CN117334352A

CN117334352A - Hypertension diagnosis and treatment decision reasoning method and device based on multiple role knowledge graph

Info

Publication number: CN117334352A
Application number: CN202311578997.4A
Authority: CN
Inventors: 鄂海红; 周庚显; 匡泽民; 汤子辰; 宋美娜; 谭玲
Original assignee: Beijing University of Posts and Telecommunications
Current assignee: Beijing University of Posts and Telecommunications
Priority date: 2023-11-24
Filing date: 2023-11-24
Publication date: 2024-01-02
Anticipated expiration: 2043-11-24
Also published as: CN117334352B

Abstract

The application provides a hypertension diagnosis and treatment decision reasoning method and device based on a multi-role knowledge graph, wherein the method comprises the following steps: constructing a hypertension diagnosis and treatment knowledge graph according to a multi-role knowledge paradigm and a generating rule, developing a plurality of single-hop inference operators based on actions of the knowledge graph induction and abstraction in the hypertension diagnosis and treatment, updating the state of a patient through a reasoning engine to realize multi-hop reasoning in the hypertension diagnosis and treatment knowledge graph, obtaining the blood pressure level and the hypertension danger layering of the patient, generating treatment measures and a plurality of candidate medication schemes according to medication decisions, recording decision reasons related to the plurality of candidate medication schemes, scoring each candidate medication scheme according to a medication recommendation mechanism by considering the indication, the tabulation and the medication scheme of the recommended medication, and obtaining a recommendation list and a warning list. The method can realize accurate diagnosis and medication of the hypertension patient, and the constructed framework is very universal and is easy to expand to the diagnosis and treatment decision field of other diseases.

Description

Hypertension diagnosis and treatment decision reasoning method and device based on multiple role knowledge graph

Technical Field

The application relates to the field of medical diagnosis, in particular to a hypertension diagnosis and treatment decision-making reasoning method and device based on a multivariate role knowledge graph.

Background

Knowledge graph is a graph-based way of representing knowledge. The graph is composed of nodes and edges, where the nodes represent entities in the real world and the edges represent relationships between the entities. Knowledge maps store rich domain knowledge based on which various knowledge-driven applications and services can be built. Therefore, the knowledge graph is widely applied to the fields of medical treatment, finance, knowledge question answering, recommendation systems and the like. In the medical field, since medical tasks often involve a great deal of medical knowledge, knowledge-bases are also naturally widely used by practitioners, and some common applications include: protein interactions, drug interactions, disease diagnosis, clinical decision support.

Hypertension is a common chronic disease worldwide. Diagnosis and treatment of hypertension involves a great deal of medical knowledge and clinical experience. Physicians need to grade the patient for hypertension, target blood pressure determination, risk level stratification, and then determine the treatment regimen to be used, depending on the patient's current condition. If the treatment scheme is drug treatment, doctors further give specific medication schemes according to the physical examination result of patients and the clinical complicated diseases, and different patients often need different antihypertensive drugs for treatment. In clinical decisions of hypertension diagnosis and treatment, each step involves corresponding clinical knowledge of hypertension, and the clinical decisions of hypertension are a knowledge-intensive task.

Although many knowledge maps have been constructed in the medical field, the breadth and depth of the knowledge stored in the knowledge maps are limited, and specific and complex hypertension diagnosis and treatment decision tasks cannot be solved. Current general medical knowledge patterns, for example: UMLS, CMeKG, although storing a great deal of general knowledge in the medical field, lack specific clinical decision knowledge in the step of hypertension diagnosis and treatment. The existing hypertension knowledge graph only stores knowledge of indication, contraindications, disease relations, medicine relations and the like, can only solve the problem of single medicine recommendation in hypertension medicine treatment, and cannot support all links of complex hypertension diagnosis and treatment decisions.

Disclosure of Invention

The present application aims to solve, at least to some extent, one of the technical problems in the related art.

Therefore, the first purpose of the application is to provide a hypertension diagnosis and treatment decision reasoning method based on a multi-role knowledge graph so as to realize the representation and storage of the hypertension diagnosis and treatment knowledge, the construction of a general reasoning framework and a comprehensive multi-factor hypertension antihypertensive drug recommendation mechanism.

The second aim of the application is to provide a hypertension diagnosis and treatment decision-making reasoning device based on a multivariate role knowledge graph.

In order to achieve the above objective, an embodiment of a first aspect of the present application provides a method for reasoning about diagnosis and treatment decisions of hypertension based on a knowledge graph of multiple roles, including:

constructing a hypertension diagnosis and treatment knowledge graph, wherein a multi-role knowledge paradigm is adopted to represent multi-element medical knowledge in the hypertension diagnosis and treatment knowledge graph, and the multi-element medical knowledge is stored based on a generating rule;

according to the hypertension diagnosis and treatment knowledge graph, inducing and abstracting actions in the hypertension diagnosis and treatment process, developing a plurality of single-hop inference operators based on the actions, and constructing an inference engine based on the plurality of single-hop inference operators;

acquiring patient input data, taking out knowledge corresponding to the patient input data from the hypertension diagnosis and treatment knowledge graph through the reasoning engine, and updating the patient state to obtain the blood pressure level and hypertension risk stratification of the patient;

if the difference value between the blood pressure level of the patient and the dangerous stratification of hypertension and the difference value between the target level and the blood pressure are larger than a preset difference value, generating a treatment measure and a plurality of candidate medication schemes according to the diagnosis and treatment knowledge graph of hypertension, and recording decision reasons related to the plurality of candidate medication schemes;

And scoring each candidate medication scheme according to a medicine recommendation mechanism to obtain a recommendation list and a warning list in the plurality of candidate medication schemes.

