Large-scale model hospital infection determination method and system based on intelligent agent

Abstract

The present application discloses a large-model hospital infection determination method and system based on an intelligent agent, which relates to the field of risk assessment technology, including: obtaining target physiological characteristics of a target patient; the target physiological characteristics include at least one of the target patient's course of disease description, examination results and test results; searching a hospital infection knowledge graph according to the target physiological characteristics to obtain a corresponding target sub-graph; wherein the hospital infection knowledge graph is constructed according to a hospital infection diagnosis standard document; generating a relationship vector set according to the target sub-graph to obtain a target relationship vector set; inputting the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model to obtain a target infection outcome, thereby reducing the difficulty of data set screening of the hospital infection prediction model and improving the prediction accuracy of hospital infection prediction.

Description

Technical Field

[0001] The present application relates to the technical field of risk assessment, and in particular to a large-scale model hospital infection determination method and system based on an intelligent agent. Background Art

[0002] Nosocomial infections (HAs) are infections acquired by hospitalized patients within a hospital setting, including those that occur during hospitalization and after discharge. However, they do not include infections that began before admission or were already latent at the time of admission. Monitoring patients for HAIs not only helps ensure timely and effective treatment and protect patient safety, but also facilitates early identification of infection sources and control of infection trends.

[0003] Methods for monitoring whether patients have hospital infections are already available in the prior art. For example, patent application number CN201910347848.4 describes an intelligent diagnosis method and system for hospital infections based on multi-model fusion. This method uses historical patient case data as a training dataset for multiple base models. After training, the recognition performance of each base model is evaluated. At least two base models with the best performance are fused using a stacking method to obtain a diagnostic model for diagnosing whether a patient has a hospital infection. This method not only monitors whether a patient has a hospital infection but also improves the accuracy of hospital infection identification using an ensemble learning method. However, this method essentially uses the data distribution characteristics in the training dataset to simulate the data distribution characteristics in the real world. The diagnostic experience learned by the diagnostic model is the diagnostic experience implicit in the data distribution characteristics in the training dataset. When the data distribution characteristics in the training dataset contain misleading information or bias, the recognition accuracy of the diagnostic model decreases as the proportion of misleading information or bias contained in the training dataset increases. This leads to problems with this method, such as difficulty in dataset selection during the model training phase and low diagnostic accuracy. Summary of the Invention

[0004] The purpose of the present invention is to solve the technical problem of how to reduce the difficulty of selecting training data sets for hospital infection diagnosis models and improve the diagnostic accuracy, and provide a large-model hospital infection determination method and system based on intelligent agents.

[0005] According to a first aspect of the present invention, the present invention claims protection for an agent-based large-model hospital infection determination method, comprising:

[0006] Acquiring target physiological characteristics of a target patient; the target physiological characteristics include at least one of a description of the course of disease, an examination result, and a test result of the target patient;

[0007] Retrieving a hospital infection knowledge graph based on the target physiological characteristics to obtain a corresponding target subgraph; wherein the hospital infection knowledge graph is constructed based on a hospital infection diagnosis standard document;

[0008] Generate a relationship vector set according to the target subgraph to obtain a target relationship vector set;

[0009] The target physiological characteristics and the target relationship vector set are input into a pre-trained hospital infection prediction model to obtain a target infection outcome.

[0010] In one embodiment of the present application, the hospital infection knowledge graph includes entities and relationships; the entity types of the entities include symptoms, test results, examination results and diagnosis results of each infection site; the relationships include the relationship between different symptoms, the relationship between different test results, the relationship between different examination results, the relationship between symptoms and test results, the relationship between symptoms and examination results, the relationship between test results and examination results, the relationship between symptoms and diagnosis results, the relationship between examination results and diagnosis results, and the relationship between test results and diagnosis results.

[0011] In one embodiment of the present application, the entity type also includes the diagnostic criteria for each infection site; the relationship between the symptoms and the diagnostic results is obtained based on the relationship between the corresponding symptoms and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the examination results and the diagnostic results is obtained based on the relationship between the corresponding examination results and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the test results and the diagnostic results is obtained based on the relationship between the corresponding test results and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results.

[0012] In one embodiment of the present application, the relationship vector set includes a premise entity, a conclusion entity and a weight coefficient, and the weight coefficient is used to evaluate the credibility of the patient's physiological characteristics belonging to the conclusion entity based on the patient's physiological characteristics belonging to the premise entity.

[0013] In one embodiment of the present application, the weight coefficient is obtained based on the proportion of positive examples; wherein, the proportion of positive examples is obtained based on the proportion of historical patients belonging to the conclusion entity among historical patients belonging to the premise entity.

[0014] In one embodiment of the present application, the relationship type of the relationship includes a subordinate relationship and an associated relationship. When the relationship type of the relationship is the subordinate relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio. When the relationship type of the relationship is the associated relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio and the negative example ratio, wherein the negative example ratio is obtained according to the ratio of historical patients belonging to the conclusion entity among the historical patients who do not belong to the premise ratio.

[0015] In one embodiment of the present application, when the relationship type is the association relationship, the weight coefficient of the corresponding relationship is obtained according to the difference between the positive example ratio and the negative example ratio.

[0016] In one embodiment of the present application, when the relationship type is the subordinate relationship, the value of the weight coefficient of the corresponding relationship is equal to the positive example ratio; when the relationship type is the association relationship, the value of the weight coefficient of the corresponding relationship is equal to the absolute value of the difference between the positive example ratio and the negative example ratio. In the retrieval of the hospital infection knowledge graph according to the target physiological characteristics, it also includes:

[0017] Matching entities in the hospital infection knowledge graph according to the target physiological characteristics, and taking the successfully matched entity as the target entity;

[0018] Determine whether the number of the target entity is 0, if so, terminate the search, if not, update the target subgraph; query the hospital infection knowledge graph for all relationships where the premise entity is the target entity, and obtain candidate relationships;

[0019] Matching each candidate relationship with the target subgraph, and removing the candidate relationship that successfully matches the target subgraph;

[0020] The conclusion entities corresponding to the remaining candidate relations are taken as candidate entities, and all the candidate relations are divided into a plurality of target relation sets according to the conclusion entities; each target relation set is divided into a plurality of target relation subsets according to the relation type;

[0021] Obtaining a screening weight corresponding to each target relationship subset; if the relationship type corresponding to the target relationship subset is a subordinate relationship, taking the maximum value of the weight coefficients in the corresponding target relationship subset as the weight coefficient; if the relationship type corresponding to the target relationship subset is an association relationship, counting the number of relationships in the target relationship subset whose positive example ratio is 0 or whose negative example ratio is 0 to obtain a first value, and calculating the sum of all the weight coefficients in the target relationship subset and the first value to obtain the screening weight;

[0022] Obtaining a screening threshold corresponding to each target relationship subset, where the value of the screening threshold is greater than 0; comparing the screening weight corresponding to each target relationship subset with the corresponding screening threshold, and when the screening weight is less than the screening threshold, setting the screening result of the corresponding target relationship subset to fail;

[0023] When all the screening results corresponding to the target relationship set are rejected, the corresponding relationship is removed from the candidate relationship, the corresponding candidate entity is removed from the target entity, the remaining candidate entity is used as the new target entity, and the step of determining whether the number of target entities is 0 is returned to continue execution.

[0024] In one embodiment of the present application, after inputting the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model, it also includes obtaining the actual infection outcome of the target patient and updating the weight coefficient in the hospital infection knowledge graph according to the actual infection outcome.

[0025] According to the second aspect of the present invention, the present invention requests protection for a large-model hospital infection determination system based on intelligent agents, including: a knowledge graph construction module, a physiological feature acquisition module, a knowledge graph retrieval module, a relationship vector generation module and an infection outcome prediction module.

