Disease prediction method based on disease diagnosis standard knowledge graph

Abstract

The application discloses a disease prediction method based on a disease diagnosis standard knowledge graph, comprising the following steps: constructing a disease diagnosis standard knowledge graph comprising clinical manifestation class entities, examination item class entities, examination result class entities, logic node class entities and disease class entities; finding a first entity set corresponding to the detection result of the examination item of a patient and the clinical manifestation of the patient; taking the entities in the first entity set as starting nodes in sequence, finding disease class entities reached through at least one joint judgment relationship; taking the disease class entities as root nodes in sequence, finding an N-ary tree composed of all entities and relationships pointing to the disease entities through the joint judgment relationship; and judging the logic state of each entity in the N-ary tree. The application has the advantages that the significance of medical examination items and examination results is represented, the whole disease diagnosis knowledge graph is more accurate and more detailed, and the disease diagnosis can be made independently without depending on the professional knowledge and experience of doctors.

Description

Technical Field

[0001] The present invention relates to the field of intelligent disease prediction, and in particular to a disease prediction method based on a knowledge graph of disease diagnosis standards. Background Art

[0002] Due to the uneven distribution of existing medical resources, there is a significant gap between the disease diagnosis capabilities of primary care hospitals and those of tertiary hospitals. This is particularly true for some rare diseases, which can only be correctly diagnosed by experts at tertiary hospitals. If a disease is misdiagnosed by a physician lacking sufficient medical expertise or experience, it can delay the onset of illness and treatment, causing significant suffering for the patient. However, many well-organized knowledge graphs in healthcare, such as the International Classification of Diseases, DrugBank (a drug database encompassing bioinformatics and cheminformatics), and clinical guidelines and consensus, contain hierarchical information and complex relationships consistent with human cognition. Leveraging these existing knowledge graphs, combined with computer technology, can equip computers with disease prediction capabilities, thereby helping to address the current imbalance in the distribution of medical resources and their heavy reliance on physician expertise and experience.

[0003] However, on the one hand, the existing knowledge graphs do not fully describe the significance of medical test items and test results, and the precise diagnosis methods for diseases are not detailed enough; on the other hand, existing computer disease prediction methods, such as graph neural network modeling based on electronic medical records, staged modeling based on knowledge representation learning and disease prediction, and binary classification machine learning methods for single diseases, require separate training and maintenance models for each disease when used clinically. This is costly, has poor interpretability, and has high uncertainty in output results. It cannot distinguish between diagnosable diseases and suspected diseases, which means that current intelligent disease prediction can only be used as a reference suggestion, and cannot truly solve the uneven distribution of medical resources, and is highly dependent on the professional knowledge and experience of doctors. Summary of the Invention

[0004] The purpose of the present invention is to provide a disease prediction method based on a disease diagnosis standard knowledge graph, which can predict the patient's diagnosable diseases and suspected diseases based on the patient's physical signs and test results data and with reference to the disease diagnosis knowledge graph.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] The disease prediction method based on the disease diagnosis standard knowledge graph of the present invention comprises the following steps:

[0007] S1, construct a disease diagnosis standard knowledge graph containing clinical manifestation entities, test item entities, test result entities, logical node entities and disease entities;

[0008] S2, finding the first entity set in the disease diagnosis standard knowledge graph that is consistent with the test results of the patient's test items and the patient's clinical manifestations;

[0009] S3, starting with the entities in the first entity set, traversing the disease diagnosis standard knowledge graph in sequence, finding the disease-related entities that the entities reach through at least one joint judgment relationship, and forming a suspected disease set A;

[0010] S4, taking the disease entities in the suspected disease set A as root nodes, traverse the disease diagnosis standard knowledge graph in turn to find an N-ary tree consisting of all entities and relationships that point to the disease entity through a joint judgment relationship;

[0011] S5, using a post-order traversal method on the N-ary tree, sequentially determining the logical state of each entity in the N-ary tree until the logical state of the root node is obtained;

[0012] S6, when the logical state of the root node is true, the root node is a diagnosable disease; otherwise, the root node is a suspected disease.

