Disease prediction method, device, electronic device and storage medium

Abstract

The present invention relates to the field of digital medical technology, and provides a disease prediction method, device, electronic device and storage medium. On the basis of real-time collection of medical records, target medical data are first extracted from the medical records, and then credible keywords and untrustworthy keywords in the target medical data are identified, so that different entity recognition models are used to identify the entities of the keywords, thereby improving the entity recognition accuracy of the keywords, and helping to improve the classification accuracy of the target medical data. After obtaining the entities of the keywords, the target medical data are classified based on matrix rank constraints and entities to obtain multiple categories of target medical data, so that disease category prediction is performed for each category of target medical data to obtain disease category prediction results, thereby improving the accuracy of disease category prediction.

Description

Technical Field

[0001] The present invention relates to the field of digital medical technology, and in particular to a disease prediction method, device, electronic equipment and storage medium. Background Art

[0002] At present, private medical institutions, such as outpatient departments and clinics, have relatively concentrated living environments and large population flows. The symptoms of many patients are similar most of the time, and most patients basically go to private outpatient clinics in the early stages to make a simple description of their symptoms and ask the outpatient staff, and then just get a medicine. If an epidemic disease occurs, it may lead to local clustered infections among people, and it is impossible to maximize the role of various medical institutions in normalizing epidemic prevention and control work.

[0003] Traditional clinic service platforms have not yet integrated epidemic prevention and control: Traditional clinic service platforms are generally used for daily clinic business to implement an integrated process of registration, billing and charging. They have not integrated medical record reporting and prevention and control policies to upgrade products, resulting in the inability to intelligently predict epidemic diseases. Summary of the invention

[0004] In view of the above, it is necessary to propose a disease prediction method, device, electronic device and storage medium to improve the accuracy of disease prediction.

[0005] A first aspect of the present invention provides a disease prediction method, the method comprising:

[0006] Receiving medical records reported by medical institutions, and extracting target medical data from the medical records;

[0007] Identify credible keywords and untrustworthy keywords in the target medical data, use a preset first entity recognition model to identify entities of the credible keywords, and use a preset second entity recognition model to identify entities of the untrustworthy keywords;

[0008] Classifying the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data;

[0009] The disease category is predicted for each type of the target medical data to obtain a disease category prediction result.

[0010] According to an optional implementation of the present invention, the identifying the trusted keywords and the untrusted keywords in the target medical data includes:

[0011] Performing word segmentation processing on the target medical data to obtain multiple keywords;

[0012] Traversing the multiple keywords, and obtaining a keyword from the multiple keywords as a target keyword in each traversal;

[0013] Obtain the first character and the last character of the target keyword;

[0014] Acquire a first neighbor character located before the first character in the target medical data, and acquire a second neighbor character located after the last character in the target medical data;

[0015] Determine whether the first character and the first neighbor character can form a word, and determine whether the last character and the second neighbor character can form a word;

[0016] When the first character and the first neighboring characters cannot form a word, and the last character and the second neighboring characters cannot form a word, determining that the target keyword is a credible keyword;

[0017] When the first character and the first neighboring character can form a word, and / or the last character and the second neighboring character can form a word, the target keyword is determined to be an untrustworthy keyword.

[0018] According to an optional implementation of the present invention, the target medical data is classified based on the matrix rank constraint and the entity to obtain multiple categories of target medical data, including:

[0019] splicing multiple entities corresponding to the target medical data into an entity vector;

[0020] Performing alignment processing on the entity vector to obtain an aligned entity vector;

[0021] Selecting a plurality of target medical data pairs from the target medical data;

[0022] Acquire the shared neighbor target medical data of each target medical data pair, and generate a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data;

[0023] Calculate a Laplace matrix based on the similarity matrix;

[0024] Finding the undirected graph connected branch structure of the Laplace matrix based on the Laplace matrix rank constraint theory;

[0025] Output multiple categories of target medical data according to the undirected graph connected branch structure.

[0026] According to an optional implementation of the present invention, the step of obtaining the shared neighbor target medical data of each target medical data pair and generating a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data comprises:

[0027] Acquire the first target medical data and the second target medical data in each of the target medical data pairs;

[0028] Finding K nearest neighbor target medical data of the first target medical data and K nearest neighbor target medical data of the second target medical data;

[0029] Acquire shared nearest neighbor target medical data from two of the K nearest neighbor target medical data;

[0030] Calculate the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data;

[0031] A similarity matrix is ​​generated according to the similarities of the plurality of target medical data pairs.

[0032] According to an optional embodiment of the present invention, the calculating the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data includes:

[0033] An average aligned entity vector is calculated based on the aligned entity vectors corresponding to the shared neighbor target medical data;

[0034] Calculate a first distance according to the aligned entity vector of the first target medical data and the average aligned entity vector;

[0035] Calculate a second distance according to the aligned entity vector of the second target medical data and the average aligned entity vector;

[0036] Calculating a third distance according to the aligned entity vector of the first target medical data and the aligned entity vector of the second target medical data;

[0037] The similarity is calculated based on the first distance, the second distance and the third distance using a preset similarity calculation model.

