Disease prediction method, device, storage medium, equipment and program product

Abstract

The present application discloses a disease prediction method, apparatus, storage medium, device and program product, which can be applied to scenarios such as artificial intelligence, machine learning, and medical treatment. The method includes: obtaining historical disease information and demographic information of the target object; determining a basic intensity that characterizes that the target object suffers from the target disease based on the demographic information; determining a time-varying conditional intensity that characterizes that the target object suffers from the target disease at a target time based on the historical disease information; and determining a predicted intensity that the target object suffers from the target disease at a target time based on the basic intensity and the time-varying conditional intensity. The embodiment of the present application determines the fixed intensity that the target object suffers from the target disease based on the demographic information, and determines the time-varying intensity that the target object suffers from the target disease at the target time based on the historical disease information, so that the intensity of the target object suffering from the target disease at the target time can be predicted, so that the disease of the target object can be accurately predicted.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and specifically to a disease prediction method, a disease prediction device, a computer-readable storage medium, a computer device, and a computer program product. Background Art

[0002] In related technologies, for the sake of people's health, it is necessary to predict diseases of the test subjects so that protection can be taken according to the prediction results. However, how to accurately predict diseases of the test subjects is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0003] The embodiments of the present application provide a disease prediction method, a disease prediction device, a computer-readable storage medium, a computer device, and a computer program product, which can predict the intensity of a target disease suffered by a target subject at a target time, thereby accurately predicting the disease of the target subject.

[0004] On the one hand, a disease prediction method is provided, which includes: obtaining historical disease information and demographic information of a target object, the demographic information including an average occurrence probability of the target disease, and at least one of the target object's age information, the target object's gender information, the target object's location information, and the target object's type of work information; determining a basic strength based on the demographic information, the basic strength representing the fixed strength of the target object suffering from the target disease; determining a time-varying conditional strength based on the historical disease information, the time-varying conditional strength representing the time-varying strength of the target object suffering from the target disease at a target time; and determining a predicted strength of the target object suffering from the target disease at the target time based on the basic strength and the time-varying conditional strength.

[0005] On the other hand, a disease prediction device is provided, which includes an acquisition unit, a first processing unit, a second processing unit, and a third processing unit. The acquisition unit is used to acquire historical disease information and demographic information of the target object, and the demographic information includes the average occurrence probability of the target disease, and at least one of the target object's age information, the target object's gender information, the target object's location information, and the target object's type of work information. The first processing unit is used to determine a basic strength based on the demographic information, and the basic strength represents the fixed strength of the target object suffering from the target disease. The second processing unit is used to determine a time-varying conditional strength based on the historical disease information, and the time-varying conditional strength represents the time-varying strength of the target object suffering from the target disease at the target time. The third processing unit is used to determine the predicted strength of the target object suffering from the target disease at the target time based on the basic strength and the time-varying conditional strength.

[0006] On the other hand, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the steps in the disease prediction method described in any of the above embodiments.

[0007] On the other hand, a computer device is provided, which includes a processor and a memory, wherein a computer program is stored in the memory, and the processor is used to execute the steps in the disease prediction method described in any of the above embodiments by calling the computer program stored in the memory.

[0008] On the other hand, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the disease prediction method as described in any of the above embodiments.

[0009] The embodiment of the present application determines the fixed intensity of the target subject suffering from the target disease based on demographic information, and determines the time-varying intensity of the target subject suffering from the target disease at the target time based on historical disease information, so as to predict the intensity of the target subject suffering from the target disease at the target time. In this way, the disease of the target subject can be accurately predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1A schematic diagram of the structure of the disease prediction system provided in an embodiment of the present application.

[0012] Figure 2 This is a schematic diagram of the first process of the disease prediction method provided in an embodiment of the present application.

[0013] Figure 3 This is a schematic diagram of the output of the disease prediction method provided in an embodiment of the present application.

[0014] Figure 4 This is a second flow chart of the disease prediction method provided in an embodiment of the present application.

[0015] Figure 5 This is a schematic diagram of the third flow chart of the disease prediction method provided in an embodiment of the present application.

[0016] Figure 6 This is a schematic diagram of the fourth process of the disease prediction method provided in an embodiment of the present application.

[0017] Figure 7 This is a fifth flow chart of the disease prediction method provided in an embodiment of the present application.

[0018] Figure 8 This is a sixth flow chart of the disease prediction method provided in an embodiment of the present application.

[0019] Figure 9 This is a seventh flow chart of the disease prediction method provided in an embodiment of the present application.

[0020] Figure 10 This is a schematic diagram of the eighth flow chart of the disease prediction method provided in an embodiment of the present application.

[0021] Figure 11 This is another output schematic diagram of the disease prediction method provided in an embodiment of the present application.

[0022] Figure 12 Schematic diagram of the self-attention layer provided in an embodiment of the present application.

[0023] Figure 13 This is a ninth flow chart of the disease prediction method provided in an embodiment of the present application.

[0024] Figure 14 A schematic diagram of obtaining the time condition strength provided in an embodiment of the present application.

[0025] Figure 15 This is the tenth flow chart of the disease prediction method provided in an embodiment of the present application.

[0026] Figure 16 A schematic diagram of a scenario of the disease prediction method provided in an embodiment of the present application.

[0027] Figure 17 A schematic diagram of another scenario of the disease prediction method provided in an embodiment of the present application.

[0028] Figure 18 This is the eleventh flow chart of the disease prediction method provided in the embodiment of the present application.

[0029] Figure 19 This is a twelfth flow chart of the disease prediction method provided in an embodiment of the present application.

[0030] Figure 20 A schematic diagram of the structure of a disease prediction device provided in an embodiment of the present application.

[0031] Figure 21 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0033] The embodiments of the present application provide a disease prediction method, a disease prediction device, a computer-readable storage medium, a computer device, and a computer program product. Specifically, the disease prediction method of the embodiments of the present application can be executed by a computer device, wherein the computer device can be a terminal or a server or other device. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart TV, a smart speaker, a wearable smart device, an intelligent voice interaction device, a smart home appliance, a smart car terminal, an aircraft or other device. The terminal can also include a client, which can be a medical service client, a client applet, a video client, a browser client, or an instant messaging client, etc. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and basic cloud computing services such as big data and artificial intelligence platforms.

[0034] The embodiments of the present application can be applied to various scenarios such as artificial intelligence, machine learning, artificial intelligence medical care, diagnosis search, and auxiliary diagnosis. Specifically, the embodiments of the present application can achieve disease prediction through artificial intelligence, machine learning, etc.

[0035] The solution provided in the embodiments of the present application determines the fixed intensity of the target disease suffered by the target subject based on demographic information, and determines the time-varying intensity of the target disease suffered by the target subject at the target time based on historical disease information, so as to predict the intensity of the target disease suffered by the target subject at the target time. In this way, the disease of the target subject can be accurately predicted, which is specifically illustrated by the following embodiments.

[0036] First, some nouns or terms that appear in the description of the embodiments of this application are explained as follows:

[0037] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0038] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0039] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0040] EHR: A digital version of a patient's paper medical record that may include a range of data, including demographics, medical history, medications and allergies, immunization status, lab test results, radiology images, vital signs, personal statistics such as age and weight, and billing information.

[0041] Self-attention: an attention neural network structure.

[0042] Recurrent neural network (RNN): A neural network structure used to process sequential inputs.

[0043] Embedding: A way to convert discrete variables into continuous vector representations.