Optionally, the constructing a knowledge graph for diagnosing and treating hypertension includes:

selecting an entity-entity relationship type according to the Chinese hypertension guideline and the medical common knowledge, wherein the entity comprises a text entity and a numerical entity, and the entity relationship type is used for expressing the knowledge type used in the one-step medical decision process;

and expressing the related entity and entity relation type as a key value pair form, splitting the multi-element medical knowledge containing or logic connecting words into a plurality of pieces of multi-element knowledge, and organizing the multi-element medical knowledge with or logic relation through an IF-THEN form rule to complete the construction of the hypertension diagnosis and treatment knowledge graph.

Optionally, the plurality of single-hop inference operators includes:

the entity adding operator is used for adding a new medical entity into the text entity of the patient state according to the conclusion of the diagnosis and treatment knowledge, and recording the reasoning conclusion of the current diagnosis and treatment link;

the entity partial order operator is used for reserving the entity with the highest priority in the priority relation and removing the entity with the lowest priority;

An entity replacement operator for replacing the spoken text entity text in the patient state with standardized medical terms;

the entity counting operator is used for counting the appointed text entities of the patient state and adding an entity for storing the counting result into the numerical value entity of the patient state;

a numerical computation operator for computing and generating new numerical entities using existing numerical entities of the patient.

Optionally, the obtaining patient input data, extracting, by the inference engine, knowledge corresponding to the patient input data from the hypertension diagnosis and treatment knowledge graph, and updating the patient state to obtain a blood pressure level and a hypertension risk stratification of the patient, including:

the data preprocessing is carried out on the patient input data, the processing process comprises term standardization, medicine taking father type speculation, numerical calculation preprocessing and diagnosis conclusion determination, the diagnosis conclusion is updated based on the disease level relation and the implication relation, and the final diagnosis conclusion is added to the patient state;

and sequentially carrying out blood pressure grading priority judgment, prognosis factor refinement judgment, prognosis factor major judgment, risk layering judgment and risk layering priority judgment according to the patient state to obtain the blood pressure grade and the hypertension risk layering of the patient.

Optionally, the step of sequentially performing blood pressure grading priority determination, prognosis factor refinement determination, prognosis factor major determination, risk layering priority determination according to the patient state to obtain a blood pressure grade and a hypertension risk layering of the patient includes:

according to the blood pressure grading priority knowledge, reserving the entity with the highest priority in blood pressure level grading entities obtained in the diagnosis conclusion;

selecting a major class and a sub-class of prognostic factors for a plurality of categories for which patient status is accounted for;

according to the major class and the sub class of the prognosis factors of the multiple classes, estimating possible risk stratification of the patient according to hypergraph rules, adding the estimated conclusion into the patient state through the entity adding operator, and updating the patient state;

according to the priority knowledge of the dangerous stratification, the entity with the highest priority in the dangerous stratification type in the state of the patient is reserved, and the entity with the lowest priority is removed, so that the blood pressure level and the dangerous stratification of the hypertension of the patient are obtained.

Optionally, the method further comprises:

and if the blood pressure level of the patient is not more than a preset threshold value, judging that the patient does not need medication, and observing the patient or carrying out non-medication on the patient.

Optionally, if the difference between the blood pressure level and the dangerous hypertension layering of the patient and the difference between the target level and the blood pressure are greater than a preset difference, generating a therapeutic measure and a plurality of candidate medication schemes according to the diagnosis and treatment knowledge graph of hypertension, including:

judging whether a patient takes antihypertensive drugs during diagnosis and treatment;

if the patient does not take the antihypertensive drug, entering a basic medication decision link, taking out all basic medication schemes suitable for the patient under the hypertension complications from the hypertension diagnosis and treatment knowledge graph, and recording the basic medication schemes and decision reasons thereof;

if the patient takes the antihypertensive drug, entering a decision-making link of the medicine adding agent, taking out the relevant medicine adding agent scheme from the hypertension diagnosis and treatment knowledge graph according to the drug taken by the patient and the illness condition, and recording the medicine adding agent scheme and the decision-making reason thereof.

Optionally, after obtaining the basic medication scheme or the dosing scheme, the method further comprises:

decision making is carried out according to the knowledge of the indication and the contraindication so as to update the basic medication scheme or the dosing scheme to obtain a primary medication scheme;

pushing the medicine types in the primary medicine use schemes to subclasses of the medicine types through the medicine hierarchical relationship in the hypertension diagnosis and treatment knowledge graph so as to obtain multiple candidate medicine use schemes, and recording decision reasons related to the multiple candidate medicine use schemes.

Optionally, the scoring each candidate medication scheme according to the medication recommendation mechanism to obtain a recommendation list and a warning list in the multiple candidate medication schemes includes:

setting different scores for basic medication, increased medication, applicable medication, inadequately evidence applicable medication, possible applicable medication, cautious medication and contraindicated medication in the candidate medication;

judging whether the candidate medication scheme is a single-drug direct medication scheme, a single-drug indirect medication scheme or a combined medication scheme, and scoring the plurality of candidate medication schemes according to a corresponding medication scheme scoring mechanism to obtain a recommendation list and a warning list in the plurality of candidate medication schemes;

the single drug direct medication scheme is characterized in that the single drug is scored, all the reasons of medication of the single drug are directly given, the single drug is scored, but part of the reasons of the medication are needed to be obtained from parents, and the medicines in the combined medication scheme are compositely used.

