Method and device for constructing health knowledge graph

Abstract

The present invention discloses a method and device for constructing a health knowledge graph. The method includes: a method for constructing a health knowledge graph, including: abstracting an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; based on the ontology expression model, constructing the health knowledge graph using the medical diagnosis theory text; wherein the health knowledge graph is used to recognize the health of the subject to be detected. The present invention solves the technical problem of inaccurate construction of the health knowledge graph.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular, to a method and device for constructing a health knowledge graph. Background Art

[0002] A method for constructing a knowledge graph of traditional Chinese medicine in the prior art uses deep learning technology to solve the problem of inconsistent technical evaluation standards for knowledge graphs of traditional Chinese medicine; another method for constructing a knowledge graph of theoretical classics of traditional Chinese medicine provides a method for extracting knowledge graphs from theoretical classics of traditional Chinese medicine through entity tables and verb tables; there is also a method for constructing a knowledge graph of traditional Chinese medicine, which describes the process from preparing equipment, assigning staff, entering data, designing knowledge graphs, developing systems, testing systems, manual proofreading, and operation.

[0003] The problem with the existing technology is that it describes the method of constructing a knowledge graph, but the method used is no different from that used in other knowledge fields and is simply applied to the field of traditional Chinese medicine. The description of the particularity of knowledge in traditional Chinese medicine is insufficient, resulting in inaccurate graph construction.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for constructing a health knowledge graph, so as to at least solve the technical problem of inaccurate construction of a health knowledge graph.

[0006] According to one aspect of an embodiment of the present invention, a health perception and cognition method is provided, comprising: abstracting an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; based on the ontology expression model, constructing the health knowledge graph using the medical diagnosis theory text; wherein the health knowledge graph is used to recognize the health of the subject to be detected.

[0007] According to another aspect of an embodiment of the present invention, a device for constructing a health knowledge graph is also provided, including: an ontology expression model construction module, configured to abstract an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; and a health knowledge graph construction module, configured to construct the health knowledge graph based on the ontology expression model and using the medical diagnosis theory text.

[0008] In an embodiment of the present invention, an ontology expression model for medical diagnosis is abstracted, and the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner, and based on the ontology expression model, the health knowledge graph is constructed using the medical diagnosis theory text, thereby achieving the technical effect of improving the accuracy of health detection and the depth of analysis, and further solving the technical problem of inaccurate construction of the health knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0010] Figure 1 is a flowchart of a method for constructing a health knowledge graph according to a first embodiment of the present invention;

[0011] Figure 2 is a flow chart of a health perception and cognition method according to a second embodiment of the present invention;

[0012] Figure 3 is a flowchart of the health perception step S202 based on deep learning according to an embodiment of the present invention;

[0013] Figure 4 is a schematic diagram of the structure of an ontology expression model according to an embodiment of the present application;

[0014] Figure 5 is a flow chart of a health perception and cognition method according to a third embodiment of the present invention;

[0015] Figure 6 is a schematic diagram of a health knowledge graph about heart disease according to an embodiment of the present application;

[0016] Figure 7 is a schematic diagram of a health knowledge graph for reasoning with the organ heart as a node according to an embodiment of the present invention;

[0017] Figure 8 is a schematic diagram of a health knowledge graph for reasoning using characteristic deep vertical lines as nodes according to an embodiment of the present invention;

[0018] Fig. 9 is a flow chart of a method for intelligent facial diagnosis based on a health knowledge graph according to an embodiment of the present application;

[0019] Fig.10 It is a flow chart of a health perception and cognition method based on a health knowledge graph according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0022] Overview

[0023] The present application provides a health knowledge graph for the field of traditional Chinese medicine facial diagnosis, including its construction method, ontology model expression, and reasoning application.

[0024] By transforming the theoretical text of TCM facial diagnosis into triples and merging entities and relationships, we completed the transformation from unstructured text to structured expression, and also obtained the basic entities and relationships required for local model abstraction; using features, regions, symptoms, diseases, and organs as entities, and detection, composition, reflection, occurrence, and inference as relationships, we constructed a facial diagnosis ontology expression model; finally, the health knowledge graph of TCM facial diagnosis constructed based on the ontology expression model can be used for various reasoning applications, including feature statistics, cause analysis, reasoning from symptoms to diseases, disease association analysis, etc., and thus replace manual TCM facial diagnosis and apply it to the intelligent facial diagnosis system.

[0025] This application also proposes a health perception and cognition method based on the combination of deep learning and health knowledge graph. This method is divided into two core steps. The first step is to use deep learning to achieve health perception and detect the single health status of the human body; the second step is to use the health knowledge graph to achieve health cognition, and input the symptoms detected in the first step into the constructed health knowledge graph in the medical field to achieve reasoning analysis, thereby providing a more accurate and complete health analysis. Through the above two steps, the health detection analysis can be upgraded from the current health perception level to the level of health cognition, thereby improving the accuracy of health detection and the depth of analysis.

[0026] Example 1

[0027] According to an embodiment of the present invention, a method for constructing a health knowledge graph is provided, such as Figure 1 As shown, the method includes:

[0028] Step S102 , abstracting an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner.

[0029] Step S104, based on the ontology expression model, the health knowledge graph is constructed using the medical diagnosis theory text.

[0030] In an exemplary embodiment, abstracting an ontology expression model for medical diagnosis includes: abstracting multiple entities and multiple relationships, wherein the multiple entities include features, regions, symptoms, diseases, and organs, and the multiple relationships include detected, composed, reflected, occurred, and inferred; through the multiple relationships, establishing connections between the multiple entities to construct the ontology expression model.