[0044] In related technologies, disease prediction can also be achieved through recommendation algorithms. These algorithms leverage a patient's historical medical records to learn the matching relationship between the patient and the disease. The model takes both patient characteristics and the disease as input. After neural network calculations, the model outputs a score corresponding to the given patient and disease. A higher score indicates a higher match between the patient and the disease. By sequentially inputting the patient and all diseases into the network and sorting by the output values, multiple predicted diseases can be generated. However, recommendation algorithms cannot output trends in disease likelihood over time, making the model's output difficult for doctors and patients to fully trust.

[0045] The embodiment of the present application provides a disease prediction method, which mainly uses the historical disease information of the target object, supplemented by demographic information, to personalize the prediction of the diseases that the target object may suffer from in the future. This application can assist doctors in diagnosing diseases, provide advice for the precise prevention and care of diseases, and explore the relationship between diseases. The disease prediction involved in this application is different from the disease diagnosis task. Disease diagnosis is based on the patient's history or current physical condition to diagnose existing diseases, while the disease prediction of this application is to predict the target object's possible future diseases through existing data. This can provide more references for disease prevention and early intervention, while disease diagnosis is more to provide assistance for the diagnosis and treatment of existing diseases. The application scenarios and technologies of the two are obviously different.

[0046] The embodiments of the present application can be implemented in combination with cloud technology or blockchain network technology. For example, in the disease prediction method disclosed in the embodiments of the present application, these data can be stored on the blockchain.

[0047] Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide or local area network (WAN) to enable data computing, storage, processing, and sharing. Cloud technology is a general term for network, information technology, integration technology, management platform technology, and application technology, all based on the cloud computing business model. It can form a resource pool that can be used flexibly and on demand. Backend services in technical network systems, such as video websites, image websites, and more portals, require extensive computing and storage resources, so cloud technology requires cloud computing as a foundation.

[0048] Blockchain is a new application model for computer technologies, including distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a series of data blocks linked using cryptographic methods. Each block contains information about a batch of online transactions, used to verify the validity of this information (to prevent counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, the platform product and service layer, and the application service layer.

[0049] The underlying blockchain platform can include processing modules such as user management, basic services, smart contracts, and operation monitoring. Among them, the user management module is responsible for the identity information management of all blockchain participants, including maintaining public and private key generation (account management), key management, and maintaining the corresponding relationship between the user's real identity and the blockchain address (authority management), etc., and under authorization, it supervises and audits the transactions of certain real identities and provides risk control rule configuration (risk control audit); the basic service module is deployed on all blockchain node devices to verify the validity of business requests, and records the valid requests to the storage after consensus is reached. For a new business request, the basic service first adapts the interface to parse and authenticate the request (interface adaptation), and then encrypts the business information through the consensus algorithm (consensus management). The smart contract module is responsible for the registration, issuance, triggering and execution of contracts. Developers can define the contract logic in a programming language and publish it to the blockchain (contract registration). According to the logic of the contract terms, the contract logic is triggered by calling keys or other events to trigger execution. The contract logic is completed, and the contract upgrade and cancellation functions are also provided. The operation monitoring module is mainly responsible for the deployment, configuration modification, contract setting, cloud adaptation and real-time status visualization output of the product during the product release process, such as alarms, network status monitoring, and node equipment health status monitoring.

[0050] The platform's product service layer provides the basic capabilities and implementation framework for typical applications. Developers can build on these basic capabilities, overlay business features, and complete the blockchain implementation of business logic. The application service layer provides application services based on blockchain solutions for business participants to use.

[0051] For example, the algorithm model and model parameters used to calculate the basic strength, and the algorithm model and model parameters used to calculate the time-varying conditional strength, can all be stored on the blockchain. In order to facilitate the storage and query of the algorithm model and model parameters used to calculate the basic strength, and the algorithm model and model parameters used to calculate the time-varying conditional strength, the disease prediction method optionally further includes: sending the algorithm model and model parameters used to calculate the basic strength, and the algorithm model and model parameters used to calculate the time-varying conditional strength to the blockchain network, so that the nodes of the blockchain network fill the algorithm model and model parameters used to calculate the basic strength, and the algorithm model and model parameters used to calculate the time-varying conditional strength into a new block, and when consensus is reached on the new block, the new block is appended to the end of the blockchain. The embodiment of the present application can store the algorithm model and model parameters used to calculate the basic strength, and the algorithm model and model parameters used to calculate the time-varying conditional strength on the chain to achieve record backup. When it is necessary to predict a disease for the target object, the corresponding algorithm model and model parameters used to calculate the basic strength, and the algorithm model and model parameters used to calculate the time-varying conditional strength can be directly and quickly obtained from the blockchain, without the need for a series of processing through the disease prediction system before obtaining the basic strength and time-varying conditional strength, thereby improving the efficiency of data processing and data acquisition.

[0052] Please refer to Figure 1 , Figure 1 The disease prediction system provided in the embodiment of the present application is a schematic diagram of the structure of the disease prediction system. The disease prediction system includes a terminal 10 and a server 20, etc. The terminal 10 and the server 20 are connected via a network, such as a wired or wireless network connection.

[0053] Among them, the terminal 10 can be used to display a graphical user interface. Among them, the terminal is used to interact with the user through a graphical user interface, for example, by downloading and installing the corresponding client and running it through the terminal, for example, by calling the corresponding applet and running it, for example, by logging into the website to present the corresponding graphical user interface, etc. In an embodiment of the present application, the terminal 10 can receive and display the predicted intensity of the target object suffering from the target disease at the target time. The server 20 can calculate the basic intensity and the time condition intensity, and determine the predicted intensity of the target object suffering from the target disease at the target time based on the basic intensity and the time condition intensity. The server 20 can also send the predicted intensity to the terminal 10.

[0054] Among them, in an embodiment of the present application, the server 20 can be specifically used to: obtain historical disease information and demographic information of the target object, the demographic information including the average occurrence probability of the target disease, and at least one of the target object's age information, target object's gender information, target object's location information, and target object's job information; determine the basic strength based on the demographic information, the basic strength characterizes the fixed strength of the target object suffering from the target disease; determine the time-varying conditional strength based on the historical disease information, the time-varying conditional strength characterizes the time-varying strength of the target object suffering from the target disease at the target time; determine the predicted strength of the target object suffering from the target disease at the target time based on the basic strength and the time-varying conditional strength.

[0055] It should be noted that the order of description of the following embodiments does not limit the priority order of the embodiments.

[0056] Each embodiment of the present application provides a disease prediction method, which can be executed by a terminal or a server, or by both the terminal and the server. The embodiments of the present application illustrate the disease prediction method by an example in which the method is executed by a server.

[0057] Figure 2 、 Figures 4 to 10 、 Figure 13 、 Figure 15 、 Figure 18 These are all flow charts of the disease prediction method provided in the embodiments of this application. The method includes:

[0058] Step 201: Obtain historical disease information and demographic information of the target object. The demographic information includes the average probability of occurrence of the target disease, and at least one of the target object's age information, target object's gender information, target object's location information, and target object's job information.

[0059] Specifically, the target subject may be a patient, a healthy subject to be tested, or the like. Historical disease information may refer to information related to diseases that the target subject has previously experienced. For example, if the target subject experienced indigestion or acute bronchitis on October 1, 2012, this information may be recorded as historical disease information. Demographic information includes the average probability of occurrence of the target disease v and at least one of the target subject's age, gender, location, and occupation. The average probability of occurrence of the target disease v can be pre-calculated by obtaining a training set comprising historical disease information of the training subject, wherein the historical disease information of the training subject may be pre-collected historical disease information of multiple patients. Based on each patient's historical disease information, the time span from the first disease record to the last disease record and the total number of times each patient has suffered from the target disease v can be determined. Based on the time span and the total number of times, the probability of each patient suffering from the target disease can be calculated. For example, the probability of each patient suffering from the target disease can be calculated by dividing the total number of times by the time span. The average probability of all patients suffering from the target disease can then be calculated to obtain the average probability of the target disease v.