In order to achieve the above objective, an embodiment of a second aspect of the present application provides a device for reasoning about diagnosis and treatment decisions of hypertension based on a knowledge graph of multiple roles, including:

The knowledge graph construction module is used for constructing a hypertension diagnosis and treatment knowledge graph, wherein a multi-role knowledge paradigm is adopted to represent multi-element medical knowledge in the hypertension diagnosis and treatment knowledge graph, and the multi-element medical knowledge is stored based on a generating rule;

the reasoning engine construction module is used for inducing and abstracting actions in the hypertension diagnosis and treatment according to the hypertension diagnosis and treatment knowledge graph, developing a plurality of single-hop reasoning operators based on the actions, and constructing a reasoning engine based on the plurality of single-hop reasoning operators;

the diagnosis and treatment module is used for acquiring patient input data, extracting knowledge corresponding to the patient input data from the hypertension diagnosis and treatment knowledge graph through the reasoning engine, and updating the patient state to obtain the blood pressure level and the hypertension danger layering of the patient;

the medication scheme generation module is used for generating treatment measures and a plurality of candidate medication schemes according to the hypertension diagnosis and treatment knowledge graph and recording decision reasons related to the plurality of candidate medication schemes if the blood pressure level of the patient is layered with the hypertension risk and the difference between the target level and the blood pressure is larger than a preset difference;

and the scoring module is used for scoring each candidate medication scheme according to a drug recommendation mechanism to obtain a recommendation list and a warning list in the plurality of candidate medication schemes.

The technical scheme provided by the embodiment of the application at least brings the following beneficial effects:

the method has the advantages that a hypergraph-model-based hypertension diagnosis and treatment knowledge graph is constructed, high-precision and high-complexity hypertension clinical knowledge is stored, knowledge construction and model reasoning speed and efficiency are improved, actions during hypertension diagnosis and treatment are induced and abstracted according to reasoning results of a hypertension reasoning rule, a universal knowledge graph single-hop reasoning operator is developed, a framework can conduct universal multi-hop reasoning along a given path in the knowledge graph, so that decision tasks of multi-hop reasoning along a knowledge graph appointed path in a universal scene are solved, a comprehensive multi-factor hypertension antihypertensive drug recommendation mechanism is designed when medication decision is conducted, different recommendation scores of recommended drugs can be given according to various factors such as indication, contraindication and medication scheme of the recommended drugs in the knowledge graph, accurate medication of a hypertension patient is achieved, and the framework provided by the method has the characteristic of being very universal and can be easily expanded to the diagnosis and treatment field of other diseases.

Additional aspects and advantages of the application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application.

Drawings

The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

fig. 1 is a flowchart of a method for reasoning about hypertension diagnosis and treatment decision based on a multivariate role knowledge graph according to an embodiment of the present application;

FIG. 2 is a schematic representation and storage of hypertension diagnostic knowledge, shown in accordance with an embodiment of the present application;

FIG. 3 is a schematic diagram of an indication and contraindication decision link shown in accordance with an embodiment of the present application;

FIG. 4 is a schematic diagram illustrating a decision-making of drug chemical names using drug hierarchy according to an embodiment of the present application;

FIG. 5 is a schematic illustration of the reasons for administration of a combination regimen according to an embodiment of the present application;

FIG. 6 is a schematic diagram illustrating a combination regimen reasoning according to an embodiment of the present application;

FIG. 7 is an overall block diagram of a system according to an embodiment of the present application;

fig. 8 is a block diagram of a hypertension diagnosis and treatment decision reasoning device based on a multivariate role knowledge graph according to an embodiment of the present application.

Detailed Description

Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and intended for the purpose of explaining the present application and are not to be construed as limiting the present application.

The following describes a flowchart method and a device of a hypertension diagnosis and treatment decision method based on a knowledge-graph general reasoning framework in the embodiment of the application with reference to the accompanying drawings.

As shown in fig. 1, the hypertension diagnosis and treatment decision-making reasoning method based on the multi-role knowledge graph comprises the following steps:

and step 101, constructing a hypertension diagnosis and treatment knowledge graph, wherein the multi-role knowledge paradigm is adopted to represent multi-element medical knowledge in the hypertension diagnosis and treatment knowledge graph, and the multi-element medical knowledge is stored based on a generation rule.

In the embodiment of the application, the hypertension diagnosis and treatment knowledge map is constructed according to the current Chinese hypertension guide commonly used by Chinese hypertension doctors, the medical common sense knowledge which is not explicitly given in the guide is provided by the hypertension doctors, and the constructed hypertension diagnosis and treatment knowledge map covers the knowledge of the following hypertension clinical decision links: clinical evaluation knowledge, hypertension classification and stratification, target blood pressure determination, antihypertensive treatment strategies, general drug treatment schemes of common patients, drug treatment schemes of common special groups, indications and contraindications of common antihypertensive drugs, drug hierarchy relations and disease hierarchy relations.

It should be noted that, the entity and entity relationship can be considered as the most basic elements constituting knowledge in the knowledge graph, and the knowledge graph constructed by the invention contains 366 medical entities in total, and these entities are divided into two major classes: the text entity and the numerical value entity are subdivided into 15 specific categories, the relation between the category of the entity and the number of the entity is shown in table 1, and the entities except the numerical value interval entity in the table are all text type entities.

TABLE 1

In addition, the entity relationship means that 2 or more entities are connected so as to express a complete medical knowledge, the knowledge graph constructed by the invention contains 705 relationships in total, the types of the relationships are 19 in total, each type of relationship can express the knowledge type used in a one-step medical decision process, and the relationship types and the quantity distribution thereof are shown in a table 2.

TABLE 2

Therefore, the description of the entity-entity relationship in the hypertension diagnosis and treatment knowledge graph is completed.

The following describes the expression mode and the storage mode in the knowledge graph of hypertension diagnosis and treatment.

In the embodiments of the present application, a role-based knowledge paradigm is used to represent multiple medical knowledge. The multi-role knowledge paradigm is a knowledge form for expressing multi-relationships in a knowledge graph, and expresses the multi-relationships as roles in the form of key value pairs, so that the role and the role of each entity in the multi-relationships can be accurately described while expressing the multi-roles. The definition of an n-gram knowledge is as follows:r represents the type of the multiplex relationship, e represents the entity in the multiplex relationship, ++>Representing the role or effect of an entity in a multiplex relationship.