[0013] The present invention adds the meaning of medical test items and test results when constructing the knowledge graph of disease diagnosis standards. The entire disease diagnosis method is more accurate and detailed, and can make disease diagnoses independently without relying on the doctor's professional knowledge and experience.

[0014] Furthermore, the clinical manifestation entity refers to abnormal changes in the body after suffering from a certain disease; the test item entity refers to medical examination items; the test result entity refers to the value or range of the test item; the logical node entity is used to express the logical conditions of the entity; and the disease entity is used to express a specific disease.

[0015] Furthermore, the logical node type entities include AND, OR, NOT, ≥n, ≤n, where n represents the number of preceding entities that meet the conditions.

[0016] Furthermore, the relationship between entities in the disease diagnosis standard knowledge graph includes the joint judgment and test result relationship; the joint judgment relationship refers to a relationship in which the logical state of the tail entity needs to be determined based on the logical state of the head entity;

[0017] The inspection result relationship refers to the relationship between the inspection item type entity and the inspection result type entity.

[0018] Furthermore, the logic state judgment rules of the entity are as follows:

[0019] If the entity type is the clinical manifestation class, and the patient's clinical manifestations meet the entity, the entity logic state is true; otherwise, the entity logic state is false;

[0020] If the entity type is the test result class, determine whether the patient's test items and the test results of the test items meet the entity combination. If they meet, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0021] If the entity type is the logical node class, determine whether the logical states of all head entities that have a direct joint judgment relationship with the entity meet the logical requirements of the entity, and determine the logical state of the entity;

[0022] Wherein, when the entity is AND, if the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0023] When the entity is OR, if any of the logical states of all the head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0024] When the entity is negative, if the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is false; otherwise, the logical state of the entity is true;

[0025] When the entity is ≥n, if the logical state of n entities or more of all head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0026] When the entity is ≤n, if the logical state of n entities or less among all head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0027] If the entity type is a disease type, when the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false.

[0028] Furthermore, the entity combination is an entity pair consisting of the inspection item entity and the inspection result entity connected by the inspection result relationship.

[0029] The advantage of this invention lies in converting disease diagnosis criteria into an N-ary tree-like structure that is easily stored and accessed in a knowledge graph. This also adds a more meaningful representation of medical test items and test results, making the entire disease diagnosis knowledge graph more accurate and detailed, allowing independent disease diagnosis without relying on the physician's professional knowledge and experience. At the same time, a unified disease prediction method is proposed that can simultaneously predict all diseases within the disease diagnosis criteria knowledge graph. This addresses the drawbacks of existing computer disease prediction methods, which require separate training and maintenance models for each disease, resulting in high costs, poor interpretability, high uncertainty in output results, and an inability to distinguish between confirmed and suspected diseases. This truly achieves intelligent disease prediction and provides a new approach and method for addressing the uneven distribution of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the method described in the present invention (to be supplemented after the first draft is finalized).

[0031] Figure 2 This is a schematic diagram of the knowledge graph of local disease diagnosis standards involved in Example 2 of the present invention.

[0032] Figure 3 Schematic diagram of the N-ary tree of multiple myeloma in Example 2 of the present invention. DETAILED DESCRIPTION

[0033] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention. Specific embodiment 1

[0035] like Figure 1 As shown, the disease prediction method based on the disease diagnosis standard knowledge graph of the present invention includes the following steps:

[0036] S1, construct a disease diagnosis standard knowledge graph containing clinical manifestation entities, test item entities, test result entities, logical node entities and disease entities;

[0037] Among them, clinical manifestation entities refer to abnormal changes that occur in the body after suffering from a certain disease; for example, cough, fever, headache, weakness and other symptoms.