[0060] It is understandable that in the specific implementation of this application, when the above embodiments of this application are applied to specific products or technologies, the relevant data such as the historical disease information and demographic information of the target object are involved, it is necessary to obtain the permission or consent of the target object, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0061] Step 202 : determining a base strength based on demographic information, where the base strength represents the fixed strength of the target subject suffering from the target disease.

[0062] In this embodiment of the present application, the average occurrence probability p of the target disease v is used v , the target object's age information, the target object's gender information, the target object's location information, and the target object's job information are taken as input, so that the fixed disease intensity based on the target object's basic information can be captured as the basic intensity μ v (f).

[0063] Step 203 : determining the time-varying condition strength based on the historical disease information, where the time-varying condition strength represents the time-varying strength of the target subject suffering from the target disease at the target time.

[0064] Specifically, the time-varying conditional intensity Can be used to describe historical processes Effect on the intensity of the target subject's having the target disease at the target time.

[0065] Step 204 : Determine the predicted intensity of the target subject suffering from the target disease at the target time based on the baseline intensity and the time-varying conditional intensity.

[0066] Specifically, the present application uses the predicted strength λ of the target subject suffering from the target disease v at the target time t as v (t) is divided into two parts and added: basic strength μ v (f) and time-varying conditional intensity The formula is as follows:

[0067]

[0068] Calculate μ separately v (μ) and We can get λ v (t), given the target time t, the occurrence intensity of the target disease v at the target time t can be obtained, that is, the predicted intensity can be obtained, wherein the target time t can be in days. Of course, in other embodiments, the target time t can also be in hours, weeks, months, quarters, years, etc. v (f) and time-varying conditional intensity Adding them together can give the predicted strength λ v (t).

[0069] The embodiment of the present application determines the fixed intensity of the target subject suffering from the target disease based on demographic information, and determines the time-varying intensity of the target subject suffering from the target disease at the target time based on historical disease information, so as to predict the intensity of the target subject suffering from the target disease at the target time. In this way, the disease of the target subject can be accurately predicted.

[0070] The embodiment of the present application can perform personalized prediction of the target time t, where the target time t can be a certain time in the future. The present application can output the distribution of the predicted intensity of the target disease over time.

[0071] Figure 3 It shows the change in the intensity of heart disease in the future for the target subjects who suffered from diarrhea, cold, emphysema, coronary heart disease, and diabetes in the past year, as well as the contribution of each historical disease to the total intensity. Figure 3 The impact trends of various diseases on the target disease can be clearly seen, helping researchers explore interrelationships between diseases. This graph can be modified by varying the target time. Furthermore, the risk trends of the disease to be predicted (target disease) can be clearly visualized, allowing medical staff to clearly identify high-risk time points, facilitating the development of personalized care plans for patients.

[0072] Optional, such as Figure 4As shown, step 202 can be implemented through steps 401 and 402, specifically:

[0073] Step 401: Preprocess demographic information.

[0074] Specifically, demographic information requires preprocessing to facilitate subsequent use. Preprocessing methods include standardization. For example, if the target subject's age information is a continuous variable, it can be standardized within the training set. The training set includes the subject's demographic information, which can be pre-collected demographic information from multiple patients. After standardization, standardized parameters are obtained. The standardized parameters can be used to standardize the target subject's age information and the test set. If the target subject's gender information is a discrete variable, preprocessing can be performed by setting male to 1 and female to 0. If the target subject's location information is a discrete variable, preprocessing can be performed by using a one-hot vector to represent the target subject's location information. Assuming the target subject's location information has three possible values, a one-hot vector can be used to represent each value. For example, if the three locations are Beijing, Shanghai, and Shenzhen, then if the target subject's location is Beijing, the one-hot vector can be [10 0]; if the target subject's location is Shanghai, the one-hot vector can be [0 1 0]; and if the target subject's location is Shenzhen, the one-hot vector can be [0 0 1]. The target object's job type information is a discrete variable. Preprocessing of the target object's job type information can be performed by using a onehot vector to represent the target object's job type information. For details, please refer to the preprocessing method of the target object's location information, which will not be repeated here.

[0075] Step 402 : determining the basic strength based on the pre-processed demographic information and the feature weights corresponding to each type of sociodemographic information.

[0076] Specifically, simple linear regression can be used to fit the basic strength, the formula is as follows:

[0077] μ v (f)=α1*age+α2*gender+α3*area+α4*job+α5*p v

[0078] Among them, age can be a parameter formed by preprocessing the target object's age information, gender can be a parameter formed by preprocessing the target object's gender information, area can be a parameter formed by preprocessing the target object's location information, job can be a parameter formed by preprocessing the target object's job type information, and p vThe average probability of occurrence of the target disease v can be formed after preprocessing. The feature weights α1, α2, α3, α4, and α5 can be initialized to obtain initial values, and then updated through the network backpropagation during the training process. Each type of information in the sociodemographic information, such as age, gender, area, job, and p, can be updated through the feature weights α1, α2, α3, α4, and α5. v After weighting and summing, we can get the basic strength μ v (f).

[0079] In other embodiments, a shallow perceptron or other simple regression models may be used to fit the base strength. In this way, demographic information may be input into different formulas to obtain the base strength, thereby combining the time-varying conditional strength to predict the disease.

[0080] Optionally, the historical disease information includes multiple disease records, each disease record includes disease information and disease record time, such as Figure 5 As shown, step 203 can be implemented through steps 501 to 503, specifically:

[0081] Step 501 : determining a time variation parameter based on the disease recording time and the target time corresponding to the plurality of disease records. The time variation parameter represents a temporal variation trend of the impact of the target disease after the diseases corresponding to the plurality of disease records occur.

[0082] Information about diseases that the target subject has experienced can be recorded in different disease records. For example, if the target subject developed indigestion and acute bronchitis on October 1, 2012, this information can be recorded in one disease record; if the target subject developed emphysema and dental caries on March 5, 2013, this information can be recorded in another disease record. Each disease record can include disease information and the time of disease recording. The disease information can specifically refer to the disease, such as indigestion, acute bronchitis, emphysema, or dental caries, and the disease recording time can refer to the time of disease occurrence, such as October 1, 2012, or March 5, 2013.

[0083] The embodiment of the present application is mainly based on the point process modeling technology of the event flow. By treating the occurrence of the target disease as the occurrence of an event, the intensity function (time-varying parameter) of each target disease is established, and the probability of occurrence of the target disease in the target object at the target time (predicted intensity) is predicted by combining the dynamic impact intensity. Among them, the time-varying parameter f(t; β v ), time-varying parameter f(t; β v ) represents the changing trend of the impact of the target disease v over time after the occurrence of the diseases corresponding to multiple disease records. The formula is as follows:

[0084] f(t;β v )=β v exp(-β v (tt i ))

[0085] The exponential distribution commonly used in point process prediction is used, t is the target time, t i is the disease recording time of the i-th disease record, β v is the exponential distribution parameter of the target disease v, and a B vector can be randomly generated to represent {β v} v∈V , V is the set of target diseases v, the elements in the B vector are initially taken as ∈(0,1], and then updated through the network back propagation during the training process, β v Take the corresponding element from vector B. Calculate the time difference between the target time and the disease recording time, and determine the time change parameter f(t; β) based on the time difference and the exponential distribution parameter v ), wherein the time variation parameter and the time difference may be negatively correlated.

[0086] In other embodiments, in addition to using the exponential distribution to represent the time decay distribution, other distributions (such as Weibull distribution) may also be used to represent the time decay distribution, which is not specifically limited here.

[0087] Step 502: determining a dynamic impact strength based on the disease information and the disease recording time corresponding to each disease record in the plurality of disease records. The dynamic impact strength represents the impact strength of the disease information on the target disease.

[0088] Dynamic impact intensity Characterizes the target object's disease record time t i The impact intensity of the corresponding disease information on the target disease v.

[0089] Step 503: Determine the time-varying condition intensity according to the time-varying parameter and the dynamic impact intensity.

[0090] Optional, such as Figure 6 As shown, step 502 can be implemented through steps 601 to 606, specifically:

[0091] Step 601: Perform dimensionality reduction processing on the disease information to obtain dimensionality reduction information.

[0092] Dimensionality reduction can be achieved through an embedding algorithm, which uses a low-dimensional, dense vector to represent disease information. Specifically, node2vec, a classic graph network embedding algorithm, can be used to generate reduced-dimensional information based on disease information. This allows for easier processing and integration.

[0093] Step 602: Use a preset algorithm model to extract effective information from the dimensionality reduction information.

[0094] Specifically, the preset algorithm model can process the dimensionality reduction information, and the output of the hidden layer in the preset dimensionality reduction model can be extracted as effective information. The effective information can represent the integrated information of multiple disease information of each disease record and represent the information of the entire disease record.

[0095] Step 603: Determine the location information of the valid information according to the disease recording time.

[0096] Specifically, the positional embedding (PE) of valid information can refer to the time point at which the disease record corresponding to each valid information occurred in the entire historical disease information. Since the input information of the self-attention layer is input simultaneously and the positional information cannot be obtained, the positional information can be added to each valid information before the valid information is input to the self-attention layer to enable the self-attention layer to perceive the input order.

[0097] Step 604: Determine the input information of the self-attention layer based on the valid information and the location information of the valid information.

[0098] Step 605: Process the input information using the self-attention layer to obtain output information, where the output information is used to represent integrated information of multiple disease records of the target object.

[0099] The self-attention layer can automatically adjust the weight of the information contained in all inputs according to the global input. The embodiment of the present application uses the self-attention layer to correct the valid information of each disease record based on the valid information of multiple disease records of historical disease information, thereby obtaining output information.

[0100] Step 606: Determine the dynamic influence strength based on the output information and the activation function.

[0101] Dynamic impact intensity The calculation formula is as follows:

[0102]

[0103] Among them, ai is the output information of the self-attention layer, σ is the activation function, such as the ReLU activation function, W and b can be initialized to get the initial value, and then updated through the network back propagation during the training process. The output information of the self-attention layer is processed using the parameters W and b, and the processed output information is input into the activation function. After the activation function is processed, the dynamic influence intensity can be obtained.

[0104] Optionally, the preset algorithm model is the RNN model, such as Figure 7 As shown, step 602 can be implemented through step 701, specifically:

[0105] Step 701: Input the dimension reduction information into the RNN model in reverse order, and the RNN model obtains valid information after processing the dimension reduction information.

[0106] Disease records (e.g., EHRs) are recorded by disease record number. Each disease record may contain multiple diseases diagnosed simultaneously, ranked by priority. The same patient may also have multiple disease records at different times. Therefore, it is necessary to process multiple disease information within the same disease record and multiple disease records. In this embodiment, an RNN model is introduced to obtain valid information from multiple disease information within the same disease record.

[0107] Since multiple disease information under a single disease record has a primary and secondary order, for example, they can be sorted according to the severity of the disease. For example, if a single disease record contains two diseases, emphysema and dental caries, emphysema can be the primary disease and dental caries can be the secondary disease. When the dimensionality reduction information of multiple disease information under the same disease record is input into the RNN model, the dimensionality reduction information can be input into the RNN model in reverse order, that is, the dimensionality reduction information corresponding to the top-ranked disease information is placed at the end of the RNN model. The RNN model processes the dimensionality reduction information in the order in which it is input. In this way, it can ensure that the output of the RNN model retains more primary disease information. Extract the hidden state h of the hidden layer when the RNN model processes the last dimensionality reduction information of the same disease record final , h final It captures the valid information of all disease information under a single disease record and represents the information of the entire disease record.

[0108] To speed up processing, when the target object includes multiple disease records, the multiple disease records can be processed in parallel by separate RNN models. Of course, in other embodiments, the multiple disease records of the target object can also be processed in parallel by a single RNN model, which is not specifically limited here.

[0109] In an embodiment of the present application, LSTM (long short-term memory) can be used in the RNN model to process the dimensionality reduction information corresponding to multiple disease information of the same disease record, wherein LSTM can also be replaced by other structures, such as bidirectional LSTM, GRU (Gate Recurrent Unit), etc., which are not specifically limited here.

[0110] Optionally, valid information includes elements of multiple dimensions, such as Figure 8 As shown, step 603 can be implemented through step 801, specifically:

[0111] Step 801, according to the disease recording time, the number of dimensions and the order of the elements in the valid information, determine the position information of each element to obtain the position information of the valid information.

[0112] Specifically, effective information may include elements of multiple dimensions, where an element may refer to a feature vector representing disease information. The dimension of the location information is the same as that of the effective information. The formula is as follows:

[0113]

[0114] Among them, t i represents the disease recording time of the i-th disease record, and d is the valid information h final , l is the position order of the elements in the valid information, and k is used to indicate the parity of l.

[0115] In this way, the position information corresponding to each element can be obtained, and the position information corresponding to each element of the set effective information can be used as the position information of the effective information.

[0116] Optionally, the valid information includes elements of multiple dimensions, and the position information of the valid information includes the position information of each element, such as Figure 9 As shown, step 604 can be implemented through step 901, specifically:

[0117] Step 901: Add each element of the valid information and the position information of the corresponding element to obtain input information.

[0118] Specifically, the i-th input information x of the self-attention layer i The calculation formula is as follows:

[0119]

[0120] Among them, h i,final is the valid information of the target object’s i-th disease record, The location information of the valid information, Each element representing valid information and the position information element of the corresponding element are added.

[0121] Optional, such as Figure 10 As shown, step 605 can be implemented through steps 1001 to 1003, specifically:

[0122] Step 1001: Use a weight matrix to convert input information into a key tensor, a query tensor, and a value tensor.

[0123] Specifically, the self-attention layer can include three weight matrices W key 、W query 、W value Through these three weight matrices, the input information x corresponding to the i-th disease record can be i Converted into three tensors, namely key tensor k i , query tensor q i Sum value tensor v i , the formula is as follows:

[0124] k i =x i *W key

[0125] q i =x i *W query

[0126] v i =x i *W value

[0127] Among them, the weight matrix W key 、W query 、W value The initial value can be initialized and then updated through the network back propagation during the training process. Input information x i With the weight matrix W key Multiplication can be performed to obtain the key tensor k i , input information x i With the weight matrix W query Multiplication can be performed to obtain the query tensor q i , input information x i With the weight matrix W value Multiplication can be performed to obtain the value tensor v i .

[0128] Step 1002: Determine an attention score based on the key tensor and the query tensor.

[0129] Specifically, through the key tensor k j , query tensor q i You can find xi For x j Attention score score i,j , the formula is as follows:

[0130] score i,j =q i *k j

[0131] Among them, x j is the input information corresponding to the jth disease record, k j is x j The corresponding key tensor. Key tensor k j With the query tensor q i Multiplying them gives the attention score i,j .

[0132] Step 1003: Determine output information based on the attention score and the value tensor.

[0133] Get score i,j After, x i Output information a i The calculation is as follows:

[0134]

[0135] Where n is the total number of disease records, v j is x j The corresponding value tensor.

[0136] In the embodiment of the present application, a self-attention layer is introduced. Using the self-attention layer is equivalent to performing bidirectional weighting. The output information of the i-th disease record fully refers to the information of other disease records, so that the present application can refer to the overall historical disease information to calculate the impact on the target disease.

[0137] The present application provides a variety of output results to demonstrate the powerful interpretability and applicability of the model, and the model can be applied to a variety of practical scenarios. In addition to outputting the trend chart of the impact of each disease on the target disease (such as Figure 3 ), it can also learn to extract the suppression or enhancement relationship of historical disease information internally, using the score of the self-attention layer i,j It can effectively visualize the interactions between diseases, help researchers or doctors understand or discover the potential joint effects between diseases, and enhance the credibility of the output. Examples of the interaction of historical disease information of the target object are as follows: Figure 11 , where the graph can be obtained by calling the softmax(score i,j )get.

[0138] The structure of the self-attention layer can be as follows Figure 12 As shown, the leftmost query is used as q i , and the three inputs at the bottom are used as x j , each x j is transformed to obtain key (k j ) and value (v j ). According to q i and k j , softmax(score i,j ) is obtained. According to softmax(score i,j ) and v j , multiplication (multiplication) is performed, and then addition (addition) is performed to obtain the output information a i .

[0139] Optionally, the self-attention layer can include multiple ones, and multiple self-attention layers form a multi-head self-attention structure (Multi head self attention). Using the multi-head self-attention structure can process and obtain the output information more accurately.

[0140] Optionally, as Figure 13 shown, step 503 can be implemented through step 1301, specifically:

[0141] Step 1301, determine the time-varying conditional intensity according to the time change parameter and the dynamic influence intensity corresponding to the target historical disease information, and the disease record time of the target historical disease information is earlier than the target time.

[0142] The time-varying conditional intensity is calculated as follows:

[0143]

[0144] where v is the target disease, used to describe the influence of the target historical disease information on the intensity of the target object having the target disease v at the target time t. t i is the disease record time of the i-th disease record, and t i <t indicates that the disease record time is earlier than the target time.

[0145] Figure 14In the process, the dimensionality reduction information (input embedding) is processed by the RNN module to obtain effective information, and the effective information and positional information (positional embedding) are input to the attention module. After being processed by the self-attention layer (selfattention), the output information is obtained and output to the aggregation module. The aggregation module calculates the time-varying conditional strength based on the output information and the B vector.

[0146] Optional, such as Figure 15 As shown, the method further includes:

[0147] Step 1501 : When the predicted intensity is greater than the intensity threshold, predict that the target subject will suffer from the target disease at the target time.

[0148] When the predicted strength is less than the strength threshold, it is predicted that the target subject will not suffer from the target disease at the target time. In this way, whether the target subject will suffer from the target disease at the target time can be determined by the predicted strength.

[0149] The intensity threshold can be obtained in advance through experiments. For example, different thresholds can be uniformly selected in advance, and predictions with intensities greater than the thresholds are considered to occur. The accuracy rate under each threshold is then calculated, and the threshold with the highest accuracy rate is taken as the intensity threshold.

[0150] Carriers of infectious diseases often have multiple complications. However, when a new infectious disease emerges or mutates, the type of complication cannot be determined promptly, nor can a patient's presence of a related complication accurately determine the presence of that infectious disease. The disease prediction method of this application automatically predicts the likely disease of a target subject by inputting their prevalence of multiple diseases. If the predicted results include this or other infectious diseases, medical personnel can be prompted to conduct a special examination or isolate the subject for observation.

[0151] The model proposed in this application can be used for preliminary retrieval of infectious diseases (such as influenza). By inputting information such as the target subject's location, age, gender, and historical and current illness conditions (such as fever and cough), the model will output a list of predicted diseases that may be present or may occur in the future (such as Figure 16 If the disease list includes influenza or specific diseases, an early warning can be issued to medical staff or disease control personnel.

[0152] The embodiment of the present application does not need to use the current physical indicators of the target object as features, and can well predict the diseases that the target object is susceptible to, so it can be applied to patient care or physical examinations. Figure 17Caregivers can input relevant information about the target patient, and the model will output possible diseases that may develop in the future. Caregivers can then design targeted care plans based on the results to help the target patient avoid or alleviate related diseases. For users themselves, when they want a physical examination, they can input their basic information and historical medical history. The model will output susceptible diseases, and users can then schedule a targeted physical examination based on the disease list.

[0153] The prevention and control of chronic diseases is very important. Because chronic diseases have the characteristics of covering a wide range of people, having many complications, lasting for a long time, and changing in a complex manner, the embodiments of this application use historical disease information to make personalized, accurate, and efficient predictions of future diseases, which can be well applied to the prevention and control of chronic diseases or chronic complications.

[0154] Due to objective conditions, EHRs generally have missing or incorrectly recorded diseases, resulting in poor quality. The present embodiment can determine whether a record has errors by calculating the predicted strength of the target disease corresponding to the disease record. When a disease record is missing, the present embodiment can also sequentially treat all diseases as target diseases and predict the predicted strength of the target disease at the missing location, selecting the target disease with the highest predicted strength to fill the missing location.

[0155] Optionally, the base intensity and the time-varying conditional intensity form a disease prediction model, which includes a set of parameters, such as Figure 18 As shown, the method further includes:

[0156] Step 1801: Obtain a training set, which includes historical disease information and demographic information of a training subject, wherein the historical disease information and demographic information of the training subject may be pre-collected historical disease information and demographic information of multiple patients.

[0157] Step 1802: Establish an objective function of the disease prediction model.

[0158] In order to enable the disease prediction model to complete multiple tasks, the objective function Loss used to train the disease prediction model is defined as follows:

[0159]

[0160] in, is the historical disease information of the training object at time T, which can be any time. θ represents a parameter set, which includes, for example, α1, α2, α3, α4, α5, B vector, W, b, W key 、W query 、W value wait, is the likelihood estimate of historical disease information, n is H T The number of disease records included.i is the disease recording time of the i-th disease record, v i The disease information recorded for the i-th disease.

[0161] Step 1803: Train the disease prediction model based on the training set and the objective function to obtain a parameter set that maximizes the objective function, so as to obtain a trained disease prediction model. The trained disease prediction model is used to process the historical disease information and sociodemographic information of the target subject to determine the predicted intensity of the target subject suffering from the target disease at the target time.

[0162] Specifically, we obtain the parameter set θ that maximizes the objective function * , the formula is as follows:

[0163] θ * =argmax(Loss)

[0164] When the objective function is maximized, the obtained parameter set can be used by the trained disease prediction model, so that the trained disease prediction model can process the historical disease information and sociodemographic information of the target subject to determine the prediction strength of the target subject suffering from the target disease at the target time.

[0165] The disease prediction model in the embodiments of this application incorporates a neural network. After improvements, the algorithm can effectively process a variety of input information, and the model's effectiveness and interpretability are greatly improved. The model proposed in this patent has the characteristics of comprehensive disease coverage, good model prediction effect, personalized prediction, and strong model interpretability.

[0166] See also Figure 19 To better illustrate the disease prediction method provided in the embodiment of the present application, the process of the disease prediction method provided in the embodiment of the present application can be summarized into the following steps:

[0167] Step 1801: Obtain a training set, which includes historical disease information and demographic information of a training subject, wherein the historical disease information and demographic information of the training subject may be pre-collected historical disease information and demographic information of multiple patients.

[0168] Step 1802: Establish an objective function of the disease prediction model.

[0169] In order to enable the disease prediction model to complete multiple tasks, the objective function Loss used to train the disease prediction model is defined as follows:

[0170]

[0171] in, is the historical disease information of the training object at time T, which can be any time. θ represents a parameter set, which includes, for example, α1, α2, α3, α4, α5, B vector, W, b, W key 、W query 、W value wait, is the likelihood estimate of historical disease information, n is H T The number of disease records included. i is the disease recording time of the i-th disease record, v i The disease information recorded for the i-th disease.

[0172] Step 1803: Train the disease prediction model based on the training set and the objective function to obtain a parameter set that maximizes the objective function, so as to obtain a trained disease prediction model. The trained disease prediction model is used to process the historical disease information and sociodemographic information of the target subject to determine the predicted intensity of the target subject suffering from the target disease at the target time.

[0173] Specifically, we obtain the parameter set θ that maximizes the objective function * , the formula is as follows:

[0174] θ * =argmax(Loss)

[0175] When the objective function is maximized, the obtained parameter set can be used by the trained disease prediction model, so that the trained disease prediction model can process the historical disease information and sociodemographic information of the target subject to determine the prediction strength of the target subject suffering from the target disease at the target time.

[0176] Step 201: Obtain historical disease information and demographic information of the target object. The demographic information includes the average probability of occurrence of the target disease, and at least one of the target object's age information, target object's gender information, target object's location information, and target object's job information.

[0177] Specifically, the target subject may be a patient, a healthy subject to be tested, or the like. Historical disease information may refer to information related to diseases that the target subject has previously experienced. For example, if the target subject experienced indigestion or acute bronchitis on October 1, 2012, this information may be recorded as historical disease information. Demographic information includes the average probability of occurrence of the target disease v and at least one of the target subject's age, gender, location, and occupation. The average probability of occurrence of the target disease v can be pre-calculated by obtaining a training set comprising historical disease information of the training subject, wherein the historical disease information of the training subject may be pre-collected historical disease information of multiple patients. Based on each patient's historical disease information, the time span from the first disease record to the last disease record and the total number of times each patient has suffered from the target disease v can be determined. Based on the time span and the total number of times, the probability of each patient suffering from the target disease can be calculated. For example, the probability of each patient suffering from the target disease can be calculated by dividing the total number of times by the time span. The average probability of all patients suffering from the target disease can then be calculated to obtain the average probability of the target disease v.

[0178] Step 401: Preprocess demographic information.

[0179] Specifically, demographic information needs to be pre-processed so that it can be used later, and the pre-processing method includes standardization.

[0180] Step 402 : determining the basic strength based on the pre-processed demographic information and the feature weights corresponding to each type of sociodemographic information.

[0181] Specifically, simple linear regression can be used to fit the basic strength, the formula is as follows:

[0182] μ v (f)=α1*age+α2*gender+α3*area+α4*job+α5*p v

[0183] Among them, age can be a parameter formed by preprocessing the target object's age information, gender can be a parameter formed by preprocessing the target object's gender information, area can be a parameter formed by preprocessing the target object's location information, job can be a parameter formed by preprocessing the target object's job type information, and p vThe average probability of occurrence of the target disease v can be formed after preprocessing. The feature weights α1, α2, α3, α4, and α5 can be initialized to obtain initial values, and then updated through the network backpropagation during the training process. Each type of information in the sociodemographic information, such as age, gender, area, job, and p, can be updated through the feature weights α1, α2, α3, α4, and α5. v After weighting and summing, we can get the basic strength μ v (f).

[0184] Step 501 : determining a time variation parameter based on the disease recording time and the target time corresponding to the plurality of disease records. The time variation parameter represents a temporal variation trend of the impact of the target disease after the diseases corresponding to the plurality of disease records occur.

[0185] Information about diseases that the target subject has experienced can be recorded in different disease records. For example, if the target subject developed indigestion and acute bronchitis on October 1, 2012, this information can be recorded in one disease record; if the target subject developed emphysema and dental caries on March 5, 2013, this information can be recorded in another disease record. Each disease record can include disease information and the time of disease recording. The disease information can specifically refer to the disease, such as indigestion, acute bronchitis, emphysema, or dental caries, and the disease recording time can refer to the time of disease occurrence, such as October 1, 2012, or March 5, 2013.

[0186] Step 601: Perform dimensionality reduction processing on the disease information to obtain dimensionality reduction information.

[0187] Dimensionality reduction can be achieved through an embedding algorithm, which uses a low-dimensional, dense vector to represent disease information. Specifically, node2vec, a classic graph network embedding algorithm, can be used to generate reduced-dimensional information based on disease information. This allows for easier processing and integration.

[0188] Step 701: Input the dimension reduction information into the RNN model in reverse order, and the RNN model obtains valid information after processing the dimension reduction information.

[0189] Disease records (e.g., EHRs) are recorded by disease record number. Each disease record may contain multiple diseases diagnosed simultaneously, ranked by priority. The same patient may also have multiple disease records at different times. Therefore, it is necessary to process multiple disease information within the same disease record and multiple disease records. In this embodiment, an RNN model is introduced to obtain valid information from multiple disease information within the same disease record.

[0190] Step 801, according to the disease recording time, the number of dimensions and the order of the elements in the valid information, determine the position information of each element to obtain the position information of the valid information.

[0191] Specifically, effective information may include elements of multiple dimensions, where an element may refer to a feature vector representing disease information. The dimension of the location information is the same as that of the effective information. The formula is as follows:

[0192]

[0193] Among them, t i represents the disease recording time of the i-th disease record, and d is the valid information h final , l is the position order of the elements in the valid information, and k is used to indicate the parity of l.

[0194] Step 901: Add each element of the valid information and the position information of the corresponding element to obtain input information.

[0195] Specifically, the i-th input information x of the self-attention layer i The calculation formula is as follows:

[0196]

[0197] Among them, h i,final is the valid information of the target object’s i-th disease record, The location information of the valid information, Each element representing valid information and the position information element of the corresponding element are added.

[0198] Step 1001: Use a weight matrix to convert input information into a key tensor, a query tensor, and a value tensor.

[0199] Specifically, the self-attention layer can include three weight matrices W key 、W query 、W value Through these three weight matrices, the input information x corresponding to the i-th disease record can be i Converted into three tensors, namely key tensor k i , query tensor qi Sum value tensor v i , the formula is as follows:

[0200] k i =x i *W key

[0201] q i =x i *W query

[0202] v i =x i *W value

[0203] Among them, the weight matrix W key 、W query 、W value The initial value can be initialized and then updated through the network back propagation during the training process. Input information x i With the weight matrix W key Multiplication can be performed to obtain the key tensor k i , input information x i With the weight matrix W query Multiplication can be performed to obtain the query tensor q i , input information x i With the weight matrix W value Multiplication can be performed to obtain the value tensor v i .

[0204] Step 1002: Determine an attention score based on the key tensor and the query tensor.

[0205] Specifically, through the key tensor k j , query tensor q i You can find x i For x j Attention score score i,j , the formula is as follows:

[0206] score i,j =q i k j

[0207] Among them, x j is the input information corresponding to the jth disease record, k j is x j The corresponding key tensor. Key tensor k j With the query tensor q i Multiplying them gives the attention score i,j .

[0208] Step 1003, determine the output information according to the attention score and the value tensor.