In one possible embodiment, the medical knowledge that "when a patient has clinical complications and SBP is 140mmHg or more, the risk of blood pressure stratification of the patient is very high-risk" can be expressed as the following ternary knowledge: risk stratification (stratification factor: clinical complications, numerical comparison: SBP (mmHg) > = 140, risk stratification of hypertension: very high risk).

It will be appreciated that in order to accurately express multiple knowledge, in the present invention, a piece of multiple knowledge can only describe medical knowledge containing and logically connected words, and the medical knowledge containing or logically connected words is split into multiple pieces of multiple knowledge to be expressed, so that the same decision result can be obtained from more than one piece of multiple knowledge, and multiple knowledge exists or logic relations exist among the multiple pieces of multiple knowledge when the decision result is expressed, and in order to facilitate management of multiple pieces of multiple knowledge corresponding to the same result, the present invention organizes the existing or logic relation multiple knowledge by using a production rule.

The production rules employed in this application are a set of rules in the form of IF-THEN, which are generally in the form of: condition→result, which describes what new conclusions can be drawn when the condition definition is satisfied, the condition and result parts in the production formula rule can be described more precisely by introducing logical connectives (and, or, not). In the rule expression designed by the invention, the condition parts organize entity parts in the rule in a Conjunctive Normal Form (CNF), and the result part defaults to only existing and relations among the entities.

For example, for medical knowledge "when a patient has clinical complications and SBP is 140mmHg or more, the risk of blood pressure of the patient is layered at high risk", an example of hypertension diagnosis knowledge storage and representation is shown in fig. 2.

102, according to the hypertension diagnosis and treatment knowledge graph, inducing and abstracting out actions in the hypertension diagnosis and treatment, developing a plurality of single-hop inference operators based on the actions, and constructing an inference engine based on the plurality of single-hop inference operators.

In the embodiment of the application, the process of making decisions along the diagnosis and treatment path can be regarded as multi-hop reasoning on the knowledge graph, one link of the reasoning path can be regarded as single-hop reasoning by utilizing the knowledge of a specific relationship type in the knowledge graph, and different decision links in the decision path often need to use the knowledge of different relationship types in the knowledge graph, but the process of updating the patient state by utilizing the knowledge is universal, so that the process of updating the patient state by an engine is generalized and abstracted, 5 basic actions are generalized, and a plurality of corresponding single-hop reasoning operators are developed. The following describes class 5 single hop inference operators used in the general medical decision phase:

(1) And the entity adding operator is used for adding a new medical entity into the text entity of the patient state according to the conclusion of the diagnosis and treatment knowledge, so that the reasoning conclusion of the current diagnosis and treatment link is recorded, and therefore, in the follow-up reasoning link, the previously added medical entity is taken as a reasoning basis, and the new reasoning is completed according to the diagnosis and treatment knowledge.

In one possible embodiment, the operator corresponding to the action of diagnosing that the patient is in period 5 of chronic kidney disease adds an operator to the entity when the diagnosis knowledge that the GFR of the patient is less than 15ml/min 1.73/m/m, and the patient can be considered to have period 5 of chronic kidney disease when the GFR of the patient meets the numerical condition.

(2) And the entity partial order operator is used for reserving the entity with the highest priority in the priority relation and removing the entity with the lowest priority.

For diagnosis and treatment knowledge with priority relation among the entities of the reasoning result, in order to facilitate the editing and storage of the knowledge, the condition of the high-priority reasoning knowledge often contains the condition of low priority, so that for the medical entity with priority relation, after the diagnosis and treatment entity is added by utilizing the entity adding operator, only the medical entity with highest priority needs to be reserved by using the priority operator.

In a possible embodiment, taking a hypertension risk stratification decision as an example, the decision process consists of 2 decision links. In the 1 st link, the inference engine takes out the judgment knowledge of risk stratification, makes preliminary judgment on the possible risk level of the patient, and adds the judgment result entity into the patient state. In the 2 nd link, the reasoning engine reserves the highest risk layering entity in the state of the patient according to the priority relation of the hypertension risk layering entities in the knowledge graph.

(3) An entity replacement operator for replacing the spoken text entity text in the patient state with standardized medical terms.

Entity replacement operators are mainly used in the patient preprocessing stage, which normalizes text entities in patient states to standard terms in the system, thereby enabling the system to use unified entity medical terms for knowledge storage and reasoning.

One possible implementation example, for example, is that the standard term for a disease stroke is cerebral stroke, the standard term for drug type MRA is aldosterone receptor antagonist, and the entity replacement operator can replace stroke in the patient state, MRA, with the standard entity name cerebral stroke and aldosterone receptor antagonist in the system.

(4) The entity counting operator counts the appointed text entity of the patient state, and then adds an entity for storing the counting result to the numerical entity of the patient state.

The operator is used to process the counting operation in the medical knowledge graph.

In the related art, the number of cardiovascular risk factors of a patient needs to be counted before assessing the risk stratification of hypertension. In the decision process, the calculation engine counts a plurality of risk factors in the patient state according to the classification knowledge of the cardiovascular risk factors in the knowledge graph, adds the counting result into a numerical entity of the patient state, and adds 3 cardiovascular risk factors into the patient state to obtain a decision result.

(5) A numerical computation operator, which utilizes the existing numerical entity of the patient to compute and generate a new numerical entity.

The operator is mainly used for formula calculation in the knowledge graph.

In a related embodiment, the BMI of the patient is calculated through the height and the weight during the pretreatment of the patient data, and before the medication decision, the difference value between the blood pressure of the patient and the target blood pressure is calculated according to the blood pressure of the patient and the target blood pressure, so that a basis is provided for the subsequent decision on using a single-drug or multi-drug treatment scheme, and the participation of a numerical calculation operator is needed.

