A design method of structured report template based on semantic association

Abstract

The application discloses a kind of based on the design method of structured report template of semasiology association, comprising the following steps: step S1, based on historical medical record big data semantics construction obtains first diagnosis and treatment knowledge graph;Step S2, based on the construction of second diagnosis and treatment knowledge graph of disease diagnosis and treatment guideline big data;Step S3, the first diagnosis and treatment knowledge graph and second diagnosis and treatment knowledge graph are fused to obtain the diagnosis and treatment structured knowledge graph of fusion representation diagnosis and treatment practical experience and diagnosis and treatment expert experience, and based on diagnosis and treatment structured knowledge graph constructs structured report template for disease category;Step S4, according to structured report template, uniform diagnosis and treatment is carried out to disease category to improve diagnosis and treatment standardization.The application makes it in accordance with structured report template to carry out disease diagnosis and treatment, that is, it is in accordance with diagnosis and treatment practical experience of doctor also in accordance with diagnosis and treatment expert experience, realizes diagnosis and treatment standardization and maneuverability, breaks through the limitation that computer-aided diagnosis method only uses diagnosis and treatment guideline driving.

Description

Technical Field

[0001] This invention relates to the field of medical report technology, specifically to a design method for a structured report template based on semantic association. Background Technology

[0002] Structured reports, compared to traditional reports, offer a more structured and standardized content. Traditional reports are often written according to individual physician habits, resulting in complex logic, diverse terminology, and difficulty in effectively extracting valuable information, thus hindering information management and utilization. This not only wastes a significant amount of valuable medical record information but also easily leads to omissions and errors in report writing. In the era of big data and artificial intelligence, structured medical record information is the most fundamental data. Therefore, the widespread adoption and application of structured reports is urgently needed.

[0003] The prior art, CN202110892281.6, discloses a method for designing a structured report for prostate MR cancer, including the following steps: logging into the user interface and providing multiple data options; filling in the prostate's PI_RADS score and inserting PACS images in the structured report template, and then uploading it to the database; the database receives the uploaded data and classifies and archives key information; the database saves the received information; a query request is entered in the user interface and sent to the database; the database receives the request and matches and filters the data information in the request with existing information in the database; the database returns all the data the user wants to query and automatically generates a structured report.

[0004] Although the aforementioned existing technologies can generate structured reports, the need for doctors to select examination indicators still results in individualized structured reports with poor standardization. Furthermore, the generated structured reports rely on the subjectivity of doctors, leading to low objectivity and credibility. Summary of the Invention

[0005] The purpose of this invention is to provide a design method for structured report templates based on semantic association, in order to solve the technical problems in the prior art that still require doctors to select examination indicators, resulting in structured reports that vary from person to person, have poor standardization, and rely on the subjectivity of doctors, thus leading to low objectivity of structured reports.

[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0007] A method for designing structured report templates based on semantic association includes the following steps:

[0008] Step S1: Based on historical medical record big data, semantically extract the first disease category and the first examination item as the first entity and the second entity, respectively, and semantically extract the relational attributes of the first disease category and the first examination item as the first entity relation. Construct a knowledge graph of the first entity, the second entity and the first entity relation to obtain the first diagnosis and treatment knowledge graph. The first entity relation is represented as the deterministic relation and the extended relation of the first disease category and the first examination item determined by clinical practice experience.

[0009] Step S2: Based on the big data of disease diagnosis and treatment guidelines, the second disease category and the second examination item are extracted as the third entity and the fourth entity, respectively. The relationship attribute of the second disease category and the second examination item is extracted as the second entity relationship. The third entity, the fourth entity and the second entity relationship are constructed into a knowledge graph to obtain the second diagnosis and treatment knowledge graph. The second entity relationship is represented as the deterministic relationship and the extended relationship of the second disease category and the second examination item determined by the experience of diagnosis and treatment experts.

[0010] Step S3: Perform entity fusion on the first and second diagnostic and treatment knowledge graphs to obtain a structured diagnostic and treatment knowledge graph that integrates practical diagnostic and treatment experience and the experience of diagnostic and treatment experts. Based on the structured diagnostic and treatment knowledge graph, construct a structured report template for the disease category.

[0011] Step S4: Standardize the diagnosis and treatment of disease categories according to the structured report template to improve the standardization of diagnosis and treatment.

[0012] If there is no corresponding structured report template for the disease category, return to step S1 and execute steps S1 to S3 in sequence to update the diagnosis and treatment structured knowledge graph;

[0013] If a corresponding structured report template exists for a disease category, then the corresponding structured report template can be used directly for diagnosis and treatment.

[0014] As a preferred embodiment of the present invention, the step of semantically extracting the first symptom category and the first examination item based on historical medical record big data as the first entity and the second entity, respectively, includes:

[0015] A set of historical medical records is randomly selected from the historical medical record big data, and the semantic text representing the first disease category and the first examination item in the historical medical records is extracted as the first entity and the second entity, respectively. The historical medical records include the semantic text representing the first disease category and the first examination item, as well as the semantic text representing the relational attributes of the first disease category and the first examination item.

[0016] The historical medical records, the first entity, and the second entity are quantized from text form into vector form to obtain the semantic vector of the historical medical records, the vector of the first entity, and the vector of the second entity. The semantic vector of the historical medical records, the vector of the first entity, and the vector of the second entity are respectively labeled as a single first entity sample and a single second entity sample. The single first entity sample is represented as [the semantic vector of the historical medical records, the vector of the first entity], and the single second entity sample is represented as [the semantic vector of the historical medical records, the vector of the second entity].