[0031] In an exemplary embodiment, connections between multiple entities are established through multiple relationships, including at least one of the following: setting the relationship between the feature and the region to be a "detected" relationship, to indicate that the feature is detected from the region; setting the relationship between the feature and the symptom to be a "composed of" relationship, to indicate that the feature constitutes the symptom; setting the relationship between the region and the symptom to be a "composed of" relationship, to indicate that the region constitutes the symptom; setting the symptom itself to be a "composed of" relationship, to indicate that the symptom constitutes a new symptom; setting the relationship between the region and the organ to be a reflection relationship, to indicate that the region reflects the organ; setting the relationship between the disease and the organ to be "occurs in", to indicate that the disease occurs in the organ; setting the relationship between the symptom and the disease to be "inferred out", to indicate that the disease is inferred from the symptom.

[0032] In an exemplary embodiment, based on the ontology expression model, the health knowledge graph is constructed using the medical diagnosis theory text, including: parsing the medical diagnosis theory text into multiple triples through grammatical analysis and semantic analysis, wherein each triple in the multiple triples includes two entities and the relationship between the two entities; merging the multiple triples, removing redundant triples in the multiple triples, so as to optimize the multiple triples; extracting an entity table including entity types and entity instances, and a relationship table including relationship types and relationship instances from the optimized multiple triples, wherein the health knowledge graph includes the entity table and the relationship table.

[0033] In an exemplary embodiment, merging the plurality of triples comprises: merging the plurality of triples with the same or similar meanings into one triple, and representing the merged triples by setting attributes for the merged triples, wherein the attributes comprise alias attributes for describing the names of the entities in the triples and / or position description attributes for describing the positions of the entities in the triples.

[0034] In an exemplary embodiment, after constructing the health knowledge graph, the method further includes: performing health perception of the object to be detected based on deep learning to perceive each individual health status of the object to be detected; based on the health knowledge graph, performing health cognition by integrating each of the individual health statuses to obtain the health status of the object to be detected by reasoning and analysis.

[0035] In an exemplary embodiment, health perception of an object to be detected based on deep learning includes: determining a partition to be detected and features corresponding to the partition based on the health knowledge graph; using a trained deep neural network model to detect whether the partition has the feature, and obtaining health symptom results for the partition.

[0036] In an exemplary embodiment, determining a partition to be detected and a feature corresponding to the partition based on the health knowledge graph includes: receiving a picture including at least a part of the object to be detected; segmenting the picture based on the area division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, searching the health knowledge graph for a feature corresponding to the partition, and detecting the partition to determine whether the partition has a feature corresponding to the partition.

[0037] In an exemplary embodiment, after constructing the health knowledge graph, the method further includes at least one of the following: taking an organ instance corresponding to an organ in the multiple entities in the ontology expression model as a node, searching for a disease associated with the organ instance; taking a feature instance corresponding to a feature in the multiple entities in the ontology expression model as a node, searching for region instances corresponding to regions in the multiple entities in which the feature instance appears, and based on the relationship between the found region instance and the feature instance, determining the symptom and inferring the disease associated with the symptom.

[0038] The health knowledge graph provided in this embodiment can be combined with the health perception and cognition method in this application. Through the construction method in this embodiment, a health knowledge graph that more accurately reflects the medical diagnosis theory text can be constructed, thereby providing the possibility for intelligent medical diagnosis.

[0039] Example 2

[0040] According to an embodiment of the present invention, a health perception and cognition method is provided, such as Figure 2 As shown, the method includes:

[0041] Step S202: health perception based on deep learning.

[0042] Take the detection and analysis of health from face images as an example. When a face image is input, the face will first be partitioned according to the regional division rules in the health knowledge graph, and then the items to be detected will be determined for each partition based on the features associated with the region in the health knowledge graph. The features of the region will be detected using a model trained with a deep neural network, and the symptom results of the region plus the features will be obtained to determine whether the region has the symptom.

[0043] In an exemplary embodiment, Figure 3 As shown, step S202 may include the following sub-steps:

[0044] Step S2022, receiving an input picture.

[0045] Take as input a picture of a person, for example, a picture including a person's face.

[0046] Step S2024, perform image segmentation according to the regional division rules of the health knowledge graph.

[0047] Image segmentation is performed according to the regional division rules of the health knowledge graph. The health knowledge graph of this embodiment contains symptoms related to various diseases. For example, the description of the symptom can be what characteristic phenomenon occurs at a certain location. In this way, the input image can be segmented based on all the location information in the health knowledge graph.

[0048] After the image is partitioned, the features that should be detected in each partition are determined based on the health knowledge graph.

[0049] Step S2026: Use the trained deep neural network to perform feature inference and prediction for each partition.

[0050] For the partitions segmented in step S2024, the trained deep neural network is used to infer and predict the features. The features to be detected at each location are different, and the specific needs depend on the features associated with the area in the health knowledge graph.

[0051] Step S2028, obtaining the health symptom results of the region plus the features.

[0052] Through the above steps S2024 and S2026, it is possible to determine whether a certain feature appears in a certain area, and the area plus the feature can determine the symptoms from the health knowledge graph.

[0053] Step S204: health cognition based on the health knowledge graph.

[0054] After the symptoms are obtained in step S202, reasoning analysis is performed through the relationship associated with the symptoms in the health knowledge graph, such as reasoning from the symptoms to related diseases. In addition, local and overall reasoning can be performed based on whether the same symptom appears in multiple locations.

[0055] In an exemplary embodiment, the ontology expression model of the health knowledge graph is a modeling method for describing the health knowledge graph, and the core is to describe nodes and relationships. The following will describe in detail a simplest ontology expression model for realizing health cognition, but it is not limited to this model, and each health business needs are different, and there may be model logic in other forms of expression.

[0056] Figure 4 Schematic diagram of the structure of the ontology expression model according to the embodiment of the present application. Figure 4 As shown, the nodes in the rectangular boxes represent entities, including features, regions, symptoms, diseases, and organs; the edges between the nodes represent the relationships between the nodes, including detected, composed, inferred, occurred, and reflected.

[0057] The entities such as characteristics, regions, symptoms, organs, and diseases are described in detail below.