[0038] Inspection item entities refer to medical inspection items, such as biochemical immunology, microbiological culture, and imaging examinations.

[0039] The test result entity refers to the value or range of the test item; such as negative, positive, etc.

[0040] Logical node type entities are used to express the logical conditions of the entities; logical node type entities include AND, OR, NOT, ≥n, ≤n, where n represents the number of header entities that meet the conditions.

[0041] Disease entities consist of specific diseases.

[0042] The relationship between entities in the disease diagnosis standard knowledge graph includes the joint judgment and test result relationship; the joint judgment relationship refers to the relationship that the logical state of the tail entity needs to be determined based on the logical state of the head entity.

[0043] The inspection result relationship refers to the relationship between the inspection item type entity and the inspection result type entity.

[0044] S2, finding the first entity set in the disease diagnosis standard knowledge graph that is consistent with the test results of the patient's test items and the patient's clinical manifestations;

[0045] S3, starting with the entities in the first entity set, traversing the disease diagnosis standard knowledge graph in sequence, finding the disease-related entities that the entities reach through at least one joint judgment relationship, and forming a suspected disease set A;

[0046] S4, taking the disease entities in the suspected disease set A as root nodes, traverse the disease diagnosis standard knowledge graph in turn to find an N-ary tree consisting of all entities and relationships that point to the disease entity through a joint judgment relationship;

[0047] S5, using a post-order traversal method on the N-ary tree, sequentially determining the logical state of each entity in the N-ary tree until the logical state of the root node is obtained;

[0048] S6, when the logical state of the root node is true, the root node is a diagnosable disease; otherwise, the root node is a suspected disease.

[0049] The logical state judgment rules of the entity are as follows:

[0050] If the entity type is the clinical manifestation class, and the patient's clinical manifestations meet the entity, the entity logic state is true; otherwise, the entity logic state is false;

[0051] If the entity type is a test result, find the test item entity that has a test result relationship with the test result entity to form a test item and test result entity combination. Then, determine whether the patient's test item and the test result of the test item match the entity combination. If so, the entity's logical state is true; otherwise, the entity's logical state is false.

[0052] If the entity type is a logical node class, determine whether the logical states of all head entities that have a direct joint judgment relationship with the entity meet the logical requirements of the entity, and determine the logical state of the entity;

[0053] Among them, when the entity is AND, if the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0054] When the entity is or, if any of the logical states of all the head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0055] When the entity is negative, if the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is false; otherwise, the logical state of the entity is true;

[0056] When the entity is ≥n, the logical state of the entity is true if the logical state of n entities or more of all head entities that have a direct joint judgment relationship with the entity is true; otherwise, the logical state of the entity is false;

[0057] When the entity is ≤n, if the logical states of n entities or less among all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false;

[0058] If the entity type is a disease type, when the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false. Specific embodiment 2

[0060] The patient data are as follows:

[0061] Name: Zhang San

[0062] Gender: Male

[0063] Age: 46

[0064] Clinical manifestations: osteolytic lesions

[0065]

[0066] According to the patient's test items, test results and clinical manifestation data, find the entities that match the test results of the patient's test items and the entities that match the patient's clinical manifestation data in the disease diagnosis standard knowledge graph to form a first entity set; Figure 2 The entities listed in the first and second rows of the local disease diagnosis standard knowledge graph are shown.

[0067] Taking the patient test item "bone marrow nucleated cell classification and counting plasma cells" as an example, the test result is 23%. The test item entity of "bone marrow nucleated cell classification and counting plasma cells" is found in the disease diagnosis standard knowledge graph, and entity 1 is obtained; entity 11 with the test result relationship with entity 1 "bone marrow nucleated cell classification and counting plasma cells" is " Entity 1 and entity 11 form a test item and test result entity combination, namely, "bone marrow nucleated cell differential count plasma cells>15%"; the test result of the patient's test item "bone marrow nucleated cell differential count plasma cells" is 23%, which conforms to the entity combination logic, so entity 11 is placed in the first entity set.