[0026] A physiological characteristic acquisition module, wherein the physiological characteristic acquisition module is used to acquire target physiological characteristics of a target patient; the target physiological characteristics include at least one of a description of the course of disease, an examination result, and a test result of the target patient;

[0027] A knowledge graph retrieval module, wherein the knowledge graph retrieval module is used to retrieve a hospital infection knowledge graph based on the target physiological characteristics to obtain a corresponding target sub-graph; wherein the hospital infection knowledge graph is constructed based on a hospital infection diagnosis standard document;

[0028] A relationship vector generation module, wherein the relationship vector generation module is used to generate a relationship vector set according to the target sub-graph to obtain a target relationship vector set;

[0029] An infection outcome prediction module is used to input the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model to obtain a target infection outcome.

[0030] In one embodiment of the present application, in the knowledge graph construction module, the hospital infection knowledge graph includes entities and relationships; the entity types of the entities include symptoms, test results, examination results and diagnosis results of each infection site; the relationships include the relationship between different symptoms, the relationship between different test results, the relationship between different examination results, the relationship between symptoms and test results, the relationship between symptoms and examination results, the relationship between test results and examination results, the relationship between symptoms and diagnosis results, the relationship between examination results and diagnosis results, and the relationship between test results and diagnosis results.

[0031] In one embodiment of the present application, in the knowledge graph construction module, the entity type also includes the diagnostic criteria for each infection site; the relationship between the symptoms and the diagnostic results is obtained based on the relationship between the corresponding symptoms and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the examination results and the diagnostic results is obtained based on the relationship between the corresponding examination results and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the test results and the diagnostic results is obtained based on the relationship between the corresponding test results and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results.

[0032] In one embodiment of the present application, the relationship vector set includes a premise entity, a conclusion entity and a weight coefficient, and the weight coefficient is used to evaluate the credibility of the patient's physiological characteristics belonging to the conclusion entity based on the patient's physiological characteristics belonging to the premise entity.

[0033] In one embodiment of the present application, the weight coefficient is obtained based on the proportion of positive examples; wherein, the proportion of positive examples is obtained based on the proportion of historical patients belonging to the conclusion entity among historical patients belonging to the premise entity.

[0034] In one embodiment of the present application, the relationship type of the relationship includes a subordinate relationship and an associated relationship. When the relationship type of the relationship is the subordinate relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio. When the relationship type of the relationship is the associated relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio and the negative example ratio, wherein the negative example ratio is obtained according to the ratio of historical patients belonging to the conclusion entity among the historical patients who do not belong to the premise ratio.

[0035] In one embodiment of the present application, when the relationship type is the association relationship, the weight coefficient of the corresponding relationship is obtained according to the difference between the positive example ratio and the negative example ratio.

[0036] In one embodiment of the present application, when the relationship type is the subordinate relationship, the value of the weight coefficient of the corresponding relationship is equal to the positive example ratio; when the relationship type is the association relationship, the value of the weight coefficient of the corresponding relationship is equal to the absolute value of the difference between the positive example ratio and the negative example ratio. The knowledge graph retrieval module further includes:

[0037] Matching entities in the hospital infection knowledge graph according to the target physiological characteristics, and taking the successfully matched entity as the target entity;

[0038] Determine whether the number of the target entity is 0, if so, terminate the search, if not, update the target subgraph; query the hospital infection knowledge graph for all relationships where the premise entity is the target entity, and obtain candidate relationships;

[0039] Matching each candidate relationship with the target subgraph, and removing the candidate relationship that successfully matches the target subgraph;

[0040] The conclusion entities corresponding to the remaining candidate relations are taken as candidate entities, and all the candidate relations are divided into a plurality of target relation sets according to the conclusion entities; each target relation set is divided into a plurality of target relation subsets according to the relation type;

[0041] Obtaining a screening weight corresponding to each target relationship subset; if the relationship type corresponding to the target relationship subset is a subordinate relationship, taking the maximum value of the weight coefficients in the corresponding target relationship subset as the weight coefficient; if the relationship type corresponding to the target relationship subset is an association relationship, counting the number of relationships in the target relationship subset whose positive example ratio is 0 or whose negative example ratio is 0 to obtain a first value, and calculating the sum of all the weight coefficients in the target relationship subset and the first value to obtain the screening weight;

[0042] Obtaining a screening threshold corresponding to each target relationship subset, where the value of the screening threshold is greater than 0; comparing the screening weight corresponding to each target relationship subset with the corresponding screening threshold, and when the screening weight is less than the screening threshold, setting the screening result of the corresponding target relationship subset to fail;

[0043] When all the screening results corresponding to the target relationship set are rejected, the corresponding relationship is removed from the candidate relationship, the corresponding candidate entity is removed from the target entity, the remaining candidate entity is used as the new target entity, and the step of determining whether the number of target entities is 0 is returned to continue execution.

[0044] In one embodiment of the present application, the system further includes a knowledge graph updating module, which is used to obtain the actual infection outcome of the target patient and update the weight coefficient in the hospital infection knowledge graph according to the actual infection outcome.

[0045] This application has the following beneficial effects:

[0046] 1. When judging the infection outcome of the target patient through the hospital infection prediction model, a hospital infection knowledge graph is generated according to the hospital infection diagnostic standard document, and the mapping result of the patient's physiological characteristics on the hospital infection knowledge graph is used as one of the diagnostic criteria of the hospital infection prediction model. When errors or biases in the training data set of the hospital infection prediction model cause the predicted infection outcome to differ from the actual infection outcome, the predicted infection outcome can be corrected through the relationship vector generated based on the knowledge graph, thereby improving the prediction accuracy while reducing the difficulty of data set screening of the hospital infection prediction model and improving the interpretability of the model prediction.

[0047] 2. By mining the relationship between different symptoms, the relationship between different test results, and the relationship between different examination results, it is helpful to improve the diagnostic scope of the hospital infection prediction model. At the same time, when constructing the conceptual model of the hospital infection knowledge graph, the relationship between symptoms and test results, the relationship between symptoms and examination results, and the relationship type between test results and examination results are established. By extracting the relationship between different categories of entities in the hospital infection diagnosis standard document, it is helpful to improve the accuracy of hospital infection identification. In particular, when it is difficult to determine whether the patient's designated part is infected based on only one of the patient's symptoms, test results, and examination results, the relationship between symptoms and test results, and the relationship between test results and examination results stored in the knowledge graph can be used to infer another entity from one entity, thereby completing the diagnosis of hospital infection and reducing the missed diagnosis rate of hospital infection.

[0048] 3. Use diagnostic criteria as entities in the hospital infection knowledge graph to store different diagnostic criteria for the same infection and the relationship between them. This will facilitate the description of the logical relationship between different diagnostic criteria when separating the corresponding subgraphs from the hospital infection knowledge graph based on the patient's physiological characteristics. For example, when judging whether a patient's urinary system is infected by the diagnostic criteria of etiological diagnosis, it is still necessary to proceed based on the diagnostic criteria of clinical diagnosis. It is difficult to confirm whether a patient has a urinary system infection if only relying on the results of etiological tests.

[0049] 4. Using the relationship vector containing the weight coefficient as one of the inputs of the hospital infection prediction model, in addition to enriching the characteristics of the relationship between entities and improving the accuracy of the hospital infection prediction model, it can also be used to identify "false" relationships between entities extracted from hospital infection diagnosis standard documents and historical patient data sets, that is, relationships in which the value of the weight coefficient is too small. The recognition results of the hospital infection prediction model obtained by training based on the relationship vector containing the weight coefficient are further reduced by the influence of errors or biases in the historical patient data set, thereby improving the recognition accuracy.

[0050] 5. Calculate the weight coefficient based on the proportion of positive examples. The weight coefficient can be used not only to evaluate the credibility of the knowledge, but also to evaluate the causal relationship between the premise entity and the conclusion entity. Under the same other conditions, the larger the value of the positive example proportion, the greater the probability that the patient's physiological characteristics belong to the conclusion entity when the patient's physiological characteristics belong to the premise entity; the smaller the value of the positive example proportion, the smaller the probability that the patient's physiological characteristics belong to the conclusion entity when the patient's physiological characteristics belong to the premise entity. In particular, when the value of the positive example proportion is 1, it means that when the patient's physiological characteristics belong to the premise entity, the patient's physiological characteristics must belong to the conclusion entity.