[0017] 70% of the total number of samples are randomly selected from all the first entity samples as the first training set, and the remaining 30% are selected as the first test set. The first training set and the first test set are used to train the CRF model to obtain the first entity extraction model.

[0018] 70% of the total number of samples are randomly selected from all the second entity samples as the second training set, and the remaining 30% are selected as the second test set. The second training set and the second test set are used to train the CRF model to obtain the second entity extraction model.

[0019] The first entity extraction model and the second entity extraction model are used to perform semantic extraction of entities in historical medical record big data to obtain the first entity and the second entity;

[0020] Preferably, before training the CRF model using the first entity samples and the second entity samples, sample redundancy needs to be removed, wherein,

[0021] The similarity between any two first entity samples / second entity samples is calculated iteratively, and one of the two first entity samples / second entity samples whose similarity exceeds a set threshold is randomly removed, until the similarity between any two first entity samples / second entity samples does not exceed the set threshold.

[0022] The formula for calculating the similarity is:

[0023] ;

[0024] In the formula, I represents the similarity value. , These are represented as the i-th and j-th first entity samples / second entity samples, respectively, where i and j are measurement constants.

[0025] As a preferred embodiment of the present invention, the semantic extraction of the relational attributes between the first disease category and the first examination item as the first entity relation includes:

[0026] After removing sample redundancy, the semantic text representing the relationship attributes between the first disease category and the first examination item in the historical medical records corresponding to the first entity sample / second entity sample is extracted as the first entity relationship.

[0027] The historical medical records and the first entity relationship are quantized from text form into vector form to obtain the semantic vector of the historical medical records and the semantic vector of the first entity relationship. The semantic vector of the historical medical records and the semantic vector of the first entity relationship are labeled as a single first entity relationship sample, wherein the single first entity relationship sample is represented as [the semantic vector of the historical medical records and the semantic vector of the first entity relationship];

[0028] 70% of the total number of samples in all first entity relation samples are randomly selected as the first relation training set, and the remaining 30% are selected as the first relation test set. The first relation training set and the first relation test set are used to train the BP neural network to obtain the first entity relation extraction model.

[0029] The first entity relationship extraction model is used to extract the semantic attributes of the relationship from historical medical record big data to obtain the first entity relationship.

[0030] As a preferred embodiment of the present invention, the step of constructing a first diagnostic and treatment knowledge graph from the first entity, the second entity, and the relationship between the first entities includes:

[0031] The first entity, the second entity, and the first entity relationship extracted from the same historical case are connected by a graph to form a subgraph of the form first entity-first entity relationship-second entity, and the priority of merging the subgraphs is set as first entity, first entity relationship and second entity in sequence;

[0032] All subgraphs obtained from historical medical record big data are merged sequentially according to their merging priority.

[0033] The nodes of the subgraphs with the same first entity are merged at the first entity. Then, the nodes of the graph structures with the same first entity relationship are merged at the first entity relationship in the merged subgraphs. Finally, the nodes with the same second entity are merged in the merged graph structure, so as to merge all the subgraphs to generate the first diagnosis and treatment knowledge graph.

[0034] Preferably, determining that the first entity is the same includes:

[0035] All semantic names of the disease categories represented by the first entity are summarized to form a standard name lookup table for diseases. Based on the standard name lookup table for diseases, the semantic names of the disease categories represented by all the first entities are converted into standard semantic names. All first entities that can be converted into the same standard semantic name are determined to be the same first entity.

[0036] Preferably, determining that the second entity is the same includes:

[0037] All semantic names of the inspection items represented by the second entity are summarized to form a standard name lookup table. Based on the standard name lookup table, the semantic names of all inspection items represented by the second entity are converted into standard semantic names. All second entities that can be converted into the same standard semantic name are determined to be the same second entity.

[0038] As a preferred embodiment of the present invention, the extraction of the second disease category and the second examination item based on big data semantics of disease diagnosis and treatment guidelines is respectively used as the third entity and the fourth entity, including:

[0039] A set of disease diagnosis and treatment guidelines is randomly selected from the big data of disease diagnosis and treatment guidelines, and the semantic text representing the second disease category and the second examination item in the disease diagnosis and treatment guidelines is extracted as the third entity and the fourth entity, respectively. The disease diagnosis and treatment guidelines include semantic text representing the second disease category and the second examination item, as well as semantic text representing the relational attribute of the second disease category and the second examination item.

[0040] The disease diagnosis and treatment guidelines, the third entity, and the fourth entity are quantized from text form into vector form to obtain the semantic vector of the disease diagnosis and treatment guidelines, the third entity vector, and the fourth entity vector. The semantic vector of the disease diagnosis and treatment guidelines, the third entity vector, and the fourth entity vector are respectively labeled as a single third entity sample and a single fourth entity sample. The single third entity sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines, the third entity vector], and the single fourth entity sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines, the fourth entity vector];

[0041] 70% of the total number of samples are randomly selected from all third entity samples as the third training set, and the remaining 30% are selected as the third test set. The third training set and the third test set are used to train the CRF model to obtain the third entity extraction model.

[0042] 70% of the total number of samples are randomly selected from all fourth entity samples as the fourth training set, and the remaining 30% are selected as the fourth test set. The fourth training set and the fourth test set are used to train the CRF model to obtain the fourth entity extraction model.

[0043] The third entity extraction model and the fourth entity extraction model are used to extract the semantics of entities from the big data of disease diagnosis and treatment guidelines to obtain the third entity and the fourth entity.

[0044] In a preferred embodiment of the present invention, the semantic extraction of the relational attributes between the second symptom category and the second examination item is used as a second entity relation.