In the embodiment of the application, the decoupling of decision knowledge and decision behaviors is realized by using operators in the reasoning engine, the universality reasoning can be carried out based on the knowledge graph, in a single link of diagnosis and treatment decision, the engine can take out knowledge from the knowledge graph only by appointing the knowledge type required by the link and the corresponding decision action, and the corresponding operators are called to modify the state of a patient according to the decision knowledge, so that the single-hop reasoning in the knowledge graph is completed. In the process, an independent reasoning engine is not required to be developed for each decision link, and the medical decision of the link can be completed only by enabling the engine to call an operator corresponding to the decision action according to decision knowledge.

In addition, the development of operators also makes the expansion of the diagnosis and treatment process easier, when a new diagnosis and treatment link is added, only diagnosis and treatment knowledge is needed to be added to the knowledge graph, and the action type of a diagnosis and treatment result is specified, if the link relates to the new diagnosis and treatment action type, only a new operator is needed to be developed for the action in an inference engine, and when a clinical decision is made, a new decision link is added to a decision path of a decision framework, so that a new diagnosis and treatment conclusion can be generated for a patient by using the newly added diagnosis and treatment knowledge in the new decision link.

103, acquiring patient input data, taking out knowledge corresponding to the patient input data from the hypertension diagnosis and treatment knowledge graph through an inference engine, and updating the patient state to obtain the blood pressure level and the hypertension danger layering of the patient.

In the related art, a decision link can be regarded as a process of reasoning by using single-hop knowledge of a knowledge graph. Through a decision link, a decision conclusion of the link can be obtained according to medical knowledge and patient state. After a plurality of decision links are connected, a decision path is formed. The process of making decisions along a path can be seen as a process of making multi-hop reasoning on a knowledge graph.

In order to solve a medical decision task, it is first converted into a medical decision path, which is made up of a series of decision links. In the decision path, different decision links can be organized by using flow control operations such as judgment, circulation and the like so as to realize more complex decision behaviors. And then, storing the knowledge of each link into a knowledge graph, and binding corresponding decision behaviors for the knowledge. And finally, after the patient state is completely framed, calling an inference engine corresponding to each inference link along the decision path, so that the inference engine obtains an inference conclusion of the current link according to the knowledge graph, and after the conclusion of the decision path is reached, returning the final decision conclusion of the medical decision task to the outside.

The medical decision task in the embodiment of the application is a clinical medication decision for hypertension. The task can be divided into 4 phases: data preprocessing, blood pressure stratification and stratification, determination of therapeutic measures and medication regimens, determination of specific medication regimens, wherein stage 4 is not an essential stage, and if it is determined in stage 3 that the patient does not require medication, the decision process is completed after the therapeutic measures are determined in stage three. Namely, in the embodiment of the present application, the specific processes of the first stage and the second stage are:

The method comprises the steps of preprocessing data of patient input data, wherein the processing process comprises term standardization, medicine taking parent class speculation, numerical calculation preprocessing and diagnosis conclusion determination, updating the diagnosis conclusion based on disease level relation and implication relation, and adding the final diagnosis conclusion to a patient state;

and sequentially carrying out blood pressure grading priority judgment, prognosis factor refinement judgment, prognosis factor major judgment, risk layering judgment and risk layering priority judgment according to the state of the patient to obtain the blood pressure grade and the hypertension risk layering of the patient.

In order to more specifically describe how the hypertension diagnosis and treatment decision method provided by the application realizes the classification and layering decision tasks of hypertension, a hypertension patient is taken as an example for illustration.

The patient's condition is described as: the patient had unstable angina disease, and the result of the physical examination was urine microalbumin (mg/24 h): 78, SBP (mmHg): 167, DBP (mmHg): 97, height (m): 1.65, weight (kg): 78.

Specifically, the specific steps for realizing the classification and layered decision task of hypertension are as follows:

(1) The term standardized link, after receiving patient input, converts unstable angina in patient description into standard term unstable angina in map through entity replacement operation, and the link is ignored in the decision task because the dangerous layering task does not involve patient medication.

(2) And a patient pretreatment link, wherein the BMI of the patient is calculated according to the height, the weight and a stored formula of the patient. In this step, there are some processes related to the calculation of the dosage amount of the patient and labeling of some missing indexes, and the use process of these knowledge is not shown in the figure because it is irrelevant to the decision task.

(3) The diagnosis knowledge is used, the link performs preliminary diagnosis on the patient according to the diagnosis knowledge in the hypertension guide and by combining the existing examination and inspection values of the patient, and the diagnosis conclusion of the link is added into the patient state through the entity adding operator.

(4) For simplicity of the schematic diagram, the height and weight of the numerical entity which are not used in the later links in the patient state are not displayed any more, and in the decision path, the links are in a circulation structure, and the knowledge of the disease level relationship and the implication relationship is continuously used for reasoning until the circulation exit condition is met, so that no new diseases can be deduced.

(5) And (3) using blood pressure grading priority knowledge, and reserving the entity with the highest priority among blood pressure level grading entities obtained in the diagnosis conclusion.

(6) And (3) carrying out refinement judgment on the prognostic factors and major class judgment on the prognostic factors, selecting major classes and subclasses of the prognostic factors of a plurality of classes which are established for the patient state, adding the presumption conclusion into the patient state through an entity adding operator, and updating the patient state.

It should be noted that, the major type of prognosis factor determination involves 2 types of decision actions: entity addition and entity count, wherein cardiovascular risk factor knowledge forms decision conclusions through entity count, and target organ damage and clinical complications decision is obtained through entity addition.

(7) The risk of hypertension is inferred in layers, and for simplicity of the schematic diagram, the unused parts of entities in the patient state are not shown.

(8) And in the cardiovascular risk stratification link, according to hypergraph rules meeting the conditions, preliminarily presuming which risk stratification the patient possibly has, adding the conclusions into the state of the patient through entity adding actions, wherein only 4 pieces of risk stratification knowledge meeting the condition of the patient are shown in the schematic diagram, and part of knowledge which is actually used is not shown.

(9) And according to the priority knowledge of the dangerous layering, reserving the entity with the highest priority in the dangerous layering types in the patient state, and removing the entity with the lower priority.