[0051] 6. In actual application scenarios, when the relationship type of the relationship is an association relationship, the relationship between the entities can be a positive correlation relationship or a negative correlation relationship. The weight coefficient calculated based on the positive example ratio and the negative example ratio can be used to evaluate the correlation between two entities with a positive association relationship or a negative association relationship.

[0052] 7. The weight coefficient is calculated by the difference between the proportion of positive examples and the proportion of negative examples, thereby evaluating the credibility of knowledge based on the degree of correlation between two entities.

[0053] 8. The technical solution of the present application achieves a targeted reduction in the number of relationships in the retrieved target subgraph, reduces the computational complexity of the hospital infection prediction model, and improves the prediction speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0055] Figure 1 This is a flow chart of the agent-based large-model hospital infection determination method involved in an embodiment of the present application;

[0056] Figure 2 A schematic diagram of an implementation of the hospital infection knowledge graph involved in the embodiment of this application;

[0057] Figure 3 This is a schematic diagram of another implementation of the hospital infection knowledge graph involved in the embodiment of this application;

[0058] Figure 4 A schematic diagram of the target subgraph retrieval process involved in the embodiment of the present application;

[0059] Figure 5This is a schematic diagram of the structure of the large-scale model hospital infection determination system based on the intelligent agent involved in the embodiment of the present application.

[0060] Figure 6 A schematic diagram of the structure of an electronic device involved in an embodiment of the present application;

[0061] Symbols in the figure: ω1-the premise entity is the clinical diagnostic standard 1, and the conclusion entity is the weight coefficient of the upper respiratory tract infection, ω2-the premise entity is the etiological diagnostic standard 1, and the conclusion entity is the weight coefficient of the upper respiratory tract infection, ω3-the premise entity is acute inflammation of the upper respiratory tract, and the conclusion entity is the weight coefficient of the clinical diagnostic standard 1, ω4-the premise entity is fever, and the conclusion entity is the weight coefficient of the clinical diagnostic standard 1, ω5-the premise entity is the secretion smear finding of significant pathogenic microorganisms related to the upper respiratory tract, and the conclusion entity is the weight coefficient of the etiological diagnostic standard 1, ω6-the premise entity is the secretion culture finding of significant pathogenic microorganisms related to the upper respiratory tract, and the conclusion entity is the weight coefficient of the etiological diagnostic standard 1, ω7-the premise entity The premise entity is acute inflammation of the upper respiratory tract caused by allergies, and the conclusion entity is the weight coefficient of upper respiratory tract infection. ω8-the premise entity is fever, and the conclusion entity is the weight coefficient of clinical diagnostic standard 2. ω9-the premise entity is clinical diagnostic standard 2, and the conclusion entity is the weight coefficient of lower respiratory tract infection. ω10-the premise entity is etiological diagnostic standard 2, and the conclusion entity is the weight coefficient of lower respiratory tract infection. ω11-the premise entity is acute inflammation of the tonsils, and the conclusion entity is the weight coefficient of acute inflammation of the upper respiratory tract. ω12-the premise entity is acute inflammation of the nasopharynx, and the conclusion entity is the weight coefficient of acute inflammation of the upper respiratory tract. ω13-the premise entity is nasal congestion, and the conclusion entity is the weight coefficient of acute inflammation of the nasopharynx. ω14-the premise entity is nasal congestion, and the conclusion entity is general The weight coefficient of the common cold is ω15, the weight coefficient of the premise entity is the common cold and the conclusion entity is the upper respiratory tract infection, ω1'-the weight coefficient of the premise entity is the upper respiratory tract infection and the conclusion entity is the clinical diagnostic standard 1, ω2'-the weight coefficient of the premise entity is the upper respiratory tract infection and the conclusion entity is the etiological diagnostic standard 1, ω3'-the weight coefficient of the premise entity is the clinical diagnostic standard 1 and the conclusion entity is the acute inflammation of the upper respiratory tract, ω4'-the weight coefficient of the premise entity is the clinical diagnostic standard 1 and the conclusion entity is the fever, ω5'-the weight coefficient of the premise entity is the etiological diagnostic standard 1 and the conclusion entity is the secretion smear finding of significant pathogenic microorganisms related to the upper respiratory tract, ω6'-the premise entity is the etiological diagnostic standard 1 and the conclusion entity is the weight coefficient of the pathogenic microorganisms related to the upper respiratory tract, The theory entity is the weight coefficient of the significant pathogenic microorganisms related to the upper respiratory tract found in secretion culture, ω7'-the premise entity is the weight coefficient of the upper respiratory tract infection and the conclusion entity is the acute inflammation of the upper respiratory tract caused by allergies, ω8'-the premise entity is the weight coefficient of the clinical diagnostic standard 2 and the conclusion entity is the fever, ω9'-the premise entity is the weight coefficient of the lower respiratory tract infection and the conclusion entity is the clinical diagnostic standard 2, ω10'-the premise entity is the lower respiratory tract infection and the conclusion entity is the etiological diagnostic standard 2, ω11'-the premise entity is the acute inflammation of the upper respiratory tract and the conclusion entity is the acute inflammation of the tonsils, ω12'-the premise entity is the acute inflammation of the upper respiratory tract and the conclusion entity is the acute inflammation of the nasopharynx,ω13'-the weight coefficient of the premise entity being acute nasopharyngeal inflammation and the conclusion entity being nasal congestion, ω14'-the weight coefficient of the premise entity being common cold and the conclusion entity being nasal congestion, ω15'-the weight coefficient of the premise entity being upper respiratory tract infection and the conclusion entity being common cold. DETAILED DESCRIPTION

[0062] The present invention provides an agent-based large-scale model hospital infection determination method and system. To make the above-mentioned objects, features, and advantages of this application more clearly understood, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The components of the embodiments of this application generally described and illustrated in the drawings herein may be arranged and designed in various different configurations. Therefore, reference to the terms "one embodiment," "some embodiments," "implementation methods," "embodiment," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., in the following detailed description of the embodiments of this application provided in the drawings, are not intended to limit the scope of the claimed application, but merely indicate that the specific features, structures, or characteristics described in conjunction with such embodiment or example are included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. All other embodiments derived by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0063] It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. At the same time, in the description of this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0064] According to the first aspect of the present invention, the present invention claims a large model based on intelligent body hospital infection determination method, with reference to the attached Figure 1 As shown, including:

[0065] S1: Obtain the physiological characteristics of historical patients and obtain historical physiological characteristics; obtain the infection outcomes of historical patients and obtain historical infection outcomes.

[0066] It should be noted that the physiological characteristics include at least one of a description of the course of disease, examination results, and test results, but are not limited thereto. The infection outcome is used to describe the patient's current infection status, such as whether an infection exists, the location of the infection, etc., but are not limited thereto.

[0067] In this embodiment, the infection outcome includes a set of infection sites of the patient. When the patient has an infection, the corresponding infection site set is the site of infection of the patient; when the patient does not have an infection, the corresponding infection site set is an empty set.

[0068] S2: Generate a hospital infection diagnosis knowledge graph based on the hospital infection diagnosis standard document.

[0069] It should be noted that the hospital infection diagnosis standard document can be developed based on the specific conditions of the medical institution's patients, such as the types of diseases they are admitted to, or can be the "Hospital Infection Diagnosis Standard" issued by the National Health Commission of the People's Republic of China, or can be obtained by integrating multiple different hospital infection diagnosis standards to obtain the hospital infection diagnosis standard document, or can be other diagnostic standard documents that can be used for hospital infection diagnosis. This application does not limit the specific number and source of hospital infection diagnosis standard documents.

