the calculation of the overall confidence score integrates the adjusted path confidence score, the corrected confidence scores of diagnostic nodes within the path, and the effectiveness assessment information of the measure nodes. Specifically, the overall confidence score of a path is a weighted fusion of the path confidence score adjusted after conflict resolution, the average of the corrected local confidence scores of all diagnostic nodes, and the average of the backward effectiveness assessment values ​​(if any) corresponding to all measure nodes (for measures without backward assessment, this can take a default value (e.g., 1 or 0.5, depending on the specific design)).

[0108] It should be noted that the sorted candidate decision path set is an ordered list obtained by arranging the resolved forward inference path set in descending order according to the calculated comprehensive confidence level; the top path of this list represents the most reliable and priority sequence of diagnosis and treatment plans recommended by the system for the current case based on the fusion analysis of forward inference and backward verification.

[0109] It should be noted that the report synthesis operation is based on the sorted candidate decision path set and the backward reasoning conclusion set, and automatically assembles the final report content according to a preset structured template. This report template specifies the organizational logic of the content, requiring the report to include at least: a primary diagnosis and a list of alternative diagnoses, with annotations on their reasoning sources (e.g., forward reasoning or backward correction); a high-priority next step in the treatment, clearly indicating its logical connection to already implemented measures (e.g., "following a certain measure" and "replacing an ineffective measure"); a systematic evaluation and summary of the effectiveness of the implemented measures; and risk warnings (e.g., drug contraindications and operational risks) derived from the dynamic evolutionary knowledge graph analysis and related to the current primary diagnosis or recommended measures.

[0110] It should be noted that the generated and output dynamic decision support report is a structured decision support document for clinicians. This report integrates the wisdom of forward prediction, backward validation, and conflict resolution, providing not only forward-looking action guidelines but also retrospective analysis and risk warnings based on actual treatment feedback, thus achieving closed-loop support and dynamic optimization of the emergency decision-making process.

[0111] Please see Figure 2 As shown, in its second aspect, this application provides a system for a knowledge graph-based decision support method for veterinary clinical emergency care.

[0112] The system 100 of the veterinary clinical emergency decision support method based on knowledge graphs described in this invention can be installed in an electronic device. Depending on the functions implemented, the system 100 may include a data construction and acquisition module 101, a dynamic evolutionary knowledge graph generation module 102, a reasoning conclusion set generation module 103, and a dynamic decision support report output module 104. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0113] In this embodiment, the functions of each module / unit are as follows: The data construction and acquisition module is used to construct an initial static knowledge graph in the field of veterinary emergency care and acquire case information of the current case.

[0114] The dynamic evolutionary knowledge graph generation module dynamically evolves the initial static knowledge graph based on case information to generate a dynamic evolutionary knowledge graph for the current case.

[0115] The reasoning conclusion set generation module analyzes and derives the forward reasoning path set and the backward reasoning conclusion set based on the dynamic evolutionary knowledge graph of the current case and the current state snapshot.

[0116] The dynamic decision support report output module is used to resolve and merge the conflict between the forward reasoning path set and the backward reasoning conclusion set in order to generate and output a dynamic decision support report.

[0117] The knowledge graph-based veterinary emergency decision support method and system provided in this application significantly improves the dynamism and accuracy of veterinary emergency decision support by constructing a static-dynamic combined knowledge graph and a dual-path parallel reasoning mechanism. It achieves case-specific dynamic evolution of the knowledge graph, integrating the actual effects of on-site measures into the graph through a state change weighting algorithm, enabling the knowledge representation to adapt to the current case in real time and enhancing the model's adaptability. Combining forward diagnostic reasoning and backward causal reasoning, it not only derives treatment suggestions from symptoms but also assesses the effectiveness of implemented measures and corrects diagnostic hypotheses in reverse, forming a decision-making closed loop and improving the reliability of reasoning. Through conflict resolution and fusion of decision paths, it generates structured dynamic reports, effectively reducing the risk of misdiagnosis and optimizing the treatment process.

[0118] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0119] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0121] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0122] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A knowledge graph-based veterinary clinical emergency decision support method, characterized in that, include: S1. Construct an initial static knowledge graph in the field of veterinary emergency care and obtain case information for the current case; S2. Based on case information, the initial static knowledge graph is dynamically evolved to generate a dynamic evolutionary knowledge graph for the current case; S3. Based on the dynamic evolutionary knowledge graph and current state snapshot of the current case, analyze and derive the forward reasoning path set and the backward reasoning conclusion set; S4. Conflict resolution and fusion are performed on the forward reasoning path set and the backward reasoning conclusion set to generate and output a dynamic decision support report.

2. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 1, characterized in that, The case information includes time-series vital signs data, chief complaint text, and a sequence of operational measures implemented.

3. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 1, characterized in that, The construction of the initial static knowledge graph in the field of veterinary emergency care includes: To obtain structured knowledge sources and unstructured medical record texts in the field of veterinary emergency care, so as to obtain multi-source knowledge data in the field of veterinary emergency care; Based on the structured knowledge source, core entities are defined as species, diseases, symptoms, vital signs, examinations, drugs, and emergency procedures; and relationships are defined as causation, manifestation, treatment, contraindication, and monitoring relationships, to obtain the entity and relationship architecture of the veterinary emergency care knowledge graph. Based on the multi-source knowledge data and entity and relation architecture, an initial static knowledge graph in the field of veterinary emergency care is constructed through knowledge fusion, and initial global confidence weights are assigned to each treatment and relief relation in the initial static knowledge graph.

4. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 1, characterized in that, The step of dynamically evolving the initial static knowledge graph based on case information to generate a dynamically evolved knowledge graph for the current case includes: Based on the aforementioned time-series vital sign data, standardization processing and trend feature extraction operations are performed to obtain vital sign feature vectors; Based on the textual complaint, entity recognition and linking operations are performed to map the identified entities to the corresponding entities in the initial static knowledge graph, thus forming an initial symptom entity set; Based on the sequence of implemented operational measures, the trend of vital sign changes after the execution of each implemented operational measure is extracted from the time-series vital sign data to obtain the effect trend sequence. Based on the effect trend sequence and the initial static knowledge graph, the weight adjustment amount of each implemented operation measure on its expected target state node in the initial static knowledge graph is calculated by the state change weight algorithm to generate a dynamic evolutionary knowledge graph for the current case; the vital sign feature vector and the initial symptom entity set are combined to form a snapshot of the current state of the current case.

5. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 4, characterized in that, The calculation of the weight adjustment of each implemented action on its expected target state node in the initial static knowledge graph to generate a dynamic evolutionary knowledge graph for the current case includes: Through calculation formula The result is the The weight adjustment amount of each implemented operational measure. This indicates the number corresponding to the operation that has been performed. , This represents the total number of operations performed. This is represented as the adjustment factor coefficient, which is a preset constant. The function representing the trend of the effect. Indicated as the trend of effect, This represents the initial global confidence weights; Through calculation formula The result is the Updated weight values ​​for each implemented operational measure This updates the initial static knowledge graph. The weights of implemented operational measures are assigned to generate a dynamic evolutionary knowledge graph for the current case.

6. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 1, characterized in that, The analysis yields a set of forward reasoning paths, including: Based on the vital sign feature vectors in the current state snapshot of the case and the initial symptom entity set, a restricted breadth-first search is performed on the dynamic evolution knowledge graph of the current case to obtain a set of candidate relation edges. Based on the relation types and updated weight values ​​in the candidate relation edge set, the confidence of each inference chain formed by connecting the initial state node to the recommendation operation node via the candidate relation edge is calculated to obtain a set of candidate inference paths with confidence. The candidate inference path set with confidence is sorted in descending order based on its path confidence, and the top few paths are selected to obtain the forward inference path set, which includes several diagnostic hypotheses, recommended measures and path confidence.

7. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 1, characterized in that, The specific analysis process for the backward reasoning conclusion set is as follows: Based on the sequence of implemented operational measures, for each implemented operational measure in the sequence, the treatment relationships starting from the implemented operational measures are traversed backward in the dynamic evolutionary knowledge graph of the current case to locate the set of target symptom entity nodes and vital sign entity nodes that are expected to be improved. Based on the current state snapshot of the case, extract the actual state data corresponding to each node in the set of target symptom entity nodes and vital signs entity nodes to obtain the actual state dataset of the target node. For each implemented operational measure, the actual state dataset of the corresponding target node is compared with the preset expected state improvement threshold of the measure. The effectiveness evaluation value of the measure is calculated by the measure effectiveness evaluation algorithm to obtain the measure effectiveness evaluation sequence, and then the confidence level of the diagnostic correction hypothesis is obtained. The backward inference conclusion set is then derived, which includes several measures effectiveness assessments, diagnostic correction hypotheses, and confidence levels of the diagnostic correction hypotheses.

8. The veterinary clinical emergency decision support method based on knowledge graphs according to claim 7, characterized in that, The confidence level for deriving the diagnostic correction hypothesis includes: Based on the effectiveness evaluation sequence of the measures and the dynamic evolutionary knowledge graph of the current case, for implemented operational measures whose effectiveness evaluation values ​​are lower than a preset ineffectiveness threshold, alternative disease entity nodes that can simultaneously connect to the initial symptom entity set and explain the ineffectiveness of the measures are searched in the dynamic evolutionary knowledge graph to generate diagnostic correction hypotheses; based on the connection strength between the diagnostic correction hypothesis and the relevant entities in the dynamic evolutionary knowledge graph and the degree of ineffectiveness of the measures, the confidence level of the diagnostic correction hypothesis is calculated.

9. The veterinary clinical emergency decision support method based on knowledge graph according to claim 1, characterized in that, The process of resolving and fusing conflicts between the forward inference path set and the backward inference conclusion set to generate and output a dynamic decision support report includes: Based on the measure effectiveness evaluation sequence in the backward reasoning conclusion set, conflict detection and confidence adjustment operations are performed on the forward reasoning path set to obtain the resolved forward reasoning path set. Based on the resolved forward inference path set and the backward inference conclusion set, a path fusion and comprehensive confidence ranking operation is performed to obtain a ranked candidate decision path set. Based on the sorted candidate decision path set and the backward reasoning conclusion set, a report synthesis operation is performed according to a preset report template to generate and output a dynamic decision support report.

10. A system for implementing the knowledge graph-based veterinary clinical emergency decision support method according to any one of claims 1-9, characterized in that, include: The data construction and acquisition module is used to build an initial static knowledge graph in the field of veterinary emergency care and to acquire case information of the current case. The dynamic evolutionary knowledge graph generation module dynamically evolves the initial static knowledge graph based on case information to generate a dynamic evolutionary knowledge graph for the current case. The reasoning conclusion set generation module analyzes and derives the forward reasoning path set and the backward reasoning conclusion set based on the dynamic evolution knowledge graph and current state snapshot of the current case. The dynamic decision support report output module is used to resolve and merge the conflict between the forward reasoning path set and the backward reasoning conclusion set in order to generate and output a dynamic decision support report.

Images