(10) The classification and layering decision task reasoning of the hypertension outputs the decision conclusion: the blood pressure level of the patient is 2-level blood pressure, and the risk of hypertension of the patient is layered at high risk.

Through the steps, the reasoning conclusions can provide reasoning basis for deciding the treatment measures and the medication strategies of the patient in the following reasoning links.

104, if the difference between the blood pressure level of the patient and the dangerous stratification of hypertension and the difference between the target level and the blood pressure are larger than a preset difference, generating treatment measures and a plurality of candidate medication schemes according to the diagnosis and treatment knowledge graph of hypertension, and recording decision reasons related to the plurality of candidate medication schemes;

in the embodiment of the application, the reason that the recommendation or the tabu of the medicine is required to be recorded during the reasoning, and a recommendation score is calculated according to each medicine scheme according to the reasons, and because the reasoning mode of the step 104 is different from the general reasoning mode of the three links, a special reasoning engine is designed for the application: a drug recommendation engine.

It will be appreciated that if the patient's blood pressure level is not greater than a predetermined threshold value, as compared to the risk stratification of hypertension and the difference between the target level and blood pressure, it is determined that the patient does not require medication, either by observation of the patient or by non-medication of the patient.

In the embodiment of the application, in the third stage, after the inference engine decides that the patient needs the medication, it needs to determine whether the patient takes the antihypertensive medication during diagnosis and treatment, and further determines whether the patient adopts a single medication or a combination medication treatment scheme after the patient does not take the antihypertensive medication.

The following classification is given.

If the patient does not take the antihypertensive drug, entering a basic medication decision link, taking out all basic medication schemes suitable for the patient under the hypertension complications from the hypertension diagnosis and treatment knowledge graph, and recording the basic medication schemes and decision reasons thereof.

An embodiment is presented below for specifically explaining the basic medication decision link, the patient state is that the patient suffers from hypertension, stable angina pectoris, coronary heart disease, has left ventricular hypertrophy symptoms, but the patient lacks heart rate, creatinine examination data, the system decides in the previous decision link that the patient needs to use the basic medication scheme of single medicine, for the sake of the visibility of the picture, the lower half of the figure only shows 3 related medication schemes and medication knowledge thereof, for the sake of the simplicity of the picture, the knowledge is shown in a regular form, wherein the hypergraph form of the upper right corner rule is also drawn, the rule in the figure points to the application medication scheme by the arrow, and when one medication scheme is pointed by a plurality of arrows, the condition that the scheme is more suitable for the patient is explained. In the figure, ACEI, dihydropyridine CCB and beta receptor blocker are basic antihypertensive schemes of hypertension, but the latter two medicaments are basic antihypertensive schemes of hypertension complicated with stable angina.

It will be appreciated that when the patient is taking a hypotensive agent, but the blood pressure is still not ideally controlled, the system will give a regimen for further controlling the blood pressure based on the agent currently being taken by the patient and the condition.

An embodiment is presented below for specifically explaining the decision-making procedure of the booster medicine, wherein the patient is suffering from hypertension complicated with stable angina, and has been administered dihydropyridine CCBs, but still cannot control blood pressure, and for strong visibility, knowledge is displayed in a regular form, and hypergraph knowledge actually used in the rules of the upper right corner is also displayed. For simplicity of the picture, the legend only shows three dosage regimens that can be formed by adding the drug. In the figure, the left part of the decision knowledge shows the antihypertensive drug which should be increased when the dihydropyridine CCB is taken under the general condition of the patients with hypertension and the angina pectoris patients with hypertension complicated with stability, and the right part of the decision knowledge shows the antihypertensive drug which should be increased when the dihydropyridine CCB is taken under the condition of the patients with hypertension and the blood pressure cannot be controlled. The decision rule points to the final regimen formed after increasing the hypotensive agent by an arrow.

It can be appreciated that the drug-enhancing drug regimen and the basic drug regimen can provide more general drug regimens, and that different drug regimens need to be distinguished by finer indications and contraindications knowledge, so as to achieve personalized and refined drug administration for the patient.

An embodiment is presented below for specifically describing the discrimination links, as shown in fig. 3, wherein the patient in fig. 3 continues the medication decision process for the patient as described above, and the knowledge of the basic medication regimen of the last link in fig. 3 is omitted, leaving only the arrows pointing to the medication regimen. For ACEI, ACEI is a suitable medicine for hypertension with left ventricular hypertrophy and coronary heart disease, but patients lacking blood creatinine should be cautiously administered ACEI medicine from the contraindication. The dihydropyridine CCB is a suitable medicine for hypertension with left ventricular hypertrophy, and the medicine has no contraindication for the condition of patients. For beta blockers, which are potentially useful drugs for hypertension with left ventricular hypertrophy, the extent of applicability is weaker than the first two drugs, but patients lacking heart rate data should be careful to apply the drugs from a contra-point of view.

In the embodiment of the application, decision is made through knowledge of indication and contraindication so as to update a basic medication scheme or an augmented medication scheme to obtain a primary medication scheme, then the types of medicines in the primary medication scheme are pushed to subclasses of the types of medicines through a medicine hierarchical relationship in a hypertension diagnosis and treatment knowledge graph so as to obtain multiple candidate medication schemes, and decision reasons related to the multiple candidate medication schemes are recorded.

An embodiment is presented below to specifically illustrate a process for drug hierarchical relationship usage, as shown in fig. 4, which uses knowledge of drug hierarchical relationships in a knowledge graph to push inferential conclusions from drug types to sub-classes of drug types, i.e., drug chemical names, because a physician needs to determine a specific drug chemical name when deciding to take a drug, and sometimes there is some difference between different drug chemical names in the same type. Fig. 4 continues the decision link of fig. 3, showing the reasoning process of the drug chemical name decision, wherein for simplicity of the picture, only 2 sub-classes of drug chemical names are shown per drug type.