[0045] In a set of disease diagnosis and treatment guidelines, the semantic text representing the relational attributes of the second disease category and the second examination item is extracted as the second entity relation;

[0046] The disease diagnosis and treatment guidelines and the second entity relationship are quantized from text form into vector form to obtain the semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship. The semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship are respectively labeled as a single second entity relationship sample, wherein the single second entity relationship sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship];

[0047] 70% of the total number of samples in all second entity relation samples are randomly selected as the second relation training set, and the remaining 30% are selected as the second relation test set. The second relation training set and the second relation test set are used to train the BP neural network to obtain the second entity relation extraction model.

[0048] The second entity relationship extraction model is used to extract the semantic attributes of the relationship in the big data of disease diagnosis and treatment guidelines to obtain the second entity relationship.

[0049] As a preferred embodiment of the present invention, the step of constructing a second diagnostic and treatment knowledge graph by performing knowledge graph construction on the relationship between the third entity, the fourth entity, and the second entity includes:

[0050] The third entity, fourth entity, and second entity relationship extracted from the same disease diagnosis and treatment guidelines are connected by a graph to form a subgraph of the form of third entity-second entity relationship-fourth entity, and the priority of merging the subgraphs is set to third entity, second entity relationship and fourth entity in sequence;

[0051] All subgraphs obtained from the big data of disease diagnosis and treatment guidelines are merged sequentially according to their merging priority.

[0052] The nodes of the subgraphs with the same third entity are merged at the third entity. Then, the nodes of the graph structures with the same second entity relationship are merged at the second entity relationship in the merged subgraph. Finally, the nodes with the same fourth entity are merged in the merged graph structure to merge all the subgraphs to generate the second diagnosis and treatment knowledge graph.

[0053] As a preferred embodiment of the present invention, the step of entity fusion of the first and second diagnostic and treatment knowledge graphs to obtain a structured diagnostic and treatment knowledge graph that integrates and represents practical diagnostic and treatment experience and expert diagnostic and treatment experience includes:

[0054] The merging priority of all graph structures in the first and second diagnostic knowledge graphs is set sequentially as first entity / third entity, first entity relation / second entity relation, and second entity / fourth entity, and the graph structures are merged according to the merging priority.

[0055] The graph structures with the same first entity and third entity are merged at the first entity's node. Then, in the merged graph structure, the graph structures with the same first entity relationship and second entity relationship are merged at the first entity relationship. Finally, the nodes with the same second entity and fourth entity are merged in the merged graph structure. This process merges all graph structures in the first and second diagnostic knowledge graphs to generate the structured diagnostic knowledge graph. This achieves a structured integration of practical diagnostic experience and expert experience, balancing standardization and flexibility in the structured diagnosis and treatment of disease categories.

[0056] As a preferred embodiment of the present invention, the step of constructing a structured report template for disease categories based on a structured diagnostic knowledge graph includes:

[0057] The system retrieves all examination items corresponding to the disease category determined by the doctor by querying the structured knowledge graph of diagnosis and treatment, and integrates all examination items into a single report template to generate a structured report template that serves as a guide for doctors in the diagnosis and treatment of the corresponding disease category.

[0058] As a preferred embodiment of the present invention, the updating of the diagnostic structured knowledge graph includes:

[0059] When there is no corresponding structured report template for a disease category, the doctor creates a medical record for the disease category and constructs a subgraph in the form of a first entity-first entity relationship and a second entity based on the medical record execution step S1. This subgraph is then added to the first medical knowledge graph for graph structure merging to update the first medical knowledge graph.

[0060] Step S3 is executed to perform entity fusion between the updated first diagnostic and treatment knowledge graph and the second diagnostic and treatment knowledge graph to update the diagnostic and treatment structured knowledge graph, thereby expanding the diagnostic and treatment structured knowledge graph.

[0061] Compared with the prior art, the present invention has the following advantages:

[0062] This invention constructs a structured knowledge graph of diagnosis and treatment based on the fusion of historical medical record big data and disease diagnosis and treatment guideline big data. This graph represents the practical experience of diagnosis and treatment and the experience of diagnosis and treatment experts. Based on the structured knowledge graph of diagnosis and treatment, structured report templates are constructed for disease categories. This allows disease diagnosis and treatment to be carried out in accordance with both the practical experience of doctors and the experience of diagnosis and treatment experts, achieving a balance between standardization and flexibility in diagnosis and treatment. This breaks through the limitation of computer-aided diagnosis methods that are only driven by diagnosis and treatment guidelines. Attached Figure Description

[0063] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0064] Figure 1 A flowchart of a design method for a structured report template based on semantic association is provided for embodiments of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] like Figure 1 As shown, this invention provides a method for designing a structured report template based on semantic association, including the following steps:

[0067] Step S1: Based on historical medical record big data, semantically extract the first disease category and the first examination item as the first entity and the second entity, respectively, and semantically extract the relational attributes of the first disease category and the first examination item as the first entity relation. Construct a knowledge graph of the first entity, the second entity and the first entity relation to obtain the first diagnosis and treatment knowledge graph. The first entity relation is represented as the deterministic relation and the extended relation of the first disease category and the first examination item determined by clinical practice experience.

[0068] Each doctor has their own practical experience in determining the necessary examinations for a condition. This self-assessment allows them to determine the required tests, which the patient then follows. After each test, the data is recorded in the patient's medical record. Mastering a doctor's practical experience in treating a condition reflects their accumulated clinical experience, enabling rapid and accurate initial diagnosis and selection of necessary tests. This allows for quick control of the patient's condition. Therefore, this embodiment constructs a first diagnostic knowledge graph based on large-scale medical record data, representing practical diagnostic experience. This allows doctors to combine practical diagnostic knowledge with patient diagnosis, demonstrating flexibility in diagnosis—selecting tests based on the severity and urgency of the condition—to provide medical assistance in diagnosis.