[0070] In this embodiment, step S2 also includes: S21, obtaining the conceptual model of the hospital infection diagnosis knowledge graph. The conceptual model includes entities and relationships. The construction of the conceptual model of the hospital infection diagnosis knowledge graph can be completed through knowledge graph open source tools such as protégé, NeOn Toolkit, Altova SemanticWorks, TopBraid Composer, etc. S22, identifying and extracting entities and relationships in the hospital infection diagnosis standard document through natural language processing technology. S23, filling the conceptual model of the hospital infection diagnosis knowledge graph according to the extraction results of the hospital infection diagnosis standard document to obtain the hospital infection diagnosis knowledge graph. S24, verifying the hospital infection diagnosis knowledge graph according to the historical physiological characteristics and the corresponding historical infection outcomes.

[0071] In this embodiment, refer to the attached Figure 2As shown, in step S21, the entities in the hospital knowledge graph can be described as entity types and entity names, but of course they are not limited to this. The entity types include symptoms, test results, examination results, diagnostic criteria for infection sites, and diagnostic results for infection sites, but of course they are not limited to this. The relationships include the relationships between different symptoms, the relationships between different test results, the relationships between different examination results, the relationships between symptoms and test results, the relationships between symptoms and examination results, the relationships between test results and examination results, the relationships between diagnostic criteria and symptoms, the relationships between diagnostic criteria and test results, the relationships between diagnostic criteria and examination results, and the relationships between diagnostic criteria and diagnostic results, but of course they are not limited to this. Among them, the diagnostic criteria include clinical diagnosis and etiological diagnosis, but of course they are not limited to this.

[0072] It should be noted that when constructing the conceptual model of the hospital infection knowledge graph, the relationships between symptoms and test results, the relationships between symptoms and examination results, and the types of relationships between test results and examination results are established. By extracting the relationships between different categories of entities in the hospital infection diagnosis standard document, it helps to improve the accuracy of hospital infection identification. In particular, when it is difficult to determine whether a patient has an infection in a specific area based on only one of the patient's symptoms, test results, or examination results, the relationships between symptoms and test results, symptoms and examination results, and test results and examination results stored in the knowledge graph can be used to infer one entity from another, thereby completing the diagnosis of hospital infection and reducing the missed diagnosis rate of hospital infection.

[0073] In this embodiment, the diagnostic criteria are used as entities in the hospital infection knowledge graph, and the relationship between different diagnostic criteria for the same infection and the diagnostic criteria is stored. When the corresponding subgraph is separated from the hospital infection knowledge graph according to the patient's physiological characteristics, it is convenient to describe the logical relationship between different diagnostic criteria. For example, when judging whether the patient's urinary system is infected by the diagnostic criteria of etiological diagnosis, it is still necessary to do so on the basis of the diagnostic criteria of clinical diagnosis. It is difficult to confirm whether the patient has a urinary system infection if only relying on the results of etiological tests.

[0074] S3: Retrieve the hospital infection diagnosis knowledge graph according to each of the historical physiological characteristics to obtain a corresponding historical sub-graph.

[0075] In this embodiment, step S3 further includes: S31, identifying entities corresponding to the historical physiological features to obtain historical entities. The historical entities can be identified from the historical physiological features using named entity recognition (NER) technology. S32, matching entities in the hospital infection diagnosis knowledge graph with the historical entities, and extracting subgraphs from the hospital infection diagnosis knowledge graph based on the successfully matched historical entities to obtain the historical subgraph. The size of the subgraph can be determined in a pre-set manner.

[0076] S4: Convert each of the historical subgraphs into a relationship vector set to obtain a corresponding historical relationship vector set.

[0077] It should be noted that the relationship vector set is the set of all relationship vectors corresponding to the subgraph. Each relationship vector describes the relationship between entities in the subgraph. The relationship vector can be represented by (first entity, relationship name, second entity). For example, the vector ("fever," "related symptoms," "clinical diagnostic criteria for upper respiratory tract infection") indicates that "fever" is one of the related symptoms of "clinical diagnostic criteria for upper respiratory tract infection."

[0078] S5: Each of the historical relationship vectors, the corresponding historical physiological characteristics, and the corresponding historical infection outcomes are used as sample data. All sample data constitute a training data set for a hospital infection prediction model. The hospital infection prediction model is trained based on the training data set.

[0079] It should be noted that the hospital infection prediction model can be a large language model built based on transformers, which is used to generate corresponding answers based on user questions. For example, the hospital infection prediction model can be established based on the GPT series, BERT, T5, etc., and the hospital infection prediction model is pre-trained based on an expected dataset of Chinese text on the Internet. The pre-trained hospital infection prediction model is supervised and fine-tuned using the training dataset. The hospital infection prediction model can also be a large model built based on a neural network model, which is used to obtain specific values ​​of preset output results based on specific values ​​of preset input features.

[0080] In this embodiment, a large model constructed based on a neural network model is used as an example for detailed description. The input of the hospital infection prediction model includes physiological characteristics and relationship vectors, and the output includes the corresponding predicted infection outcomes. When training the hospital infection prediction model, the historical physiological characteristics and the historical relationship vectors in each of the sample data are used as the input of the hospital infection prediction model, and the output corresponds to a set of sites where infection exists, that is, a predicted set of infection sites. The hospital infection prediction model is trained according to a preset loss function, and the loss function is used to calculate the difference between the predicted infection outcomes and the historical infection outcomes of all sample data. The loss function can select the intersection-union ratio between the predicted set of infection sites and the actual set of infection sites.

[0081] S6: Acquire physiological characteristics of the target patient to obtain target physiological characteristics.

[0082] S7: Retrieve the hospital infection knowledge graph based on the target physiological characteristics to obtain a corresponding target subgraph. Generate the relationship vector set based on the target subgraph to obtain a target relationship vector set. The method for obtaining target relationship vectors based on target physiological characteristics is similar to that of steps S4 and S5 and will not be further described here.

[0083] S8: Input the target relationship vector and the target physiological characteristics into the trained hospital infection prediction model to obtain a target infection outcome.

[0084] It should be noted that when the hospital infection prediction model is used to judge the infection outcome of the target patient, a hospital infection knowledge graph is generated based on the hospital infection diagnostic standard document, and the mapping results of the patient's physiological characteristics on the hospital infection knowledge graph are used as one of the diagnostic criteria of the hospital infection prediction model. When errors or biases in the training data set of the hospital infection prediction model cause the predicted infection outcome to differ from the actual infection outcome, the predicted infection outcome can be corrected by using the relationship vector generated based on the knowledge graph, thereby improving the prediction accuracy while reducing the difficulty of data set screening of the hospital infection prediction model and improving the interpretability of the model prediction.

[0085] In a feasible embodiment, refer to the attached Figure 3As shown, the relationship types of the relationship include subordinate relationships and associated relationships, of course, it is not limited to these. Among them, the subordinate relationship is used to indicate that there is an inclusion relationship between two entities that refer to different objects. For example, the subordinate relationship between "acute tonsillitis" and "acute inflammation of the upper respiratory tract" indicates that "acute tonsillitis" belongs to "acute inflammation of the upper respiratory tract". Among them, the associated relationship is used to describe the existence of a correlation between two entities that refer to different objects. For example, the associated relationship between "acute inflammation of the nasopharynx" and "nasal congestion" indicates that although "nasal congestion" is one of the main symptoms of "acute inflammation of the nasopharynx", patients with "nasal congestion" do not necessarily have "acute inflammation of the nasopharynx", and patients with "acute inflammation of the nasopharynx" do not necessarily have "nasal congestion" but may also have other symptoms of "acute inflammation of the nasopharynx". In particular, when establishing the conceptual model of the hospital infection knowledge graph, it is pre-set that when the equivalence relationship between two entities in the conceptual model is the subordinate relationship or the associated relationship, such as the scientific name and alias of the same symptom, etc. In this embodiment, the equivalence relationship is defined as a special subordinate relationship, that is, when two entities are equivalent, the relationship type of their corresponding relationship is the subordinate relationship.