[0038] According to an optional implementation of the present invention, predicting the disease category for each type of target medical data includes:

[0039] Inputting each type of target medical data into a first disease prediction framework in a preset disease probability prediction model to obtain a first predicted disease;

[0040] Inputting the first predicted disease into the selective connection layer in the disease probability prediction model, and obtaining a second disease prediction architecture selected and specified by the selective connection layer;

[0041] Inputting the first predicted disease and the designated medical data into the designated second disease prediction framework to obtain a second predicted disease;

[0042] Mapping the designated medical data, the first predicted disease, and the second predicted disease into a disease classification vector;

[0043] The disease category of each type of target medical data is determined according to the preset standard classification vector and the disease classification vector.

[0044] According to an optional implementation of the present invention, extracting target medical data from the medical record includes:

[0045] Acquire a plurality of initial medical data from the medical record, wherein the initial medical data includes patient description information and medical description information;

[0046] Sampling the multiple initial medical data to obtain multiple medical samples having the same data distribution as the initial medical data;

[0047] Determining a medical description value in the plurality of medical samples such that a ratio of the number of initial medical data including the medical description value to the number of the plurality of medical samples is greater than a first preset threshold;

[0048] Acquire a patient description value corresponding to the medical description value from the plurality of medical samples, so that a ratio of the number of initial medical data including the patient description value to the number of the plurality of medical samples is greater than a second preset threshold;

[0049] Searching the plurality of initial medical data for initial medical data that includes the medical description value but does not include the patient description value;

[0050] The found initial medical data is used as the target medical data.

[0051] A second aspect of the present invention provides a disease prediction device, the device comprising:

[0052] A receiving module, used for receiving medical records reported by medical institutions and extracting target medical data from the medical records;

[0053] An identification module, used to identify credible keywords and untrustworthy keywords in the target medical data, using a preset first entity recognition model to identify entities of the credible keywords, and using a preset second entity recognition model to identify entities of the untrustworthy keywords;

[0054] A classification module, used for classifying the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data;

[0055] The prediction module is used to predict the disease category of each type of target medical data to obtain a disease category prediction result.

[0056] A third aspect of the present invention provides an electronic device, the electronic device comprising a processor, the processor being configured to implement the disease prediction method when executing a computer program stored in a memory.

[0057] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the disease prediction method when executed by a processor.

[0058] In summary, the disease prediction method, device, electronic device and storage medium described in the present invention, based on the real-time collection of medical records, first extract the target medical data from the medical records, and then identify the trusted keywords and untrusted keywords in the target medical data, so as to adopt different entity recognition models to identify the entities of the keywords, thereby improving the entity recognition accuracy of the keywords, and helping to improve the classification accuracy of the target medical data. After obtaining the entities of the keywords, the target medical data are classified based on matrix rank constraints and entities to obtain multiple categories of target medical data, so as to perform disease category prediction for each category of target medical data, obtain disease category prediction results, and improve the accuracy of disease category prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of the disease prediction method provided in Example 1 of the present invention.

[0060] Figure 2 It is a structural diagram of a disease prediction device provided in Embodiment 2 of the present invention.

[0061] Figure 3 It is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0062] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing embodiments in an optional embodiment and are not intended to limit the present invention.

[0064] The disease prediction method provided by the embodiment of the present invention is executed by an electronic device, and accordingly, the disease prediction device runs in the electronic device.

[0065] Embodiment 1

[0066] Figure 1 This is a flow chart of a disease prediction method provided in Embodiment 1 of the present invention. The disease prediction method specifically includes the following steps. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted.

[0067] S11, receiving medical records reported by medical institutions, and extracting target medical data from the medical records.

[0068] When a patient goes to a medical institution for treatment, the doctor of the medical institution writes the patient's medical record through the medical terminal, and the medical terminal reports the medical record to the regulatory agency in real time. Medical records refer to medical texts that include patient description information and medical description information. Patient description information refers to the relevant description of the patient's own information, which may include patient identification, gender and age, etc. Among them, the patient identification can be information such as the patient identification social security number, patient identification name or ID number. Medical description information refers to the relevant information about the patient identification's visit and the medical items and methods used during the visit, etc., which may include primary diagnosis information, drug information, diagnosis and treatment project information and medical service facility information.

[0069] The regulatory agency is the superior department of the medical institution and can connect with multiple medical institutions. The regulatory terminal corresponding to the regulatory agency receives the medical records reported by the medical terminal corresponding to each medical institution and stores these medical records in the local database. The regulatory terminal can obtain medical records from the local database at preset time periods, such as weekly or monthly, and extract target medical data from the obtained medical records, so as to perform epidemic prediction and monitoring based on the extracted target medical data.

[0070] In an optional implementation, extracting target medical data from the medical record includes:

[0071] Acquire a plurality of initial medical data from the medical record, wherein the initial medical data includes patient description information and medical description information;

[0072] Sampling the multiple initial medical data to obtain multiple medical samples having the same data distribution as the initial medical data;

[0073] Determining a medical description value in the plurality of medical samples such that a ratio of the number of initial medical data including the medical description value to the number of the plurality of medical samples is greater than a first preset threshold;

[0074] Acquire a patient description value corresponding to the medical description value from the plurality of medical samples, so that a ratio of the number of initial medical data including the patient description value to the number of the plurality of medical samples is greater than a second preset threshold;

[0075] Searching the plurality of initial medical data for initial medical data that includes the medical description value but does not include the patient description value;

[0076] The found initial medical data is used as the target medical data.