[0069] The first symptom category and the first examination item extracted from historical medical record big data are respectively designated as the first entity and the second entity, including:

[0070] A set of historical medical records is randomly selected from the historical medical record big data, and the semantic text representing the first disease category and the first examination item in the historical medical records is extracted as the first entity and the second entity, respectively. The historical medical records include the semantic text representing the first disease category and the first examination item, as well as the semantic text representing the relational attributes of the first disease category and the first examination item.

[0071] The historical medical records, the first entity, and the second entity are quantized from text form into vector form to obtain the semantic vector of the historical medical records, the vector of the first entity, and the vector of the second entity. The semantic vector of the historical medical records, the vector of the first entity, and the vector of the second entity are respectively labeled as a single first entity sample and a single second entity sample. The single first entity sample is represented as [the semantic vector of the historical medical records, the vector of the first entity], and the single second entity sample is represented as [the semantic vector of the historical medical records, the vector of the second entity].

[0072] 70% of the total number of samples are randomly selected from all the first entity samples as the first training set, and the remaining 30% are selected as the first test set. The first training set and the first test set are used to train the CRF model to obtain the first entity extraction model.

[0073] 70% of the total number of samples are randomly selected from all the second entity samples as the second training set, and the remaining 30% are selected as the second test set. The second training set and the second test set are used to train the CRF model to obtain the second entity extraction model.

[0074] In this embodiment, all 70% and 30% are adjustable data, and users can customize and modify them according to their actual needs.

[0075] The first entity extraction model and the second entity extraction model are used to perform semantic extraction of entities in historical medical record big data to obtain the first entity and the second entity;

[0076] Preferably, before training the CRF model using the first entity samples and the second entity samples, sample redundancy needs to be removed to avoid redundant data consuming computational resources for model training and to improve the efficiency of model training.

[0077] The similarity between any two first entity samples / second entity samples is calculated iteratively, and one of the two first entity samples / second entity samples whose similarity exceeds a set threshold is randomly removed, until the similarity between any two first entity samples / second entity samples does not exceed the set threshold.

[0078] The formula for calculating the similarity is:

[0079] ;

[0080] In the formula, I represents the similarity value. , These are represented as the i-th and j-th first entity samples / second entity samples, respectively, where i and j are measurement constants.

[0081] The higher the similarity between two first entity samples / second entity samples, the more similar they are. In this way, one of them can be randomly used for representation. This retains the entity sample type while removing redundant items in that entity sample type, achieving data simplification from multiple to one, and ultimately achieving the goal of retaining sample diversity while removing sample redundancy.

[0082] The semantic extraction of the relational attributes between the first symptom category and the first examination item as the first entity relation includes:

[0083] After removing sample redundancy, the semantic text representing the relationship attributes between the first disease category and the first examination item in the historical medical records corresponding to the first entity sample / second entity sample is extracted as the first entity relationship.

[0084] The historical medical records and the first entity relationship are quantized from text form into vector form to obtain the semantic vector of the historical medical records and the semantic vector of the first entity relationship. The semantic vector of the historical medical records and the semantic vector of the first entity relationship are labeled as a single first entity relationship sample, wherein the single first entity relationship sample is represented as [the semantic vector of the historical medical records and the semantic vector of the first entity relationship];

[0085] 70% of the total number of samples in all first entity relation samples are randomly selected as the first relation training set, and the remaining 30% are selected as the first relation test set. The first relation training set and the first relation test set are used to train the BP neural network to obtain the first entity relation extraction model.

[0086] The first entity relationship extraction model is used to extract the semantic attributes of the relationship from historical medical record big data to obtain the first entity relationship.

[0087] The step of constructing a first diagnostic and treatment knowledge graph from the first entity, the second entity, and the relationships between the first entities includes:

[0088] The first entity, the second entity, and the first entity relationship extracted from the same historical case are connected by a graph to form a subgraph of the form first entity-first entity relationship-second entity, and the priority of merging the subgraphs is set as first entity, first entity relationship and second entity in sequence;

[0089] All subgraphs obtained from historical medical record big data are merged sequentially according to their merging priority.

[0090] The nodes of the subgraphs with the same first entity are merged at the first entity. Then, the nodes of the graph structures with the same first entity relationship are merged at the first entity relationship in the merged subgraphs. Finally, the nodes with the same second entity are merged in the merged graph structure, so as to merge all the subgraphs to generate the first diagnosis and treatment knowledge graph.

[0091] Preferably, determining that the first entity is the same includes:

[0092] All semantic names of the disease categories represented by the first entity are summarized to form a standard name lookup table for diseases. Based on the standard name lookup table for diseases, the semantic names of the disease categories represented by all the first entities are converted into standard semantic names. All first entities that can be converted into the same standard semantic name are determined to be the same first entity.

[0093] Preferably, determining that the second entity is the same includes:

[0094] All semantic names of the inspection items represented by the second entity are summarized to form a standard name lookup table. Based on the standard name lookup table, the semantic names of all inspection items represented by the second entity are converted into standard semantic names. All second entities that can be converted into the same standard semantic name are determined to be the same second entity.

[0095] The need to determine entity similarity is because doctors may use different aliases instead of standard semantic names (scientific names) to refer to disease categories due to different habits. Therefore, determining similarity can identify entities representing the same disease category and improve the accuracy of knowledge graph construction.