[0086] In this embodiment, the relationship vector set includes a premise entity, a conclusion entity and a weight coefficient, and the weight coefficient is used to evaluate the credibility of the patient's physiological characteristics belonging to the conclusion entity based on the patient's physiological characteristics belonging to the premise entity.

[0087] It should be noted that the relationship between the weight coefficient and the degree of credibility is preset. For example, a relationship with a higher degree of credibility may have a larger weight coefficient, or a relationship with a lower degree of credibility may have a smaller weight coefficient, or other feasible setting methods may be used.

[0088] It should be noted that the relationship vector can be obtained based on the vector (premise entity, relationship type, conclusion entity, weight coefficient), or based on the vector (premise entity, conclusion entity, weight coefficient), and of course can also be obtained through other feasible methods.

[0089] In this embodiment, a detailed description is provided using the example of a vector (precondition entity, relationship type, conclusion entity, weight coefficient) where a higher confidence level corresponds to a larger weight coefficient. For example, the subordination relationship between acute tonsillitis and acute upper respiratory tract inflammation can be converted into a vector ("acute tonsillitis," "subordination relationship," "acute upper respiratory tract inflammation," 1). This means that the confidence level of 1 is 1 based on the patient's physiological characteristics belonging to "acute tonsillitis" and therefore also deriving that the patient's physiological characteristics also belong to "acute upper respiratory tract inflammation." Alternatively, the subordination relationship between acute tonsillitis and acute upper respiratory tract inflammation can be described using a vector ("acute upper respiratory tract inflammation," "subordination relationship," "acute tonsillitis," first weight coefficient). The first weight coefficient is less than 1. This means that the confidence level of 1 is 1 based on the patient's physiological characteristics belonging to "acute upper respiratory tract inflammation" and therefore deriving that the patient's physiological characteristics also belong to "acute tonsillitis."

[0090] It should be noted that using the relationship vector containing the weight coefficient as one of the inputs to the hospital infection prediction model not only enriches the characteristics of the relationships between entities and improves the accuracy of the hospital infection prediction model, but also can be used to identify "false" relationships between entities extracted from hospital infection diagnostic standard documents and historical patient datasets, i.e., relationships in which the weight coefficient value is too small. The output results of the hospital infection prediction model trained based on the relationship vector containing the weight coefficient are further reduced from being affected by errors or biases in the historical patient dataset, thereby improving recognition accuracy.

[0091] It should be noted that the weight coefficient can be preset according to the relationship between the two entities, or can be obtained according to a cross neural network model, ResourceRank algorithm, PTCA model, grid search algorithm, etc.

[0092] It should be noted that when the relationship type of the relationship is the subordinate relationship, the weight coefficient of the corresponding relationship is obtained based on the positive example ratio. The positive example ratio is obtained based on the proportion of historical patients belonging to the conclusion entity among the historical patients belonging to the premise entity, and the positive example ratio is recorded as p(B|A). A represents the event in which the patient's physiological characteristics belong to the premise entity, and B represents the event in which the patient's physiological characteristics belong to the conclusion entity. The value of the weight coefficient of the corresponding relationship can be equal to the positive example ratio, or it can be the result of further processing of the positive example ratio, for example, the value of the weight coefficient is equal to the product of a preset ratio and the positive example ratio, or the value of the weight coefficient is equal to the sum of a preset value and the positive example ratio. The weight coefficient can also be other feasible specific value-taking methods.

[0093] It should be noted that when the relationship type of the relationship is the association relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio and the negative example ratio. Among them, the negative example ratio is obtained according to the ratio of historical patients belonging to the conclusion entity among the historical patients who do not belong to the premise ratio, which is recorded as .in, This indicates an event in which the patient's physiological characteristics do not correspond to the premise entity. The weight coefficient corresponding to this relationship can be expressed as a vector (positive example percentage, negative example percentage), or can be obtained by calculating the linear weighted result of the positive example percentage and the negative example percentage, or other feasible representation methods.

[0094] It should be noted that the weight coefficient is calculated based on the positive example percentage. This weight coefficient can be used not only to assess the credibility of the knowledge but also to evaluate the causal relationship between the premise entity and the conclusion entity. All other conditions being equal, a larger positive example percentage indicates a greater probability that the patient's physiological characteristics belong to the conclusion entity when the patient's physiological characteristics belong to the premise entity. A smaller positive example percentage indicates a lower probability that the patient's physiological characteristics belong to the conclusion entity when the patient's physiological characteristics belong to the premise entity. For example, fever is not only one of the clinical diagnostic criteria for upper respiratory tract infection, but also one of the clinical diagnostic criteria for lower respiratory tract infection. In the absence of other symptoms, examination results, or test results, if the value of p(meets the clinical diagnostic criteria for upper respiratory tract infection | fever) is greater than the value of p(meets the clinical diagnostic criteria for lower respiratory tract infection | fever), the patient's probability of meeting the clinical diagnostic criteria for upper respiratory tract infection should be considered higher if they have a fever. In particular, when the value of the positive example ratio is 1, it indicates that when the patient's physiological characteristics belong to the premise entity, the patient's physiological characteristics must belong to the conclusion entity. For example, the value of p (meets the etiological diagnostic criteria for vascular-related infection | Simple catheter culture isolates meaningful pathogenic microorganisms corresponding to vascular-related infection) is 1, indicating that when the patient's test result is that the catheter tip culture isolates meaningful pathogenic microorganisms corresponding to vascular-related infection, it indicates that the patient meets the etiological diagnostic criteria for vascular-related infection. Furthermore, the value of p (presence of vascular-related infection | meets the etiological diagnostic criteria for vascular-related infection) is 1, indicating that when the patient's physiological characteristics meet the etiological diagnostic criteria for vascular-related infection, it can be concluded that the patient has vascular-related infection. The above two positive example ratios related to vascular-related infection indicate that when the patient's physiological characteristics meet the criteria for the isolation of meaningful pathogenic microorganisms corresponding to vascular-related infection by catheter tip culture, they are symptom markers for vascular-related infection.

[0095] It should be noted that in actual application scenarios, when the relationship type of a relationship is an association relationship, the relationship between entities can be a positive correlation or a negative correlation. For example, in the hospital infection diagnosis standard document, acute diarrhea is one of the clinical diagnostic criteria for infectious diarrhea. That is, when a patient has acute diarrhea, there is a probability that the patient has infectious diarrhea. Therefore, there is a positive correlation between the entity "acute diarrhea" and the entity "clinical diagnostic criteria for infectious diarrhea." When a patient meets the diagnostic criteria for infectious diarrhea, it should be ruled out that the patient's diarrhea is caused by an acute attack of chronic diarrhea or non-infectious factors such as diagnostic and treatment reasons, underlying diseases, and psychological stress. That is, when the patient's physiological characteristics fall within the entity "acute attack of chronic diarrhea", the patient should not be diagnosed with "infectious diarrhea". Only when the patient's physiological characteristics do not fall within the entity "acute attack of chronic diarrhea" can the patient be diagnosed with "infectious diarrhea". Therefore, there is a negative correlation between the entity "acute attack of chronic diarrhea" and the entity "infectious diarrhea". At the same time, the weight coefficient calculated based on the positive example ratio and the negative example ratio can be used to evaluate the correlation between two entities with a positive correlation relationship, and can also be used to evaluate the correlation between two entities with a negative correlation relationship.

[0096] In this embodiment, the weight coefficient ω is calculated as follows:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] Wherein, p represents the calculation proportion; B represents the conclusion entity; A represents the premise entity; represents the complement of the premise entity; that is, it is not the premise entity; n(A) represents the number of historical patients whose physiological characteristics belong to the conclusion entity, represents the number of historical patients who do not belong to the conclusion entity, N represents the total number of historical patients, and n(AB) represents the number of historical patients who belong to both the premise entity and the conclusion entity. Represents the number of historical patients who do not belong to the premise entity but belong to the conclusion entity.