[0077] The monitoring terminal can identify medical data from each medical record based on OCR recognition technology as initial medical data. One medical record corresponds to one initial medical data, each initial medical data corresponds to a patient's one-time medical settlement information, and each initial medical data includes patient description information and medical description information.

[0078] By sampling multiple initial medical data, multiple medical samples are obtained, so that target medical data can be subsequently extracted based on the medical samples instead of from a large amount of initial medical data, thereby improving the efficiency of extracting the target medical data. The determined medical description value and the determined patient description value have a strong correlation, and accordingly, the initial medical data that does not satisfy the strong correlation is the risky medical data. Therefore, among the multiple initial medical data, the initial medical data that includes the medical description value but does not include the patient description value is the target medical data.

[0079] The above implementation method can identify high-risk target medical data from massive initial medical data, first determine strongly correlated patient description values ​​and medical description values ​​for multiple initial medical data themselves, and then use the determined strong correlation to judge the high-risk medical data in the initial medical data, and use it as the target medical data to predict epidemics with high accuracy.

[0080] S12, identifying credible keywords and untrustworthy keywords in the target medical data, using a preset first entity recognition model to identify entities of the credible keywords, and using a preset second entity recognition model to identify entities of the untrustworthy keywords.

[0081] The medical description information in the target medical data may include, but is not limited to, descriptions of the main symptoms in the medical records, medicines purchased by patients, etc. However, the descriptions of the main symptoms in the medical records and the medicines purchased by patients are unstructured text data, which cannot be understood by computers. Therefore, entity recognition technology must be used to extract necessary information from the text data to form structured data.

[0082] Before performing entity recognition on the target medical data, it is first necessary to identify which keywords in the target medical data are optional keywords and which are untrustworthy keywords, so as to adopt different entity recognition models for recognition, thereby improving the entity recognition accuracy of the keywords.

[0083] In the scenario of epidemic prediction and monitoring, the entities that need to be identified may include diseases, symptoms, patient age, etc. The entity corresponding to prescription identification is medicine. Various methods such as dictionary-based and model-based methods can be used for entity recognition tasks.

[0084] In an optional embodiment, the identifying the credible keywords and the uncredible keywords in the target medical data includes:

[0085] Performing word segmentation processing on the target medical data to obtain multiple keywords;

[0086] Traversing the multiple keywords, and obtaining a keyword from the multiple keywords as a target keyword in each traversal;

[0087] Obtain the first character and the last character of the target keyword;

[0088] Acquire a first neighbor character located before the first character in the target medical data, and acquire a second neighbor character located after the last character in the target medical data;

[0089] Determine whether the first character and the first neighbor character can form a word, and determine whether the last character and the second neighbor character can form a word;

[0090] When the first character and the first neighboring characters cannot form a word, and the last character and the second neighboring characters cannot form a word, determining that the target keyword is a credible keyword;

[0091] When the first character and the first neighboring character can form a word, and / or the last character and the second neighboring character can form a word, the target keyword is determined to be an untrustworthy keyword.

[0092] The supervision terminal can use the Jieba word segmentation tool to segment each target medical data to obtain multiple keywords. The multiple keywords of each target medical data are traversed sequentially, that is, the first keyword is first obtained as the target keyword, then the second keyword is obtained as the target keyword, and then the third keyword is obtained as the target keyword, and so on, until the last keyword is obtained as the target keyword.

[0093] For example, assuming that a target medical data is M1M2M3M4M5M6M7, Mi is a character, and i belongs to [1, 7]. M3M4M5 is the target keyword to be obtained, then the first character in the target keyword is M3, the last character is M5, the first neighbor character before the first character in the target medical data is M2, and the second neighbor character after the last character M5 is M6.

[0094] When the first character M3 and the first neighbor character M2 cannot form a word, and the last character M5 and the second neighbor character M6 cannot form a word, it means that the characters M2M3M4M5M6 cannot form a keyword, so the target keyword M3M4M5 is a credible keyword.

[0095] When the first character M3 and the first neighbor character M2 can form a word, but the last character M5 and the second neighbor character M6 cannot form a word, it means that the characters M2M3M4M5 can form a keyword, so the target keyword M3M4M5 is an untrustworthy keyword.

[0096] When the first character M3 and the first neighbor character M2 cannot form a word, but the last character M5 and the second neighbor character M6 can form a word, it means that the characters M3M4M5M6 can form a keyword, so the target keyword M3M4M5 is an untrustworthy keyword.

[0097] When the first character M3 and the first neighbor character M2 can form a word, and the last character M5 and the second neighbor character M6 can form a word, it means that the characters M2M3M4M5M6 can form a keyword, so the target keyword M3M4M5 is an untrustworthy keyword.

[0098] Trusted keywords and untrusted keywords are relative. Since the Jieba word segmentation tool performs word segmentation on the data, the keywords obtained are all regular words, which are independent minimum units. Therefore, in order to avoid inaccurate keywords obtained by word segmentation, this optional implementation method can be used to divide the keywords into regular trusted keywords and unconventional untrusted keywords, so as to improve the accuracy of entity recognition when performing entity recognition on the keywords in the subsequent process.

[0099] In an optional implementation, a dictionary tree may be constructed based on a dictionary data set as a preset first entity recognition model, and then a dictionary longest match algorithm is used to perform entity matching on the dictionary tree and the multiple keywords to obtain entities of credible keywords.