[0096] This embodiment provides an example of constructing a first diagnostic knowledge graph. For instance, there are four subgraphs (for ease of description, we set it to four; in reality, the number of subgraphs far exceeds four): First Entity A - First Entity Relation 1 - Second Entity a, First Entity B - First Entity Relation 2 - Second Entity c, First Entity C - First Entity Relation 1 - Second Entity b, and First Entity A - First Entity Relation 2 - Second Entity b. Firstly, we determine the subgraphs with the same first entity. Then, we convert the first and second entities of all subgraphs into standard semantic names, resulting in: First Entity A - First Entity Relation 1 - Second Entity a, First Entity B - First Entity Relation 2 - Second Entity c, First Entity A - First Entity Relation 1 - Second Entity b, and First Entity A - First Entity Relation 2 - Second Entity c. We find that three of these have the same first entity, which are then merged to obtain: (or In the merged graph structure, two identical first entity relations are found. Merging these first entity relations yields: (or Since there are no identical third entities, the final first diagnostic knowledge graph is: (or ).

[0097] Entity relationships 1 and 2 represent deterministic and extended relationships, respectively. Deterministic relationships refer to the examinations that must be performed for this disease category, while extended relationships refer to the examinations that can be performed for this disease category, such as those used to determine whether complications exist.

[0098] Step S2: Based on the big data of disease diagnosis and treatment guidelines, the second disease category and the second examination item are extracted as the third entity and the fourth entity, respectively. The relationship attribute of the second disease category and the second examination item is extracted as the second entity relationship. The third entity, the fourth entity and the second entity relationship are constructed into a knowledge graph to obtain the second diagnosis and treatment knowledge graph. The second entity relationship is represented as the deterministic relationship and the extended relationship of the second disease category and the second examination item determined by the experience of diagnosis and treatment experts.

[0099] Doctors' practical clinical experience is somewhat limited, while disease diagnosis and treatment guidelines are guidelines developed by medical experts from various fields to guide physicians in clinical diagnosis. Therefore, this embodiment constructs a second diagnostic and treatment knowledge graph based on big data from disease diagnosis and treatment guidelines to represent the experience of diagnostic and treatment experts. This allows for the integration of diagnostic and treatment experts' experience and knowledge into patient diagnosis, compensating for the limitations of doctors' practical experience in diagnosis. This enables a more comprehensive and standardized diagnosis of patients, thereby improving the standardization and comprehensiveness of diagnosis and providing medical assistance for patient diagnosis.

[0100] The second disease category and the second examination item, extracted from big data semantics based on disease diagnosis and treatment guidelines, are respectively designated as the third and fourth entities, including:

[0101] A set of disease diagnosis and treatment guidelines is randomly selected from the big data of disease diagnosis and treatment guidelines, and the semantic text representing the second disease category and the second examination item in the disease diagnosis and treatment guidelines is extracted as the third entity and the fourth entity, respectively. The disease diagnosis and treatment guidelines include semantic text representing the second disease category and the second examination item, as well as semantic text representing the relational attribute of the second disease category and the second examination item.

[0102] The disease diagnosis and treatment guidelines, the third entity, and the fourth entity are quantized from text form into vector form to obtain the semantic vector of the disease diagnosis and treatment guidelines, the third entity vector, and the fourth entity vector. The semantic vector of the disease diagnosis and treatment guidelines, the third entity vector, and the fourth entity vector are respectively labeled as a single third entity sample and a single fourth entity sample. The single third entity sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines, the third entity vector], and the single fourth entity sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines, the fourth entity vector];

[0103] 70% of the total number of samples are randomly selected from all third entity samples as the third training set, and the remaining 30% are selected as the third test set. The third training set and the third test set are used to train the CRF model to obtain the third entity extraction model.

[0104] 70% of the total number of samples are randomly selected from all fourth entity samples as the fourth training set, and the remaining 30% are selected as the fourth test set. The fourth training set and the fourth test set are used to train the CRF model to obtain the fourth entity extraction model.

[0105] The third entity extraction model and the fourth entity extraction model are used to extract the semantics of entities from the big data of disease diagnosis and treatment guidelines to obtain the third entity and the fourth entity.

[0106] The semantic extraction of the relational attributes between the second symptom category and the second examination item is used as the second entity relation.

[0107] In a set of disease diagnosis and treatment guidelines, the semantic text representing the relational attributes of the second disease category and the second examination item is extracted as the second entity relation;

[0108] The disease diagnosis and treatment guidelines and the second entity relationship are quantized from text form into vector form to obtain the semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship. The semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship are respectively labeled as a single second entity relationship sample, wherein the single second entity relationship sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship];

[0109] 70% of the total number of samples in all second entity relation samples are randomly selected as the second relation training set, and the remaining 30% are selected as the second relation test set. The second relation training set and the second relation test set are used to train the BP neural network to obtain the second entity relation extraction model.

[0110] The second entity relationship extraction model is used to extract the semantic attributes of the relationship in the big data of disease diagnosis and treatment guidelines to obtain the second entity relationship.

[0111] The step of constructing a second diagnostic and treatment knowledge graph by performing a knowledge graph on the relationships between the third entity, the fourth entity, and the second entity includes:

[0112] The third entity, fourth entity, and second entity relationship extracted from the same disease diagnosis and treatment guidelines are connected by a graph to form a subgraph of the form of third entity-second entity relationship-fourth entity, and the priority of merging the subgraphs is set to third entity, second entity relationship and fourth entity in sequence;

[0113] All subgraphs obtained from the big data of disease diagnosis and treatment guidelines are merged sequentially according to their merging priority.