And 105, scoring each candidate medication scheme according to a medicine recommendation mechanism to obtain a recommendation list and a warning list in a plurality of candidate medication schemes.

After completion of the use of the drug hierarchy, the medication regimen the patient is involved in, and the decision reasons for the regimen, have been recorded. A scoring mechanism is needed at present, different medication schemes are distinguished according to decision reasons related to the medication schemes, and the medication scheme most suitable for a patient is given the highest score, so that recommendation is preferentially carried out when recommendation is carried out, tabulated medicines are given negative scores, and a warning list is entered when recommendation is carried out, so that a doctor is reminded of avoiding medicines in the warning list when medication is carried out.

Firstly, a mapping relation is established between the reasons and the scores of a medication scheme, and different scores are corresponding to different medication reasons, so that the score of a single medication knowledge pointing to the medication scheme can be calculated.

In one possible embodiment, the basic medication is set to 8 minutes, the incremental medication is set to 8 minutes, the applicable medication is set to 3 minutes, the evidence of insufficient applicable medication is set to 2 minutes, the possible applicable medication is set to 1 minute, the cautious medication is set to-4 minutes, and the contraindicated medication is set to-40 minutes.

It will be appreciated that where appropriate, the base medication regimen and the reasons for adding the medication regimen are given the highest score of 8 points, allowing them to predominate at the time of medication recommendation. The recommended score for the applicable drug, the evidence-deficient application, and the potentially applicable drug decreases in turn. The scores corresponding to the three kinds of knowledge are not high in the scores corresponding to the basic and prescription adding schemes, so that the medicine recommendation scores can only be modified when the medicine recommendation scores are calculated, and in the same basic scheme, the personalized medicine application scheme which is most suitable for a patient is distinguished, and personalized medicine application is realized.

In unsuitable cases, knowledge of contraindicated drugs is given the negative score with the highest absolute value, allowing it to play a dominant role in medication decisions. If one medication scheme contains a tabu, the recommended score of the drug is buckled by a large score, so that the recommended score is difficult to be arranged in front during recommendation and is more likely to appear in a warning list. The cautious drug is given a negative score slightly higher than the applicable drug, which also acts as a modified component in the decision to take the drug, but only slightly more strongly in the modified effect than the indication.

Furthermore, the case of candidate regimen score calculation can be categorized into three cases: the single-drug direct administration scheme, the single-drug indirect administration scheme or the combined administration scheme, wherein the single-drug direct administration scheme is characterized in that the single drug is scored, all the drug reasons of the drugs are directly given, the single drug score in the single-drug indirect administration scheme is obtained from a father class, and part of the drug reasons are obtained from the father class, and the drugs in the combined administration scheme are combined.

For a single-drug direct medication scheme, all decision reasons of the medication scheme are taken out, the scores of all the reasons are obtained through mapping, and then the scores are summed to be the final score of the medication scheme.

For an indirect dosing regimen for a single drug, the score for the drug may be calculated after the parent score for the drug is calculated. The specific calculation mode is as follows: by this calculation formula, the parent drug type score can be transferred to the drug chemical name of the sub-class along the drug hierarchy by this mechanism. In addition, if an indication or a base solution is applicable to a chemical name, the score of the chemical name of the medicine will be high in all subclasses of the parent class, so that the chemical name of the medicine will be distinguished.

In one possible embodiment, the single-drug direct regimen includes one applicable drug, one possible applicable drug, one basic drug regimen, and one cautious drug, where the final score of the single-drug direct regimen is 3+1+8-4=8 points.

In one possible embodiment, the indirect single drug regimen includes one applicable drug, one possible applicable drug, one basic drug regimen, and one cautious drug, and the score of the recommended score of the parent drug is 3 points, and the final score of the indirect single drug regimen is 11 points.

For a combination regimen, the scoring process of the combination regimen needs to take into account more factors, and the reasons for the administration of a general combination regimen are shown in fig. 5. The scoring of the combination scheme needs to consider the medication reasons of the scheme itself and the reasons of the parent combination scheme of the scheme, wherein the parent scheme may have a plurality of scores and the score of each drug member in the combination scheme.

It can be appreciated that the final score of the combination regimen is calculated as:

the calculation formula of the member score is as follows:

in the calculation formula of the final score, the score of the combined medication scheme is composed of three parts, the score of the scheme itself and the score of the parent scheme take the dominant role, the member scores can play a modifying role, and under the same adaptation condition, different combined medication schemes are distinguished according to the adaptation degree of the members in the scheme. If the multiple parents have drug reasons, after all the parent scheme scores are calculated, taking the average value of the parent scheme scores as the parent scheme score. In the calculation formula of the membership score, the membership score is composed of two parts according to the scores of the drug members. For drug members with positive scores, the average of all drug scores is taken as the positive score, and for drug members with negative scores, the scores of all drug members are summed as the negative score. The basis for adopting different aggregation strategies for members with different scores is as follows: the effect of drug members with negative scores (unsuitable medication) on the final score is highlighted. If a combination regimen includes negative scoring members, the regimen is given a lower score.

An embodiment is presented below to introduce a process for calculating a patient combination score. The patient suffers from hypertension complicated with stable angina and hyperuricemia, is taking dihydropyridines CCBs but has unsatisfactory blood pressure control. Through the previous decision links, three combination regimens and reasons for the patient are shown in fig. 6. From a parent perspective, these co-administration schemes are not pushed by the system, so the parent score is 0. From a subclass perspective, one of these regimens has a medication reason for one member drug. The ACEI+dihydropyridine CCB scheme and the thiazine diuretic+dihydropyridine CCB scheme have two medication reasons obtained by knowledge of the dosing schemes. However, the ACEI of the former member is responsible for the indication, while the thiazide diuretic of the latter member is responsible for the caution, so that the score of the former scheme is higher after calculation by this mechanism. The "beta blocker+dihydropyridine CCB" regimen is based on the fact that there is only one dosing regimen knowledge to obtain and the member beta blocker is carefully used, and the regimen score is the lowest after score calculation.