[0102] It should be noted that, with respect to the subordination relationship, the larger the value of the weight coefficient, the greater the probability that the patient's physiological characteristics belong to the premise entity, in the absence of additional evidence. In particular, when the weight coefficient is 1, the premise entity is included in the conclusion entity, and when the patient's physiological characteristics belong to the premise entity, the patient's physiological characteristics must belong to the conclusion entity.

[0103] It should be noted that, for the association relationship, when the value of the positive example ratio is closer to the value of the negative example ratio, it indicates that whether the patient's physiological characteristics belong to the premise entity has a smaller impact on whether the patient's physiological characteristics belong to the conclusion entity, that is, the correlation between the premise entity and the conclusion entity is smaller. In particular, when the value of the positive example ratio is equal to the value of the negative example ratio, it indicates that whether the patient's physiological characteristics belong to the premise entity has no effect on whether the patient belongs to the conclusion entity, and the correlation between the premise entity and the conclusion entity is the smallest. When the difference between the value of the positive example ratio and the value of the negative example ratio is greater, it indicates that whether the patient's physiological characteristics belong to the premise entity has a greater impact on whether the patient's physiological characteristics belong to the conclusion entity, that is, the correlation between the premise entity and the conclusion entity is greater. In particular, for the association relationship in which the value of the positive example ratio is 1 and the value of the negative example ratio is 0, when the patient's physiological characteristics belong to the premise entity, the patient's physiological characteristics must also belong to the conclusion entity, and when the patient's physiological characteristics do not belong to the premise entity, the patient's physiological characteristics must also not belong to the conclusion entity, that is, the premise entity and the conclusion entity are positively correlated, and the correlation is the largest. For example, a disease with a unique marker, etc. Similarly, for the relationship in which the value of the positive example ratio is 0 and the value of the negative example ratio is 1, the relationship between the corresponding premise entity and the corresponding conclusion entity is a negative correlation, and the correlation is the largest. The weight coefficient is calculated by the difference between the positive example ratio and the negative example ratio, so that the credibility of knowledge is evaluated according to the degree of correlation between the two entities.

[0104] In this embodiment, refer to the attached Figure 4 As shown, step S7 also includes the following sub-steps:

[0105] S70. Initialize the target subgraph to an empty graph, and initialize the candidate relationship set, target entity set, and target relationship set to empty sets respectively.

[0106] S71. Match entities in the hospital infection knowledge graph based on the target physiological characteristics, and use the successfully matched entities as target entities. Identify entities based on the target physiological characteristics using named entity recognition (NER). Match the identified entities with entities in the hospital infection diagnosis knowledge graph, and use the successfully matched entities as target entities in the target entity set.

[0107] S72. Determine whether the target entity set is an empty set. If the target entity set is an empty set, terminate the search. If the target entity set is not an empty set, add each target entity in the target entity set to the target subgraph, and proceed to step S73.

[0108] S73: Update the target subgraph based on the relations in the candidate relation set. If the number of relations in the candidate relation set is 0, directly execute step S74; if the number of relations in the candidate relation set is not 0, add the relations in the candidate relation set to the target subgraph.

[0109] S74: Query the hospital infection knowledge graph for all relationships where the premise entity is the target entity, and update the candidate relationship set based on the query result. The updated candidate relationship set is the relationship obtained by the query.

[0110] S75. Match each candidate relationship with the target subgraph, and remove the candidate relationship that successfully matches the target subgraph. If and only if there is a relationship in the target subgraph that is identical to the candidate relationship, the corresponding candidate relationship is considered to be a successful match, to avoid repeated searches and computational redundancy. If all candidate relationships successfully match the target subgraph, the search terminates. If there is a candidate relationship that fails to match the target subgraph, continue to step S76.

[0111] S76. The conclusion entities corresponding to the remaining candidate relations are used as candidate entities. All the remaining candidate relations are divided into several sets according to the conclusion entities, and the target relation set is updated according to the division results. The conclusion entities of all relations in the same target relation set are the same. Each target relation set is divided into several sets according to the relation type to obtain target relation subsets. The conclusion entities of all relations in the same target relation subset are the same, and the relation types are also the same.

[0112] S77, obtaining the screening weight corresponding to each target relationship subset. i The screening weight α i The calculation methods include:

[0113] ;

[0114] ;

[0115] ;

[0116] Among them, max represents the function of taking the maximum value; ω i，j Represents the target relation subset R i The weight coefficient of the jth relationship in ; m represents the target relationship subset R i The total number of relationships in C i，j Represents the target relation subset R i The corresponding first value; I (preset condition) represents an indicative function. When the preset condition is met, the value of the indicative function is 1, and when the preset condition is not met, the value of the indicative function is 0; A i，j Represents the target relation subset R i The premise entity of the jth relationship in B i，j Represents the target relation subset R i The conclusion entity of the jth relation in A i，j Represents the target relation subset R i The complement of the premise entity of the j-th relation in .

[0117] S78. Obtain a screening threshold corresponding to each target relationship subset. Compare the screening weight corresponding to each target relationship subset with the corresponding screening threshold. If the screening weight is less than the screening threshold, set the screening result of the corresponding target relationship subset to be a failure; if the screening weight is not less than the screening threshold, set the screening result of the corresponding target relationship subset to be a pass.

[0118] It should be noted that the screening threshold can be preset according to the relationship type, or other feasible methods. The screening thresholds corresponding to different target relationship subsets can be the same or different. The larger the value of the screening threshold, the smaller the target sub-graph retrieved from the hospital infection knowledge graph according to the target physiological characteristics, the fewer target relationship vectors generated according to the target sub-graph, and the less computational complexity the hospital infection prediction model requires during the prediction process; the closer the value of the screening threshold is to 0, the larger the sub-graph retrieved from the hospital infection knowledge graph according to the target physiological characteristics, the more target relationship vectors generated according to the target sub-graph, and the greater the computational complexity the hospital infection prediction model requires during the prediction process.

[0119] In this embodiment, the target relation subset R i The screening threshold β i The calculation methods include:

[0120] .

[0121] In this embodiment, the target relationship subset with a screening result of pass and a relationship type of subordinate relationship is the relationship in which the premise entity is included in the conclusion entity, that is, the premise entity is a subset of the conclusion entity. The target relationship subset with a screening result of pass and a relationship type of associated relationship is the relationship in which the patient's physiological characteristics belong to the conclusion entity and all necessary premise entities are met. For example, when the hospital infection diagnosis standard document clearly states that the clinical diagnostic standard for infectious diarrhea is considered to meet the clinical diagnostic standard if it meets one of the following three conditions: 1. Acute diarrhea, with routine fecal microscopy showing no fewer than 10 white blood cells per high-power field; 2. Acute diarrhea, or fever, nausea, vomiting, abdominal pain, etc.; 3. Acute diarrhea more than 3 times a day for 2 consecutive days, or more than 5 watery diarrheas in 1 day. As can be seen from the above, acute diarrhea is a necessary condition for meeting the clinical diagnostic standard for infectious diarrhea. In other words, if and only if the patient has acute diarrhea, it is possible to meet the clinical diagnostic standard for infectious diarrhea. If the patient does not have acute diarrhea, the patient can be preliminarily ruled out as not meeting the clinical diagnostic standard for infectious diarrhea. In addition, because the clinical diagnostic criteria for acute pelvic inflammatory disease include diarrhea (acute diarrhea is a type of diarrhea), when a patient's physiological characteristics indicate acute diarrhea, the patient may have both infectious diarrhea and acute pelvic inflammatory disease. If the patient also has the physiological characteristic of a stool microscopic examination with a white blood cell count of at least 10 per high-power field of view, and does not have other necessary clinical diagnostic criteria for acute pelvic inflammatory disease, the relationship ("diarrhea," "association," "acute pelvic inflammatory disease," second weight coefficient) should not be included in the retrieved target subgraph. This method achieves a targeted reduction in the number of relationships in the retrieved target subgraph, reduces the computational complexity of the hospital infection prediction model, and improves prediction speed.

