Method, device, storage medium and electronic device for constructing critical disease knowledge base

Abstract

The present disclosure relates to a method, device, storage medium and electronic device for constructing a knowledge base of serious diseases, so as to construct a knowledge base of serious diseases with high reference value in clinical diagnosis. The construction method includes: obtaining relevant data of multiple types of serious diseases, the relevant data including first structured data and unstructured data; inputting the unstructured data into a preprocessing model to obtain second structured data output by the preprocessing model, the preprocessing model is trained based on labeled unstructured sample data of serious diseases; constructing a storage structure based on the attribute items included in the first structured data and the second structured data, and filling the attribute values ​​included in the first structured data and the second structured data into the corresponding positions in the storage structure to obtain the knowledge base of serious diseases, wherein the structured data includes a mapping relationship between attribute items and attribute values.

Description

Technical Field

[0001] The present disclosure relates to the technical field of critical disease knowledge base, and in particular, to a method, device, storage medium and electronic device for constructing a critical disease knowledge base. Background Art

[0002] The important role of critical care medicine in the treatment of critically ill patients has been widely recognized. Especially in the past decade, critical care medicine has developed rapidly and has become one of the most important clinical medical specialties. Critical care medicine is highly professional and comprehensive. It integrates high-end technologies from multiple disciplines such as internal medicine, surgery, anesthesia, nursing, and nutrition. In addition to placing higher demands on medical staff engaged in critical care medicine, these high-end technologies also place high demands on information systems that assist in critical care diagnosis and treatment, especially knowledge bases that play an auxiliary supporting role for information systems. At present, although there are many pharmaceutical and clinical knowledge bases, there is no knowledge base specifically for critical illness. Therefore, a method for constructing a knowledge base for critical illness is needed to achieve the purpose of constructing a knowledge base for critical illness. Summary of the invention

[0003] The purpose of the present disclosure is to provide a method, device, storage medium and electronic device for constructing a critical disease knowledge base, so as to construct a critical disease knowledge base with high reference value in clinical diagnosis.

[0004] In order to achieve the above-mentioned object, the present disclosure provides a method for constructing a critical disease knowledge base in a first aspect, comprising:

[0005] Acquire relevant data of multiple types of serious diseases, wherein the relevant data includes first structured data and unstructured data;

[0006] Inputting the unstructured data into a preprocessing model to obtain second structured data output by the preprocessing model, wherein the preprocessing model is trained based on labeled unstructured sample data of severe diseases;

[0007] A storage structure is constructed based on the attribute items included in the first structured data and the second structured data, and the attribute values ​​included in the first structured data and the second structured data are filled into corresponding positions in the storage structure to obtain the serious disease knowledge base, wherein the structured data includes a mapping relationship between attribute items and attribute values.

[0008] Optionally, the preprocessing model includes a labeling module and an output module. Accordingly, the training process of the preprocessing model includes:

[0009] The unstructured sample data in the preset database is divided into training sample data and verification sample data;

[0010] The training sample data is used as an input parameter, and the labeled result of the training sample data is used as an output parameter to train the labeling module until the labeling accuracy of the labeling module for the verification sample data reaches a preset value, and the training is terminated to obtain the labeling module;

[0011] Performing sequence labeling on the unstructured sample data through the trained labeling module;

[0012] The conversion parameters of the output module are determined according to the sequence annotation results output by the annotation module, the predefined association relationship and the second structured data corresponding to the unstructured sample data, so that the sequence annotation results and the predefined association relationship can obtain the second structured data after the conversion parameters are passed.

[0013] Optionally, performing sequence labeling on the unstructured sample data by the trained labeling module includes:

[0014] Annotating critical care medical knowledge attribute items and disease-related attribute items of the unstructured sample data by using the trained annotation module;

[0015] The critical care medical knowledge attribute items include at least one of the following: clinical medical concepts, medical terms subordinate to the clinical medical concepts, associations between medical terms under the same clinical medical concept, and associations between medical terms under different clinical medical concepts;

[0016] The disease-related attribute items include attribute items related to serious diseases and / or attribute items related to individual patients.

[0017] Optionally, constructing a storage structure based on the attribute items included in the first structured data and the second structured data includes:

[0018] Constructing a data dictionary storage structure based on the critical care medicine knowledge attribute items included in the first structured data and the second structured data;

[0019] A tree storage structure is constructed based on the disease-related attribute items included in the first structured data and the second structured data.

[0020] Optionally, the constructing a tree storage structure based on the disease-related attribute items included in the first structured data and the second structured data includes:

[0021] For each type of serious disease, a tree storage structure is constructed based on the disease-related attribute items included in the first structural data and the second structural data of the serious disease.

[0022] Optionally, the disease-related attribute items include attribute items related to serious diseases and attribute items related to individual patients, and the attribute values ​​included in the first structured data and the second structured data are filled into corresponding positions in the storage structure to obtain the serious disease knowledge base, including:

[0023] For each type of severe disease, determine the high-risk factors of the severe disease according to the attribute items related to the individual patient included in the first structural data and the second structural data of the severe disease, fill the high-risk factors into the corresponding positions in the tree storage structure, and obtain the severe disease model of the severe disease;

[0024] The critical illness knowledge base is obtained according to the critical illness model of each type of critical illness.

[0025] Optionally, before constructing a storage structure based on the attribute items included in the first structured data and the second structured data, the method further includes:

[0026] According to a preset update relationship, the identifiers of the attribute items included in the first structured data and the second structured data are updated to update different identifiers representing the same attribute item to the same identifier.