[0114] The nodes of the subgraphs with the same third entity are merged at the third entity. Then, the nodes of the graph structures with the same second entity relationship are merged at the second entity relationship in the merged subgraph. Finally, the nodes with the same fourth entity are merged in the merged graph structure to merge all the subgraphs to generate the second diagnosis and treatment knowledge graph.

[0115] The construction process of the second diagnostic and treatment knowledge graph is similar to that of the first diagnostic and treatment knowledge graph. However, the second diagnostic and treatment knowledge graph does not require the removal of sample redundancy and the determination of entity similarity, because redundancy removal was carried out during the compilation process of the disease diagnosis and treatment guidelines. In addition, the third and fourth entities are presented with standard semantic names (scientific names).

[0116] Step S3: Perform entity fusion on the first and second diagnostic and treatment knowledge graphs to obtain a structured diagnostic and treatment knowledge graph that integrates practical diagnostic and treatment experience and the experience of diagnostic and treatment experts. Based on the structured diagnostic and treatment knowledge graph, construct a structured report template for the disease category.

[0117] The step of fusing the first and second diagnostic knowledge graphs to obtain a structured diagnostic knowledge graph that integrates practical diagnostic experience and expert diagnostic experience includes:

[0118] The merging priority of all graph structures in the first and second diagnostic knowledge graphs is set sequentially as first entity / third entity, first entity relation / second entity relation, and second entity / fourth entity, and the graph structures are merged according to the merging priority.

[0119] The graph structures with the same first entity and third entity are merged at the first entity's node. Then, in the merged graph structure, the graph structures with the same first entity relationship and second entity relationship are merged at the first entity relationship. Finally, the nodes with the same second entity and fourth entity are merged in the merged graph structure. This process merges all graph structures in the first and second diagnostic knowledge graphs to generate the structured diagnostic knowledge graph. This achieves a structured integration of practical diagnostic experience and expert experience, balancing standardization and flexibility in the structured diagnosis and treatment of disease categories.

[0120] This embodiment provides an example of constructing a structured knowledge graph for diagnosis and treatment. The first knowledge graph for diagnosis and treatment is: or The second diagnostic knowledge graph is as follows: or Merge the nodes of the same graph structure in the first entity and the third entity at the first entity to obtain: In the merged graph structure, nodes with the same first entity relation and second entity relation are merged at the first entity relation, resulting in: In the merged graph structure, the second entity and the fourth identical node are merged to obtain the following structured knowledge graph for diagnosis and treatment: .

[0121] Step S4: Standardize the diagnosis and treatment of disease categories according to the structured report template to improve the standardization of diagnosis and treatment.

[0122] If there is no corresponding structured report template for the disease category, return to step S1 and execute steps S1 to S3 in sequence to update the diagnosis and treatment structured knowledge graph;

[0123] If a corresponding structured report template exists for the disease category, then the corresponding structured report template will be used directly for diagnosis and treatment.

[0124] Assuming the doctor determines the patient's condition to be category A, inputting "A" into a computer terminal containing a structured diagnostic knowledge graph will generate all examination items for lesion category A, and produce a corresponding structured report template: Patients can directly use the corresponding structured report template to examine their symptoms and obtain various indicators of their condition.

[0125] The structured report templates built based on the diagnostic and treatment structured knowledge graph for disease categories include:

[0126] The system retrieves all examination items corresponding to the disease category determined by the doctor by querying the structured knowledge graph of diagnosis and treatment, and integrates all examination items into a single report template to generate a structured report template that serves as a guide for doctors in the diagnosis and treatment of the corresponding disease category.

[0127] The update of the diagnostic structured knowledge graph includes:

[0128] When there is no corresponding structured report template for a disease category, the doctor creates a medical record for the disease category and constructs a subgraph in the form of a first entity-first entity relationship and a second entity based on the medical record execution step S1. This subgraph is then added to the first medical knowledge graph for graph structure merging to update the first medical knowledge graph.

[0129] Step S3 is executed to perform entity fusion between the updated first diagnostic and treatment knowledge graph and the second diagnostic and treatment knowledge graph to update the diagnostic and treatment structured knowledge graph, thereby expanding the diagnostic and treatment structured knowledge graph.

[0130] This invention constructs a structured knowledge graph of diagnosis and treatment based on the fusion of historical medical record big data and disease diagnosis and treatment guideline big data. This graph represents the practical experience of diagnosis and treatment and the experience of diagnosis and treatment experts. Based on the structured knowledge graph of diagnosis and treatment, structured report templates are constructed for disease categories. This allows disease diagnosis and treatment to be carried out in accordance with both the practical experience of doctors and the experience of diagnosis and treatment experts, achieving a balance between standardization and flexibility in diagnosis and treatment. This breaks through the limitation of computer-aided diagnosis methods that are only driven by diagnosis and treatment guidelines.