[0068] Similarly, for example, if a patient's test result for "Bone Marrow Nucleated Cell Differential Count Plasma Cells" is positive, the disease diagnosis standard knowledge graph searches for the test item entity "Bone Marrow Nucleated Cell Differential Count Plasma Cells," yielding entity 2. Entity 12, which has a test result relationship with entity 2 ("Bone Marrow Nucleated Cell Differential Count Plasma Cells"), is "Positive." Entities 2 and 12 form the test item and test result entity combination "Bone Marrow Nucleated Cell Differential Count Plasma Cells Positive." The patient's test result for "Bone Marrow Nucleated Cell Differential Count Plasma Cells" is positive, which conforms to the entity combination logic, so entity 12 is placed in the first entity set.

[0069] Taking the patient's clinical manifestation "osteolytic lesion" as an example, the clinical manifestation entity of "osteolytic lesion" is searched in the disease diagnosis standard knowledge graph, and entity 22 is obtained, and entity 22 is placed in the first entity set.

[0070] Similarly, all clinical manifestations and test item data of the patient are traversed to obtain the first entity set {entity 11, entity 12, entity 14, entity 15, entity 16, entity 18, entity 19, entity 22}.

[0071] Then, starting from the entities in the first entity set, we traverse the disease diagnosis standard knowledge graph and find the diseases that can be reached by the entity through at least one joint judgment relationship, thus forming a suspected disease set A. Figure 2 The three suspected diseases are listed in the last row of the local disease diagnosis standard knowledge graph shown.

[0072] Taking entity 11 as an example, entity 24 has a joint judgment relationship with entity 11, entity 25 has a joint judgment relationship with entity 24, entity 26 has a joint judgment relationship with entity 25, and the entity with a joint judgment relationship with entity 26 is disease entity 27 "multiple myeloma". Entity 27 is placed in suspected disease set A;

[0073] Similarly, taking entity "12" as an example, entity 23 has a joint judgment relationship with entity 12, entity 24 has a joint judgment relationship with entity 23, entity 25 has a joint judgment relationship with entity 24, entity 26 has a joint judgment relationship with entity 25, and the entity with a joint judgment relationship with entity 26 is disease entity 27 "multiple myeloma". At this time, entity 27 already exists in A and does not need to be processed again.

[0074] Taking entity 15 as an example, entity 28 has a joint judgment relationship with entity 15, and entities 26, 29, and 31 have a joint judgment relationship with entity 28. The entity with a joint judgment relationship with entity 26 is disease entity 27 "Multiple Myeloma"; the entity with a joint judgment relationship with entity 29 is disease entity 30 "Progressive Multiple Myeloma"; entity 33 has a joint judgment relationship with entity 31, and the entity with a joint judgment relationship with entity 33 is disease entity 34 "Non-secretory Multiple Myeloma". Entity 34 is placed in suspected disease set A;

[0075] Traverse the first entity set in the same way to obtain the suspected disease set A{entity 27, entity 30, entity 34}

[0076] Next, take the diseases in the suspected disease set A as the root nodes, traverse the disease diagnosis standard knowledge graph in turn, find all entities pointing to the disease through the joint judgment relationship, and form an N-ary tree with the root node and the joint judgment relationship, denoted as T; use the post-order traversal method on the N-ary tree T to judge the logical state of each entity in the N-ary tree T in turn until the logical state of the root node is obtained; when the logical state of the root node is true, the root node is a diagnosable disease; otherwise, the root node is a suspected disease.

[0077] like Figure 2 The local disease diagnosis standard knowledge graph shown takes the suspected disease of multiple myeloma as an example to illustrate the process as follows. All entities are replaced by serial numbers.