[0058] Features: Features can be understood as phenomena. Taking the TCM facial diagnosis as an example, features mainly include four categories: color, shape, points, and lines. For example, looking at the color, checking the color characteristics of a certain area, TCM pays attention to whether it is red, blue, white, black, yellow, etc.; shape is to look at the shape of the area or the whole, such as fatness, thinness, bumps, etc.; points are to look at acne, spots, moles, thorns, etc.; lines are wrinkles, horizontal lines, vertical lines, crab claw lines, etc.; this embodiment takes TCM as an example, but is not limited to this. In other embodiments, features can also be features summarized according to the basis of Western medicine.

[0059] Region: Different medical theories have different regional divisions, but the core is still to divide according to the location of the symptom. This embodiment takes TCM inspection as an example. On the human face, according to TCM division, there are dozens of division areas such as the front, upper que, middle que, lower que, nose bridge, nose tip, nose wing, philtrum, cheek, and mouth periorbital area.

[0060] Symptom: Symptom is also a symptom, which is a manifestation of a certain health condition or disease, and is generally described by what features appear in which area. For example, in traditional Chinese medicine, purple lips are a symptom of heart disease; then this symptom is that the area is "lips", and the feature is "purple color", which forms the symptom of purple lips.

[0061] Diseases: including physiological, psychological, traditional Chinese medicine, Western medicine, etc. For example, heart disease, coronary heart disease, hypertension, pharyngitis, etc.;

[0062] Organ: includes organs and systems, such as the heart, lungs, stomach, kidneys, spleen, etc.;

[0063] Specific description of the relationship:

[0064] It should be noted that all relationships can have their inverse relationships. In addition, the types on the relationships are for the purpose of being more concrete. In reality, there may be no types, which will make it more generalized.

[0065] The following is an explanation of the marked relationships:

[0066] 1) Detected: Features are detected from regions, which expresses which regions the feature is detected from. The reverse relationship is which features are detected in a region; for example, acne is detected from the forehead.

[0067] 2) Composition: Symptoms are composed of features and areas. In addition, symptoms can also be composed of other elements, which are not elaborated here. For example, symptoms can be further combined into more complex symptoms, that is, the composition relationship is transitive and introspective; for example, a black forehead and a yellow face.

[0068] 3) Inference: Diseases can be inferred from symptoms, and related symptoms can be inferred from diseases. For example, if vertical lines and redness appear between the eyebrows, it can be inferred that there may be insufficient blood supply to the cardiovascular system, and insufficient blood supply to the cardiovascular system can also be reversely inferred as the above symptoms, as well as the possibility that the user may experience symptoms such as neurasthenia, frequent dreams, poor sleep, palpitations, and irritability.

[0069] 4) Occurs in: A disease occurs in a certain organ, for example, heart disease occurs in the heart and cholecystitis occurs in the gallbladder.

[0070] 5) Reflection: A certain area of ​​the face reflects a certain organ. For example, the cheek reflects the kidney in TCM theory. When the cheek has undesirable features, it means that there is a problem with the kidney. For example, if the cheek has red bloodshot, acne, or spots, it reflects that the kidney may be deficient.

[0071] The above ontology expression model is of great significance for correctness reasoning at a high level, designing business models, and external interfaces.

[0072] Finally, based on the pre-built health knowledge graph model, reasoning analysis is performed to obtain the health status of the object to be tested.

[0073] This embodiment performs reasoning from multiple aspects to achieve more accurate and comprehensive health cognition analysis.

[0074] Example 3

[0075] According to an embodiment of the present invention, a health perception and cognition method is provided, such as Figure 5 As shown, the method includes:

[0076] Step S502, pre-constructing a health knowledge graph.

[0077] This embodiment takes heart disease as an example to construct a health knowledge graph.

[0078] Figure 6 is a health knowledge graph about heart disease according to an embodiment of the present application, such as Figure 6 As shown, the entity nodes in the health knowledge graph include features, regions, symptoms, diseases, and organs; the edges between nodes represent the relationships between nodes, including detected, composed, inferred, occurred, and reflected.

[0079] In this embodiment, the features mainly include deep horizontal lines, deep vertical lines and purple. The areas mainly include the subvalvular region, tongue and lips.

[0080] From the relationship between features and areas, we can get the following three symptoms, for example: deep horizontal lines under the forehead, deep vertical lines on the tongue, and purple lips.

[0081] The combination of deep horizontal lines under the chin and deep vertical lines on the tongue can indicate the possibility of heart disease. Purple lips can also indicate the possibility of heart disease.

[0082] In an exemplary embodiment, more and richer reasoning can be made through the health knowledge graph. For example, heart disease occurs in the heart, so take the heart as the node and search for diseases associated with "happenIn" to obtain the following: Figure 7 The following results are shown: coronary heart disease, heart disease, congenital deficiency, myocardial weakness, psychological stress, arrhythmia, etc.

[0083] In an exemplary embodiment, a feature can be used as the center to find out in which areas the feature appears and what symptoms it represents, and then infer the possible associated diseases. Figure 8 As shown, when deep vertical lines appear between the eyebrows, it means insufficient blood supply to the cardiovascular system; when deep vertical lines appear beside the nose, it means there may be problems with the gallbladder; when deep vertical lines appear on the tongue, it means there may be heart disease; when deep vertical lines appear beside the ears, it may mean kidney deficiency; when deep vertical lines appear on the forehead, it may be a sign of high blood pressure or cerebral thrombosis.

[0084] In other embodiments, similar reasoning, or even more complex and comprehensive reasoning and analysis, can be performed on the health knowledge graph. It is only necessary to match the health perception results to the symptoms of a certain feature + area in the health knowledge graph, and various reasoning and analysis can be performed around this.

[0085] Step S504, detecting the input image and obtaining the health symptom result of the region and features.

[0086] Receive the input image, segment the received image based on the health knowledge graph, and determine the features that should be detected in each partition after segmentation. Use the neural network model to detect the corresponding features determined in each partition, determine whether a certain feature appears in a certain area, and obtain the health symptom results of the area plus the feature.

[0087] Step S506, inferring a disease corresponding to the symptom result based on the symptom result.