It may be understood that the recommendation list may be a medication scheme with a final score exceeding a preset recommendation threshold, or may be a medication scheme with top N ranks after the final score is ranked, which is not limited herein, and the alert list may be similar to a medication scheme with a final score lower than a preset alert threshold, or may be a medication scheme with bottom M ranks after the final score is ranked.

The overall system structure diagram of the system is shown in fig. 7, when the decision system interacts with the outside through the agent, and diagnosis and treatment decision is started, the agent receives the patient condition description and a decision path on the knowledge graph, the patient description is used for generating an initial patient state, the initial patient state is composed of a text state and a numerical value state, the patient state is used as an reasoning basis of a reasoning engine, and the reasoning engine is updated after reasoning is completed, so that the current diagnosis and treatment reasoning conclusion is recorded. The decision path is composed of a series of links for clinical decisions of hypertension. Each decision link specifies a type of inferred knowledge in the knowledge graph. When the hypertension medication decision is made, the agent firstly uses the reasoning engine to complete the universal clinical reasoning task on the knowledge graph, and then uses the medication recommendation engine to complete the medication recommendation task by using the disease-medication and medication knowledge of the knowledge graph. In a single decision link on the decision path, the engine completes the decision of the link and updates the patient state as the decision input of the next link according to the patient state and the action type of the decision knowledge. After finishing the diagnosis and treatment decision, the agent returns the diagnosis and treatment conclusion to the outside according to the finally updated patient state, and therefore, diagnosis and treatment is finished.

According to the embodiment of the application, the hypergraph-model-based hypertension diagnosis and treatment knowledge graph is constructed, the high-precision and high-complexity hypertension clinical knowledge is stored, the speed and efficiency of knowledge construction and model reasoning are improved, actions of a hypertension patient in diagnosis and treatment are induced and abstracted according to the reasoning result of the hypertension reasoning rule, a universal knowledge graph single-hop reasoning operator is developed, the framework can conduct universal multi-hop reasoning along the reasoning of a given path in the knowledge graph, the decision task of conducting multi-hop reasoning along the knowledge graph appointed path in a universal scene is solved, a comprehensive multi-factor hypertension antihypertensive drug recommendation mechanism is designed when a drug decision is conducted, different recommendation scores of recommended drugs can be given according to various factors such as indication, contraindication and drug use scheme in the knowledge graph, and accurate drug administration of the hypertension patient is achieved.

Fig. 8 is a block diagram of a hypertension diagnosis and treatment decision reasoning apparatus 100 based on a multivariate role knowledge graph according to an embodiment of the present application, including:

The knowledge graph construction module 110 is configured to construct a knowledge graph of hypertension diagnosis and treatment, wherein the knowledge graph of hypertension diagnosis and treatment is represented by a multi-role knowledge paradigm, and the multi-role knowledge is stored based on a generation rule;

the inference engine construction module 120 is configured to generalize and abstract actions in the case of hypertension diagnosis and treatment according to the hypertension diagnosis and treatment knowledge graph, develop a plurality of single-hop inference operators based on the actions, and construct an inference engine based on the plurality of single-hop inference operators;

the diagnosis and treatment module 130 is configured to obtain patient input data, extract knowledge corresponding to the patient input data from the hypertension diagnosis and treatment knowledge graph through the inference engine, and update the patient state to obtain a blood pressure level and a hypertension risk stratification of the patient;

the medication scheme generating module 140 generates a treatment measure and a plurality of candidate medication schemes according to the hypertension diagnosis and treatment knowledge graph and records decision reasons related to the plurality of candidate medication schemes if the blood pressure level of the patient is greater than a preset difference value with the hypertension risk stratification and the difference value between the target level and the blood pressure;

the scoring module 150 is configured to score each candidate medication scheme according to the medication recommendation mechanism, so as to obtain a recommendation list and a warning list in the multiple candidate medication schemes.

The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps recited in the present disclosure may be performed in parallel, sequentially, or in a different order, provided that the desired results of the technical solutions of the present disclosure are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present disclosure are intended to be included within the scope of the present disclosure.

Claims

1. A hypertension diagnosis and treatment decision reasoning method based on a multivariate role knowledge graph is characterized by comprising the following steps:

2. The method according to claim 1, wherein the constructing a knowledge graph of hypertension diagnosis and treatment comprises:

3. The method of claim 2, wherein the plurality of single-hop inference operators comprises:

4. The method of claim 3, wherein the obtaining patient input data, by the inference engine, retrieving knowledge corresponding to the patient input data from the knowledge graph of hypertension diagnosis and treatment, and updating the patient status, to obtain a blood pressure level and a hypertension risk stratification of the patient, comprises:

5. The method according to claim 4, wherein the sequentially performing blood pressure grading priority determination, prognosis factor refinement determination, prognosis factor major category determination, risk stratification priority determination according to the patient status, obtaining a blood pressure grade and a hypertension risk stratification of the patient comprises:

6. The method as recited in claim 1, further comprising:

7. The method according to claim 1, wherein if the difference between the blood pressure level and the risk stratification of hypertension and the difference between the target level and the blood pressure of the patient are greater than a preset difference, generating a therapeutic measure and a plurality of candidate medication schemes according to the diagnosis and treatment knowledge graph of hypertension, comprising:

8. The method of claim 7, further comprising, after obtaining the base regimen or the enhancement regimen:

9. The method of claim 1, wherein scoring each candidate medication regimen according to a medication recommendation mechanism to obtain a recommendation list and a warning list in the plurality of candidate medication regimens comprises:

10. The utility model provides a hypertension diagnosis and treatment decision-making reasoning device based on multiple role knowledge graph which characterized in that includes: