CN116682551A

CN116682551A - Disease prediction method, disease prediction model training method and device

Info

Publication number: CN116682551A
Application number: CN202310930789.XA
Authority: CN
Inventors: 孙继超; 吴贤; 郑冶枫
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2023-07-27
Filing date: 2023-07-27
Publication date: 2023-09-01
Anticipated expiration: 2043-07-27
Also published as: CN116682551B

Abstract

The application provides a disease prediction method, a disease prediction model training method and a disease prediction model training device, and relates to the field of artificial intelligence. The disease prediction method comprises the following steps: obtaining diagnosis knowledge information of a subject; the diagnostic knowledge information is used to determine a disease type; matching the diagnosis knowledge information with at least one logic rule to obtain a matching result of the diagnosis knowledge information and the at least one logic rule; inputting the diagnosis knowledge information into a neural network model for feature extraction to obtain a first disease feature of the diagnosis knowledge information; inputting the matching result and the first disease characteristics into a knowledge enhancement network, and carrying out data enhancement on the characteristics of the dimension corresponding to at least one disease category of the first disease characteristics to obtain second disease characteristics; based on the second disease characteristic, a disease category of the subject is predicted. The method can realize the organic fusion of the disease reasoning scheme based on expert rules and the disease prediction scheme based on data driving, and is beneficial to improving the accuracy and the interpretability of disease prediction.

Description

Disease prediction method, disease prediction model training method and device

Technical Field

The embodiment of the application relates to the technical field of artificial intelligence, in particular to a disease prediction method, a disease prediction model training method and a device.

Background

Along with the development of artificial intelligence, disease prediction functions are widely applied to the field of disease auxiliary diagnosis. The disease prediction function can be integrated in an intelligent medical disease auxiliary decision-making system, and the possible disease condition of the user can be intelligently estimated according to the disease condition information of the user. The disease prediction schemes mainly comprise two types, namely a disease reasoning scheme based on expert rules, and relevant diagnosis and treatment rules are arranged mainly according to authoritative medical knowledge or medical experience, so that an experience decision system is constructed for disease diagnosis; the other is a disease prediction scheme based on data driving, namely, a machine learning or deep learning model is built based on large-scale electronic physiological data to perform disease prediction.

The disease reasoning scheme based on expert rules has the advantages of being relatively simple and direct, and the rules have certain evidence-based and explanatory properties, and the disadvantages are that the arrangement of expert rules requires a great deal of manual work, and the rules cannot be automatically updated along with the update of medical knowledge. In addition, the background and experience of different medical institutions and specialists are quite different, so that rule decision making has certain limitation. The data-driven disease prediction scheme can achieve higher accuracy based on the existing pre-trained model and a large amount of data, but at the same time sacrifices a certain interpretability. In addition, in the case of diseases with low morbidity, the traditional neural network model may have poor prediction effect due to insufficient training data. Currently, there is no disease prediction scheme that can fuse these two schemes.

Disclosure of Invention

The application provides a disease prediction method, a disease prediction model training method and a disease prediction model training device, which can realize the organic fusion of a disease reasoning scheme based on expert rules and a disease prediction scheme based on data driving, and are beneficial to improving the accuracy and the interpretability of disease prediction.

In a first aspect, an embodiment of the present application provides a disease prediction method, including:

obtaining diagnosis knowledge information of a subject; the diagnostic knowledge information is used for determining the disease type;

matching the diagnosis knowledge information with at least one logic rule to obtain a matching result of the diagnosis knowledge information and the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category;

inputting the diagnosis knowledge information into a neural network model for feature extraction to obtain a first disease feature of the diagnosis knowledge information; wherein the dimension of the first disease feature is equal to the number of the at least one disease category;

inputting the matching result and the first disease feature into a knowledge enhancement network, and enhancing data of the feature of the dimension corresponding to the at least one disease category of the first disease feature to obtain a second disease feature;

Predicting a disease category of the subject based on the second disease characteristic.

In a second aspect, an embodiment of the present application provides a disease prediction model training method, including:

acquiring a physiological data sample, wherein the physiological data sample comprises diagnosis knowledge information and disease category labels of a subject sample;

inputting the diagnosis knowledge information into a neural network model for feature extraction to obtain a first disease feature of the diagnosis knowledge information; the dimension of the first disease feature is equal to the number of the at least one disease category;

inputting the matching result and the first disease feature into a knowledge enhancement network to enhance data of the feature of the dimension corresponding to the at least one disease category of the first disease feature, so as to obtain a second disease feature;

obtaining a predicted disease category for the subject based on the second disease characteristic;

and according to the predicted disease category and the disease category label, carrying out parameter adjustment on the neural network model and the knowledge enhancement network to obtain the trained disease prediction model.

In a third aspect, an embodiment of the present application provides a disease prediction apparatus, including:

an acquisition unit configured to acquire diagnostic knowledge information of an object; the diagnostic knowledge information is used for determining the disease type;

the matching unit is used for matching the diagnosis knowledge information with at least one logic rule to obtain a matching result of the diagnosis knowledge information and the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category;

the neural network model is used for extracting features of the diagnosis knowledge information to obtain first disease features of the diagnosis knowledge information; wherein the dimension of the first disease feature is equal to the number of the at least one disease category;

the knowledge enhancement network is used for inputting the matching result and the first disease characteristics, and carrying out data enhancement on the characteristics of the dimension corresponding to the at least one disease category of the first disease characteristics to obtain second disease characteristics;

and a prediction unit for predicting a disease category of the subject based on the second disease characteristic.

In a fourth aspect, an embodiment of the present application provides a disease prediction model training apparatus, including:

An acquisition unit for acquiring a physiological data sample including diagnostic knowledge information of a subject sample and a disease category label;

the neural network model is used for extracting the characteristics of the diagnosis knowledge information to obtain first disease characteristics of the diagnosis knowledge information; the dimension of the first disease feature is equal to the number of the at least one disease category;

the knowledge enhancement network is used for inputting the matching result and the first disease characteristics so as to carry out data enhancement on the characteristics of the dimension corresponding to the at least one disease category of the first disease characteristics and obtain second disease characteristics;

a prediction unit for obtaining a predicted disease category of the subject based on the second disease feature;

and the parameter adjustment unit is used for carrying out parameter adjustment on the neural network model and the knowledge enhancement network according to the predicted disease category and the disease category label to obtain the trained disease prediction model.

In a fifth aspect, an embodiment of the present application provides an electronic device, including: a processor and a memory for storing a computer program, the processor being for invoking and running the computer program stored in the memory for performing the method as in the first or second aspect.

In a sixth aspect, embodiments of the application provide a computer readable storage medium comprising instructions which, when run on a computer, cause the computer to perform a method as in the first or second aspect.

In a seventh aspect, embodiments of the present application provide a computer program product comprising computer program instructions for causing a computer to perform the method as in the first or second aspect.

In an eighth aspect, embodiments of the present application provide a computer program that causes a computer to perform the method as in the first or second aspect.

According to the technical scheme, the diagnosis knowledge information of the object is matched with at least one logic rule, and the matching result is used as an expert rule to carry out data enhancement on the first disease feature of the diagnosis knowledge information extracted by the neural network, so that the enhanced second disease feature is fused with the expert rule and the disease feature extracted by the neural network, and the disease prediction based on the expert rule can be organically fused with the disease reasoning scheme based on the expert rule and the disease prediction scheme based on the data driving according to the second disease feature, so that the knowledge experience rule formed in the industry for a long time can be utilized, and the disease prediction can be carried out by utilizing the calculation model based on the data driving, thereby being beneficial to improving the accuracy and the interpretability of the disease prediction.

Furthermore, for the situation that the traditional neural network model has poor prediction effect on some disease types with lower morbidity due to insufficient training samples, the embodiment of the application can greatly improve the prediction accuracy of the disease types by combining the prior expert rules to strengthen the knowledge of the output of the neural network model.

In addition, the embodiment of the application carries out data enhancement on the first disease characteristic of the diagnosis knowledge information extracted by the neural network through expert rules, and can realize knowledge enhancement on the output of the original neural network model without changing the structure and the parameter quantity of the original neural network model.

Drawings

Fig. 1 is a schematic diagram of an application scenario of an embodiment of the present application;

FIG. 2 is a schematic illustration of an interface of a disease-assisted diagnostic product;

FIG. 3 is a schematic flow chart of a disease prediction method according to an embodiment of the present application;

FIG. 4 is a schematic flow chart of another disease prediction method according to an embodiment of the present application;

FIG. 5 is a schematic diagram of a process for data enhancement according to an embodiment of the present application;

FIG. 6 is a schematic flow chart diagram of another disease prediction method according to an embodiment of the present application;

FIG. 7 is a schematic diagram of the structure of an enhanced knowledge network, in accordance with an embodiment of the application;

FIG. 8 is a schematic flow chart diagram of another disease prediction method according to an embodiment of the present application;

FIG. 9 is a schematic flow chart diagram of a disease prediction model training method according to an embodiment of the present application;

fig. 10 is a schematic block diagram of a disease prediction apparatus according to an embodiment of the present application;

FIG. 11 is a schematic block diagram of a disease prediction model training apparatus according to embodiments hereof;

fig. 12 is a schematic block diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application.

It should be understood that in embodiments of the present application, "B corresponding to a" means that B is associated with a. In one implementation, B may be determined from a. It should also be understood that determining B from a does not mean determining B from a alone, but may also determine B from a and/or other information.

In the description of the present application, unless otherwise indicated, "at least one" means one or more, and "a plurality" means two or more. In addition, "and/or" describes an association relationship of the association object, and indicates that there may be three relationships, for example, a and/or B may indicate: a alone, a and B together, and B alone, wherein a, B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one (one) of a, b, or c may represent: a, b, c, a-b, a-c, b-c, or a-b-c, wherein a, b, c may be single or plural.

It should be further understood that the description of the first, second, etc. in the embodiments of the present application is for illustration and distinction of descriptive objects, and is not intended to represent any limitation on the number of devices in the embodiments of the present application, nor is it intended to constitute any limitation on the embodiments of the present application.

It should also be appreciated that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

The embodiment of the application is applied to the technical field of artificial intelligence.

Among these, artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

With research and advancement of artificial intelligence technology, research and application of artificial intelligence technology is being developed in various fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned, automatic driving, unmanned aerial vehicles, robots, smart medical treatment, smart customer service, etc., and it is believed that with the development of technology, artificial intelligence technology will be applied in more fields and with increasing importance value.

Embodiments of the present application may relate to natural language processing (Nature Language processing, NLP) in artificial intelligence technology, an important direction in the computer science and artificial intelligence fields. It is studying various theories and methods that enable effective communication between a person and a computer in natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Thus, the research in this field will involve natural language, i.e. language that people use daily, so it has a close relationship with the research in linguistics. Natural language processing techniques typically include text processing, semantic understanding, machine translation, robotic questions and answers, knowledge graph techniques, and the like.

The embodiment of the application can also relate to Machine Learning (ML) in the artificial intelligence technology, wherein ML is a multi-domain interdisciplinary, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, confidence networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

First, related terms related to the embodiments of the present application will be described.

First order logic: first-order logic (FOL), also known as first-order predicate logic (first-order predicate logic), is a formal inference logic system. In predicate logic, predicates can be divided into two parts, predicate names and individuals, wherein an individual is a subject in a proposition, and is used for representing a certain independent thing or a certain abstract concept, and predicate names are predicates of the proposition, and are used for representing properties, states or relations among individuals, and the like. The first-order predicate logic is a formal language that approximates natural language, knowledge represented by it is easily accepted, and the predicate logic is a binary logic, only true or false scores, with which precise knowledge can be expressed. Illustrating: the x represents an electronic physiological data, the predicate Symptom (x) represents that the physiological data x contains typical symptoms of influenza, and the predicate Disease (x) represents that the diagnosis of x is influenza. Expressed as Symptom (x) = > Disease (x): if physiological data x contains typical symptoms of influenza, then the disease to which x corresponds is diagnosed as influenza.

Fuzzy logic: fuzzy logic (fuzzy logic) is a mathematical method of processing uncertainty and ambiguity information, and aims to solve the problem that conventional binary logic (i.e. true or false) is difficult to process. Fuzzy logic allows us to use fuzzy sets to represent uncertainties and ambiguities to better address complex problems in the real world. For example, some physiological data x contains many symptoms, some are typical symptoms of influenza, some are not, and when the physiological data x contains typical symptoms of influenza, the logic symptomm (x) has a certain uncertainty, the fuzzy logic represents the uncertainty by membership (between 0 and 1), and the uncertainty can be used for synthesizing a plurality of fuzzy logics, so as to perform fuzzy reasoning and decision.

Convolutional neural network: (Convolutional Neural Network, CNN) is a feed-forward neural network that covers a portion of the information of the peripheral neurons by a convolutional layer and a pooling layer. Convolutional neural networks consist of one or more convolutional layers and a top fully-connected layer, which also includes associated weights and pooling layers. This architecture enables convolutional neural networks to be trained using a two-dimensional structure of input data, as well as using back-propagation algorithms. Convolutional neural networks require fewer parameters to consider and have wide application in both image and text processing.

Cyclic neural network: (Recurrent Neural Network, RNN) is a very classical sequence-oriented network model that models natural language sentences or other timing signals. It has only one physical RNN unit, but the RNN unit may be expanded in time steps, with each time step receiving the input of the current time step and the output of the previous time step, and then performing the calculation to obtain the output of the current time step.

At present, two disease prediction schemes are mainly adopted, one is a disease reasoning scheme based on expert rules, and related diagnosis and treatment rules are mainly arranged according to authoritative medical knowledge or medical experience, so that an experience decision system is constructed for disease diagnosis; the other is a disease prediction scheme based on data driving, namely, a machine learning or deep learning model is built based on large-scale electronic physiological data to perform disease prediction.

The disease reasoning scheme based on expert rules has the advantages of being relatively simple and direct, and the rules have certain evidence-based and explanatory properties, and the disadvantages are that the arrangement of expert rules requires a great deal of manual work, and the rules cannot be automatically updated along with the update of medical knowledge. In addition, the background and experience of different medical institutions and specialists are quite different, so that rule decision making has certain limitation. The data-driven disease prediction scheme can achieve higher accuracy based on the existing pre-trained model and a large amount of data, but at the same time sacrifices a certain interpretability. In addition, in the case of diseases with low morbidity, the traditional neural network model may have poor prediction effect due to insufficient training data.

Currently, there is no disease prediction scheme that can fuse these two schemes. The disease reasoning scheme based on expert rules and the disease prediction scheme based on data driving have advantages and disadvantages, and how to combine the two schemes to conduct disease prediction becomes a difficulty in the field of disease diagnosis.

In view of the above, the embodiment of the application provides a disease prediction method, which can realize the organic fusion of a disease reasoning scheme based on expert rules and a disease prediction scheme based on data driving, and is beneficial to improving the accuracy and the interpretability of disease prediction.

Specifically, diagnostic knowledge information of the subject may be acquired; matching the diagnosis knowledge information with at least one logic rule to obtain a matching result of the diagnosis knowledge information and the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category; inputting the diagnosis knowledge information into a neural network model for feature extraction to obtain a first disease feature of the diagnosis knowledge information; the first disease feature has a dimension equal to the number of the at least one disease category; inputting the matching result and the first disease characteristics into a knowledge enhancement network, and carrying out data enhancement on the characteristics of the dimension corresponding to at least one disease category of the first disease characteristics to obtain second disease characteristics; based on the second disease characteristic, a disease category of the subject is predicted.

According to the embodiment of the application, the diagnosis knowledge information of the object is matched with at least one logic rule, and the matching result is used as an expert rule to carry out data enhancement on the first disease feature of the diagnosis knowledge information extracted by the neural network, so that the enhanced second disease feature is fused with the expert rule and the disease feature extracted by the neural network, and the disease prediction based on the second disease feature can realize the organic fusion of the disease reasoning scheme based on the expert rule and the disease prediction scheme based on the data drive, so that the knowledge experience rule formed in the industry for a long time can be utilized, and the disease prediction can be carried out by using the calculation model based on the data drive, thereby being beneficial to improving the accuracy and the interpretability of the disease prediction.

Fig. 1 shows a schematic diagram of an application scenario according to an embodiment of the present application.

As shown in fig. 1, the application scenario involves a server 1 and a terminal device 2, and the terminal device 2 may communicate data with the server 1 through a communication network. The server 1 may be a background server of the terminal device 2.

The terminal device 2 may be, for example, a device with rich man-machine interaction, internet access capability, various operating systems, and strong processing capability. The terminal device may be a terminal device such as a smart phone, a tablet computer, a portable notebook computer, a desktop computer, a wearable device, a vehicle-mounted device, etc., but is not limited thereto. Alternatively, in the embodiment of the present application, the terminal device 2 is installed with an application program for disease prediction or an application program with a disease prediction function.

The server 1 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs (Content Delivery Network, content delivery networks), basic cloud computing services such as big data and artificial intelligent platforms, and the like. Servers may also become nodes of the blockchain.

The server may be one or more. Where the servers are multiple, there are at least two servers for providing different services and/or there are at least two servers for providing the same service, such as in a load balancing manner, as embodiments of the application are not limited in this respect.

The terminal device and the server may be directly or indirectly connected through wired or wireless communication, which is not limited in the present application. The present application does not limit the number of servers or terminal devices. The scheme provided by the application can be independently completed by the terminal equipment, can be independently completed by the server, and can be completed by the cooperation of the terminal equipment and the server, and the application is not limited to the scheme.

It should be understood that fig. 1 is only an exemplary illustration, and does not specifically limit the application scenario of the embodiment of the present application. For example, fig. 1 illustrates one terminal device, one server, and may actually include other numbers of terminal devices and servers, which the present application is not limited to.

The embodiment of the application can be applied to any application scene of disease auxiliary diagnosis. Fig. 2 shows a schematic diagram of the interface of the disease-assisted diagnostic product. As shown in fig. 2, a user may input patient visit information, such as including basic information and symptom information of a patient, in a patient information input box 210. For example, the basic information of the patient may include information that can be used for disease diagnosis, such as patient age, patient sex, and registration department. As a specific example, in the patient information input box 210, the patient age is 30, the sex is female, the department of registration is gynaecology, the patient complaint is "obesity", and the current medical history (i.e., symptom information) is: purple skin with lines, hair enlargement, slow or stagnant growth, shorter length, etc. The disease auxiliary diagnostic system can predict the possible disease type of the patient according to the patient's visit information. With continued reference to fig. 2, the disease prediction box 220 displays a list of predicted disease types for the current patient, pre-predicted 4: obesity, polycystic ovary syndrome, hypercortisolism, and amenorrhea. The doctor determines that the predicted disease list meets the medical rationality.

In the disease prediction process, the original end-to-end neural network model fails to predict the disease type because of the low incidence of hypercortisolism and the corresponding insufficient sample size of training data. However, according to the embodiment of the application, typical symptoms of hypercortisolism can be obtained through the rule from priori treatment information to disease types, such as hirsutism, central obesity, skin purple streak and the like, so that the disease types of hypercortisolism can be accurately predicted through enhancing the output of the neural network model through priori knowledge.

The following describes the technical scheme of the embodiments of the present application in detail through some embodiments. The following embodiments may be combined with each other, and some embodiments may not be repeated for the same or similar concepts or processes.

Fig. 3 is a schematic flow chart of a disease prediction method 300 according to an embodiment of the present application, where the disease prediction method 300 may be performed by any electronic device having data processing capabilities, for example, the electronic device may be implemented as a server or a terminal device, for example, may be implemented as the server 1 or the terminal device 2 in fig. 1, which is not limited in this regard. As shown in fig. 3, the disease prediction method 300 includes steps 310 to 350.

At 310, diagnostic knowledge information of the subject is obtained, the diagnostic knowledge information being used to determine a disease type.

Illustratively, the entered diagnostic knowledge information of the patient may be obtained through an interactive interface of the disease-assisted diagnostic system, such as the interface of FIG. 2. The diagnosis knowledge information is used to determine the disease type of the subject, and may be referred to as diagnosis knowledge, diagnosis information, or the like, without limitation.

Illustratively, the diagnostic knowledge information includes, but is not limited to, at least one of symptom information, medical history information, disease-inducing factor information. Alternatively, the diagnosis knowledge information may further include at least one of information for diagnosing a disease, such as age, sex, morbidity, etc., without limitation.

In some embodiments, the diagnosis text data may be acquired, the diagnosis knowledge information may be identified by the diagnosis text data, and the identified diagnosis knowledge information may be normalized to obtain normalized diagnosis knowledge information. For example, when the visit text data X includes: the man, 25 years old, the high fever accompanies debilitation, coughs, aversion to cold three days … and other text data, and the diagnosis knowledge information can be obtained by identifying and standardizing the diagnosis knowledge information: men, 25 years old, high fever, debilitation, cough, aversion to cold, etc.; wherein, men and 25 years old are basic information of patients, and high fever, hypodynamia, cough and aversion to cold are symptom information. By standardizing the diagnostic knowledge information, rapid and accurate data processing, such as data storage or data matching, may be facilitated.

320, matching the diagnosis knowledge information with at least one logic rule to obtain a matching result of the diagnosis knowledge information and the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category.

The at least one logic rule may be obtained prior to step 320. For example, a medical logic knowledge base may be constructed including the at least one logic rule. As shown in fig. 4, the at least one logic rule may be obtained through the following steps 321 and 322.

321, a medical knowledge statement is acquired, which medical knowledge statement is used to characterize the relationship of the diagnostic knowledge information and the disease type.

Specifically, medical knowledge sentences can be obtained according to authoritative medical knowledge or medical experience. The medical Knowledge may include Knowledge experience formed in the medical field for a long time, and may be "priori Knowledge (Prior knowledges)", in the embodiment of the present application. By way of example, the medical knowledge statement may include a declarative medical knowledge statement describing the relationship of diagnostic knowledge information (e.g., symptom information, age information, gender, etc.) to a disease type. The medical knowledge sentence may be also called a disease diagnosis sentence, a disease diagnosis knowledge, or the like, without limitation.

Illustratively, the medical knowledge statement is usually in the form of a natural language description, for example, a, b, c, etc. are typical symptoms of a disease a, and a person with high incidence of a disease is preschool children. As a specific example, the medical knowledge statement may be: the symptoms of high fever with debilitation are common in influenza, cold intolerance and polycystic ovary syndrome is a common disease caused by endocrine and metabolic abnormality of women of childbearing age, and the polycystic ovary syndrome is mainly clinically manifested by irregular menstrual cycle, infertility, hirsutism and/or acne and the like.

322, performing diagnosis knowledge information identification and disease type identification on the medical knowledge statement to obtain at least one logic rule.

By way of example, combining diagnostic information identified from a medical knowledge statement with a disease category may result in a logical rule of the diagnostic information to the disease category. The logic rule may be a first order logic or a first order predicate logic, where an individual may be a disease category and a predicate name may be visit information.

As one implementation, the logic rules may be expressed in the form of "IF … THEN …". Wherein the part after the IF before the THEN is a front piece of the logic rule, i.e. the condition of the IF, and the part after the THEN is a result of the logic rule, i.e. the front piece after the IF can produce the result after the THEN.

For example, for the medical knowledge statement "high fever with systemic debilitation is common to influenza", symptom information (an example of diagnosis knowledge information) can be identified as high fever, debilitation, disease category as influenza is identified, AND the corresponding obtained logic rule is "IF fever AND debilitation THEN influenza"; for the medical knowledge statement "cold intolerance state of influenza", the symptom information can be identified as cold intolerance, the disease category is identified as influenza, and the corresponding obtained logic rule is "IF cold intolerance THEN influenza"; for the medical knowledge statement "polycystic ovary syndrome is a disease caused by endocrine AND metabolic abnormality common to women of childbearing age", the age (one example of diagnosis knowledge information) can be identified as the childbearing age, the sex (one example of diagnosis knowledge information) as the female, the disease-inducing factor as endocrine AND metabolic abnormality, the disease category as polycystic ovary syndrome, the corresponding obtained logic rule as "IF childbearing age AND female AND endocrine AND metabolic abnormality THEN polycystic ovary syndrome"; for the medical knowledge statement "polycystic ovary syndrome is mainly clinically represented by irregular menstrual cycle, infertility, hirsutism AND/or acne", the symptom information can be identified as irregular menstrual cycle, infertility, hirsutism AND/or acne, the disease category is identified as polycystic ovary syndrome, AND the corresponding obtained logic rule is "IF irregular menstrual cycle AND infertility AND hirsutism AND acne chen polycystic ovary syndrome".

Alternatively, the diagnostic knowledge information identified from the medical knowledge sentence may be subjected to a normalization process to obtain normalized diagnostic knowledge information. For example, for the medical knowledge statement "high fever with systemic debilitation commonly found in influenza", the identified symptom information is "high fever with systemic debilitation", and after standardized processing, the symptom information is "high fever, debilitation", and the characters or words without corresponding actual symptoms are removed.

Alternatively, the disease type identified from the medical knowledge statement may also be subjected to a normalization process to obtain a normalized disease type.

Further, based on standardized diagnostic information and/or disease type, corresponding logic rules may be derived. By standardizing the diagnostic knowledge information and/or the disease type, a fast and accurate data processing, such as data storage or data matching, may be facilitated.

Alternatively, the weight (also referred to as confidence) of each logic rule may also be obtained. Illustratively, the weights may be between 0-1. In general, a medical knowledge statement is described in more detail for disease diagnosis knowledge, and the more certain the medical knowledge statement is used for, the higher the credibility of the medical knowledge statement is, the larger the weight is, and the larger the corresponding weight is for obtaining a logic rule according to the medical knowledge statement. As an achievable way, the medical knowledge statement may be labeled with weights by a specialist, which weights may also be weights of the logic rules derived from the medical knowledge statement.

For example, "high fever with debilitation commonly found in influenza" has a weight of 0.7, AND the corresponding "IF fever AND debilitation THEN influenza" has a weight of 0.7; the weight of "aversion to cold" is usually found in influenza "is 0.4, and the weight of" IF aversion to cold THEN influenza "is 0.4.

In some embodiments, after the at least one logical rule is obtained, the at least one logical rule may be stored in a database. Alternatively, the weight of each logic rule may also be stored in the database. The database may be referred to as, without limitation, a medical logic rule base, or a medical logic knowledge base.

Alternatively, assuming that the number of disease categories corresponding to the disease prediction is N, the database may include logic rules corresponding to the N disease categories. For example, medical knowledge sentences related to 270 kinds of diseases can be collected currently, and logic rules corresponding to the 270 kinds of diseases are obtained according to the medical knowledge sentences, wherein each logic rule corresponds to one disease type and corresponding diagnosis knowledge. It should be appreciated that a disease type may correspond to one or more logical rules, as the application is not limited in this regard.

In some embodiments, in step 320, the diagnostic knowledge information may be matched with at least one logic rule in the knowledge base or rule base to obtain a matching result of the diagnostic knowledge information with each logic rule. The matching result of the diagnosis knowledge information and the logic rule can be determined according to the matching degree of the diagnosis knowledge information and the front piece in the logic rule.

Illustratively, when part or all of the diagnostic knowledge information is completely matched with the front piece in one logic rule, the degree of matching of the diagnostic knowledge information with the logic rule is 100%; when part or all of the diagnostic knowledge information is matched with the front piece 50% in one logic rule, the matching degree of the diagnostic knowledge information and the logic rule is 50%; when part or all of the information in the diagnosis knowledge information is completely unmatched with the front piece in one logic rule, the degree of matching of the diagnosis knowledge information with the logic rule is 0.

330, inputting the diagnosis knowledge information into a neural network model for feature extraction to obtain first disease features of the diagnosis knowledge information; the dimension of the first disease feature is equal to the number of the at least one disease category.

Wherein each dimension of the first disease feature may correspond to a disease category. For example, when the number of disease types that the neural network model can predict is N, the first disease feature may be an N-dimensional feature, each dimension of which may correspond to one disease type, and the feature of the corresponding dimension corresponds to a feature component of the diagnostic knowledge information in the corresponding disease type. Optionally, the disease type that the neural network model can predict is the same as the disease type corresponding to the logic rules in the database.

By way of example, the neural network model may include, without limitation, a CNN or RNN, or other neural network having a feature extraction function or encoding function, such as a transducer encoder, etc.

In some embodiments, the first disease feature may be an output layer feature of the neural network model, the vector dimension of the output layer feature being the same as the number of predicted disease types. As one implementation, the output layer feature is the output feature prior to activation.

And 340, inputting the matching result and the first disease characteristics into a knowledge enhancement module, and carrying out data enhancement on the characteristics of the dimensions corresponding to at least one disease category of the first disease characteristics to obtain second disease characteristics.

For example, if the diagnosis knowledge information is matched (e.g., the matching degree is between 0 and 100%) to one or more logic rules corresponding to a certain disease category, data enhancement may be performed on the feature of the dimension corresponding to the disease category in the first disease feature, and the enhanced first disease feature is the second disease feature.

In some embodiments, FIG. 5 shows a schematic diagram of a process for data enhancement. As shown in fig. 5, after the diagnostic knowledge information X is input into the neural network model 501, an output layer feature Y is obtained, and the output layer feature Y may be further input into the knowledge-enhanced neural network element 502. The matching result of the diagnostic knowledge information and at least one logic rule obtained at the same time can be used as priori knowledge 503 to be input into knowledge coding 504, the coding result is input into knowledge enhancement neural network element 502, and data enhancement is performed on the output layer feature Y to obtain an enhanced feature Y'.

In some embodiments, referring to fig. 6, the second disease feature may be obtained through the following steps 341 and 342.

And 341, synthesizing at least one logic rule according to the matching result to obtain a synthetic knowledge rule, wherein the synthetic knowledge rule comprises at least one first-order logic of at least one disease category corresponding to the diagnosis knowledge information.

For example, the at least one disease category to which the diagnostic knowledge information corresponds may be predicted N disease categories.

Wherein the composite knowledge rule may comprise at least one weighted first order logic. This first order logic can be expressed as [ Symptom (X) ], i.e. the diagnostic knowledge information X contains typical symptoms of the disease Y. For N disease categories, the synthetic knowledge rules may include N categories of this first order logic.

In some embodiments, the synthetic knowledge rules may be obtained through the following steps 3411 through 3413.

3411 determining M logic rules matched with the diagnosis knowledge information according to the matching result.

For example, when the degree of matching of the diagnostic knowledge information with the front piece of the logic rule is greater than 0, the logic rule may be determined to be the logic rule matching the diagnostic knowledge.

3412 performing fuzzy logic processing on the logic rules corresponding to the same disease category in the M logic rules to obtain the first-order logic of the same disease category in the synthetic knowledge rule.

Specifically, when the M logic rules include logic rules of the same disease category, fuzzy logic processing may be performed according to logic rules corresponding to the same disease category in the M logic rules, so as to obtain first-order logic of the same disease category in the synthetic knowledge rule. Optionally, the first order logic is weighted first order logic.

For example, for s logic rules corresponding to the same disease type, fuzzy logic processing, such as extraction operation, may be performed on the front part and the weight of the s logic rules, so as to obtain a calculation result as the first-order logic of the disease type in the synthetic knowledge rule. The disjunctive operation corresponds to an OR operation in the binary logic and is used for calculating the union of two fuzzy sets: d (x) =max (a (x), B (x)). Optionally, the front pieces of the s logic rules are subjected to fuzzy logic processing, for example, a union of the front pieces of the s logic rules is taken as typical diagnosis information (such as typical symptoms) of the disease type in the synthetic knowledge rule. The more typical diagnostic information for a disease type in the synthetic knowledge rules, the greater the likelihood that the diagnostic knowledge information corresponds to that disease type. Alternatively, the result of fuzzy logic processing by the weight of the s logic rules may be the maximum weight in the s logic rules.

As a specific example, when diagnosing knowledge information: when a man is 25 years old, the man is 100% matched with the front part of logic rule #1 of IF high heat AND hypodynamia THEN influenza AND is 100% matched with the front part of logic rule #2 of IF high heat AND hypodynamia THEN influenza AND is not matched with other logic rules corresponding to influenza diseases, the two logic rules are subjected to extraction operation to obtain the weight of the first-order logic of the diagnosis knowledge information corresponding to the influenza in the synthetic rule of 0.7, namely the weight of the diagnosis knowledge information corresponding to typical symptoms of the influenza in the synthetic knowledge rule of 0.7. Here, typical symptoms of influenza are the extracted arithmetic results of these two logical rules, namely including "high fever, hypodynamia, aversion to cold".

3413 obtaining the first order logic of the other disease categories in the synthetic knowledge rule according to the logic rules corresponding to the other disease categories except the same disease category in the M logic rules.

For example, for a disease type #1 other than the same disease type among the M logic rules, the diagnosis knowledge information is matched to one logic rule #1 corresponding to the disease type #1, and at this time, the first order logic of the disease type #1 in the synthetic knowledge rule can be obtained according to the logic rule #1, which indicates that the diagnosis knowledge information includes typical symptoms of the disease type # 1. Alternatively, the weight of the first order logic may be determined according to the weight of the logic rule # 1.

Optionally, the weight of the first-order logic of the diagnosis knowledge information corresponding to the logic rule corresponding to the disease type in the synthetic knowledge rule may be determined according to the matching degree of the diagnosis knowledge information and the front piece of the logic rule and the weight of the logic rule. For example, the weight of the first order logic corresponding to the type of disease in the composite knowledge rule may be determined based on the product of the degree of matching of the diagnostic knowledge information with the front piece of the logic rule and the weight of the logic rule.

For example, when the matching degree of the diagnosis knowledge information and the front piece of one logic rule is 0, the weight of the first-order logic corresponding to the disease category in the diagnosis knowledge information in the synthetic knowledge rule is 0; when the matching degree of the diagnosis knowledge information and the front piece of one logic rule is 100%, the weight of the first-order logic corresponding to the diagnosis knowledge information in the synthesized knowledge rule is the weight of the logic rule.

342, inputting the synthetic knowledge rule and the first disease feature into a knowledge enhancement network, and performing data enhancement on the feature of the dimension corresponding to at least one disease category of the first disease feature to obtain a second disease feature.

Illustratively, with continued reference to fig. 5, the synthetic knowledge rules may be input as a priori knowledge 503 to the knowledge coding 504, and the resulting coding result may be input to the knowledge enhancement neural network element 502 to perform data enhancement on the input layer feature Y, resulting in an enhanced feature Y'.

Fig. 7 shows a schematic diagram of the structure of the enhanced knowledge network. Assuming n disease categories, the vector dimension of the Neural Network (NN) output layer is n. In the embodiment of the application, the neuron of the output layer of the neural network can be modified to strengthen the output layer. As shown in fig. 7, a new Knowledge Enhancement (KE) module may be introduced, comprising n knowledge enhancement neurons k, updating the original output layer feature y to y'.

For one computational graph, assume that there is a non-circular conditional statement: K→Y, where K is an expression in conjunctive or disjunctive form and Y is a single predicate expression. The computational graph of neurons associated with Y can be defined as:

where g is the activation function and where,for network parameters, x is the middle layer input of y. In addition, vector K may also be defined as a neuron associated with predicate K. In fig. 7, vector K, predicate K, may be used as a priori knowledge to augment y. When predicate K is expressed as true, the value of y increases. Alternatively, the value of y may be reduced when predicate K is expressed as false. To achieve this goal, an enhanced neuronal structure may be defined, expressed as the following relationship (1):

（1）

Wherein,,the method is used for capturing information whether the priori knowledge K is true or not as a distance function;Is a super-parameter scalar for indicating the degree of data enhancement, < ->。

Illustratively, with continued reference to FIG. 5, knowledge encoding 504 may perform the encoding process to derive a distance functionIs input into a knowledge-enhanced neural network element 502. />

According to the embodiment of the application, the implicit priori knowledge is encoded into the neural network by defining the enhanced neurons of the output layer, namely by a distance function, so that the neural network can adjust the downstream output value according to the output value of the upstream network, thereby realizing the knowledge enhancement of the output value.

In some embodiments, the second disease feature may be obtained by data enhancement of the feature of the dimension corresponding to at least one disease category of the first disease feature according to steps 3421 and 3422 below.

3421 determining, by the encoding unit, a distance function from the at least one first order logic and the feature of the dimension corresponding to the at least one disease category in the first disease feature; the dimension of the distance function is the same as the dimension of the first disease feature.

Specifically, the encoding unit may encode according to at least one first-order logic of at least one disease category corresponding to the diagnosis knowledge information obtained in step 341 and the dimension feature corresponding to each disease category, and determine the distance function . Here, the first order logic corresponding to each disease type in the synthetic knowledge rule and the feature of the dimension corresponding to each disease type in the first disease feature may be encoded to obtain a value of each dimension of the distance function.

It should be noted that, the key step of constructing the knowledge enhancement neural network element is to obtain a good distance function, and the distance function can return a positive value to enhance the output layer characteristics of the neural network when the first order logic is true. Optionally, when the first order logic is false, the distance function returns a value of 0, and the output layer characteristics of the neural network are not enhanced. Based on this, the embodiment of the present application adopts the following two calculation modes to acquire the distance function.

As a first calculation method, the encoding unit may perform a trigonometric norm (T-norm) operation on at least one first order logic and a feature of a dimension corresponding to at least one disease category in the first disease feature to obtain the distance function.

By way of example, the trigonometric norm operation may be expressed as the following expression:

wherein,,representation->The state of the corresponding neuron. Exemplary, ->The value of (2) is in the range of 0 to 1.

Wherein,,and the feature of the dimension corresponding to the ith disease category in the first order logic of the ith disease category in the synthetic knowledge rule or the first disease feature is represented. / >May be referred to as a precondition predicate. Alternatively, the synthetic knowledge rule may include, without limitation, first order logic of various diagnostic knowledge information (such as symptom information, evoked factor information, medical history information, etc.) corresponding to influenza, or first order logic of other types of diagnostic information.

Specifically, the distance function obtained by the trigonometric norm operation is a distance function in a conjunctive form, which means that as long as one of the plurality of preconditions predicates is false, the return value of the distance function is 0. When all of the plurality of preconditions predicates are true, the return value of the distance function is。

As a specific example, for a certain diagnostic knowledge information x, if the output layer feature of the neural network corresponds to the predicted value y=0.5 of influenza, and the weight of the first order logic corresponding to the typical symptom of influenza in the synthetic knowledge rule is 0.7, that is, the knowledge enhancement neuron k=0.7, the distance function in the conjunctive form is adopted to obtain d=0.5+0.7-2+1=0.2.

As a second calculation method, the encoding unit may perform a trigonometric residual norm (T-norm) operation on at least one first order logic and a feature of a dimension corresponding to at least one disease category in the first disease feature to obtain the distance function.

By way of example, the trigonometric residual norm operation may be expressed as the following expression:

wherein,,representation->The state of the corresponding neuron. Exemplary, ->Is in the range of 0 to 1。The meaning of (c) may be found in the description above. />

Specifically, the distance function obtained by the trigonometric-residual-norm operation is a distance function in a disjunctive form, which means that the return value of the distance function is 0 only when all of the plurality of preconditions predicates are false. When one of the plurality of preconditions predicates is true, the return value of the distance function is。

As a specific example, for a certain diagnostic knowledge information x, if the output layer feature of the neural network corresponds to the predicted value y=0.5 of influenza, and the weight of the first order logic corresponding to the typical symptom of influenza in the synthetic knowledge rule is 0.7, that is, the knowledge enhancement neuron k=0.7, d=min (1, (0.5+0.7))=1 is obtained by using the distance function in the above extraction form.

In some embodiments, the distance function comprises a differentiable distance function. When the distance function is distinguishable, a gradient descent algorithm can be adopted in the model training stage to update the model parameter value by solving the gradient of the objective function.

3422 inputting the distance function and the first disease feature into a knowledge enhancement network, and performing data enhancement on the feature of the dimension corresponding to at least one disease category of the first disease feature to obtain a second disease feature.

For example, the knowledge enhancement network may superimpose a value of a dimension of at least one disease category of the distance function with a dimension feature corresponding to at least one disease category of the first disease feature, to achieve data enhancement of the dimension feature corresponding to the at least one disease category of the first disease feature, resulting in the second disease feature.

Continuing with the example above, superimposing the distance function d=0.2 in the form of a conjunctive onto the predicted value y=0.5 of the output layer characteristics of the neural network for influenza, resulting in an enhanced neuron output y' =0.5+0.2=0.7; and adding the distance function d=1 in the extracted form to the predicted value y=0.5 of the output layer characteristic corresponding to the influenza of the neural network, so as to obtain the output y' =0.5+1=1.5 of the enhanced neuron.

350, predicting a disease class of the subject based on the second disease signature.

For example, a second disease feature may be entered into the classification layer to predict a disease category of the subject. As a specific example, when the first disease feature is a feature of the output layer (before being inactivated), the enhanced first disease feature, i.e. the second disease feature, may be input to the activation layer to obtain the disease type of the subject.

Therefore, the embodiment of the application performs data enhancement on the first disease feature of the diagnosis knowledge information extracted by the neural network by matching the diagnosis knowledge information of the object with at least one logic rule and taking the matching result as an expert rule, so that the enhanced second disease feature is fused with the expert rule and the disease feature extracted by the neural network, thereby performing disease prediction according to the second disease feature, realizing the organic fusion of a disease reasoning scheme based on the expert rule and a disease prediction scheme based on data driving, not only utilizing knowledge experience rules formed in the industry for a long time, but also utilizing a calculation model based on data driving to perform disease prediction, and being beneficial to improving the accuracy and the interpretability of the disease prediction.

Fig. 8 shows a schematic flow chart of another disease prediction method provided by an embodiment of the present application. It should be understood that fig. 8 illustrates steps or operations of a disease prediction method, but these steps or operations are merely examples, and that embodiments of the present application may also perform other operations or variations of the individual operations in fig. 8. Furthermore, the various steps in fig. 8 may be performed in a different order than presented in fig. 8, and it is possible that not all of the operations of fig. 8 are to be performed.

801, medical knowledge is acquired.

In particular, the medical knowledge, i.e. medical knowledge sentences, is used to characterize the relationship of diagnostic knowledge information and disease types. As an example, medical knowledge sentences such as "high fever with systemic debilitation is common in influenza", "influenza has cold intolerance symptoms". In particular, reference may be made to the relevant description as in fig. 4.

802, labeling medical knowledge weights.

In particular, medical knowledge weights may be annotated by a physician. For example, it is possible to label "high fever with general debilitation frequently occurring in influenza" with a weight of 0.7 and "influenza with cold intolerance symptoms" with a weight of 0.4.

803, constructing a logic rule base.

Specifically, the medical knowledge can be subjected to keyword (such as symptom information) identification and standardization processing to obtain diagnosis knowledge information in the medical knowledge as an IF (intermediate frequency) front piece, and the obtained disease category as the THEN result, so as to construct a logic rule base. As a specific example, the logical rule for "high fever with systemic debilitation is commonly found in influenza" IF high fever AND debilitation THEN influenza "AND its weight is 0.7, AND the logical rule for" influenza with chilly symptoms "is" IF chilly THEN influenza "AND its weight is 0.4.

In particular, the process of constructing a logical rule base can be seen from the relevant description as in fig. 4.

Keyword recognition and normalization 804.

Here, the keywords such as diagnosis knowledge information, such as symptom information, disease induction information, medical history information, age, sex, and the like, are not limited. Optionally, the keywords may also include disease category information. In addition, the identified information is subjected to standardization processing, so that rapid and accurate data processing such as data storage or data matching can be facilitated.

805, inputting diagnostic knowledge information.

For example, diagnostic knowledge information of a patient may be entered at a disease assisted diagnostic system interface. The diagnosis knowledge information may include, but is not limited to, information such as symptom information, disease induction information, medical history information, age, sex, and the like. As a specific example, the diagnosis knowledge information x may be: men, 25 years old, have high fever with debilitation, cough and aversion to cold for three days.

Optionally, the input diagnostic information can be standardized, which is beneficial to rapid and accurate data processing, such as data storage or data matching.

806, logic rule matching.

Specifically, the diagnosis knowledge information (e.g., after normalization) can be matched with the logic rules in the logic rule base, so as to obtain a matching result of the diagnosis knowledge information and each logic rule. Specifically, the matching process may refer to the related description in step 320 in fig. 3, which is not repeated here.

As a specific example, matching the logic rule base according to the diagnosis knowledge information x may be extracted by matching the logic rule: "IF hyperthermia AND debilitation THEN influenza" AND its weight is 0.7, "IF aversion to cold THEN influenza" AND its weight is 0.4.

807, a synthetic knowledge rule is obtained.

Wherein the synthetic knowledge rule comprises at least one first order logic for diagnosing at least one disease category to which the knowledge information corresponds. Specifically, the process of obtaining the synthetic knowledge rule may refer to the description in step 341 in fig. 6, which is not repeated herein. As a specific example, the synthesis of "IF hyperthermia AND hypodynamia THEN influenza" (weight of 0.7) AND "IF aversion to cold THEN influenza" (weight of 0.4) were performed, AND the resulting synthetic knowledge rules: the physiological data X contains influenza typical symptoms {0.7}.

808, inputting diagnostic knowledge information into the neural network.

809, a first disease signature is acquired.

In particular, the diagnostic knowledge information is input into the neural network, and the output layer of the neural network, or the output of a layer in between, can be used as the first disease feature. Specifically, the process of obtaining the first disease feature through the neural network may refer to the description of step 330 in fig. 3, which is not repeated herein.

810, inputting the composite knowledge rules and the first disease feature into a knowledge enhancement network.

811, obtaining a second disease feature.

Specifically, the synthetic knowledge rule and the first disease feature are input into a knowledge enhancement network, and the data enhancement is performed on the feature of the dimension corresponding to at least one disease category of the first disease feature through the synthetic knowledge rule, so that the second disease feature can be obtained. Specifically, the process of obtaining the second disease feature through the knowledge-enhanced network may refer to the description of step 342 in fig. 6, which is not repeated here.

The embodiment of the application can input the second disease characteristics into the classification layer to predict the disease category of the object. As a specific example, when the first disease feature is a feature of the output layer (before being inactivated), the second disease feature may be input to the activation layer, resulting in a disease type of the subject.

Fig. 9 is a schematic flow chart of a disease prediction model training method 900 according to an embodiment of the present application, where the disease prediction model training method 900 may be performed by any electronic device having data processing capabilities, for example, the electronic device may be implemented as a server or a terminal device, for example, may be implemented as the server 1 or the terminal device 2 in fig. 1, which is not limited in this regard. As shown in fig. 9, the disease prediction model training method 900 includes steps 910 to 960.

At 910, a physiological data sample is obtained, the physiological data sample including diagnostic knowledge information and a disease category label of the subject sample.

The physiological data sample may be, for example, a raw physiological data sample. Physiological data is a record of the course of medical activity, such as the occurrence, development, and prognosis of a patient's disease, checking, cutting, and treatment. The physiological data sample may include, among other things, diagnostic knowledge information of the subject and a disease category label. As a specific example, training sample data may include 2 tens of thousands of electrophysiological data, corresponding to 270 disease categories.

920, matching the diagnosis knowledge information with at least one logic rule to obtain a matching result of the diagnosis knowledge information and the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category.

930, inputting the diagnosis knowledge information into a neural network model for feature extraction to obtain a first disease feature of the diagnosis knowledge information; the first disease feature has a dimension equal to the number of the at least one disease category.

By way of example, the neural network model may be TextCNN or TextRNN, or a transducer encoder, etc., without limitation.

And 940, inputting the matching result and the first disease feature into a knowledge enhancement network to perform data enhancement on the feature of the dimension corresponding to at least one disease category of the first disease feature, thereby obtaining a second disease feature.

In some embodiments, the at least one logic rule may be synthesized based on the matching result to obtain a synthesized knowledge rule including at least one first order logic for diagnosing the at least one disease category with knowledge information. And then, inputting the synthetic knowledge rule and the first disease feature into a knowledge enhancement network, and carrying out data enhancement on the feature of the dimension corresponding to at least one disease category of the first disease feature to obtain a second disease feature.

In some embodiments, the distance function may be determined by the encoding unit from the at least one first order logic and a feature of a dimension corresponding to the at least one disease category; the dimension of the distance function is the same as the dimension of the first disease feature. And then, inputting the distance function and the first disease characteristics into a knowledge enhancement network, and carrying out data enhancement on the characteristics of the dimension corresponding to at least one disease category of the first disease characteristics to obtain second disease characteristics.

In some embodiments, the distance function may be obtained by performing a trigonometric norm operation on the feature of the dimension corresponding to the at least one first order logic and the at least one disease category by the encoding unit.

In some embodiments, the distance function may be obtained by performing a trigonometric residual norm operation on the feature of the dimension corresponding to the at least one first order logic and the at least one disease category by the encoding unit.

In some embodiments, the distance function comprises a differentiable distance function.

In some embodiments, M logic rules that match the diagnostic knowledge information may be determined based on the matching results; performing fuzzy logic processing on logic rules corresponding to the same disease category in the M logic rules to obtain first-order logic of the same disease category in the synthetic knowledge rule; and obtaining the first-order logic of other disease categories in the synthetic knowledge rule according to the logic rules corresponding to other disease categories except the same disease category in the M logic rules.

950, deriving a predicted disease category for the subject based on the second disease characteristic.

Specifically, steps 920 to 950 may refer to descriptions of steps 320 to 350 in fig. 3, and are not repeated here.

And 960, performing parameter adjustment on the neural network model and the knowledge enhancement network according to the predicted disease type and the disease type label to obtain a trained disease prediction model.

Specifically, parameters of the neural network model and the knowledge enhancement network can be adjusted according to the predicted disease type and the disease type label until the training stopping condition is met, and the neural network model and the knowledge enhancement network determined by meeting the training stopping condition are output as a trained disease prediction model. That is, the disease prediction model may include a neural network model with parameters adjusted and a knowledge-enhancement network.

The training stop condition may be a preset training stop condition, for example, the maximum training number is reached or the accuracy of the model meets a preset value, which is not limited by the present application.

In some embodiments, the matching result and the first disease feature may be obtained by adopting a joint extraction and extraction coding manner to obtain a distance function, and further, according to the distance function, data enhancement is performed on the feature of the dimension corresponding to at least one disease category of the first disease feature, so as to test the effects of different coding methods. Through experiments, the extracted distance function has obvious improvement on the disease prediction result.

Table 1 shows a specific example of the prediction effect obtained by knowledge enhancement of the outputs of different neural network models. Among them, table 1 shows the predicted effect obtained by training with a small sample size (5000). Therefore, when the sample size is small, the disease prediction effect of the model can be remarkably improved. For example, as shown in table 1, knowledge enhanced TextCNN (i.e., textcnn+ke) can increase accuracy from 37% to 41%, i.e., 4 percentage points, relative to the original text-based CNN (TextCNN); knowledge-enhanced TextRNN (i.e., textrnn+ke) can increase accuracy from 36% to 39%, i.e., 3 percentage points, relative to the original TextRNN.

TABLE 1

Therefore, the embodiment of the application carries out data enhancement on the first disease feature of the diagnosis knowledge information extracted by the neural network by matching the diagnosis knowledge information of the object with at least one logic rule and taking the matching result as an expert rule, so that the enhanced second disease feature is fused with the expert rule and the disease feature extracted by the neural network, thereby carrying out disease prediction according to the second disease feature, and carrying out parameter updating on the neural network and the knowledge enhancement network through the disease type label corresponding to the prediction disease type and the diagnosis knowledge information to obtain a disease prediction model.

The specific embodiments of the present application have been described in detail above with reference to the accompanying drawings, but the present application is not limited to the specific details of the above embodiments, and various simple modifications can be made to the technical solution of the present application within the scope of the technical concept of the present application, and all the simple modifications belong to the protection scope of the present application. For example, the specific features described in the above embodiments may be combined in any suitable manner, and in order to avoid unnecessary repetition, various possible combinations are not described further. As another example, any combination of the various embodiments of the present application may be made without departing from the spirit of the present application, which should also be regarded as the disclosure of the present application.

It should be further understood that, in the various method embodiments of the present application, the sequence numbers of the foregoing processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic of the processes, and should not constitute any limitation on the implementation process of the embodiments of the present application. It is to be understood that the numbers may be interchanged where appropriate such that the described embodiments of the application may be practiced otherwise than as shown or described.

The method embodiments of the present application are described above in detail, and the apparatus embodiments of the present application are described below in detail with reference to fig. 10 to 12.

Fig. 10 is a schematic block diagram of a disease prediction apparatus 10 according to an embodiment of the present application. As shown in fig. 10, the disease prediction apparatus 10 may include an acquisition unit 11, a matching unit 12, a neural network model 13, a knowledge enhancement network 14, and a prediction unit 15.

An acquisition unit 11 for acquiring diagnostic knowledge information of the subject; the diagnostic knowledge information is used for determining the disease type;

a matching unit 12, configured to match the diagnosis knowledge information with at least one logic rule, so as to obtain a matching result of the diagnosis knowledge information with the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category;

The neural network model 13 is used for extracting features of the diagnosis knowledge information to obtain first disease features of the diagnosis knowledge information; wherein the dimension of the first disease feature is equal to the number of the at least one disease category;

a knowledge enhancement network 14, configured to input the matching result and the first disease feature, and perform data enhancement on a feature of a dimension corresponding to the at least one disease category of the first disease feature, so as to obtain a second disease feature;

a prediction unit 15 for predicting a disease category of the subject based on the second disease characteristic.

In some embodiments, knowledge enhancement network 14 is specifically configured to:

synthesizing the at least one logic rule according to the matching result to obtain a synthetic knowledge rule, wherein the synthetic knowledge rule comprises at least one first-order logic corresponding to the at least one disease category by the diagnosis knowledge information;

and inputting the synthetic knowledge rules and the first disease characteristics into the knowledge enhancement network, and carrying out data enhancement on the characteristics of the dimension corresponding to the at least one disease category of the first disease characteristics to obtain the second disease characteristics.

determining a distance function by an encoding unit according to the at least one first order logic and the feature of the dimension corresponding to the at least one disease category; the dimension of the distance function is the same as the dimension of the first disease feature;

and inputting the distance function and the first disease feature into the knowledge enhancement network, and carrying out data enhancement on the feature of the dimension corresponding to the at least one disease category of the first disease feature to obtain the second disease feature.

In some embodiments, the knowledge enhancement network 14 is specifically configured to:

and performing triangular norm operation on the features of the dimensions corresponding to the at least one first-order logic and the at least one disease category through the coding unit to obtain the distance function.

and performing trigonometric residual norm operation on the features of the dimensions corresponding to the at least one first-order logic and the at least one disease category through the coding unit to obtain the distance function.

According to the matching result, M logic rules matched with the diagnosis knowledge information are determined;

performing fuzzy logic processing on logic rules corresponding to the same disease category in the M logic rules to obtain first-order logic of the same disease category in the synthetic knowledge rule;

and obtaining first-order logic of other disease categories in the synthetic knowledge rule according to logic rules corresponding to other disease categories except the same disease category in the M logic rules.