[0088] Based on the health detection map, the corresponding disease is inferred.

[0089] The health cognition ability in this embodiment is a capability that no existing method has. It should be noted that this cognitive ability will become more powerful as the health knowledge graph is enriched. The perception is becoming more and more accurate, and the cognition is becoming more and more powerful, thus realizing the real artificial intelligence health perception and cognition.

[0090] Example 4

[0091] According to an embodiment of the present invention, a health knowledge graph for the field of traditional Chinese medicine facial diagnosis is provided, including its construction method, ontology model expression, and reasoning application. By converting the theoretical text of traditional Chinese medicine facial diagnosis into triples and merging entities and relationships, the transformation from unstructured text to structured expression is completed, and the basic entities and relationships required for local model abstraction are obtained.

[0092] This embodiment uses features, regions, symptoms, diseases, and organs as entities, and detects, forms, reflects, occurs in, and is inferred as relationships to construct a facial diagnosis ontology model expression; the health knowledge graph of TCM facial diagnosis that is finally constructed can be used for various reasoning applications, including feature statistics, etiology analysis, reasoning from symptoms to diseases, disease association analysis, etc., thereby replacing manual TCM facial diagnosis and applying it to the intelligent facial diagnosis system.

[0093] According to an embodiment of the present invention, a method for intelligent facial diagnosis based on a health knowledge graph is provided. This embodiment takes the theory of facial diagnosis in traditional Chinese medicine as an example, but is not limited thereto. Fig. 9 As shown, the method includes:

[0094] Step S902: construct an ontology expression model.

[0095] The entities and relationships of face-to-face consultation are abstracted at a high level to form an ontology expression model of Chinese medicine face-to-face consultation. The local expression model describes the health knowledge graph in a more abstract dimension, which is convenient for checking the data standardization in the card-building knowledge graph and external business docking.

[0096] Ontology expression model such as Figure 4 The embodiments described above will not be repeated here.

[0097] Step S904, constructing a health knowledge graph based on the ontology expression model.

[0098] Constructing a health knowledge graph mainly includes the following steps: extracting three principles and merging entities and relationships.

[0099] 1) Extract triples.

[0100] Parse medical diagnosis theory texts, such as TCM facial diagnosis theory texts, into (entity, relationship, entity) triples. The health knowledge graph needs to take the existing TCM facial diagnosis theory text data as input. These are theoretical descriptive texts, which are composed of sentences. They need to be converted into (entity, relationship, entity) triples through grammatical analysis and semantic analysis.

[0101] In an exemplary embodiment, parsing the TCM facial diagnosis theory text into structured triples can be constructed according to the following method.

[0102] Listed below are several typical examples of TCM facial diagnosis theory texts.

[0103] "The heart reflex zone is on the bridge of the nose between the two corners of the eyes. If there are horizontal lines here, it means that the heart rhythm is irregular in the front."

[0104] From the above sentences, the triplets that can be extracted are: (the bridge of the nose between the two corners of the eyes, reflection, heart); (horizontal lines, appearing on, the bridge of the nose between the two corners of the eyes); (horizontal lines appear on the bridge of the nose between the two corners of the eyes, inference, arrhythmia).

[0105] It can be seen that the description of TCM facial diagnosis has very complex relationships, and there are often multiple triples extracted from a sentence.

[0106] Through the conversion from text to triples similar to the above, typical conversion examples of TCM facial diagnosis theory text can be sorted into the following table:

[0107]

[0108]

[0109] Table 1

[0110] In the above process of converting text into triples, it is not difficult to find that due to the imprecise description of the TCM facial diagnosis theory text, the constructed triples have the following problems:

[0111] 1) The constructed triples can be further split, that is, the nodes in the constructed triples can be further split into one or more triples;

[0112] 2) Both entities and relationships are complex, redundant, and inconsistent;

[0113] In order to solve the above problems, it is necessary to merge entities and relationships to extract triples that can represent the original meaning of the text.

[0114] 2) Merge entities and relationships.

[0115] Due to the imprecise description of TCM facial diagnosis theory, there may be multiple words describing the same entity, and the essence of multiple relationships is the same relationship. Therefore, it is necessary to perform synonym merging on both entities and relationships to obtain standardized entity tables and relationship tables.

[0116] In an exemplary embodiment, the merging of entities and relationships may be performed as follows.

[0117] Merging is to merge similar items, for example: Sentence 1: Quezhong (Yintang, eyebrow center) is the lung; Sentence 2: The reflex area of ​​the lung is between the 1 / 2 of the two eyebrows and the part below 1 / 3 of the forehead; The triplets that can be extracted from sentence 1 are: ("Quezhong (Yintang, eyebrow center), is, lung), and the triplets after further decomposition are: (Quezhong, is, lung), (Yintang, is, lung), (eyebrow center, is, lung). The triplets that can be extracted from sentence 2 are: ("1 / 2 between the two eyebrows, the part below 1 / 3 of the forehead", reflex, lung).

[0118] It can be seen that the above statements and triples actually have the same meaning, but entities and relationships have more than one expression, so such entities and relationships can be merged and their diversity can be expressed with attributes. For example, the above logic can be expressed as follows: (quezhong, reflection, lung)

[0119] Among them, the Quezhong entity has an alias attribute and a location description attribute; entity name: Quezhong; aliases: Yintang, eyebrow center. Location description: the area between the two eyebrows and below 1 / 3 of the forehead.

[0120] For the above relations "reflection" and "for", considering that the meaning of "for" is too abstract, "reflection" is directly used as the relationship between the above entities. Other sentences and triples can also be optimized and merged through similar merging processing.

[0121] In addition, in order to improve the rationality and consistency of entity and relationship design, when merging, it is also necessary to standardize entities and relationships based on the higher-level abstract model of the ontology expression model, so that the entities and relationships in the entity table and relationship table conform to the structure and logic of the ontology expression model.