[0131] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A design method for a structured report template based on semantic association, characterized in that: Includes the following steps: Step S1: Based on historical medical record big data, semantically extract the first disease category and the first examination item as the first entity and the second entity, respectively, and semantically extract the relational attributes of the first disease category and the first examination item as the first entity relation. Construct a knowledge graph of the first entity, the second entity and the first entity relation to obtain the first diagnosis and treatment knowledge graph. The first entity relation is represented as the deterministic relation and the extended relation of the first disease category and the first examination item determined by clinical practice experience. Step S2: Based on the big data of disease diagnosis and treatment guidelines, the second disease category and the second examination item are extracted as the third entity and the fourth entity, respectively. The relationship attribute of the second disease category and the second examination item is extracted as the second entity relationship. The third entity, the fourth entity and the second entity relationship are constructed into a knowledge graph to obtain the second diagnosis and treatment knowledge graph. The second entity relationship is represented as the deterministic relationship and the extended relationship of the second disease category and the second examination item determined by the experience of diagnosis and treatment experts. Step S3: Perform entity fusion on the first and second diagnostic and treatment knowledge graphs to obtain a structured diagnostic and treatment knowledge graph that integrates practical diagnostic and treatment experience and the experience of diagnostic and treatment experts. Based on the structured diagnostic and treatment knowledge graph, construct a structured report template for the disease category. Step S4: Standardize the diagnosis and treatment of disease categories according to the structured report template to improve the standardization of diagnosis and treatment. If there is no corresponding structured report template for the disease category, return to step S1 and execute steps S1 to S3 in sequence to update the diagnosis and treatment structured knowledge graph; If a corresponding structured report template exists for a disease category, then the corresponding structured report template can be used directly for diagnosis and treatment. Based on a structured knowledge graph of diagnosis and treatment, structured report templates are constructed for disease categories, including: The system retrieves all examination items corresponding to the disease category determined by the doctor by querying the structured knowledge graph of diagnosis and treatment, and integrates all examination items into the same report template to generate a structured report template that serves as a guide for doctors in the diagnosis and treatment of the corresponding disease category. Updates to the structured knowledge graph of diagnosis and treatment include: When there is no corresponding structured report template for a disease category, the doctor creates a medical record for the disease category and constructs a subgraph in the form of a first entity-first entity relationship and a second entity based on the medical record execution step S1. This subgraph is then added to the first medical knowledge graph for graph structure merging to update the first medical knowledge graph. Step S3 is executed to perform entity fusion between the updated first diagnostic and treatment knowledge graph and the second diagnostic and treatment knowledge graph to update the diagnostic and treatment structured knowledge graph, thereby expanding the diagnostic and treatment structured knowledge graph.

2. The design method for a structured report template based on semantic association according to claim 1, characterized in that: The first symptom category and the first examination item extracted from historical medical record big data are respectively designated as the first entity and the second entity, including: A set of historical medical records is randomly selected from the historical medical record big data, and the semantic text representing the first disease category and the first examination item in the historical medical records is extracted as the first entity and the second entity, respectively. The historical medical records include the semantic text representing the first disease category and the first examination item, as well as the semantic text representing the relational attributes of the first disease category and the first examination item. The historical medical records, the first entity, and the second entity are quantized from text form into vector form to obtain the semantic vector of the historical medical records, the vector of the first entity, and the vector of the second entity. The semantic vector of the historical medical records, the vector of the first entity, and the vector of the second entity are respectively labeled as a single first entity sample and a single second entity sample. The single first entity sample is represented as [the semantic vector of the historical medical records, the vector of the first entity], and the single second entity sample is represented as [the semantic vector of the historical medical records, the vector of the second entity]. 70% of the total number of samples are randomly selected from all the first entity samples as the first training set, and the remaining 30% are selected as the first test set. The first training set and the first test set are used to train the CRF model to obtain the first entity extraction model. 70% of the total number of samples are randomly selected from all the second entity samples as the second training set, and the remaining 30% are selected as the second test set. The second training set and the second test set are used to train the CRF model to obtain the second entity extraction model. The first entity extraction model and the second entity extraction model are used to perform semantic extraction of entities in historical medical record big data to obtain the first entity and the second entity; Before training the CRF model using the first and second entity samples, sample redundancy needs to be removed. The similarity between any two first entity samples / second entity samples is calculated iteratively, and one of the two first entity samples / second entity samples whose similarity exceeds a set threshold is randomly removed, until the similarity between any two first entity samples / second entity samples does not exceed the set threshold. The formula for calculating the similarity is: ； In the formula, I represents the similarity value. , These are represented as the i-th and j-th first entity samples / second entity samples, respectively, where i and j are measurement constants.

3. The method for designing a structured report template based on semantic association according to claim 2, characterized in that: The semantic extraction of the relational attributes between the first symptom category and the first examination item as the first entity relation includes: After removing sample redundancy, the semantic text representing the relationship attributes between the first disease category and the first examination item in the historical medical records corresponding to the first entity sample / second entity sample is extracted as the first entity relationship. The historical medical records and the first entity relationship are quantized from text form into vector form to obtain the semantic vector of the historical medical records and the semantic vector of the first entity relationship. The semantic vector of the historical medical records and the semantic vector of the first entity relationship are labeled as a single first entity relationship sample, wherein the single first entity relationship sample is represented as [the semantic vector of the historical medical records and the semantic vector of the first entity relationship]; 70% of the total number of samples in all first entity relation samples are randomly selected as the first relation training set, and the remaining 30% are selected as the first relation test set. The first relation training set and the first relation test set are used to train the BP neural network to obtain the first entity relation extraction model. The first entity relationship extraction model is used to extract the semantic attributes of the relationship from historical medical record big data to obtain the first entity relationship.