[0078] Taking entity 27 "multiple myeloma" as the root node, traverse the disease diagnosis standard knowledge graph, and through the joint judgment relationship, the entity that directly points to the disease entity 27 "multiple myeloma" is entity 26;

[0079] For entity 26, the entities that directly point to entity 25 through the joint judgment relationship are entity 25 and entity 28;

[0080] For entity 25, the entities that directly point to entity 25 through the joint judgment relationship are entity 24 and entity 14;

[0081] For entity 24, the entities that directly point to entity 24 through the joint judgment relationship are entity 11 and entity 23;

[0082] For entity 11, no entity points to entity 11 through the joint judgment relationship, and entity 11 is a leaf node. However, the entity that directly points to entity 11 through the detection result relationship is entity 1;

[0083] For entity 23, the entities that directly point to entity 23 through the joint judgment relationship are entity 12 and entity 13;

[0084] For entities 12 and 13, there is no entity pointing to them through the joint judgment relationship, so entities 12 and 13 are leaf nodes. However, the entities that directly point to entities 12 and 13 through the detection result relationship are entities 2 and 3;

[0085] For entity 14, no entity points to entity 14 through the joint judgment relationship, and entity 14 is a leaf node. However, the entity that directly points to entity 14 through the detection result relationship is 4;

[0086] For entity 28, the entities that directly point to entity 28 through the joint judgment relationship are entities 15, 16, 17, 18, 19, and 20;

[0087] For entities 15, 16, 17, 18, 19, and 20, no entity points to entities 15, 16, 17, 18, 19, and 20 through the joint judgment relationship. Entities 15, 16, 17, 18, 19, and 20 are leaf nodes; however, the entities that directly point to entities 15, 16, 17, 18, 19, and 20 through the detection result relationship are entities 5, 6, 7, 8, 9, and 10.

[0088] All the relationships and entities found are formed into an N-ary tree T, such as Figure 3 As shown. Then, the post-order traversal method is used to determine the logical state of each entity in the N-ary tree T in turn. The subsequent traversal method is as follows: starting from the root node, for each node, the first step is to visit the child nodes of the node if the node has child nodes; the second step is to put the node into the return list.

[0089] First, for the root node entity 27, the child node of entity 27 is entity 26;

[0090] For entity 26, the child nodes of entity 26 are entity 25 and entity 28;

[0091] For entity 25, the child nodes of entity 25 are entity 24 and entity 14;

[0092] For entity 24, the child nodes of entity 24 are entity 11 and entity 23;

[0093] For entity 11, entity 11 is a leaf node and has no child nodes, so entity 11 is put into the return list;

[0094] For entity 23, the child nodes of entity 23 are entity 12 and entity 13;

[0095] For entity 12, entity 12 is a leaf node and has no child nodes, so entity 12 is put into the return list;

[0096] For entity 13, entity 13 is a leaf node and has no child nodes, so entity 13 is put into the return list;

[0097] For entity 23, the child node traversal is completed and entity 23 is placed in the return list;

[0098] For entity 24, the child node traversal is completed and entity 24 is placed in the return list;

[0099] Similarly, until the root node entity 27 is put into the return list, the subsequent traversal of the N-ary tree is completed. The return list is {entity 11, 12, 13, 23, 24, 14, 25, 15, 16, 17, 18, 19, 20, 29, 26, 27}

[0100] Determine the logical status of each entity according to the order of entities in the returned list.

[0101] For example, entity 11 is a test result entity. Entity 1 is connected to entity 11 through the test result relationship, forming the entity combination "Bone marrow nucleated cell differential count plasma cells > 15%." The patient's test item data for "Bone marrow nucleated cell differential count plasma cells" has a value of 23%, satisfying this condition. Therefore, the logical state of entity 11 is true. Similarly, the logical state of entity 12 is true, while the logical state of entity 13 is false.

[0102] Taking entity 23 as an example, entity 23 is a logical node entity. The entity is "OR", and the logical state of entity 12, which has a direct joint judgment relationship with entity 23, is true, the logical state of entity 13 is false, and the logical state of entity 23 is true.