[0209] Obtain score i,j After that, the output information a of x i is calculated as follows: i

[0210]

[0211] where n is the total number of disease records, and v j is the value tensor corresponding to x j

[0212]

[0213] Step 606, determine the dynamic influence intensity according to the output information and the activation function. The calculation formula of the dynamic influence intensity

[0214] is as follows:

[0215] where a i is the output information of the self-attention layer, σ is the activation function, and the activation function is, for example, the ReLU activation function. W and b can be initialized to obtain initial values, and then their values are updated through the backward propagation of the network during the training process. The output information of the self-attention layer is processed using the parameters W and b, and the processed output information is input into the activation function. After being processed by the activation function, the dynamic influence intensity can be obtained.

[0216]

[0217] Step 1301, determine the time-varying conditional intensity according to the time change parameter corresponding to the target historical disease information and the dynamic influence intensity. The disease record time of the target historical disease information is earlier than the target time. The calculation formula of the time-varying conditional intensity

[0218] is as follows:

[0219] where v is the target disease, which is used to describe the influence of the target historical disease information

[0220] on the intensity of the target object having the target disease v at the target time t. t i is the disease record time of the i-th disease record, and t i < t indicates that the disease record time is earlier than the target time.

[0221] Step 204, determine the predicted intensity of the target object having the target disease at the target time according to the basic intensity and the time-varying conditional intensity.

[0221] Specifically, the present application uses the predicted strength λ of the target subject suffering from the target disease v at the target time t as v (t) is divided into two parts and added: basic strength μ v (f) and time-varying conditional intensity The formula is as follows:

[0222]

[0223] Calculate μ separately v (f) and We can get λ v (t), given the target time t, the occurrence intensity of the target disease v at the target time t can be obtained, that is, the predicted intensity can be obtained, wherein the target time t can be in days. Of course, in other embodiments, the target time t can also be in hours, weeks, months, quarters, years, etc. v (f) and time-varying conditional intensity Adding them together can give the predicted strength λ v (t).

[0224] Step 1501 : When the predicted intensity is greater than the intensity threshold, predict that the target subject will suffer from the target disease at the target time.

[0225] When the predicted strength is less than the strength threshold, it is predicted that the target subject will not suffer from the target disease at the target time. In this way, whether the target subject will suffer from the target disease at the target time can be determined by the predicted strength.

[0226] All of the above technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0227] The embodiment of the present application determines the fixed intensity of the target subject suffering from the target disease based on demographic information, and determines the time-varying intensity of the target subject suffering from the target disease at the target time based on historical disease information, so as to predict the intensity of the target subject suffering from the target disease at the target time. In this way, the disease of the target subject can be accurately predicted.

[0228] In order to better implement the disease prediction method of the embodiment of the present application, the embodiment of the present application also provides a disease prediction device. Figure 20 , Figure 20 This is a schematic diagram of the structure of a disease prediction device provided in an embodiment of the present application. The disease prediction device 2000 may include:

[0229] An acquisition unit 2010 is configured to acquire historical disease information and demographic information of a target subject, wherein the demographic information includes an average probability of occurrence of the target disease, and at least one of the target subject's age, gender, location, and occupation.

[0230] A first processing unit 2020 is configured to determine a base strength based on demographic information, where the base strength represents a fixed strength of the target subject suffering from the target disease;

[0231] The second processing unit 2030 is configured to determine a time-varying condition strength based on historical disease information, where the time-varying condition strength represents a time-varying strength of the target subject suffering from the target disease at the target time;

[0232] The third processing unit 2040 is used to determine the predicted intensity of the target subject suffering from the target disease at the target time according to the basic intensity and the time-varying conditional intensity.

[0233] Optionally, the first processing unit 2020 may be configured to: pre-process the demographic information; and determine the basic strength based on the pre-processed demographic information and the feature weights corresponding to each type of sociodemographic information.

[0234] Optionally, the historical disease information includes multiple disease records, each disease record includes disease information and disease record time, and the second processing unit 2030 can be used to: determine a time change parameter based on the disease record time and target time corresponding to the multiple disease records, the time change parameter characterizing the time-dependent trend of the impact of the target disease after the occurrence of the diseases corresponding to the multiple disease records; determine the dynamic impact intensity based on the disease information and disease record time corresponding to each disease record in the multiple disease records, the dynamic impact intensity characterizing the impact intensity of the disease information on the target disease; determine the time-varying condition intensity based on the time change parameter and the dynamic impact intensity.

[0235] Optionally, the second processing unit 2030 can be used to: perform dimensionality reduction processing on disease information to obtain dimensionality reduction information; extract effective information of the dimensionality reduction information using a preset algorithm model; determine the location information of the effective information based on the disease record time; determine the input information of the self-attention layer based on the effective information and the location information of the effective information; use the self-attention layer to process the input information to obtain output information, and the output information is used to represent the integrated information of multiple disease records of the target object; determine the dynamic influence intensity based on the output information and the activation function.

[0236] Optionally, the preset algorithm model is an RNN model, and the second processing unit 2030 can be used to: input the dimensionality reduction information into the RNN model in reverse order, and the RNN model obtains valid information after processing the dimensionality reduction information.

[0237] Optionally, the valid information includes elements of multiple dimensions, and the second processing unit 2030 can be used to determine the position information of each element according to the disease recording time, the number of dimensions and the order of the elements in the valid information to obtain the position information of the valid information.

[0238] Optionally, the valid information includes elements of multiple dimensions, and the position information of the valid information includes position information of each element. The second processing unit 2030 can be used to add each element of the valid information and the position information of the corresponding element to obtain input information.

[0239] Optionally, the second processing unit 2030 can be used to: convert the input information into a key tensor, a query tensor and a value tensor using a weight matrix; determine an attention score based on the key tensor and the query tensor; and determine output information based on the attention score and the value tensor.

[0240] Optionally, the second processing unit 2030 may be configured to determine the time-varying condition intensity based on the time-varying parameters and dynamic impact intensity corresponding to the target historical disease information, where the disease recording time of the target historical disease information is earlier than the target time.

[0241] Optionally, the apparatus 2000 includes a fourth processing unit 2050 , which may be configured to: predict that the target subject will suffer from the target disease at the target time when the predicted intensity is greater than an intensity threshold.

[0242] Optionally, the base strength and the time-varying conditional strength form a disease prediction model, the disease prediction model including a parameter set, and the apparatus 2000 includes an establishment unit 2060, a collection unit 2070, and a training unit 2080. The establishment unit 2060 may be configured to establish an objective function for the disease prediction model. The collection unit 2070 may be configured to obtain a training set, the training set including historical disease information and demographic information of the training subject. The training unit 2080 may be configured to train the disease prediction model based on the training set and the objective function to obtain a parameter set that maximizes the objective function, thereby obtaining a trained disease prediction model. The trained disease prediction model is used to process the historical disease information and sociodemographic information of the target subject to determine the predicted intensity of the target subject suffering from the target disease at the target time.

[0243] Each module and unit in the aforementioned disease prediction device may be implemented in whole or in part through software, hardware, or a combination thereof. Each module and unit may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module and unit.

[0244] The disease prediction device 2000 may be integrated into a terminal or server that has a storage device and a processor and has computing capabilities, or the disease prediction device 2000 may be the terminal or server.

[0245] Optionally, the present application also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0246] Figure 21 A schematic diagram of the structure of a computer device provided in an embodiment of the present application, which may be Figure 1 The terminal or server shown. Figure 21 As shown, the computer device 2100 may include: a communication interface 2101, a memory 2102, a processor 2103, and a communication bus 2104. The communication interface 2101, the memory 2102, and the processor 2103 communicate with each other via the communication bus 2104. The communication interface 2101 is used to communicate data with external devices. The memory 2102 can be used to store software programs and modules. The processor 2103 executes the software programs and modules stored in the memory 2102, such as the software programs for the corresponding operations in the aforementioned method embodiments.

[0247] Optionally, the processor 2103 can call the software program and module stored in the memory 2102 to perform the following operations: obtain historical disease information and demographic information of the target object, the demographic information includes the average occurrence probability of the target disease, and at least one of the target object's age information, target object's gender information, target object's location information, and target object's job information; determine the basic strength based on the demographic information, the basic strength represents the fixed strength of the target object suffering from the target disease; determine the time-varying conditional strength based on the historical disease information, the time-varying conditional strength represents the time-varying strength of the target object suffering from the target disease at the target time; determine the predicted strength of the target object suffering from the target disease at the target time based on the basic strength and the time-varying conditional strength.

[0248] This application also provides a computer-readable storage medium for storing a computer program. This computer-readable storage medium can be applied to a computer device, and the computer program causes the computer device to execute the corresponding process of the disease prediction method in the embodiment of this application. For the sake of brevity, it is not further described here.

[0249] This application also provides a computer program product comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding processes of the disease prediction method in the embodiments of this application. For the sake of brevity, these processes are not further described here.

[0250] This application also provides a computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding process of the disease prediction method in the embodiments of this application. For the sake of brevity, these processes are not further described here.

[0251] It should be understood that the processor of the embodiments of the present application may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiment can be completed by hardware integrated logic circuits in the processor or software instructions. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The various methods, steps, and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly implemented as a hardware decoding processor, or can be implemented by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0252] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0253] It should be understood that the above-mentioned memories are exemplary but not restrictive. For example, the memories in the embodiments of the present application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM RAM (DR RAM), etc. In other words, the memories in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories.

[0254] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0255] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0256] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0257] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0258] In addition, each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0259] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0260] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A disease prediction method, characterized in that: The method comprises: Acquiring historical disease information and demographic information of a target subject, wherein the demographic information includes an average probability of occurrence of the target disease and at least one of the target subject's age information, the target subject's gender information, the target subject's location information, and the target subject's job type information, wherein the historical disease information includes multiple disease records, each of which includes disease information and a time when the disease was recorded; determining a base strength based on the demographic information, wherein the base strength represents a fixed strength of the target subject suffering from the target disease; Determining a time variation parameter based on the disease recording time and target time corresponding to the multiple disease records, wherein the time variation parameter represents a time variation trend of the impact of the target disease after the diseases corresponding to the multiple disease records occur; determining a dynamic impact strength according to the disease information corresponding to each of the plurality of disease records and the disease recording time, wherein the dynamic impact strength represents the impact strength of the disease information on the target disease; determining a time-varying conditional strength according to the time-varying parameter and the dynamic impact strength, wherein the time-varying conditional strength represents the time-varying strength of the target subject suffering from the target disease at the target time; The predicted intensity of the target subject suffering from the target disease at the target time is determined according to the basic intensity and the time-varying conditional intensity.

2. The method according to claim 1, characterized in that Determining the base strength according to the demographic information includes: preprocessing the demographic information; The basic strength is determined according to the pre-processed demographic information and the feature weight corresponding to each type of information in the demographic information.

3. The method according to claim 1, characterized in that The determining of the dynamic impact intensity according to the disease information corresponding to each of the plurality of disease records and the disease recording time includes: Performing dimensionality reduction processing on the disease information to obtain dimensionality reduction information; Extracting effective information from the dimensionality reduction information using a preset algorithm model; Determining the location information of the valid information according to the disease recording time; Determining input information of a self-attention layer according to the valid information and the position information of the valid information; Processing the input information using the self-attention layer to obtain output information, wherein the output information is used to represent integrated information of the multiple disease records of the target subject; The dynamic influence intensity is determined according to the output information and the activation function.

4. The method according to claim 3, characterized in that The preset algorithm model is an RNN model, and the method of extracting effective information from the dimensionality reduction information using the preset algorithm model includes: The dimensionality reduction information is input into the RNN model in reverse order, and the RNN model obtains the valid information after processing the dimensionality reduction information.

5. The method according to claim 3, characterized in that The valid information includes elements of multiple dimensions, and the location information of the valid information is determined according to the disease recording time, including: According to the disease recording time, the number of the dimensions and the order of the elements in the valid information, the position information of each element is determined to obtain the position information of the valid information.

6. The method according to claim 3, characterized in that The effective information includes elements of multiple dimensions, the position information of the effective information includes position information of each element, and determining the input information of the self-attention layer according to the effective information and the position information of the effective information includes: The input information is obtained by adding each of the elements of the valid information and the position information of the corresponding elements.

7. The method according to claim 3, characterized in that The processing of the input information by the self-attention layer to obtain output information includes: Converting the input information into a key tensor, a query tensor, and a value tensor using a weight matrix; determining an attention score based on the key tensor and the query tensor; The output information is determined according to the attention score and the value tensor.

8. The method according to claim 1, characterized in that The determining the time-varying condition intensity according to the time-varying parameter and the dynamic influence intensity includes: The time-varying condition intensity is determined based on the time-varying parameter and the dynamic impact intensity corresponding to the target historical disease information, where the disease recording time of the target historical disease information is earlier than the target time.

9. The method according to any one of claims 1 to 8, characterized in that The method further comprises: When the prediction strength is greater than a strength threshold, it is predicted that the target subject will suffer from the target disease at the target time.

10. The method according to any one of claims 1 to 8, characterized in that The basic intensity and the time-varying conditional intensity form a disease prediction model, the disease prediction model includes a parameter set, and the method further includes: Obtaining a training set, wherein the training set includes historical disease information and demographic information of a training subject; Establishing an objective function of the disease prediction model; The disease prediction model is trained according to the training set and the objective function to obtain the parameter set that maximizes the objective function, so as to obtain a trained disease prediction model. The trained disease prediction model is used to process the historical disease information and demographic information of the target subject to determine the predicted intensity of the target subject suffering from the target disease at the target time.

11. A disease prediction device, characterized in that: The device comprises: an acquisition unit, the acquisition unit being configured to acquire historical disease information and demographic information of a target subject, the demographic information including an average probability of occurrence of a target disease, and at least one of the target subject's age information, the target subject's gender information, the target subject's location information, and the target subject's type of work information, the historical disease information including a plurality of disease records, each of the disease records including disease information and a time when the disease was recorded; a first processing unit configured to determine a base strength according to the demographic information, wherein the base strength represents a fixed strength of the target subject suffering from the target disease; a second processing unit, the second processing unit being configured to determine a time-varying parameter based on the disease recording time and target time corresponding to the multiple disease records, the time-varying parameter representing a temporal trend of the impact of the target disease after the disease corresponding to the multiple disease records occurs; determine a dynamic impact strength based on the disease information and the disease recording time corresponding to each of the multiple disease records, the dynamic impact strength representing the intensity of the impact of the disease information on the target disease; and determine a time-varying conditional strength based on the time-varying parameter and the dynamic impact strength, the time-varying conditional strength representing the time-varying intensity of the target subject suffering from the target disease at the target time; A third processing unit is configured to determine, based on the basic intensity and the time-varying conditional intensity, a predicted intensity of the target subject suffering from the target disease at the target time.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the steps in the disease prediction method according to any one of claims 1 to 10.

13. A computer device, characterized in that: The computer device includes a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to execute the steps of the disease prediction method according to any one of claims 1 to 10 by calling the computer program stored in the memory.

14. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the steps of the disease prediction method according to any one of claims 1 to 10 are implemented.

Images