4. The design method for a structured report template based on semantic association according to claim 3, characterized in that: The step of constructing a first diagnostic and treatment knowledge graph from the first entity, the second entity, and the relationships between the first entities includes: The first entity, the second entity, and the first entity relationship extracted from the same historical case are connected by a graph to form a subgraph of the form first entity-first entity relationship-second entity, and the priority of merging the subgraphs is set as first entity, first entity relationship and second entity in sequence; All subgraphs obtained from historical medical record big data are merged sequentially according to their merging priority. The nodes of the subgraphs with the same first entity are merged at the first entity. Then, the nodes of the graph structures with the same first entity relationship are merged at the first entity relationship in the merged subgraphs. Finally, the nodes with the same second entity are merged in the merged graph structure, so as to merge all the subgraphs to generate the first diagnosis and treatment knowledge graph. Determining that the first entity is the same includes: All semantic names of the disease categories represented by the first entity are summarized to form a standard name lookup table for diseases. Based on the standard name lookup table for diseases, the semantic names of the disease categories represented by all the first entities are converted into standard semantic names. All first entities that can be converted into the same standard semantic name are determined to be the same first entity. Determining that the second entity is the same includes: All semantic names of the inspection items represented by the second entity are summarized to form a standard name lookup table. Based on the standard name lookup table, the semantic names of all inspection items represented by the second entity are converted into standard semantic names. All second entities that can be converted into the same standard semantic name are determined to be the same second entity.

5. The design method for a structured report template based on semantic association according to claim 4, characterized in that: The second disease category and the second examination item, extracted from big data semantics based on disease diagnosis and treatment guidelines, are respectively designated as the third and fourth entities, including: A set of disease diagnosis and treatment guidelines is randomly selected from the big data of disease diagnosis and treatment guidelines, and the semantic text representing the second disease category and the second examination item in the disease diagnosis and treatment guidelines is extracted as the third entity and the fourth entity, respectively. The disease diagnosis and treatment guidelines include semantic text representing the second disease category and the second examination item, as well as semantic text representing the relational attribute of the second disease category and the second examination item. The disease diagnosis and treatment guidelines, the third entity, and the fourth entity are quantized from text form into vector form to obtain the semantic vector of the disease diagnosis and treatment guidelines, the third entity vector, and the fourth entity vector. The semantic vector of the disease diagnosis and treatment guidelines, the third entity vector, and the fourth entity vector are respectively labeled as a single third entity sample and a single fourth entity sample. The single third entity sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines, the third entity vector], and the single fourth entity sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines, the fourth entity vector]; 70% of the total number of samples are randomly selected from all third entity samples as the third training set, and the remaining 30% are selected as the third test set. The third training set and the third test set are used to train the CRF model to obtain the third entity extraction model. 70% of the total number of samples are randomly selected from all fourth entity samples as the fourth training set, and the remaining 30% are selected as the fourth test set. The fourth training set and the fourth test set are used to train the CRF model to obtain the fourth entity extraction model. The third entity extraction model and the fourth entity extraction model are used to extract the semantics of entities from the big data of disease diagnosis and treatment guidelines to obtain the third entity and the fourth entity.

6. The design method for a structured report template based on semantic association according to claim 5, characterized in that: The semantic extraction of the relational attributes between the second symptom category and the second examination item is used as the second entity relation. In a set of disease diagnosis and treatment guidelines, the semantic text representing the relational attributes of the second disease category and the second examination item is extracted as the second entity relation; The disease diagnosis and treatment guidelines and the second entity relationship are quantized from text form into vector form to obtain the semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship. The semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship are respectively labeled as a single second entity relationship sample, wherein the single second entity relationship sample is represented as [the semantic vector of the disease diagnosis and treatment guidelines and the semantic vector of the second entity relationship]; 70% of the total number of samples in all second entity relation samples are randomly selected as the second relation training set, and the remaining 30% are selected as the second relation test set. The second relation training set and the second relation test set are used to train the BP neural network to obtain the second entity relation extraction model. The second entity relationship extraction model is used to extract the semantic attributes of the relationship in the big data of disease diagnosis and treatment guidelines to obtain the second entity relationship.

7. The method for designing a structured report template based on semantic association according to claim 6, characterized in that, The step of constructing a second diagnostic and treatment knowledge graph by performing a knowledge graph on the relationships between the third entity, the fourth entity, and the second entity includes: The third entity, fourth entity, and second entity relationship extracted from the same disease diagnosis and treatment guidelines are connected by a graph to form a subgraph of the form of third entity-second entity relationship-fourth entity, and the priority of merging the subgraphs is set to third entity, second entity relationship and fourth entity in sequence; All subgraphs obtained from the big data of disease diagnosis and treatment guidelines are merged sequentially according to their merging priority. The nodes of the subgraphs with the same third entity are merged at the third entity. Then, the nodes of the graph structures with the same second entity relationship are merged at the second entity relationship in the merged subgraph. Finally, the nodes with the same fourth entity are merged in the merged graph structure to merge all the subgraphs to generate the second diagnosis and treatment knowledge graph.

8. The method for designing a structured report template based on semantic association according to claim 7, characterized in that, The step of fusing the first and second diagnostic knowledge graphs to obtain a structured diagnostic knowledge graph that integrates practical diagnostic experience and expert diagnostic experience includes: The merging priority of all graph structures in the first and second diagnostic knowledge graphs is set sequentially as first entity / third entity, first entity relation / second entity relation, and second entity / fourth entity, and the graph structures are merged according to the merging priority. The graph structures with the same first entity and third entity are merged at the first entity's node. Then, in the merged graph structure, the graph structures with the same first entity relationship and second entity relationship are merged at the first entity relationship. Finally, the nodes with the same second entity and fourth entity are merged in the merged graph structure. This process merges all graph structures in the first and second diagnostic knowledge graphs to generate the structured diagnostic knowledge graph. This achieves a structured integration of practical diagnostic experience and expert experience, balancing standardization and flexibility in the structured diagnosis and treatment of disease categories.

Images