[0122] Finally, the following summary entity table is obtained:

[0123]

[0124]

[0125] Table 2

[0126] In addition, the relationships are also merged and summarized as follows:

[0127]

[0128] Table 3

[0129] By further decomposing and merging the triples, a better health knowledge graph can be obtained. That is, through the above merging and sorting of entities and relationships, a clearly structured entity and relationship is obtained, and a lot of complexity, redundancy, and inconsistency are removed.

[0130] Step S906, using the health knowledge graph to perform health checks on the subject to be checked.

[0131] After the health knowledge graph of TCM consultation is constructed, various reasoning applications can be performed based on the health knowledge graph, such as predictive analysis from symptoms to diseases, correlation analysis of symptoms, etc.

[0132] Receive a picture including at least part of the object to be detected, for example, a face picture; segment the picture based on the regional division rules in the health knowledge graph to obtain multiple partitions. For each of the multiple partitions, find the features corresponding to the partition from the health knowledge graph, that is, the features that should be detected in the partition, and detect the partition to determine whether the partition has the found features. For example, assuming that the partition is the lip, according to the health knowledge graph, it is determined that the feature to be detected is color. Then, using a deep learning neural network model, the color of the lips is detected, and it is detected that the lips are purple. Based on the region plus features (i.e., lips + purple), the symptom of purple lips can be obtained, which is a single health state. After determining the single health state, all these single health states are combined to infer the possible diseases of the object to be detected.

[0133] Example 5

[0134] Fig.10It is a method of intelligent face diagnosis based on health knowledge graph according to an embodiment of the present application, such as Fig.10 As shown, the method includes:

[0135] Step S1002, parse the TCM facial diagnosis theory text into entity, relationship, entity triples.

[0136] Parse medical diagnosis theory texts, such as TCM facial diagnosis theory texts, into (entity, relationship, entity) triples. The health knowledge graph needs to take the existing TCM facial diagnosis theory text data as input. These are theoretical descriptive texts, which are composed of sentences. They need to be converted into (entity, relationship, entity) triples through grammatical analysis and semantic analysis.

[0137] The construction of TCM facial diagnosis theory text into triples will be described in detail below and will not be repeated here.

[0138] Step S1004: merge entities and relationships.

[0139] Due to the imprecise description of TCM facial diagnosis theory, there may be multiple words describing the same entity, and the essence of multiple relationships is the same relationship. Therefore, it is necessary to perform synonym merging on both entities and relationships to obtain standardized entity tables and relationship tables.

[0140] Step S1006, abstracting the ontology expression model of TCM facial diagnosis.

[0141] A higher level of abstraction is performed on the entities and relationships of face-to-face consultation to form an ontology expression model of Chinese medicine face-to-face consultation. The local expression model describes the health knowledge graph in a more abstract dimension, which is convenient for checking the data standardization in the knowledge graph of card building and external business docking.

[0142] Step S1008, reasoning application of health knowledge graph based on traditional Chinese medicine facial diagnosis.

[0143] After the health knowledge graph of TCM consultation is constructed, various reasoning applications can be performed based on the health knowledge graph, such as predictive analysis from symptoms to diseases, correlation analysis of symptoms, etc.

[0144] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0145] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention. More examples are not listed here one by one.

[0146] Example 6

[0147] An embodiment of the present application provides a health perception and cognition device, including: a health perception module, which is configured to perform health perception of an object to be detected based on deep learning, and perceive each individual health status of the object to be detected; a health cognition module, which is configured to perform health cognition based on a pre-constructed health knowledge graph, integrating each of the individual health statuses, so as to obtain the health status of the object to be detected through reasoning and analysis.

[0148] In an exemplary embodiment, the health perception module is further configured to: receive a picture including at least part of the object to be detected; segment the picture based on the area division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, use a trained deep neural network model to perform feature inference and prediction to obtain the health symptom results of the partition, so as to perceive each individual health status of the object to be detected.

[0149] In an exemplary embodiment, the health perception module is further configured to: according to each of the multiple partitions, find out the features associated with the partition from the health knowledge graph; based on the found features, use the deep neural network model to perform inference prediction on the features, determine whether the partition has the features, and obtain the health symptom results of the partition.

[0150] In an exemplary embodiment, the device also includes a health knowledge graph construction module, which is configured to: abstract an ontology expression model, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; based on the ontology expression model, construct the health knowledge graph using the medical diagnosis theory text.

[0151] In an exemplary embodiment, the device also includes an ontology model construction module, which is configured to: abstract multiple entities and multiple relationships, wherein the multiple entities include features, regions, symptoms, diseases, organs, and the multiple relationships include detected, composed, reflected, occurred, and inferred; through the multiple relationships, establish connections between the multiple entities to construct the ontology expression model.

[0152] In an exemplary embodiment, the ontology expression model construction module is also configured to: establish connections between multiple entities through multiple relationships, including at least one of the following: setting the relationship between the feature and the region to be a "detected" relationship to indicate that the feature is detected from the region; setting the relationship between the feature and the symptom to be a "composed of" relationship to indicate that the feature constitutes the symptom; setting the relationship between the region and the symptom to be a "composed of" relationship to indicate that the region constitutes the symptom; setting the symptom itself to be a "composed of" relationship to indicate that the symptom constitutes a new symptom; setting the relationship between the region and the organ to be a reflection relationship to indicate that the region reflects the organ; setting the relationship between the disease and the organ to be "occurs in" to indicate that the disease occurs in the organ; setting the relationship between the symptom and the disease to be "inferred from" to indicate that the disease is inferred from the symptom.

[0153] In an exemplary embodiment, the health knowledge graph construction module is further configured to: based on the ontology expression model, parse the medical diagnosis theory text into multiple triples through grammatical analysis and semantic analysis, wherein each triple in the multiple triples includes two entities and the relationship between the two entities; merge the multiple triples, remove redundant triples in the multiple triples, so as to optimize the multiple triples; extract an entity table including entity types and entity instances, and a relationship table including relationship types and relationship instances from the optimized multiple triples, wherein the health knowledge graph includes the entity table and the relationship table.