[0122] S79. When all the screening results corresponding to the target relationship set are rejected, remove the relationships in the target relationship set for which all the screening results are rejected from the candidate relationship set, remove the candidate entities corresponding to the target relationship set for which all the screening results are rejected from the candidate entity set, take the remaining candidate entities as the new target entities, and return to step S72 to continue execution.

[0123] In this embodiment, the method further includes step S8, obtaining the actual infection outcome of the target patient, and updating the weight coefficient in the hospital infection knowledge graph according to the actual infection outcome to complete the update of the hospital infection knowledge graph.

[0124] According to the second aspect of the present invention, the present invention claims protection for a large model hospital infection determination system based on an intelligent agent, Figure 5As shown, it includes: knowledge graph construction module, physiological feature acquisition module, knowledge graph retrieval module, relationship vector generation module and infection outcome prediction module.

[0125] A physiological characteristic acquisition module, wherein the physiological characteristic acquisition module is used to acquire target physiological characteristics of a target patient; the target physiological characteristics include at least one of a description of the course of disease, an examination result, and a test result of the target patient;

[0126] A knowledge graph retrieval module, wherein the knowledge graph retrieval module is used to retrieve a hospital infection knowledge graph based on the target physiological characteristics to obtain a corresponding target sub-graph; wherein the hospital infection knowledge graph is constructed based on a hospital infection diagnosis standard document;

[0127] A relationship vector generation module, wherein the relationship vector generation module is used to generate a relationship vector set according to the target sub-graph to obtain a target relationship vector set;

[0128] An infection outcome prediction module is used to input the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model to obtain a target infection outcome.

[0129] In a feasible embodiment, in the knowledge graph construction module, the hospital infection knowledge graph includes entities and relationships; the entity types of the entities include symptoms, test results, examination results and diagnosis results of each infection site; the relationships include the relationship between different symptoms, the relationship between different test results, the relationship between different examination results, the relationship between symptoms and test results, the relationship between symptoms and examination results, the relationship between test results and examination results, the relationship between symptoms and diagnosis results, the relationship between examination results and diagnosis results, and the relationship between test results and diagnosis results.

[0130] In a feasible embodiment, in the knowledge graph construction module, the entity type also includes the diagnostic criteria for each infection site; the relationship between the symptoms and the diagnostic results is obtained based on the relationship between the corresponding symptoms and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the examination results and the diagnostic results is obtained based on the relationship between the corresponding examination results and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the test results and the diagnostic results is obtained based on the relationship between the corresponding test results and the diagnostic criteria and the relationship between the diagnostic criteria and the corresponding diagnostic results.

[0131] In a feasible embodiment, the relationship vector set includes a premise entity, a conclusion entity and a weight coefficient, and the weight coefficient is used to evaluate the credibility of the patient's physiological characteristics belonging to the conclusion entity based on the patient's physiological characteristics belonging to the premise entity.

[0132] In a feasible implementation, the weight coefficient is obtained based on the proportion of positive examples; wherein, the proportion of positive examples is obtained based on the proportion of historical patients belonging to the conclusion entity among historical patients belonging to the premise entity.

[0133] In a feasible embodiment, the relationship type of the relationship includes a subordinate relationship and an associated relationship. When the relationship type of the relationship is the subordinate relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio. When the relationship type of the relationship is the associated relationship, the weight coefficient of the corresponding relationship is obtained according to the positive example ratio and the negative example ratio, wherein the negative example ratio is obtained according to the ratio of historical patients belonging to the conclusion entity among the historical patients who do not belong to the premise ratio.

[0134] In a feasible implementation, when the relationship type is the association relationship, the weight coefficient of the corresponding relationship is obtained according to the difference between the positive example ratio and the negative example ratio.

[0135] In a feasible implementation, when the relationship type is the subordinate relationship, the value of the weight coefficient of the corresponding relationship is equal to the positive example ratio; when the relationship type is the association relationship, the value of the weight coefficient of the corresponding relationship is equal to the absolute value of the difference between the positive example ratio and the negative example ratio; the knowledge graph retrieval module further includes:

[0136] Matching entities in the hospital infection knowledge graph according to the target physiological characteristics, and taking the successfully matched entity as the target entity;

[0137] Determine whether the number of the target entity is 0, if so, terminate the search, if not, update the target subgraph; query the hospital infection knowledge graph for all relationships where the premise entity is the target entity, and obtain candidate relationships;

[0138] Matching each candidate relationship with the target subgraph, and removing the candidate relationship that successfully matches the target subgraph;

[0139] The conclusion entities corresponding to the remaining candidate relations are taken as candidate entities, and all the candidate relations are divided into a plurality of target relation sets according to the conclusion entities; each target relation set is divided into a plurality of target relation subsets according to the relation type;

[0140] Obtaining a screening weight corresponding to each target relationship subset; if the relationship type corresponding to the target relationship subset is a subordinate relationship, taking the maximum value of the weight coefficients in the corresponding target relationship subset as the weight coefficient; if the relationship type corresponding to the target relationship subset is an association relationship, counting the number of relationships in the target relationship subset whose positive example ratio is 0 or whose negative example ratio is 0 to obtain a first value, and calculating the sum of all the weight coefficients in the target relationship subset and the first value to obtain the screening weight;

[0141] Obtaining a screening threshold corresponding to each target relationship subset, where the value of the screening threshold is greater than 0; comparing the screening weight corresponding to each target relationship subset with the corresponding screening threshold, and when the screening weight is less than the screening threshold, setting the screening result of the corresponding target relationship subset to fail;

[0142] When all the screening results corresponding to the target relationship set are rejected, the corresponding relationship is removed from the candidate relationship, the corresponding candidate entity is removed from the target entity, the remaining candidate entity is used as the new target entity, and the step of determining whether the number of target entities is 0 is returned to continue execution.

[0143] In a feasible embodiment, the system further includes a knowledge graph updating module, which is used to obtain the actual infection outcome of the target patient and update the weight coefficient in the hospital infection knowledge graph according to the actual infection outcome.

[0144] Refer to the attached Figure 6 As shown, an embodiment of the present application provides an electronic device, including: a processor and a memory, the processor and the memory are interconnected and communicate with each other through a communication bus and / or other forms of connection mechanisms (not shown), the memory stores a computer program executable by the processor, and when the computing device is running, the processor executes the computer program to execute the system in any optional implementation mode of the above embodiment.

[0145] An embodiment of the present application provides a storage medium, wherein when the computer program is executed by a processor, the system of any optional implementation of the above embodiment is executed. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0146] In the embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of the modules is only a logical function division, and can be implemented in another way. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, indirect coupling or communication connection of the system or unit, which can be electrical, mechanical or other forms.

[0147] In addition, the units described as separate components may or may not be physically separate, and the components shown as units 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 units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] Furthermore, the functional modules in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0149] Flowcharts are used herein to illustrate the steps of the methods of the embodiments of the present disclosure. It should be understood that the preceding or following steps do not necessarily need to be performed in exact order. Instead, the various steps may be evaluated in reverse order or simultaneously. Furthermore, other operations may be added to these processes.

[0150] Unless otherwise defined, all terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and should not be interpreted in an idealized or highly formal sense unless expressly defined as such herein.

[0151] The above describes in detail the provided agent-based large-model hospital infection determination method and system. This article uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are merely examples of this application and are only used to help understand the agent-based large-model hospital infection determination method and system of this application, and are not intended to limit the scope of protection of this application. At the same time, for those skilled in the art, this application may be subject to various modifications and variations. Any modifications or equivalent substitutions made within the spirit and principles of this application shall be included in the scope of protection of this application.