In some embodiments, the acquisition unit 11 is further configured to:

acquiring medical knowledge sentences, wherein the medical knowledge sentences are used for representing the relationship between diagnosis knowledge information and disease types;

and carrying out diagnosis knowledge information identification and disease type identification on the medical knowledge statement to obtain the at least one logic rule.

In some embodiments, the diagnostic knowledge information includes at least one of symptom information, medical history information, disease-inducing factor information.

It should be understood that apparatus embodiments and method embodiments may correspond with each other and that similar descriptions may refer to the method embodiments. To avoid repetition, no further description is provided here. Specifically, the disease prediction apparatus 10 shown in fig. 10 may perform the above method embodiments, and each module or function in the disease prediction apparatus 10 is respectively for implementing the corresponding flow in the above disease prediction method 300, which is not described herein for brevity.

Fig. 11 is a schematic block diagram of a disease prediction model training apparatus 20 according to an embodiment of the present application. As shown in fig. 11, the disease prediction model training apparatus 20 may include an acquisition unit 21, a matching unit 22, a neural network model 23, a knowledge enhancement network 24, a prediction unit 25, and a parameter adjustment unit 26.

An acquisition unit 21 for acquiring a physiological data sample including diagnostic knowledge information of a subject sample and a disease category label;

a matching unit 22, configured to match the diagnosis knowledge information with at least one logic rule, so as to obtain a matching result of the diagnosis knowledge information with the at least one logic rule; wherein the at least one logic rule includes rules for diagnosing knowledge information into at least one disease category;

the neural network model 23 is used for extracting features of the diagnosis knowledge information to obtain first disease features of the diagnosis knowledge information; the dimension of the first disease feature is equal to the number of the at least one disease category;

a knowledge enhancement network 24, configured to input the matching result and the first disease feature, so as to perform data enhancement on the feature of the dimension corresponding to the at least one disease category of the first disease feature, so as to obtain a second disease feature;

A prediction unit 25 for deriving a predicted disease category for the subject based on the second disease characteristic;

and a parameter adjustment unit 26, configured to perform parameter adjustment on the neural network model and the knowledge enhancement network according to the predicted disease category and the disease category label, so as to obtain the trained disease prediction model.

It should be understood that apparatus embodiments and method embodiments may correspond with each other and that similar descriptions may refer to the method embodiments. To avoid repetition, no further description is provided here. Specifically, the disease prediction model training apparatus 20 shown in fig. 11 may perform the above-described method embodiments, and each module or function in the disease prediction model training apparatus 20 is for implementing the corresponding flow in the above-described disease prediction model training method 900, and for brevity, will not be described herein.

The apparatus of the embodiments of the present application is described above in terms of functional modules with reference to the accompanying drawings. It should be understood that the functional module may be implemented in hardware, or may be implemented by instructions in software, or may be implemented by a combination of hardware and software modules. Specifically, each step of the method embodiment in the embodiment of the present application may be implemented by an integrated logic circuit of hardware in a processor and/or an instruction in a software form, and the steps of the method disclosed in connection with the embodiment of the present application may be directly implemented as a hardware decoding processor or implemented by a combination of hardware and software modules in the decoding processor. Alternatively, the software modules may be located in a well-established storage medium in the art such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, and the like. The storage medium is located in a memory, and the processor reads information in the memory, and in combination with hardware, performs the steps in the above method embodiments.

Fig. 12 is a schematic block diagram of an electronic device 30 provided by an embodiment of the present application.

As shown in fig. 12, the electronic device 30 may include:

a memory 31 and a processor 32, the memory 31 being for storing a computer program and for transmitting the program code to the processor 32. In other words, the processor 32 may call and run a computer program from the memory 31 to implement the method in the embodiment of the present application.

For example, the processor 32 may be configured to perform the above-described method embodiments according to instructions in the computer program.

In some embodiments of the present application, the processor 32 may include, but is not limited to:

a general purpose processor, digital signal processor (Digital Signal Processor, DSP), application specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like.

In some embodiments of the present application, the memory 31 includes, but is not limited to:

volatile memory and/or nonvolatile memory. The nonvolatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable EPROM (EEPROM), or a flash Memory. The volatile memory may be random access memory (Random Access Memory, RAM) which acts as an external cache. By way of example, and not limitation, many forms of RAM are available, such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (Double Data Rate SDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), and Direct memory bus RAM (DR RAM).

In some embodiments of the present application, the computer program may be divided into one or more modules, which are stored in the memory 31 and executed by the processor 32 to perform the methods provided by the present application. The one or more modules may be a series of computer program instruction segments capable of performing the specified functions, which are used to describe the execution of the computer program in the electronic device.

As shown in fig. 12, the electronic device 30 may further include:

a transceiver 33, the transceiver 33 being connectable to the processor 32 or the memory 31.

The processor 32 may control the transceiver 33 to communicate with other devices, and in particular, may send information or data to other devices or receive information or data sent by other devices. The transceiver 33 may include a transmitter and a receiver. The transceiver 33 may further include antennas, the number of which may be one or more.

It will be appreciated that the various components in the electronic device are connected by a bus system that includes, in addition to a data bus, a power bus, a control bus, and a status signal bus.

The present application also provides a computer storage medium having stored thereon a computer program which, when executed by a computer, enables the computer to perform the method of the above-described method embodiments. Alternatively, embodiments of the present application also provide a computer program product comprising instructions which, when executed by a computer, cause the computer to perform the method of the method embodiments described above.

When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a digital video disc (digital video disc, DVD)), or a semiconductor medium (e.g., a Solid State Disk (SSD)), or the like.

It will be appreciated that in the specific implementation of the present application, when the above embodiments of the present application are applied to specific products or technologies and relate to data related to user information and the like, user permission or consent needs to be obtained, and the collection, use and processing of the related data needs to comply with the relevant laws and regulations and standards.

Those of ordinary skill in the art will appreciate that the various illustrative modules and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or modules, which may be in electrical, mechanical, or other forms.

The modules illustrated as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. For example, functional modules in various embodiments of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module.

The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily appreciate variations or alternatives within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method of disease prediction comprising:

2. The disease prediction method according to claim 1, wherein inputting the matching result and the first disease feature into a knowledge enhancement network, and performing data enhancement on the feature of the dimension corresponding to the at least one disease category of the first disease feature to obtain a second disease feature, includes:

3. The disease prediction method according to claim 2, wherein inputting the synthetic knowledge rules and the first disease features into the knowledge enhancement network, and performing data enhancement on the feature of the dimension corresponding to the at least one disease category of the first disease features to obtain the second disease features, comprises:

4. A disease prediction method according to claim 3, wherein the determining, by the encoding unit, a distance function from the at least one first order logic and the feature of the dimension corresponding to the at least one disease category comprises:

5. A disease prediction method according to claim 3, wherein the determining, by the encoding unit, a distance function from the at least one first order logic and the feature of the dimension corresponding to the at least one disease category comprises:

6. A disease prediction method according to claim 3, wherein the distance function comprises a differentiable distance function.

7. The disease prediction method according to claim 2, wherein synthesizing the at least one logic rule according to the matching result results to obtain a synthetic knowledge rule comprises:

8. The disease prediction method according to claim 1, further comprising:

9. The disease prediction method of any one of claims 1 to 8, wherein the diagnosis knowledge information includes at least one of symptom information, medical history information, disease-inducing factor information.

10. A disease prediction model training method, comprising:

11. A disease prediction apparatus, comprising:

12. A disease prediction model training device, comprising:

13. An electronic device comprising a processor and a memory, the memory having instructions stored therein that when executed by the processor cause the processor to perform the method of any of claims 1-10.

14. A computer storage medium for storing a computer program, the computer program comprising instructions for performing the method of any one of claims 1-10.

15. A computer program product comprising computer program code which, when run by an electronic device, causes the electronic device to perform the method of any one of claims 1-10.