[0027] The second aspect of the present disclosure further provides a device for constructing a critical disease knowledge base, comprising:

[0028] A first acquisition module is used to acquire relevant data of multiple types of serious diseases, wherein the relevant data includes first structured data and unstructured data;

[0029] A second acquisition module is used to input the unstructured data into a preprocessing model to obtain second structured data output by the preprocessing model, wherein the preprocessing model is trained based on labeled unstructured sample data of severe diseases;

[0030] A generation module is used to construct a storage structure based on the attribute items included in the first structured data and the second structured data, and fill the attribute values ​​included in the first structured data and the second structured data into corresponding positions in the storage structure to obtain the serious disease knowledge base, wherein the structured data includes a mapping relationship between attribute items and attribute values.

[0031] The third aspect of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods provided in the first aspect of the present disclosure.

[0032] A fourth aspect of the present disclosure further provides an electronic device, comprising:

[0033] a memory having a computer program stored thereon;

[0034] A processor is used to execute the computer program in the memory to implement the steps of any one of the methods provided in the first aspect of the present disclosure.

[0035] Through the above technical solution, the unstructured data in the relevant data of multiple types of serious diseases can be converted into the second structured data through the preprocessing model. After that, a storage structure is constructed according to the relevant data including the first structured data and the attribute items included in the second structured data, and the attribute values ​​included in the first structured data and the second structured data are filled into the corresponding positions in the storage structure, so as to achieve the purpose of building a knowledge base for serious diseases. In addition, the constructed knowledge base for serious diseases includes multiple types of serious diseases, which provides a rich model for the scientific research and teaching of serious diseases and has a high reference value in clinical diagnosis.

[0036] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0038] Figure 1 The present invention is a flowchart of a method for constructing a critical disease knowledge base according to an exemplary embodiment.

[0039] Figure 2 The figure is a flowchart of a method for training a preprocessing model according to an exemplary embodiment.

[0040] Figure 3 The diagram is a schematic diagram showing a clinical medical concept and medical terms subordinate to the clinical medical concept according to an exemplary embodiment.

[0041] Figure 4 is a schematic diagram showing a critical illness model of a critical illness ARDS according to an exemplary embodiment.

[0042] Figure 5 It is a schematic diagram of a device for constructing a critical disease knowledge base according to an exemplary embodiment.

[0043] Figure 6 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0044] In the related art, some software systems (e.g., YimaiTong) have a critical disease knowledge base, but they lack modeling for critical diseases, lack disease-related information, and have few types of critical diseases, and cannot be directly used by other information systems. In order to solve the problem of the lack of critical disease knowledge base in the related art and the lack of support for the diagnosis, treatment, scientific research and teaching of critical diseases, the present disclosure provides a method, device, storage medium and electronic device for constructing a critical disease knowledge base.

[0045] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0046] Figure 1 FIG. 1 is a flowchart of a method for constructing a critical disease knowledge base according to an exemplary embodiment. Figure 1 As shown, the construction method may include the following steps.

[0047] In step 101, relevant data of multiple types of serious diseases are obtained, and the relevant data include first structured data and unstructured data.

[0048] In the present disclosure, multiple approaches can be used to collect data related to multiple categories of critical illnesses for generating a critical illness knowledge base, including but not limited to: collecting data from NMDIC (Neusoft Medical Data-sets for Intensive Care), collecting data from critical illness clinical data of medical institutions, collecting data from medical literature, and crawling from the Internet. Among them, NMDIC is a Chinese data set developed by Neusoft Clinical Medical Critical Care R&D Group, including data related to approximately 10,000 visits to intensive care units, which may include demographics, vital signs, laboratory tests, drugs, etc. All data resources in NMDIC have been strictly processed to remove identity information.

[0049] It is worth noting that, no matter which of the above methods is used, the data related to critical illness obtained includes first structured data and unstructured data. For example, the data related to critical illness is obtained from NMDIC, which usually contains doctor's advice information, which is unstructured data.

[0050] In addition, the types of data obtained will also be different due to different acquisition methods. In order to obtain relevant data of different types of serious diseases, in one possible embodiment, the electronic device that executes the construction method of the serious disease knowledge base has multiple types of data interfaces, for example, an interface for obtaining character type (String) data, an interface for obtaining stream (Stream) type data, an interface for obtaining file (File) type data, an interface for obtaining JavaScript object representation (JavaScript Object Notation) data, and an interface for obtaining binary type data. In another possible embodiment, the electronic device that executes the construction method of the serious disease knowledge base only has an interface for obtaining target type data. In this embodiment, first, the data type conversion technology of the relevant technology is used to convert non-target type data into target type data, and then the converted target type data is obtained through the interface for obtaining target type data. Among them, the target type data can be any of the above-mentioned types of data. The present disclosure does not specifically limit the specific method for obtaining relevant data of multiple types of serious diseases.

[0051] After obtaining the relevant data of serious diseases, the structured data and unstructured data in the relevant data can be further identified. For example, the format of the data can be used to identify whether the data is structured data or unstructured data. For example, if the relevant data of serious diseases are obtained from the database of a medical institution, entity attributes such as table names (or data set names) and field names can be identified by database data model extraction (such as JSON format data). Afterwards, some non-numeric fields in the identified entity attributes are identified as unstructured data, or some data from medical papers are also identified as unstructured data. Among them, JSON data model extraction technology is relatively mature, such as using Google's gson, Ali's fastjson, etc., which can be extracted quickly and accurately, and the present disclosure does not make specific limitations on this.

[0052] In one embodiment, the recognition result can also be output so that the user can judge whether the recognition of structured data and unstructured data is accurate. If it is inaccurate, the user's modification can be received to accurately identify structured data and unstructured data, and so on.

[0053] In addition, the relationship can be identified through the relationship recognition of the database, and the identified relationship can be used to maintain the uniqueness of the data to ensure that the subject of the data after subsequent processing remains unchanged.

[0054] In the present disclosure, the relationship identification of the database is obtained by reading the records of the tables, fields, and types in the database. Different databases store different tables, so the locations of the table, field, and type records are also different. For example: Mysql database, the table, field, and type are recorded in INFORMATION_SCHEMA.COLUMNS, then the table, field, and type need to be read from NFORMATION_SCHEMA.COLUMNS. For example, postgresql database, the table information is recorded in PG_CLASS, the field information is recorded in PG_ATTRIBUTE, the field type information is recorded in PG_TYPE, and the comment information is recorded in PG_DESCRIPTION, then the table, field, and type need to be read from the corresponding position. For example, Oracle database, the table, field, and type are recorded in USER_TAB_COLUMNS, then the table, field, and type need to be read from USER_TAB_COLUMNS. For example, in the SQL Server database, information is recorded in SYSOBJECTS, field information is recorded in SYSCOLUMNS, field type information is recorded in SYSTYPES, and comment information is recorded in SYSCOMMENTS. In this case, you need to read tables, fields, and types from the corresponding locations. For example, in the DB2 database, tables, fields, and types are recorded in SYSIBM.COLUMNS. In this case, you need to read tables, fields, and types from SYSIBM.COLUMNS. After reading tables, fields, and types in the above manner, perform relationship recognition based on the read tables, fields, and types.

[0055] After identifying structured data, unstructured data and relationships in the above manner, additional unique information can be maintained for the unstructured data, that is, the identified relationship is associated with the unstructured data to ensure that the relationship with the original subject remains unchanged after the unstructured data is subsequently processed. For example, to ensure that the examination report data of patient number 001 can still be retrieved as belonging to patient number 001 after subsequent processing.

[0056] In step 102, the unstructured data is input into a preprocessing model to obtain second structured data output by the preprocessing model, where the preprocessing model is trained based on labeled unstructured sample data of critical illnesses.

[0057] In order to distinguish the structured data from the related data which is already structured, the structured data included in the related data is called first structured data, and the structured data obtained by converting the unstructured data is called second structured data.

[0058] In the present disclosure, only structured data can construct a critical disease knowledge base, and in practical applications, the relevant data of critical diseases usually also include unstructured data, so it is also necessary to convert the unstructured data into structured data. For example, the unstructured data can be converted by a preprocessing model. Among them, the preprocessing model is used to convert unstructured data into structured data, so only unstructured data needs to be input into the preprocessing model for conversion, and the specific training method of the preprocessing model will be described in detail below.

[0059] In step 103, a storage structure is constructed based on the attribute items included in the first structured data and the second structured data, and the attribute values ​​included in the first structured data and the second structured data are filled into corresponding positions in the storage structure to obtain a knowledge base of serious diseases, wherein the structured data includes a mapping relationship between attribute items and attribute values.

[0060] By adopting the above technical solution, the unstructured data in the relevant data of multiple types of serious diseases can be converted into the second structured data through the preprocessing model. After that, a storage structure is constructed according to the relevant data including the first structured data and the attribute items included in the second structured data, and the attribute values ​​included in the first structured data and the second structured data are filled into the corresponding positions in the storage structure, so as to achieve the purpose of building a knowledge base for serious diseases. In addition, the constructed knowledge base for serious diseases includes multiple types of serious diseases, which provides a rich model for the scientific research and teaching of serious diseases and has a high reference value in clinical diagnosis.

[0061] In order to better understand the method for constructing the critical disease knowledge base provided by the present disclosure, the construction method is described below with a complete example.

[0062] First, the training method of the preprocessing model is explained. Figure 2 The flowchart of a training method of a preprocessing model according to an exemplary embodiment is shown. The preprocessing model may include a labeling module and an output module, wherein the labeling module is used to perform sequence labeling on the input non-structured data, and the output module is used to convert the non-structured data into structured data according to the sequence labeling result. Figure 2 As shown, the training method may include the following steps.

[0063] In step 201, the unstructured sample data in the preset database is divided into training sample data and verification sample data, wherein the training sample data is used to train the annotation module, and the verification sample data is used to verify the accuracy of the annotation by the annotation module.

[0064] For example, the unstructured sample data in NMDIC can be used to train and verify the annotation module. For example, the unstructured sample data of 6,000 patients among the 10,000 patients with severe diseases in NMDIC are used as training sample data, and the unstructured sample data of 4,000 patients are used as verification sample data.

[0065] In step 202, the training sample data is used as an input parameter, and the labeled result of the training sample data is used as an output parameter to train the labeling module until the labeling accuracy of the labeling module for the verification sample data reaches a preset value. The training is terminated and the labeling module is obtained.

[0066] In the present disclosure, the training sample data can be sequence labeled in advance. Sequence labeling refers to the classification of the input sequence, which is a common method in natural language processing (NLP) and can be used for word segmentation, part-of-speech tagging, named entity recognition, relationship extraction and other processing of the input sequence. In the present disclosure, the BIO labeling method can be used to label each element in the sequence as "BX", "IX" or "O". "BX" means that the segment where this element is located belongs to type X and this element is at the beginning of this segment, "IX" means that the segment where this element is located belongs to type X and this element is in the middle of this segment, and "O" means that it does not belong to any type. For example, X represents label classification, and the unstructured sample data is "Xiao Ming lives in Shenyang City". The main information of concern in this text is people and places. Then, the labels of "Person" and "Location" can be established in advance, and the labeling results of the unstructured sample data are as follows:

[0067] Xiao Ming lives in Shenyang

[0068] BP IP OOO BL IL IL

[0069] The annotation can be layered, and each layer represents a type of processing. For example, the second layer processes the part of speech. When the second layer annotation is performed on the unstructured sample data, X is represented as the part of speech, noun phrase (NP), verb phrase (VP), and the two-layer annotation results of the unstructured sample data are as follows:

[0070] Xiao Ming lives in Shenyang

[0071] BP IP OOO BL IL IL

[0072] B-NP I-NP B-VP I-VP O B-NP I-NP I-NP

[0073] It is worth noting that the present disclosure does not limit the number of layers annotated by the annotation module.

[0074] Among them, Bi-LSTM (Bi-directional Long Short-Term Memory) is very powerful in sequence modeling, can capture long-term contextual information, and also has the ability to fit nonlinear neural networks. CRF (Conditional Random Field) considers more the linear weighted combination of local features of the entire sentence (scanning the entire unstructured data through feature templates). Therefore, in this disclosure, the annotation module uses a joint model of Bi-LSTM (Bi-directional Long Short-Term Memory) + CRF (Conditional Random Field) to annotate unstructured data.

[0075] In the present disclosure, the annotation module is first trained using the training sample data, wherein the annotation module can be trained with reference to the method for model training in the related art, and the present disclosure does not elaborate on the specific implementation process of the training. During the training process, after each training sample data has been used, the annotation module is verified using the verification sample data. That is, the verification sample data is sequence-annotated using the annotation module, and when it is determined based on the annotation results that the annotation accuracy reaches a preset value, the training is terminated to obtain the annotation module. When it is determined that the annotation accuracy does not reach the preset value, the training sample data is continued to be used for training until the annotation accuracy reaches the preset value.

[0076] In step 203, the unstructured sample data is sequence labeled using the trained labeling module.

[0077] In the present disclosure, when training the annotation module, the annotated results of the training sample data may include annotations for critical care medical knowledge attribute items and / or annotations for disease-related attribute items. Accordingly, the trained annotation module can perform sequence annotations on critical care medical knowledge attribute items and / or disease-related attribute items of unstructured sample data.

[0078] Therefore, in one possible manner, the above-mentioned sequence labeling of the unstructured sample data through the trained labeling module may further include: labeling the critical care medical knowledge attribute items and disease-related attribute items of the unstructured sample data through the trained labeling module.

[0079] Among them, the attribute items of critical care medicine knowledge include at least one of the following: clinical medical concepts, medical terms subordinate to clinical medical concepts, the association between medical terms under the same clinical medical concept, and the association between medical terms under different clinical medical concepts; disease-related attribute items include attribute items related to critical diseases and / or attribute items related to individual patients.

[0080] Clinical medicine concepts can be objective objects, such as a ventilator, or abstract things, such as diagnostic and treatment ideas. In the present disclosure, clinical medicine concepts can be regarded as classifications of certain things and corresponded using unique identifiers.

[0081] Medical terms subordinate to clinical medical concepts refer to the language references of clinical medical concepts. There can be multiple terms under a medical concept, and each term has a unique identifier. There is only one first term under each concept. The first term is the term most commonly used to express this clinical medical concept in clinical practice. Other terms under this clinical medical concept are defined as available terms. For example, Figure 3 FIG. 1 is a schematic diagram showing a clinical medical concept and medical terms subordinate to the clinical medical concept according to an exemplary embodiment. Figure 3 As shown, the identifier of the clinical medical concept is "10123547", and the medical terms subordinate to the clinical medical concept include "glucagonoma", "α-cell adenoma", "alpha cell adenoma", "glucagonoma", and "glucagonoma". Among them, "glucagonoma" is the first term, and the other terms are called available terms.

[0082] Association relationships include inheritance relationships and attribute relationships. Among them, inheritance relationships are hierarchical relationships. A clinical medical concept can be a subclass of another or multiple clinical medical concepts. For example, the disease "congenital plagiocephaly with pelvic tilt" is a subclass of the disease "congenital malformation of the skull", and is also a subclass of the disease "oblique head". Attribute relationships are used to characterize the characteristics of clinical medical concepts. For example, the pathological process of the disease "congenital plagiocephaly with pelvic tilt" is a "pathological developmental process", the anatomical location of the occurrence is the "skull", and the morphological change is a "malformation".

[0083] During the annotation process, the association relationship is represented by a pre-established classification set, so there is no need to additionally identify the association relationship during annotation. For example, the unstructured data is "pneumonia and developed sepsis" in "past medical history", the "relationship" classification is disease (D, Disease), and T (Term) is used to represent the term, then the annotation result is:

[0084] Pneumonia progressing to sepsis

[0085] BD ID OOOO BD ID ID

[0086] BT IT OOOO BT IT IT

[0087] It can be seen from the annotation results that "pneumonia" and "sepsis" are both the next level relationship of "disease", that is, "pneumonia" and "sepsis" are subclasses of "disease".

[0088] For another example, the unstructured data is "small patchy shadows can be seen in the left lung", and the annotation results are as follows, where the first layer of annotation is attribute relationship (position) annotation, P represents the detection position, the second layer of annotation is term annotation, and the third layer of annotation is attribute relationship (examination result), R represents the examination result:

[0089]

[0090] In addition, in practical applications, the critical care medicine knowledge attribute items also include term mapping, which refers to establishing an association between a clinical medical concept or term of a term and another term with the same or similar semantic coding, clinical medical concept or term. Through mapping, a term set can be linked to other terms. Each mapping can be represented by a unique identifier. It is worth noting that the term mapping is manually established, and the present disclosure does not specifically limit this.

[0091] Attributes related to critical illnesses may include, but are not limited to: disease category, disease characteristics, disease causes, disease clinical manifestations, disease examinations, disease diagnosis, disease differential diagnosis, and disease treatment. Attributes related to individual patients may include, but are not limited to: patient identification, height, weight, age, blood type, permanent residence, disease type, past medical history, dietary habits, lifestyle habits, and prognosis.

[0092] In step 204, the conversion parameters of the output module are determined according to the sequence annotation results output by the annotation module, the predefined association relationship and the second structured sample data corresponding to the non-structured sample data of the training sample, so that the sequence annotation results and the predefined association relationship can obtain the second structured sample data after the conversion parameters.

[0093] In the present disclosure, the labeling module can be trained first, and after the labeling training is completed, the trained labeling module is used to perform sequence labeling on the unstructured sample data, and then the output module is trained according to the sequence labeling results and the predefined association relationship.

[0094] For example, the sequence annotation results and the predefined association relationship can be used as input parameters, and the second structured data corresponding to the unstructured sample data can be used as output parameters to train the output module to determine the conversion parameters of the output module. The conversion parameters refer to the parameters used when converting the sequence annotation results and the predefined association relationship into the second structured data.

[0095] For example, the critical care medicine knowledge attribute items of the non-structured sample data annotated by the annotation module are used as an example for explanation. The non-structured data is "small patchy shadows can be seen in the left lung", the annotation results are as above, and the converted second structured data is shown in Table 1. The determined conversion parameters may include: term identification field, association relationship category identification field, association relationship identification field and term field, as well as, the term identification of the term "left lung" is "21104537", the term identification of the term "small patchy shadow" is "22016744", the attribute relationship identification is "11102312", and the attribute relationship can be subdivided under this attribute relationship. The attribute relationship can be divided into a scanning position relationship and an examination result relationship, the scanning position relationship is identified as "R0102186", and the examination result relationship is identified as "R0102195". That is to say, if the pre-defined association relationship is a scanning position relationship, the corresponding conversion parameters are the attribute relationship identifier "11102312" and the scanning position relationship identifier "R0102186"; if the pre-defined association relationship is an inspection result relationship, the corresponding conversion parameters are the attribute relationship identifier "11102312" and the inspection result relationship "R0102195".

[0096] Table 1

[0097] T_ID CATEGORY_ID RELATION_ID TERM 21104537 11102312 R0102186 Left lung 22016744 11102312 R0102195 Small patchy shadow

[0098] By adopting the above scheme, the purpose of training the labeling module and the output module in the preprocessing model can be achieved, and the trained preprocessing model can have the function of converting unstructured data into second structured data.

[0099] After converting the unstructured data into the second structured data according to the trained preprocessing model, a critical disease knowledge base is constructed based on the first structured data and the second structured data.

[0100] It is worth noting that in the present disclosure, a critical care medicine knowledge system model and a critical disease model can be established separately.

[0101] First, the critical care medicine knowledge system model is explained. In the present disclosure, a structured data dictionary can be used to represent the critical care medicine knowledge system model. By way of example, the specific method of establishing the critical care medicine knowledge system model is as follows: first, based on the critical care medicine knowledge class attribute items included in the first structured data and the second structured data, a data dictionary storage structure is constructed, and then, the corresponding attribute values ​​of the critical care medicine knowledge class attribute items included in the first structured data and the second structured data are filled into the corresponding positions in the data dictionary storage structure.

[0102] For example, Table 2 is a type of structured data corresponding to clinical medical concepts.

[0103] Table 2

[0104] ID CONCEPT 10101001 Serious illness

[0105] Table 3 is a structural data corresponding to the medical terms belonging to the clinical medical concept.

[0106] Table 3

[0107] T_ID CONCEPT_ID TERM 10112315 10101001 ARDS

[0108] Table 4-1 is a type of structured data corresponding to the inheritance relationship. Table 4-2 is a type of second structured data corresponding to the association relationship of the unstructured data "Small cell carcinoma belongs to neuroendocrine tumors". The "parent node" and "child node" attributes are used to represent the inheritance relationship. In Tables 4-1 and 4-2, the identifier "R0102113" represents the inheritance relationship, and the character "small cell carcinoma" is a subclass of the character "neuroendocrine tumors".

[0109] Table 4-1

[0110] R_ID RELATION R0102113 Inheritance

[0111] Table 4-2

[0112] T_ID RELATION_ID PARENT_ID TERM 19008764 R0102113 10101001 Neuroendocrine tumors 19008785 R0102113 19008764 Small cell carcinoma

[0113] Table 5-1 is a type of structured data corresponding to an attribute relationship. Table 5-2 is a type of second structured data corresponding to the association relationship of the unstructured data "small patchy shadows can be seen in the left lung". Among them, the "composition" attribute is used to represent the attribute relationship. In Table 5-1 and Table 5-2, the identifier "R0102186" and the identifier "R0102195" both belong to the attribute relationship, among which the identifier "R0102186" represents the scanning position relationship, and the identifier "R0102195" represents the inspection result relationship. That is, the character "left lung" is the scanning position of the identifier "R0102186", and the character "small patchy shadow" is the inspection result of the identifier "R0102195".

[0114] Table 5-1

[0115] R_ID RELATION R0102186 Scan location R0102195 Check the results

[0116] Table 5-2

[0117] T_ID CATEGORY_ID RELATION_ID TERM 21104537 11102312 R0102186 Left lung 22016744 11102312 R0102195 Small patchy shadow

[0118] In the above table, the first row is an attribute item of critical care medicine knowledge, and the other rows except the first row are attribute values ​​corresponding to the attribute items of critical care medicine knowledge.

[0119] In this way, according to the above method, a critical care medicine knowledge system model can be constructed, and the critical care medicine knowledge system model can be stored in the critical care disease knowledge base in the form of a data dictionary structure. It is worth noting that since the critical care medicine knowledge system model includes clinical medical concepts, medical terms subordinate to the clinical medical concepts, the association between medical terms under the same clinical medical concept, and the association between medical terms under different clinical medical concepts, for the same critical disease, even if expressed in different medical terms, it can be determined that these different medical terms belong to the same disease. In this way, when analyzing a certain type of critical disease, the disease-related class attributes of different medical terms subordinate to the clinical medical concept of the critical disease can be analyzed to enrich disease-related information.

[0120] Next, the critical illness model is explained. In the present disclosure, a structured tree diagram can be used to characterize the critical illness model. By way of example, the specific method of establishing the critical illness model is as follows: first, based on the disease-related attribute items included in the first structured data and the second structured data, a tree storage structure is constructed, and then the attribute values ​​corresponding to the disease-related attribute items included in the first structured data and the second structured data are filled into the corresponding positions in the storage structure to obtain a critical illness knowledge base.

[0121] It is worth noting that in order to facilitate disease classification and subsequent search for relevant knowledge of a certain type of serious disease, in the present disclosure, for each type of serious disease, a tree storage structure can be constructed based on the disease-related class attribute items included in the first structured data and the second structured data of the serious disease, and the attribute values ​​corresponding to the disease-related class attribute items included in the first structured data and the second structured data of the serious disease are filled into the corresponding positions in the tree storage structure to construct a serious disease model for the serious disease.

[0122] For example, first, the unstructured data is converted into the second structured data using the sequence annotation results of the disease-related attribute items output by the pre-defined association relationship and the annotation model. For example, the unstructured data is "right lower lung: small cell carcinoma, bronchial resection margin (-), lymph nodes (6 / 17) see metastatic carcinoma. IHC: TTF-1+++, Cg-A+, Syn++, CD56+++, Ki-67+++70%, ALK(V)+". Through the annotation module, the annotation of the relationship such as "disease location" and "immunohistochemistry" can be converted into the second structured data as shown in Table 6:

[0123] Table 6

[0124] property result Pathological type Histological pathology Pathological results Neuroendocrine tumors, small cell carcinoma Tumor location Right lower lung Nerve invasion Negative Residual tumor Negative Bronchial stump Negative Number of transfers 6 Number of samples 17 TTF-1 +++ Cg-A + Syn ++ CD56 +++ Ki-67 +++70％ ALK(V) +

[0125] Next, a tree storage structure is constructed according to the second structural data obtained by the conversion.

[0126] For example, a tree structure instance is constructed according to the pre-set disease-related attribute items, and then the converted second structure data and the first structure data of the type of serious disease are used to fill the tree structure instance layer by layer according to the second structure data and the first structure data. When filling each layer, the attribute items in the second structure data and the first structure data can be compared with the attribute items in the tree structure instance, and the attribute items that do not exist in the second structure data and the first structure data can be deleted from the tree structure instance, and the attribute items that exist in the second structure data and / or the first structure data but do not exist in the tree structure instance are added to the layer, and finally the tree storage structure of the type of serious disease is obtained.

[0127] Finally, the attribute values ​​corresponding to the disease-related attribute items included in the first structured data and the second structured data are filled into the corresponding positions in the storage structure to obtain a severe disease model of this type of severe disease represented by a tree structure.

[0128] It is worth noting that the second structured data is obtained by converting the unstructured data. Therefore, there may be some attribute items in the second structured data that have no attribute values, that is, the attribute values ​​are empty. For example, blood type data belongs to structured data, that is, the blood type data has not been processed by the preprocessing model. Therefore, the attribute values ​​corresponding to the attribute items of the blood type information in the second structured data are empty. In this case, it is necessary to fill the attribute values ​​corresponding to the attribute items of the blood type information in the first structured data of the patient into the position where the attribute items of the blood type information in the tree storage structure are located.

[0129] In addition, in actual applications, different medical institutions may use different names for the same disease-related attribute items. For example, medical institution A uses identifier 1 to represent blood type information, and medical institution B uses identifier 2 to represent blood type information. Therefore, before building a storage structure based on the attribute items included in the first structured data and the second structured data, the identifiers of the relevant attribute items need to be updated.

[0130] For example, according to a preset update relationship, the attribute items included in the first structured data and the second structured data are updated to update different identifiers representing the same attribute item to the same identifier. For example, the preset update relationship may be to update identifiers 1 and 2 representing blood type information to identifier 3. In this way, the identifiers of the same attribute item in the first structured data and the second structured data are the same. When constructing a storage structure, the same attribute item is avoided from being mistaken for two different attribute items.

[0131] In one embodiment, the disease-related attribute items include attribute items related to serious diseases and attribute items related to individual patients. First, a tree framework is established using the attribute items related to serious diseases. Each node of the tree framework is the name of an attribute item related to a serious disease. Some attribute items are also divided into multiple subcategories. Then, fill in the attribute values ​​corresponding to the attribute items related to serious diseases. In addition, attribute items related to individual patients can also be used to fill in. For example, an attribute item related to a serious disease is health level, and a subcategory of the health level is BMI (Body Mass Index, body mass index). In this case, the BMI calculation formula needs to be added: BMI = weight (kg) / height (m) 2 , and use it as the attribute value of the child node "Calculation Formula" of BMI.

[0132] In another embodiment, the disease-related attribute items include attribute items related to serious diseases and attribute items related to individual patients. The above-mentioned filling the attribute values ​​included in the first structured data and the second structured data into corresponding positions in the storage structure to obtain a serious disease knowledge base may further include:

[0133] For each type of serious disease, the high-risk factors of the serious disease are determined based on the attribute items related to individual patients included in the first structured data and the second structured data of the serious disease, and the high-risk factors are filled into the corresponding positions in the tree storage structure to obtain the serious disease model of the serious disease. In addition, based on the serious disease model of each type of serious disease, a serious disease knowledge base is obtained.

[0134] For example, for sepsis, the attribute items related to individual patients of sepsis are counted. For example, the height, blood type, gender, age, treatment location, etc. of 500 sepsis patients are analyzed, extracted and sorted to obtain the high-risk population of sepsis. For example, the probability of males and people over 65 years old suffering from sepsis is higher, that is, males and people over 65 years old are high-risk factors for sepsis. Therefore, the high-risk factor can be translated into the formula "Person.gender=1ANDPerson.age>=65" and added to the corresponding position in the tree storage structure.

[0135] It is worth noting that when analyzing attribute items related to individual patients, discrete data are usually analyzed. Therefore, before analysis, it is necessary to determine whether an attribute item related to an individual patient is continuous. If it is continuous, it needs to be converted into discrete data.

[0136] For example, first obtain the attribute items related to the individual patient of this type of serious illness. Then, for each attribute item, determine whether the attribute item is discrete data. For example, if the attribute item is stored in the data dictionary, the attribute item is discrete data, otherwise it is continuous data. For attribute items of discrete data, such as height, wavecluster (wavelet clustering, a network-based clustering algorithm) can be used for classification, and the classification result replaces the continuous classification value of the attribute item, which is equivalent to discretizing the continuous attribute value first. Afterwards, use the CPAR (a classification algorithm based on association rules) algorithm to classify all data to determine the high-risk factors for serious diseases.

[0137] For example, Figure 4 FIG. 1 is a schematic diagram showing a critical illness model of ARDS according to an exemplary embodiment. Figure 4 As shown in Figure 1, a tree structure is used to characterize the severe disease model. Figure 4 In the tree structure shown, each "+" node can be further expanded. For each type of severe disease, the following Figure 4 The critical illness model shown in FIG. And the critical illness model of each type of critical illness is used to build a critical illness knowledge base.

[0138] By adopting the above scheme, the development process of this type of serious disease can be displayed using a tree structure, and the high-risk factors of this type of serious disease can be determined based on the attribute items of this type of serious disease related to the individual patient, further improving the reference value in clinical diagnosis.

[0139] Based on the same inventive concept, the present disclosure also provides a device for constructing a critical disease knowledge base. Figure 5 FIG. 1 is a schematic diagram of a device for constructing a knowledge base of serious diseases according to an exemplary embodiment. Figure 5 As shown, the critical disease knowledge base construction device 500 may include:

[0140] A first acquisition module 501 is used to acquire relevant data of multiple types of serious diseases, wherein the relevant data includes first structured data and unstructured data;

[0141] A second acquisition module 502 is used to input the unstructured data into a preprocessing model to obtain second structured data output by the preprocessing model, wherein the preprocessing model is trained based on labeled unstructured sample data of severe diseases;

[0142] A creation module 503 is used to construct a storage structure based on the attribute items included in the first structured data and the second structured data, and fill the attribute values ​​included in the first structured data and the second structured data into corresponding positions in the storage structure to obtain the serious disease knowledge base, wherein the structured data includes a mapping relationship between attribute items and attribute values.

[0143] Optionally, the preprocessing model includes a labeling module and an output module, and accordingly, the device further includes:

[0144] A classification module, used for dividing the unstructured sample data in the preset database into training sample data and verification sample data;

[0145] A training module, used to take the training sample data as an input parameter and the labeled result of the training sample data as an output parameter, to train the labeling module until the labeling accuracy of the labeling module for the verification sample data reaches a preset value, to terminate the training, and to obtain the labeling module;

[0146] A sequence labeling module, used for performing sequence labeling on the unstructured sample data through the trained labeling module;

[0147] A determination module is used to determine the conversion parameters of the output module according to the sequence annotation results output by the annotation module, the predefined association relationship and the second structured data corresponding to the unstructured sample data, so that the sequence annotation results and the predefined association relationship can obtain the second structured data after the conversion parameters.

[0148] Optionally, the sequence labeling module is used to: label the critical care medical knowledge attribute items and disease-related attribute items of the unstructured sample data through the trained labeling module;

[0149] The critical care medical knowledge attribute items include at least one of the following: clinical medical concepts, medical terms subordinate to the clinical medical concepts, associations between medical terms under the same clinical medical concept, and associations between medical terms under different clinical medical concepts;

[0150] The disease-related attribute items include attribute items related to serious diseases and / or attribute items related to individual patients.

[0151] Optionally, the creation module 503 includes:

[0152] A first creation submodule is used to construct a data dictionary storage structure based on the critical care medicine knowledge attribute items included in the first structured data and the second structured data;

[0153] The second creation submodule is used to construct a tree storage structure based on the disease-related attribute items included in the first structured data and the second structured data.

[0154] Optionally, the second creation submodule is used to: for each type of serious disease, construct a tree storage structure according to the disease-related attribute items included in the first structural data and the second structural data of the serious disease.

[0155] Optionally, the disease-related attribute items include attribute items related to serious diseases and attribute items related to individual patients, and the creation module 503 includes:

[0156] A first determination submodule is used to determine, for each type of severe disease, the high-risk factors of the severe disease according to the first structural data of the disease and the attribute items related to the individual patient included in the second structural data, and fill the high-risk factors into the corresponding positions in the tree storage structure to obtain a severe disease model of the severe disease;

[0157] A submodule is created to obtain the critical disease knowledge base according to the critical disease model of each type of critical disease.

[0158] Optionally, the device further comprises:

[0159] The updating module is used to update the identifiers of the attribute items included in the first structured data and the second structured data according to a preset updating relationship, so as to update different identifiers representing the same attribute item to the same identifier.

[0160] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0161] Based on the same inventive concept, the present disclosure also provides an electronic device, including:

[0162] a memory having a computer program stored thereon;

[0163] A processor is used to execute the computer program in the memory to implement the steps of any one of the methods provided by the present disclosure.

[0164] For example, Figure 6 FIG. 6 is a block diagram of an electronic device 6 according to an exemplary embodiment. Figure 6 As shown, the electronic device 600 may include: a processor 601 and a memory 602. The electronic device 600 may also include one or more of a multimedia component 603, an input / output (I / O) interface 604, and a communication component 605.

[0165] Among them, the processor 601 is used to control the overall operation of the electronic device 600 to complete all or part of the steps in the above-mentioned method for constructing a knowledge base of serious diseases. The memory 602 is used to store various types of data to support the operation of the electronic device 600, and these data may include, for example, instructions for any application or method used to operate on the electronic device 600, and application-related data, such as contact data, messages sent and received, pictures, audio, video, etc. The memory 602 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 603 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 602 or sent through the communication component 605. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 604 provides an interface between the processor 601 and other interface modules, and the other interface modules may be keyboards, mice, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 605 is used for wired or wireless communication between the electronic device 600 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or a combination of one or more of them, is not limited here. Therefore, the corresponding communication component 605 may include: Wi-Fi module, Bluetooth module, NFC module, etc.

[0166] In an exemplary embodiment, the electronic device 600 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned method for constructing a critical disease knowledge base.

[0167] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the method for constructing the critical disease knowledge base described above are implemented. For example, the computer-readable storage medium may be the memory 602 including the program instructions, and the program instructions may be executed by the processor 601 of the electronic device 600 to complete the method for constructing the critical disease knowledge base described above.

[0168] In another exemplary embodiment, a computer program product is also provided, which includes a computer program that can be executed by a programmable device, and the computer program has a code portion for executing the above-mentioned method for constructing a critical illness knowledge base when executed by the programmable device.

[0169] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0170] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0171] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A method for constructing a critical disease knowledge base, characterized in that: include: Acquire relevant data of multiple types of serious diseases, wherein the relevant data includes first structured data and unstructured data; Inputting the unstructured data into a preprocessing model to obtain second structured data output by the preprocessing model, wherein the preprocessing model is trained based on labeled unstructured sample data of severe diseases; Building a storage structure based on the attribute items included in the first structured data and the second structured data, and filling the attribute values ​​included in the first structured data and the second structured data into corresponding positions in the storage structure to obtain the critical disease knowledge base, wherein the structured data includes a mapping relationship between attribute items and attribute values; The preprocessing model includes a labeling module and an output module. Accordingly, the training process of the preprocessing model includes: The unstructured sample data in the preset database is divided into training sample data and verification sample data; The training sample data is used as an input parameter, and the labeled result of the training sample data is used as an output parameter to train the labeling module until the labeling accuracy of the labeling module for the verification sample data reaches a preset value, and the training is terminated to obtain the labeling module; Performing sequence labeling on the unstructured sample data through the trained labeling module; Determining the conversion parameters of the output module according to the sequence annotation results output by the annotation module, the predefined association relationship and the second structured data corresponding to the unstructured sample data, so that the sequence annotation results and the predefined association relationship can obtain the second structured data after the conversion parameters are passed; The step of constructing a storage structure based on the attribute items included in the first structured data and the second structured data includes: For each type of serious disease, a tree storage structure is constructed based on the disease-related attribute items included in the first structural data and the second structural data of the serious disease.

2. The method according to claim 1, characterized in that The step of performing sequence labeling on the unstructured sample data by the trained labeling module includes: Annotating critical care medical knowledge attribute items and disease-related attribute items of the unstructured sample data by using the trained annotation module; The critical care medical knowledge attribute items include at least one of the following: clinical medical concepts, medical terms subordinate to the clinical medical concepts, associations between medical terms under the same clinical medical concept, and associations between medical terms under different clinical medical concepts; The disease-related attribute items include attribute items related to serious diseases and / or attribute items related to individual patients.

3. The method according to any one of claims 1 to 2, characterized in that: The step of constructing a storage structure based on the attribute items included in the first structured data and the second structured data further includes: A data dictionary storage structure is constructed based on the critical care medicine knowledge attribute items included in the first structured data and the second structured data.

4. The method according to claim 3, characterized in that The disease-related attribute items include attribute items related to serious diseases and attribute items related to individual patients, and the attribute values ​​included in the first structured data and the second structured data are filled into corresponding positions in the storage structure to obtain the serious disease knowledge base, including: For each type of severe disease, determine the high-risk factors of the severe disease according to the attribute items related to the individual patient included in the first structural data and the second structural data of the severe disease, fill the high-risk factors into the corresponding positions in the tree storage structure, and obtain the severe disease model of the severe disease; The critical illness knowledge base is obtained according to the critical illness model of each type of critical illness.

5. The method according to claim 1, characterized in that Before constructing a storage structure based on the attribute items included in the first structured data and the second structured data, the method further includes: According to a preset update relationship, the identifiers of the attribute items included in the first structured data and the second structured data are updated to update different identifiers representing the same attribute item to the same identifier.

6. A device for constructing a critical disease knowledge base, characterized in that: include: A first acquisition module is used to acquire relevant data of multiple types of serious diseases, wherein the relevant data includes first structured data and unstructured data; A second acquisition module is used to input the unstructured data into a preprocessing model to obtain second structured data output by the preprocessing model, wherein the preprocessing model is trained based on labeled unstructured sample data of severe diseases; A creation module, used to construct a storage structure based on the attribute items included in the first structured data and the second structured data, and fill the attribute values ​​included in the first structured data and the second structured data into corresponding positions in the storage structure to obtain the critical disease knowledge base, wherein the structured data includes a mapping relationship between attribute items and attribute values; The preprocessing model includes a labeling module and an output module. Accordingly, the device further includes: A classification module, used for dividing the unstructured sample data in the preset database into training sample data and verification sample data; A training module, used to take the training sample data as an input parameter and the labeled result of the training sample data as an output parameter, to train the labeling module until the labeling accuracy of the labeling module for the verification sample data reaches a preset value, to terminate the training, and to obtain the labeling module; A sequence labeling module, used for performing sequence labeling on the unstructured sample data through the trained labeling module; a determination module, configured to determine a conversion parameter of the output module according to the sequence annotation result output by the annotation module, the predefined association relationship and the second structured data corresponding to the unstructured sample data, so that the sequence annotation result and the predefined association relationship can obtain the second structured data after being converted by the conversion parameter; The creation module includes: The second creation submodule is used to construct a tree storage structure for each type of serious disease according to the disease-related attribute items included in the first structural data and the second structural data of the serious disease.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

8. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 5.

Images