[0154] In an exemplary embodiment, the health knowledge graph construction module is further configured to: merging the plurality of triples comprises: merging the plurality of triples with the same or similar meanings into one triple, and representing the merged triples by setting attributes for the merged triples, wherein the attributes comprise alias attributes for describing the names of the entities in the triples and / or position description attributes for describing the positions of the entities in the triples.

[0155] In an exemplary embodiment, the health cognition module is further configured as at least one of the following: taking an organ instance corresponding to an organ in the multiple entities in the ontology expression model as a node, searching for a disease associated with the organ instance; taking a feature instance corresponding to a feature in the multiple entities in the ontology expression model as a node, searching for region instances corresponding to regions in the multiple entities in which the feature instance appears, and based on the relationship between the found region instance and the feature instance, determining the symptom and inferring the disease associated with the symptom.

[0156] The device provided in this embodiment can implement the methods in embodiments 1 to 4, which will not be described in detail here.

[0157] Example 7

[0158] According to an embodiment of the present invention, a method for detecting a picture of an object to be detected is provided. The method comprises:

[0159] Step S1102, segmenting the image based on the health knowledge graph.

[0160] A picture of the object to be detected is received. In this embodiment, a face picture is taken as an example. Of course, in other embodiments, the picture may also include other parts of the object to be detected.

[0161] Segment the image based on the health knowledge graph, for example, segment the face image

[0162] Step S1104: performing feature detection on each segmented region.

[0163] Take a square area of ​​N*N pixels on the image with a sliding window, calculate the grayscale value of the pixels in the square area, and obtain the grayscale value matrix of the square area. Determine the credibility value of the feature according to each grayscale value in the grayscale value matrix corresponding to the area, and take the point with a credibility value greater than the credibility threshold as the feature point. Features, for example, can be color, patch, texture, etc.

[0164] The credibility value can be calculated by the following formula:

[0165]

[0166] Where M represents the credibility value, N represents the number of pixels, i represents the row of pixels in the matrix, j represents the column of pixels in the matrix, and P ij Represents the gray value, represents the average gray value, μ represents the correction value, a ij Represents the compensation factor for each pixel in the matrix.

[0167] The step size N of the sliding window may be 5, or a natural number less than 10. The sliding window is slid with a preset step size to traverse the corresponding area in the sample image, thereby obtaining the credibility values ​​of all pixels in the corresponding area, and then determining the number of feature points in the corresponding area.

[0168] According to the preset threshold of the number of feature points, the features in the area can be accurately identified. For example, the area without wrinkles, slight wrinkles, moderately severe wrinkles and severe wrinkles can be accurately identified; or the color of yellow, white, blue, dark, red, purple and black can be accurately identified; or the plaque type can be accurately identified as acne, white spots, spots, acne, mole, freckle and rosacea; or the texture can be accurately identified as wrinkles, horizontal lines, vertical lines, crow's feet and red blood streaks.

[0169] The feature detection method in this embodiment can be combined with the health perception and cognition method in any of the above embodiments. The feature detection method in this embodiment can make the features detected from the image more accurate. In this way, when the health perception and cognition method is executed, the symptoms obtained by adding regions and features are more accurate because the detected features are more accurate, so that the disease of the object to be detected can be inferred more accurately.

[0170] Example 8

[0171] According to an embodiment of the present invention, a health perception and cognition method is provided, the method comprising:

[0172] Step S1202: Perform health perception on the object to be detected based on deep learning to perceive each individual health status of the object to be detected.

[0173] In an exemplary embodiment, health perception of an object to be detected is performed based on deep learning to perceive each individual health status of the object to be detected, including: determining a partition to be detected and a feature corresponding to the partition based on the health knowledge graph; using a trained deep neural network model to detect whether the feature exists in the partition, and obtaining a health symptom result for the partition.

[0174] For example, a picture including at least part of the object to be detected is received; the picture is segmented based on the regional division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, the features that should be detected in each partition are determined, and the trained deep neural network model is used to detect the corresponding features of each partition, and the health symptoms of the partition are inferred and predicted to perceive the individual health states of the object to be detected.

[0175] In an exemplary embodiment, for each of the multiple partitions, a trained deep neural network model is used to perform feature inference and prediction to obtain the health symptom results of the partition, including: according to each of the multiple partitions, searching for features associated with the partition from the health knowledge graph; based on the found features, using the model trained by the deep neural network to perform feature inference and prediction to determine whether the partition has the features, and obtain the health symptom results of the partition.

[0176] Step S1204, based on the pre-constructed health knowledge graph, health cognition is performed by integrating each of the single health states to obtain the health status of the object to be detected through reasoning and analysis.

[0177] This embodiment mainly realizes health cognition through the health knowledge graph, and determines the area and features to be detected in step S1202 through the knowledge of the health knowledge graph.

[0178] In an exemplary embodiment, a health knowledge graph needs to be pre-built. Building a health knowledge graph includes: abstracting an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; and building the health knowledge graph based on the ontology expression model and using the medical diagnosis theory text.

[0179] By constructing an ontology expression model, unstructured medical diagnostic theory text can be converted into a structured expression, making it possible to construct a health knowledge graph.

[0180] In an exemplary embodiment, abstracting an ontology expression model for medical diagnosis includes: abstracting multiple entities and multiple relationships, wherein the multiple entities include features, regions, symptoms, diseases, and organs, and the multiple relationships include detected, composed, reflected, occurred, and inferred; through the multiple relationships, establishing connections between the multiple entities to construct the ontology expression model.

[0181] By abstracting the entities in the ontology expression model into features, regions, symptoms, diseases, and organs, and abstracting the relationships into detected, composed, reflected, occurred, and inferred, entities can be connected through relationships.