Claims

1. A large-scale model hospital infection determination method based on an intelligent agent, characterized in that: include: Acquire target physiological characteristics of target patients; The target physiological characteristic includes at least one of a description of the disease course, an examination result, and a test result of the target patient; Retrieving a hospital infection knowledge graph based on the target physiological characteristics to obtain a corresponding target subgraph; wherein the hospital infection knowledge graph is constructed based on a hospital infection diagnosis standard document; Generate a relationship vector set according to the target subgraph to obtain a target relationship vector set; Inputting the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model to obtain a target infection outcome; The hospital infection knowledge graph includes entities and relationships; the entity types of the entities include symptoms, test results, examination results, and diagnosis results of each infection site; the relationships include relationships between different symptoms, relationships between different test results, relationships between different examination results, relationships between symptoms and test results, relationships between symptoms and examination results, relationships between test results and examination results, relationships between symptoms and diagnosis results, relationships between examination results and diagnosis results, and relationships between test results and diagnosis results; The relationship vector set includes a premise entity, a conclusion entity, and a weight coefficient, wherein the weight coefficient is used to evaluate the credibility of the patient's physiological characteristics belonging to the conclusion entity based on the patient's physiological characteristics belonging to the premise entity. A relationship with a higher credibility has a corresponding larger weight coefficient. When the relationship type is a subordinate relationship, the value of the weight coefficient of the corresponding relationship is equal to the positive example ratio; wherein, the positive example ratio is obtained based on the ratio of historical patients belonging to the conclusion entity among historical patients belonging to the premise entity; when the relationship type is an association relationship, the value of the weight coefficient of the corresponding relationship is equal to the absolute value of the difference between the positive example ratio and the negative example ratio; wherein, the negative example ratio is obtained based on the ratio of historical patients belonging to the conclusion entity among historical patients who do not belong to the premise entity; The step of searching the hospital infection knowledge graph according to the target physiological characteristics further includes: Matching entities in the hospital infection knowledge graph according to the target physiological characteristics, and taking the successfully matched entity as the target entity; Determine whether the number of the target entity is 0, if so, terminate the search, if not, update the target subgraph; query the hospital infection knowledge graph for all relationships where the premise entity is the target entity, and obtain candidate relationships; Matching each candidate relationship with the target subgraph, and removing the candidate relationship that successfully matches the target subgraph; The conclusion entities corresponding to the remaining candidate relations are taken as candidate entities, and all the candidate relations are divided into a plurality of target relation sets according to the conclusion entities; each target relation set is divided into a plurality of target relation subsets according to the relation type; Obtaining a screening weight corresponding to each target relationship subset; if the relationship type corresponding to the target relationship subset is a subordinate relationship, taking the maximum value of the weight coefficients in the corresponding target relationship subset as the weight coefficient; if the relationship type corresponding to the target relationship subset is an association relationship, counting the number of relationships in the target relationship subset whose positive example ratio is 0 or whose negative example ratio is 0 to obtain a first value, and calculating the sum of all the weight coefficients in the target relationship subset and the first value to obtain the screening weight; Obtaining a screening threshold corresponding to each target relationship subset, where the value of the screening threshold is greater than 0; comparing the screening weight corresponding to each target relationship subset with the corresponding screening threshold, and when the screening weight is less than the screening threshold, setting the screening result of the corresponding target relationship subset to fail; When all the screening results corresponding to the target relationship set are rejected, the corresponding relationship is removed from the candidate relationship, the corresponding candidate entity is removed from the target entity, the remaining candidate entity is used as the new target entity, and the step of determining whether the number of target entities is 0 is returned to continue execution.

2. The agent-based large-model hospital infection determination method according to claim 1, characterized in that: The entity type also includes the diagnostic criteria for each infection site; the relationship between the symptoms and the diagnostic results is obtained based on the relationship between the corresponding symptoms and the diagnostic criteria, and the relationship between the diagnostic criteria and the corresponding diagnostic results; the relationship between the examination results and the diagnostic results is obtained based on the relationship between the corresponding examination results and the diagnostic criteria, and the relationship between the diagnostic criteria and the corresponding diagnostic results; The relationship between the test results and the diagnosis results is obtained based on the relationship between the corresponding test results and the diagnostic criteria, and the relationship between the diagnostic criteria and the corresponding diagnosis results.

3. The large-scale model hospital infection determination method based on an agent according to claim 2, characterized in that: After inputting the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model, the method further includes obtaining the actual infection outcome of the target patient and updating the weight coefficient in the hospital infection knowledge graph according to the actual infection outcome.

4. The large-scale model hospital infection determination system based on intelligent agent is characterized by: It includes a knowledge graph construction module, a physiological feature acquisition module, a knowledge graph retrieval module, a relationship vector generation module, and an infection outcome prediction module; The knowledge graph construction module is used to construct a hospital infection knowledge graph based on the hospital infection diagnosis standard document; The physiological characteristic acquisition module is used to acquire target physiological characteristics of a target patient; the target physiological characteristics include at least one of a description of the course of disease, an examination result, and a test result of the target patient; The knowledge graph retrieval module is used to retrieve the hospital infection knowledge graph according to the target physiological characteristics to obtain the corresponding target sub-graph; The relationship vector generation module is used to generate a relationship vector set according to the target subgraph to obtain a target relationship vector set; The infection outcome prediction module is used to input the target physiological characteristics and the target relationship vector set into a pre-trained hospital infection prediction model to obtain the target infection outcome; In the knowledge graph construction module, the hospital infection knowledge graph includes entities and relationships; the entity types of the entities include symptoms, test results, examination results, and diagnosis results of each infection site; the relationships include relationships between different symptoms, relationships between different test results, relationships between different examination results, relationships between symptoms and test results, relationships between symptoms and examination results, relationships between test results and examination results, relationships between symptoms and diagnosis results, relationships between examination results and diagnosis results, and relationships between test results and diagnosis results; The relationship vector set includes a premise entity, a conclusion entity, and a weight coefficient, wherein the weight coefficient is used to evaluate the credibility of the patient's physiological characteristics belonging to the conclusion entity based on the patient's physiological characteristics belonging to the premise entity. A relationship with a higher credibility has a corresponding larger weight coefficient. When the relationship type is a subordinate relationship, the value of the weight coefficient of the corresponding relationship is equal to the positive example ratio; wherein, the positive example ratio is obtained based on the ratio of historical patients belonging to the conclusion entity among historical patients belonging to the premise entity; when the relationship type is an association relationship, the value of the weight coefficient of the corresponding relationship is equal to the absolute value of the difference between the positive example ratio and the negative example ratio; wherein, the negative example ratio is obtained based on the ratio of historical patients belonging to the conclusion entity among historical patients who do not belong to the premise entity; The knowledge graph retrieval module also includes: Matching entities in the hospital infection knowledge graph according to the target physiological characteristics, and taking the successfully matched entity as the target entity; Determine whether the number of the target entity is 0, if so, terminate the search, if not, update the target subgraph; query the hospital infection knowledge graph for all relationships where the premise entity is the target entity, and obtain candidate relationships; Matching each candidate relationship with the target subgraph, and removing the candidate relationship that successfully matches the target subgraph; The conclusion entities corresponding to the remaining candidate relations are taken as candidate entities, and all the candidate relations are divided into a plurality of target relation sets according to the conclusion entities; each target relation set is divided into a plurality of target relation subsets according to the relation type; Obtaining a screening weight corresponding to each target relationship subset; if the relationship type corresponding to the target relationship subset is a subordinate relationship, taking the maximum value of the weight coefficients in the corresponding target relationship subset as the weight coefficient; if the relationship type corresponding to the target relationship subset is an association relationship, counting the number of relationships in the target relationship subset whose positive example ratio is 0 or whose negative example ratio is 0 to obtain a first value, and calculating the sum of all the weight coefficients in the target relationship subset and the first value to obtain the screening weight; Obtaining a screening threshold corresponding to each target relationship subset, where the value of the screening threshold is greater than 0; comparing the screening weight corresponding to each target relationship subset with the corresponding screening threshold, and when the screening weight is less than the screening threshold, setting the screening result of the corresponding target relationship subset to fail; When all the screening results corresponding to the target relationship set are rejected, the corresponding relationship is removed from the candidate relationship, the corresponding candidate entity is removed from the target entity, the remaining candidate entity is used as the new target entity, and the step of determining whether the number of target entities is 0 is returned to continue execution.

Images