[0103] Similarly, it can be concluded that the logical state of entity 27 is true.

[0104] Similarly, in suspected disease set A, entity 30 has a logical state of false, while entity 34 has a logical state of true. Therefore, entity 27, multiple myeloma, is a confirmed disease, entity 30, progressive multiple myeloma, is a suspected disease, and entity 34, nonsecretory multiple myeloma, is a confirmed disease. Nonsecretory multiple myeloma is a more detailed subtype of multiple myeloma, so diagnosing both at the same time is not contradictory.

Claims

1. A disease prediction method based on a disease diagnosis standard knowledge graph, characterized by: The following steps are involved: S1, construct a disease diagnosis standard knowledge graph containing clinical manifestation entities, test item entities, test result entities, logical node entities and disease entities; logical node entities are used to express the logical conditions of the entities; logical node entities include AND, OR, NOT, ≥n, ≤n, where n represents the number of head entities that meet the conditions; S2, finding the first entity set in the disease diagnosis standard knowledge graph that is consistent with the test results of the patient's test items and the clinical manifestations of the patient; S3, starting with the entities in the first entity set, traversing the disease diagnosis standard knowledge graph in sequence to find the disease-related entities that the entities reach through at least one joint judgment relationship, thereby forming a suspected disease set A; a joint judgment relationship refers to a relationship in which the logical state of the tail entity is determined based on the logical state of the head entity; S4, taking the disease entities in the suspected disease set A as root nodes, traverse the disease diagnosis standard knowledge graph in turn to find an N-ary tree consisting of all entities and relationships that point to the disease entity through a joint judgment relationship; S5, using a post-order traversal method on the N-ary tree, sequentially determining the logical state of each entity in the N-ary tree until the logical state of the root node is obtained; S6, when the logical state of the root node is true, the root node is a diagnosable disease; otherwise, the root node is a suspected disease; The logic state judgment rules of the entity are as follows: If the entity type is the clinical manifestation class, and the patient's clinical manifestations meet the entity, the entity logic state is true; otherwise, the entity logic state is false; If the entity type is the test result class, determine whether the patient's test items and the test results of the test items meet the entity combination. If they do, the logical state of the entity is true; Otherwise, the entity's logical state is false; If the entity type is the logical node class, determine whether the logical states of all head entities that have a direct joint judgment relationship with the entity meet the logical requirements of the entity, and determine the logical state of the entity; Wherein, when the entity is AND, if the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false; When the entity is OR, if any of the logical states of all the head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false; When the entity is negative, if the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is false; otherwise, the logical state of the entity is true; When the entity is ≥n, if the logical state of n entities or more of all head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false; When the entity is ≤n, if the logical state of n entities or less among all head entities that have a direct joint judgment relationship with the entity is true, the logical state of the entity is true; otherwise, the logical state of the entity is false; If the entity type is a disease type, when the logical states of all head entities that have a direct joint judgment relationship with the entity are true, the logical state of the entity is true; otherwise, the logical state of the entity is false.

2. The disease prediction method based on the disease diagnosis standard knowledge graph according to claim 1 is characterized by: The clinical manifestation entity refers to the abnormal changes that occur in the body after suffering from a certain disease; the test item entity refers to the medical examination item; the test result entity refers to the value or range of the test item; and the disease entity is used to describe a specific disease.

3. The disease prediction method based on the disease diagnosis standard knowledge graph according to claim 1 is characterized in that: The relationship between entities in the disease diagnosis standard knowledge graph includes the joint judgment and test result relationship; the test result relationship refers to the relationship between the test item type entity and the test result type entity.

4. The disease prediction method based on the disease diagnosis standard knowledge graph according to claim 1 is characterized in that: The entity combination is an entity pair consisting of the inspection item entity and the inspection result entity connected by the inspection result relationship.

Images