[0182] In an exemplary embodiment, connections between multiple entities are established through multiple relationships, including at least one of the following: setting the relationship between the feature and the region to be a "detected" relationship, to indicate that the feature is detected from the region; setting the relationship between the feature and the symptom to be a "composed of" relationship, to indicate that the feature constitutes the symptom; setting the relationship between the region and the symptom to be a "composed of" relationship, to indicate that the region constitutes the symptom; setting the symptom itself to be a "composed of" relationship, to indicate that the symptom constitutes a new symptom; setting the relationship between the region and the organ to be a reflection relationship, to indicate that the region reflects the organ; setting the relationship between the disease and the organ to be "occurs in", to indicate that the disease occurs in the organ; setting the relationship between the symptom and the disease to be "inferred out", to indicate that the disease is inferred from the symptom.

[0183] Through the above method, symptoms can be obtained by adding features to regions, and diseases can be inferred from symptoms. The disease can be inferred back to the organ where the disease occurs, and the organ reflected by the region can also be inferred from the region, thus providing a logical basis for reasoning about the disease.

[0184] In an exemplary embodiment, the health knowledge graph is constructed based on the ontology expression model using the medical diagnosis theory text, including: based on the ontology expression model, parsing the medical diagnosis theory text into multiple triples through grammatical analysis and semantic analysis, wherein each triple in the multiple triples includes two entities and the relationship between the two entities; merging the multiple triples, removing redundant triples in the multiple triples, so as to optimize the multiple triples; extracting an entity table including entity types and entity instances, and a relationship table including relationship types and relationship instances from the optimized multiple triples, wherein the health knowledge graph includes the entity table and the relationship table.

[0185] Through the above method, the specific medical diagnosis theory text is extracted into triple expressions according to the structure of the ontology expression model, that is, the entity, relationship, entity expression, so as to convert the unstructured specific medical diagnosis theory text into a structured expression. And based on multiple triple data, a health knowledge graph including entity tables and relationship tables can be constructed, so that the corresponding disease can be queried from the health knowledge graph based on the detected features and the area corresponding to the features, so that the reasoning of the disease is more accurate.

[0186] In an exemplary embodiment, health perception of an object to be detected is performed based on deep learning to perceive each individual health status of the object to be detected, including: receiving a picture including at least a part of the object to be detected; segmenting the picture based on the area division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, finding a feature corresponding to the partition from the health knowledge graph, detecting the partition to determine whether the partition exists the found feature; determining each individual health status of the object to be detected based on the detection result, wherein the individual health status includes symptoms determined based on the partition plus the features existing in the partition.

[0187] Through the above method, the symptoms of the area can be detected according to the features and corresponding areas in the input image, so that potential diseases can be further deduced based on the features and areas.

[0188] In an exemplary embodiment, after constructing the health knowledge graph, the method further includes at least one of the following: taking an organ instance corresponding to an organ in the multiple entities in the ontology expression model as a node, searching for a disease associated with the organ instance; taking a feature instance corresponding to a feature in the multiple entities in the ontology expression model as a node, searching for region instances corresponding to regions in the multiple entities in which the feature instance appears, and based on the relationship between the found region instance and the feature instance, determining the symptom and inferring the disease associated with the symptom.

[0189] Through the above method, relevant divergent reasoning can be carried out with specific organs or specific characteristics as the central node, thereby providing more ways for disease diagnosis.

[0190] Through this embodiment, health detection and analysis can be well upgraded from the current health perception level to the level of health cognition, thereby improving the accuracy of health detection and the depth of analysis.

[0191] Example 9

[0192] This embodiment provides a device for constructing a health knowledge graph, including an ontology expression model construction module and a health knowledge graph construction module.

[0193] The ontology expression model construction module is configured to abstract an ontology expression model for medical diagnosis, where the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner.

[0194] The health knowledge graph construction module is configured to construct the health knowledge graph based on the ontology expression model and using the medical diagnosis theory text.

[0195] In an exemplary embodiment, the ontology expression model construction module is configured to: abstract multiple entities and multiple relationships, wherein the multiple entities include features, regions, symptoms, diseases, organs, and the multiple relationships include detected, composed, reflected, occurred, and inferred; through the multiple relationships, establish connections between the multiple entities to construct the ontology expression model.

[0196] In an exemplary embodiment, the ontology expression model construction module is configured as at least one of the following: setting the relationship between the feature and the region to be a "detected" relationship to indicate that the feature is detected from the region; setting the relationship between the feature and the symptom to be a "composed of" relationship to indicate that the feature constitutes the symptom; setting the relationship between the region and the symptom to be a "composed of" relationship to indicate that the region constitutes the symptom; setting the symptom itself to be a "composed of" relationship to indicate that the symptom constitutes a new symptom; setting the relationship between the region and the organ to be a reflection relationship to indicate that the region reflects the organ; setting the relationship between the disease and the organ to be "occurred in" to indicate that the disease occurs in the organ; setting the relationship between the symptom and the disease to be "inferred out" to indicate that the disease is inferred from the symptom.

[0197] In an exemplary embodiment, the health knowledge graph construction module is configured to: parse the medical diagnosis theory text into multiple triples through grammatical analysis and semantic analysis, wherein each triple in the multiple triples includes two entities and the relationship between the two entities; merge the multiple triples, remove redundant triples in the multiple triples, so as to optimize the multiple triples; extract an entity table including entity types and entity instances, and a relationship table including relationship types and relationship instances from the optimized multiple triples, wherein the health knowledge graph includes the entity table and the relationship table.

[0198] In an exemplary embodiment, the health knowledge graph construction module is configured to merge multiple triples with the same or similar meanings into one triple, and represent the merged triple by setting attributes for the merged triple, wherein the attributes include an alias attribute for describing the name of the entity in the triple and / or a position description attribute for describing the position of the entity in the triple.

[0199] In an exemplary embodiment, the device further includes a health perception module and a health cognition module. The health perception module is configured to: perform health perception of the object to be detected based on deep learning, and perceive each individual health status of the object to be detected; the health cognition module is configured to perform health cognition based on the health knowledge graph and integrate each individual health status to obtain the health status of the object to be detected through reasoning and analysis.

[0200] In an exemplary embodiment, the health perception module is configured to: determine the partition to be detected and the features corresponding to the partition based on the health knowledge graph; use a trained deep neural network model to detect whether the partition has the feature, and obtain the health symptom results of the partition.

[0201] In an exemplary embodiment, the health perception module is further configured to: receive a picture including at least a portion of the object to be detected; segment the picture based on the area division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, find out the features corresponding to the partition from the health knowledge graph, and detect the partition to determine whether the partition has the features corresponding to the partition.

[0202] In an exemplary embodiment, the health cognition module is further configured as at least one of the following: taking an organ instance corresponding to an organ in the multiple entities in the ontology expression model as a node, searching for a disease associated with the organ instance; taking a feature instance corresponding to a feature in the multiple entities in the ontology expression model as a node, searching for region instances corresponding to regions in the multiple entities in which the feature instance appears, and based on the relationship between the found region instance and the feature instance, determining the symptom and inferring the disease associated with the symptom.

[0203] The health knowledge graph provided in this embodiment can be combined with the health perception and cognition method in this application. Through the construction method in this embodiment, a health knowledge graph that more accurately reflects the medical diagnosis theory text can be constructed, thereby providing the possibility for intelligent medical diagnosis.

[0204] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a health knowledge graph, characterized in that: include: Abstracting an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; Based on the ontology expression model, the health knowledge graph is constructed using the medical diagnosis theory text; The health knowledge graph is used to understand the health of the subject to be tested; After constructing the health knowledge graph, the method further includes: based on deep learning, performing health perception on the subject to be detected, and perceiving each individual health status of the subject to be detected; based on the health knowledge graph, performing health cognition by integrating each individual health status, and obtaining the health status of the subject to be detected by reasoning and analysis; Among them, the health perception of the object to be detected based on deep learning includes: determining the partition to be detected and the features corresponding to the partition based on the health knowledge graph; using the trained deep neural network model to detect whether the partition has the feature, and obtaining the health symptom result of the partition; Among them, determining the partition to be detected and the features corresponding to the partition based on the health knowledge graph includes: receiving a picture including at least a part of the object to be detected; segmenting the picture based on the area division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, searching the health knowledge graph for the features corresponding to the partition, and detecting the partition to determine whether the partition has the features corresponding to the partition.

2. The method according to claim 1, characterized in that The ontology expression model for medical diagnosis is abstracted as follows: Abstracting multiple entities and multiple relationships, wherein the multiple entities include features, regions, symptoms, diseases, organs, and the multiple relationships include detected, composed, reflected, occurred, and inferred; Through the multiple relationships, connections between the multiple entities are established to construct the ontology expression model.

3. The method according to claim 2, characterized in that Establishing a connection between the multiple entities through the multiple relationships includes at least one of the following: Setting a "detected" relationship between the feature and the region to indicate that the feature is detected from the region; The relationship between the feature and the symptom is set to be "composed of" to indicate that the feature constitutes the symptom; A "compose of" relationship is set between the region and the symptom, to indicate that the region constitutes the symptom; Set the symptom itself to a "compose of" relationship to indicate that the symptom constitutes a new symptom; Setting a reflection relationship between the region and the organ to indicate that the region reflects the organ; Set the relationship between the disease and the organ as "occurred in" to indicate that the disease occurred in the organ; The relationship between the symptom and the disease is set to "inferred" to indicate that the disease is inferred from the symptom.

4. The method according to any one of claims 2 to 3, characterized in that Based on the ontology expression model, the health knowledge graph is constructed using the medical diagnosis theory text, including: Parsing the medical diagnosis theory text into a plurality of triples through grammatical analysis and semantic analysis, wherein each triple in the plurality of triples includes two entities and a relationship between the two entities; Merging the plurality of triples, removing redundant triples from the plurality of triples, so as to optimize the plurality of triples; An entity table including entity types and entity instances, and a relationship table including relationship types and relationship instances are extracted from the optimized multiple triples, wherein the health knowledge graph includes the entity table and the relationship table.

5. The method according to claim 4, characterized in that Merging multiple triples includes: merging multiple triples with the same or similar meanings into one triple, and representing the merged triples by setting attributes for the merged triples, wherein the attributes include alias attributes for describing the names of the entities in the triples and / or position description attributes for describing the positions of the entities in the triples.

6. The method according to claim 2, characterized in that After constructing the health knowledge graph, the method further includes at least one of the following: Taking an organ instance corresponding to an organ in the plurality of entities in the ontology expression model as a node, searching for a disease associated with the organ instance; Taking the feature instances corresponding to the features in the multiple entities in the ontology expression model as nodes, find out which region instances corresponding to the regions in the multiple entities the feature instances appear in, and based on the relationship between the found region instances and the feature instances, determine the symptoms and infer the diseases associated with the symptoms.

7. A device for constructing a health knowledge graph, characterized in that: include: An ontology expression model building module is configured to abstract an ontology expression model for medical diagnosis, wherein the ontology expression model is used to abstractly describe unstructured medical diagnosis theory text in a structured expression manner; A health knowledge graph construction module is configured to construct the health knowledge graph based on the ontology expression model and using the medical diagnosis theory text; After constructing the health knowledge graph, the device is further configured to: perform health perception of the object to be detected based on deep learning, and perceive each individual health status of the object to be detected; based on the health knowledge graph, perform health cognition by integrating each individual health status, and obtain the health status of the object to be detected by reasoning and analysis; The device is further configured to: determine the partition to be detected and the feature corresponding to the partition based on the health knowledge graph; use the trained deep neural network model to detect whether the partition has the feature, and obtain the health symptom result of the partition; Wherein, the device is also configured to: receive a picture including at least a part of the object to be detected; segment the picture based on the area division rules in the health knowledge graph to obtain multiple partitions; for each of the multiple partitions, find out the features corresponding to the partition from the health knowledge graph, and detect the partition to determine whether the partition has the features corresponding to the partition.

Images