[0100] In an optional implementation, an entity recognition model can be trained based on a deep neural network as a preset second entity recognition model, and then the preset second entity recognition model is used to identify entities of untrusted keywords. The training process of the entity recognition model is not described in detail.

[0101] S13, classifying the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data.

[0102] Classification is to classify multiple entities identified from the target medical data, thereby achieving classification of the target medical data.

[0103] The monitoring terminal can map multiple entities of each target medical data to a high-dimensional space, thereby taking multiple entities of each target medical data as a coordinate point in the high-dimensional space, and then calculating the distance between any two target medical data based on the coordinates, and then determining the similarity between any two target medical data based on the distance. The smaller the distance, the higher the similarity between the corresponding two target medical data. The larger the distance, the lower the similarity between the corresponding two target medical data.

[0104] The similarity of the target medical data is determined through spatial mapping and distance calculation, thereby achieving classification of the target medical data. However, this method does not take into account the coefficient density of the target medical data in the high-dimensional space. Therefore, the similarity matrix constructed based on similarity is not accurate. In order to improve the classification accuracy of the target medical data, classification can be performed based on matrix rank constraint theory.

[0105] In an optional implementation, the target medical data is classified based on the matrix rank constraint and the entity to obtain multiple categories of target medical data including:

[0106] splicing multiple entities corresponding to the target medical data into an entity vector;

[0107] Performing alignment processing on the entity vector to obtain an aligned entity vector;

[0108] Selecting a plurality of target medical data pairs from the target medical data;

[0109] Acquire the shared neighbor target medical data of each target medical data pair, and generate a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data;

[0110] Calculate a Laplace matrix based on the similarity matrix;

[0111] Finding the undirected graph connected branch structure of the Laplace matrix based on the Laplace matrix rank constraint theory;

[0112] Output multiple categories of target medical data according to the undirected graph connected branch structure.

[0113] Each target medical data corresponds to multiple entities, and the multiple entities are used as elements of the entity vector of the target medical data. However, the number of multiple entities corresponding to different target medical data varies. In order to facilitate the subsequent calculation of the similarity between any two target medical data, the entity vectors of the target medical data need to be aligned. Alignment processing refers to unifying the number of elements of the entity vectors of all target medical data. If the number of elements is insufficient, preset element values, such as 0, can be used for filling.

[0114] The similarity between two aligned entity vectors is calculated to obtain the similarity between two target medical data corresponding to the two aligned entity vectors, thereby generating a similarity matrix according to the similarity between any two target medical data, and the generated similarity matrix is ​​a symmetric matrix.

[0115] In order to reduce the amount of calculation and improve the classification efficiency, after the similarity matrix is ​​generated, the Laplace matrix is ​​also calculated according to the similarity matrix, so as to classify and output multiple classes based on the Laplace matrix. In the specific implementation, the eigenvalues ​​of the similarity matrix and the eigenvectors corresponding to each eigenvalue are calculated, the eigenvalues ​​are sorted, and the first L smallest eigenvalues ​​are obtained, and the matrix composed of the eigenvectors corresponding to the first L smallest eigenvalues ​​is used as the Laplace matrix.

[0116] Laplace matrix rank constraint theory: The number of connected branches of a similarity matrix is ​​equal to the number of eigenvalues ​​of its corresponding Laplace matrix L that are 0. If the condition rank(L) = nc is satisfied (n is the number of target medical data), that is, the sum of the first c smallest eigenvalues ​​of L is equal to 0, then the final c classes can be directly obtained through the Laplace matrix.

[0117] You can use the graphconncomp function in the Matlab toolbox to find the undirected graph connected branch structure of the Laplacian matrix.

[0118] This optional implementation manner can improve the accuracy of similarity matrix construction by selecting multiple target medical data pairs from the target medical data, obtaining shared neighbor target medical data of each target medical data pair, and generating a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data; calculating the Laplace matrix through the similarity matrix, and finding the undirected graph connected branch structure of the Laplace matrix based on the Laplace matrix rank constraint theory, thereby reducing the computational time spent on finding the undirected graph connected branch structure, simplifying the solution difficulty, and reducing the time complexity.

[0119] In an optional embodiment, the acquiring of the shared neighbor target medical data of each target medical data pair and generating a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data comprises:

[0120] Acquire the first target medical data and the second target medical data in each of the target medical data pairs;

[0121] Finding K nearest neighbor target medical data of the first target medical data and K nearest neighbor target medical data of the second target medical data;

[0122] Acquire shared nearest neighbor target medical data from two of the K nearest neighbor target medical data;

[0123] Calculate the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data;

[0124] A similarity matrix is ​​generated according to the similarities of the plurality of target medical data pairs.

[0125] Assuming that the number of target medical data is N, the number of target medical data pairs obtained from the target medical data is N*(N-1) / 2.

[0126] Each target medical data captures the information of several target medical data around it. Assuming that target medical data A and target medical data B each have 8 nearest neighbor target medical data, of which 4 are shared by target medical data A and target medical data B, then these 4 target medical data are called shared neighbor target medical data of target medical data A and target medical data B.

[0127] In an optional embodiment, the calculating the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data includes:

[0128] An average aligned entity vector is calculated based on the aligned entity vectors corresponding to the shared neighbor target medical data;

[0129] Calculate a first distance according to the aligned entity vector of the first target medical data and the average aligned entity vector;

[0130] Calculate a second distance according to the aligned entity vector of the second target medical data and the average aligned entity vector;

[0131] Calculating a third distance according to the aligned entity vector of the first target medical data and the aligned entity vector of the second target medical data;

[0132] The similarity is calculated based on the first distance, the second distance and the third distance using a preset similarity calculation model.

[0133] The preset similarity calculation model can be as follows: S = exp(-S3*(S1S2(S3 1 / 2 )+1)) 1 / 2 , S1 is the first distance, S2 is the second distance, and S3 is the third distance.

[0134] K generally does not exceed 20.

[0135] By calculating the first distance and the second distance, the sparseness or density of the target medical data distribution in the field where the two target medical data are located can be captured, and the situation where the two target medical data are located in different density clusters can be better dealt with, thereby making it easier to obtain correct classification results.

[0136] If the distance between two target medical data is close, the calculated third distance is small, and the similarity between the two target medical data is large. If two target medical data are in the same cluster, but one of the two target medical data and the third target medical data are in different clusters, the greater the number of shared neighbor target medical data, the greater the corresponding similarity, so that the two target medical data in the same cluster have a higher similarity.

[0137] S14, predicting the disease category for each type of the target medical data to obtain a disease category prediction result.

[0138] After the target medical data is classified into categories, the distribution of the target medical data can be determined according to the categories.

[0139] In an optional implementation, predicting the disease category for each type of target medical data includes:

[0140] Inputting each type of target medical data into a first disease prediction framework in a preset disease probability prediction model to obtain a first predicted disease;

[0141] Inputting the first predicted disease into the selective connection layer in the disease probability prediction model, and obtaining a second disease prediction architecture selected and specified by the selective connection layer;

[0142] Inputting the first predicted disease and the designated medical data into the designated second disease prediction framework to obtain a second predicted disease;

[0143] Mapping the designated medical data, the first predicted disease, and the second predicted disease into a disease classification vector;

[0144] The disease category of each type of target medical data is determined according to the preset standard classification vector and the disease classification vector.

[0145] The preset disease probability prediction model is formed by sequentially connecting a first disease prediction architecture, a selective connection layer, and a plurality of second disease prediction architectures. The first disease prediction architecture may be a long short-term memory architecture, and the designated second disease prediction architecture is trained based on a neural network model.

[0146] The standard classification vector is labeled with the first predicted disease and the second predicted disease, and the standard classification vector is labeled as a specified category.

[0147] The designated medical data refers to the historical target medical data under the category corresponding to the target medical data.

[0148] The distance value between the standard classification vector and the disease classification vector is calculated. The larger the distance value, the less similar the standard classification vector is to the disease classification vector, and the smaller the distance value, the more similar the standard classification vector is to the disease classification vector. The designated category corresponding to the annotated classification vector with the smallest distance is determined as the disease category of the target medical data.

[0149] In an optional implementation, after obtaining each type of target medical data and the corresponding disease category prediction results, the monitoring terminal can also report all types of target medical data and the corresponding disease category prediction results to the epidemiological center for disease control and prevention, and can also report the highest disease category prediction result and its corresponding target medical data to the epidemiological center for disease control and prevention. The staff of the epidemiological center for disease control and prevention analyzes the reported target medical data and the corresponding disease category prediction results. If the level category prediction result belongs to an epidemic disease, an alarm instruction can be sent to the monitoring terminal, so that the monitoring terminal sends an alarm instruction to each medical institution, and then each medical institution supervises patients with epidemic diseases, such as centralized treatment.

[0150] The disease prediction method of the present invention, based on the real-time collection of medical records, first extracts the target medical data from the medical records, then identifies the trusted keywords and untrusted keywords in the target medical data, thereby using different entity recognition models to identify the entities of the keywords, improving the accuracy of the entity recognition of the keywords, and helping to improve the classification accuracy of the target medical data. After obtaining the entities of the keywords, the target medical data is classified based on the matrix rank constraint and the entity to obtain multiple types of target medical data, thereby predicting the disease category for each type of target medical data, obtaining the disease category prediction result, and improving the accuracy of the disease category prediction. Real-time prediction of the outbreak trend of the epidemic helps the CDC to grasp the possible risk of the outbreak of the epidemic at the first time, take corresponding preventive measures in advance, and indirectly improve the effect and reliability of epidemic monitoring.

[0151] Embodiment 2

[0152] Figure 2 It is a structural diagram of a disease prediction device provided in Embodiment 2 of the present invention.

[0153] In some embodiments, the disease prediction device 20 may include multiple functional modules composed of computer program segments. The computer programs of the various program segments in the disease prediction device 20 may be stored in a memory of an electronic device and executed by at least one processor to perform (see Figure 1 Description) Function of disease prediction.

[0154] In this embodiment, the disease prediction device 20 can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a receiving module 201, an identification module 202, a classification module 203 and a prediction module 204. The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0155] The receiving module 201 is used to receive medical records reported by medical institutions and extract target medical data from the medical records.

[0156] When a patient goes to a medical institution for treatment, the doctor of the medical institution writes the patient's medical record through the medical terminal, and the medical terminal reports the medical record to the regulatory agency in real time. Medical records refer to medical texts that include patient description information and medical description information. Patient description information refers to the relevant description of the patient's own information, which may include patient identification, gender and age, etc. Among them, the patient identification can be information such as the patient identification social security number, patient identification name or ID number. Medical description information refers to the relevant information about the patient identification's visit and the medical items and methods used during the visit, etc., which may include primary diagnosis information, drug information, diagnosis and treatment project information and medical service facility information.

[0157] The regulatory agency is the superior department of the medical institution and can connect with multiple medical institutions. The regulatory terminal corresponding to the regulatory agency receives the medical records reported by the medical terminal corresponding to each medical institution and stores these medical records in the local database. The regulatory terminal can obtain medical records from the local database at preset time periods, such as weekly or monthly, and extract target medical data from the obtained medical records, so as to perform epidemic prediction and monitoring based on the extracted target medical data.

[0158] In an optional implementation, the receiving module 201 extracts the target medical data from the medical record including:

[0159] Acquire a plurality of initial medical data from the medical record, wherein the initial medical data includes patient description information and medical description information;

[0160] Sampling the multiple initial medical data to obtain multiple medical samples having the same data distribution as the initial medical data;

[0161] Determining a medical description value in the plurality of medical samples such that a ratio of the number of initial medical data including the medical description value to the number of the plurality of medical samples is greater than a first preset threshold;

[0162] Acquire a patient description value corresponding to the medical description value from the plurality of medical samples, so that a ratio of the number of initial medical data including the patient description value to the number of the plurality of medical samples is greater than a second preset threshold;

[0163] Searching the plurality of initial medical data for initial medical data that includes the medical description value but does not include the patient description value;

[0164] The found initial medical data is used as the target medical data.

[0165] The monitoring terminal can identify medical data from each medical record based on OCR recognition technology as initial medical data. One medical record corresponds to one initial medical data, each initial medical data corresponds to a patient's one-time medical settlement information, and each initial medical data includes patient description information and medical description information.

[0166] By sampling multiple initial medical data, multiple medical samples are obtained, so that target medical data can be subsequently extracted based on the medical samples instead of from a large amount of initial medical data, thereby improving the efficiency of extracting the target medical data. The determined medical description value and the determined patient description value have a strong correlation, and accordingly, the initial medical data that does not satisfy the strong correlation is the risky medical data. Therefore, among the multiple initial medical data, the initial medical data that includes the medical description value but does not include the patient description value is the target medical data.

[0167] The above implementation method can identify high-risk target medical data from massive initial medical data, first determine strongly correlated patient description values ​​and medical description values ​​for multiple initial medical data themselves, and then use the determined strong correlation to judge the high-risk medical data in the initial medical data, and use it as the target medical data to predict epidemics with high accuracy.

[0168] The identification module 202 is used to identify credible keywords and untrustworthy keywords in the target medical data, using a preset first entity recognition model to identify entities of the credible keywords, and using a preset second entity recognition model to identify entities of the untrustworthy keywords.

[0169] The medical description information in the target medical data may include, but is not limited to, descriptions of the main symptoms in the medical records, medicines purchased by patients, etc. However, the descriptions of the main symptoms in the medical records and the medicines purchased by patients are unstructured text data, which cannot be understood by computers. Therefore, entity recognition technology must be used to extract necessary information from the text data to form structured data.

[0170] Before performing entity recognition on the target medical data, it is first necessary to identify which keywords in the target medical data are optional keywords and which are untrustworthy keywords, so as to adopt different entity recognition models for recognition, thereby improving the entity recognition accuracy of the keywords.

[0171] In the scenario of epidemic prediction and monitoring, the entities that need to be identified may include diseases, symptoms, patient age, etc. The entity corresponding to prescription identification is medicine. Various methods such as dictionary-based and model-based methods can be used for entity recognition tasks.

[0172] In an optional embodiment, the identification module 202 identifies the trusted keywords and untrusted keywords in the target medical data including:

[0173] Performing word segmentation processing on the target medical data to obtain multiple keywords;

[0174] Traversing the multiple keywords, and obtaining a keyword from the multiple keywords as a target keyword in each traversal;

[0175] Obtain the first character and the last character of the target keyword;

[0176] Acquire a first neighbor character located before the first character in the target medical data, and acquire a second neighbor character located after the last character in the target medical data;

[0177] Determine whether the first character and the first neighbor character can form a word, and determine whether the last character and the second neighbor character can form a word;

[0178] When the first character and the first neighboring characters cannot form a word, and the last character and the second neighboring characters cannot form a word, determining that the target keyword is a credible keyword;

[0179] When the first character and the first neighboring character can form a word, and / or the last character and the second neighboring character can form a word, the target keyword is determined to be an untrustworthy keyword.

[0180] The supervision terminal can use the Jieba word segmentation tool to segment each target medical data to obtain multiple keywords. The multiple keywords of each target medical data are traversed sequentially, that is, the first keyword is first obtained as the target keyword, then the second keyword is obtained as the target keyword, and then the third keyword is obtained as the target keyword, and so on, until the last keyword is obtained as the target keyword.

[0181] For example, assuming that a target medical data is M1M2M3M4M5M6M7, Mi is a character, and i belongs to [1, 7]. M3M4M5 is the target keyword to be obtained, then the first character in the target keyword is M3, the last character is M5, the first neighbor character before the first character in the target medical data is M2, and the second neighbor character after the last character M5 is M6.

[0182] When the first character M3 and the first neighbor character M2 cannot form a word, and the last character M5 and the second neighbor character M6 cannot form a word, it means that the characters M2M3M4M5M6 cannot form a keyword, so the target keyword M3M4M5 is a credible keyword.

[0183] When the first character M3 and the first neighbor character M2 can form a word, but the last character M5 and the second neighbor character M6 cannot form a word, it means that the characters M2M3M4M5 can form a keyword, so the target keyword M3M4M5 is an untrustworthy keyword.

[0184] When the first character M3 and the first neighbor character M2 cannot form a word, but the last character M5 and the second neighbor character M6 can form a word, it means that the characters M3M4M5M6 can form a keyword, so the target keyword M3M4M5 is an untrustworthy keyword.

[0185] When the first character M3 and the first neighbor character M2 can form a word, and the last character M5 and the second neighbor character M6 can form a word, it means that the characters M2M3M4M5M6 can form a keyword, so the target keyword M3M4M5 is an untrustworthy keyword.

[0186] Trusted keywords and untrusted keywords are relative. Since the Jieba word segmentation tool performs word segmentation on the data, the keywords obtained are all regular words, which are independent minimum units. Therefore, in order to avoid inaccurate keywords obtained by word segmentation, this optional implementation method can be used to divide the keywords into regular trusted keywords and unconventional untrusted keywords, so as to improve the accuracy of entity recognition when performing entity recognition on the keywords in the subsequent process.

[0187] In an optional implementation, a dictionary tree may be constructed based on a dictionary data set as a preset first entity recognition model, and then a dictionary longest match algorithm is used to perform entity matching on the dictionary tree and the multiple keywords to obtain entities of credible keywords.

[0188] In an optional implementation, an entity recognition model can be trained based on a deep neural network as a preset second entity recognition model, and then the preset second entity recognition model is used to identify entities of untrusted keywords. The training process of the entity recognition model is not described in detail.

[0189] The classification module 203 is used to classify the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data.

[0190] Classification is to classify multiple entities identified from the target medical data, thereby achieving classification of the target medical data.

[0191] The monitoring terminal can map multiple entities of each target medical data to a high-dimensional space, thereby taking multiple entities of each target medical data as a coordinate point in the high-dimensional space, and then calculating the distance between any two target medical data based on the coordinates, and then determining the similarity between any two target medical data based on the distance. The smaller the distance, the higher the similarity between the corresponding two target medical data. The larger the distance, the lower the similarity between the corresponding two target medical data.

[0192] The similarity of the target medical data is determined through spatial mapping and distance calculation, thereby achieving classification of the target medical data. However, this method does not take into account the coefficient density of the target medical data in the high-dimensional space. Therefore, the similarity matrix constructed based on similarity is not accurate. In order to improve the classification accuracy of the target medical data, classification can be performed based on matrix rank constraint theory.

[0193] In an optional implementation, the classification module 203 classifies the target medical data based on the matrix rank constraint and the entity, and obtains multiple categories of target medical data including:

[0194] splicing multiple entities corresponding to the target medical data into an entity vector;

[0195] Performing alignment processing on the entity vector to obtain an aligned entity vector;

[0196] Selecting a plurality of target medical data pairs from the target medical data;

[0197] Acquire the shared neighbor target medical data of each target medical data pair, and generate a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data;

[0198] Calculate a Laplace matrix based on the similarity matrix;

[0199] Finding the undirected graph connected branch structure of the Laplace matrix based on the Laplace matrix rank constraint theory;

[0200] Output multiple categories of target medical data according to the undirected graph connected branch structure.

[0201] Each target medical data corresponds to multiple entities, and the multiple entities are used as elements of the entity vector of the target medical data. However, the number of multiple entities corresponding to different target medical data varies. In order to facilitate the subsequent calculation of the similarity between any two target medical data, the entity vectors of the target medical data need to be aligned. Alignment processing refers to unifying the number of elements of the entity vectors of all target medical data. If the number of elements is insufficient, preset element values, such as 0, can be used for filling.

[0202] The similarity between two aligned entity vectors is calculated to obtain the similarity between two target medical data corresponding to the two aligned entity vectors, thereby generating a similarity matrix according to the similarity between any two target medical data, and the generated similarity matrix is ​​a symmetric matrix.

[0203] In order to reduce the amount of calculation and improve the classification efficiency, after the similarity matrix is ​​generated, the Laplace matrix is ​​also calculated according to the similarity matrix, so as to classify and output multiple classes based on the Laplace matrix. In the specific implementation, the eigenvalues ​​of the similarity matrix and the eigenvectors corresponding to each eigenvalue are calculated, the eigenvalues ​​are sorted, and the first L smallest eigenvalues ​​are obtained, and the matrix composed of the eigenvectors corresponding to the first L smallest eigenvalues ​​is used as the Laplace matrix.

[0204] Laplace matrix rank constraint theory: The number of connected branches of a similarity matrix is ​​equal to the number of eigenvalues ​​of its corresponding Laplace matrix L that are 0. If the condition rank(L) = nc is satisfied (n is the number of target medical data), that is, the sum of the first c smallest eigenvalues ​​of L is equal to 0, then the final c classes can be directly obtained through the Laplace matrix.

[0205] You can use the graphconncomp function in the Matlab toolbox to find the undirected graph connected branch structure of the Laplacian matrix.

[0206] This optional implementation manner can improve the accuracy of similarity matrix construction by selecting multiple target medical data pairs from the target medical data, obtaining shared neighbor target medical data of each target medical data pair, and generating a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data; calculating the Laplace matrix through the similarity matrix, and finding the undirected graph connected branch structure of the Laplace matrix based on the Laplace matrix rank constraint theory, thereby reducing the computational time spent on finding the undirected graph connected branch structure, simplifying the solution difficulty, and reducing the time complexity.

[0207] In an optional embodiment, the acquiring of the shared neighbor target medical data of each target medical data pair and generating a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data comprises:

[0208] Acquire the first target medical data and the second target medical data in each of the target medical data pairs;

[0209] Finding K nearest neighbor target medical data of the first target medical data and K nearest neighbor target medical data of the second target medical data;

[0210] Acquire shared nearest neighbor target medical data from two of the K nearest neighbor target medical data;

[0211] Calculate the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data;

[0212] A similarity matrix is ​​generated according to the similarities of the plurality of target medical data pairs.

[0213] Assuming that the number of target medical data is N, the number of target medical data pairs obtained from the target medical data is N*(N-1) / 2.

[0214] Each target medical data captures the information of several target medical data around it. Assuming that target medical data A and target medical data B each have 8 nearest neighbor target medical data, of which 4 are shared by target medical data A and target medical data B, then these 4 target medical data are called shared neighbor target medical data of target medical data A and target medical data B.

[0215] In an optional embodiment, the calculating the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data includes:

[0216] An average aligned entity vector is calculated based on the aligned entity vectors corresponding to the shared neighbor target medical data;

[0217] Calculate a first distance according to the aligned entity vector of the first target medical data and the average aligned entity vector;

[0218] Calculate a second distance according to the aligned entity vector of the second target medical data and the average aligned entity vector;

[0219] Calculating a third distance according to the aligned entity vector of the first target medical data and the aligned entity vector of the second target medical data;

[0220] The similarity is calculated based on the first distance, the second distance and the third distance using a preset similarity calculation model.

[0221] The preset similarity calculation model can be as follows: S = exp(-S3*(S1S2(S3 1 / 2 )+1)) 1 / 2 , S1 is the first distance, S2 is the second distance, and S3 is the third distance.

[0222] K generally does not exceed 20.

[0223] By calculating the first distance and the second distance, the sparseness or density of the target medical data distribution in the field where the two target medical data are located can be captured, and the situation where the two target medical data are located in different density clusters can be better dealt with, thereby making it easier to obtain correct classification results.

[0224] If the distance between two target medical data is close, the calculated third distance is small, and the similarity between the two target medical data is large. If two target medical data are in the same cluster, but one of the two target medical data and the third target medical data are in different clusters, the greater the number of shared neighbor target medical data, the greater the corresponding similarity, so that the two target medical data in the same cluster have a higher similarity.

[0225] The prediction module 204 is used to predict the disease category of each type of the target medical data to obtain a disease category prediction result.

[0226] After the target medical data is classified into categories, the distribution of the target medical data can be determined according to the categories.

[0227] In an optional implementation, the prediction module 204 predicts the disease category for each type of the target medical data, including:

[0228] Inputting each type of target medical data into a first disease prediction framework in a preset disease probability prediction model to obtain a first predicted disease;

[0229] Inputting the first predicted disease into the selective connection layer in the disease probability prediction model, and obtaining a second disease prediction architecture selected and specified by the selective connection layer;

[0230] Inputting the first predicted disease and the designated medical data into the designated second disease prediction framework to obtain a second predicted disease;

[0231] Mapping the designated medical data, the first predicted disease, and the second predicted disease into a disease classification vector;

[0232] The disease category of each type of target medical data is determined according to the preset standard classification vector and the disease classification vector.

[0233] The preset disease probability prediction model is formed by sequentially connecting a first disease prediction architecture, a selective connection layer, and a plurality of second disease prediction architectures. The first disease prediction architecture may be a long short-term memory architecture, and the designated second disease prediction architecture is trained based on a neural network model.

[0234] The standard classification vector is labeled with the first predicted disease and the second predicted disease, and the standard classification vector is labeled as a specified category.

[0235] The designated medical data refers to the historical target medical data under the category corresponding to the target medical data.

[0236] The distance value between the standard classification vector and the disease classification vector is calculated. The larger the distance value, the less similar the standard classification vector is to the disease classification vector, and the smaller the distance value, the more similar the standard classification vector is to the disease classification vector. The designated category corresponding to the annotated classification vector with the smallest distance is determined as the disease category of the target medical data.

[0237] In an optional implementation, after obtaining each type of target medical data and the corresponding disease category prediction results, the monitoring terminal can also report all types of target medical data and the corresponding disease category prediction results to the epidemiological center for disease control and prevention, and can also report the highest disease category prediction result and its corresponding target medical data to the epidemiological center for disease control and prevention. The staff of the epidemiological center for disease control and prevention analyzes the reported target medical data and the corresponding disease category prediction results. If the level category prediction result belongs to an epidemic disease, an alarm instruction can be sent to the monitoring terminal, so that the monitoring terminal sends an alarm instruction to each medical institution, and then each medical institution supervises patients with epidemic diseases, such as centralized treatment.

[0238] The disease prediction device of the present invention, based on the real-time collection of medical records, first extracts the target medical data from the medical records, and then identifies the trusted keywords and untrusted keywords in the target medical data, thereby using different entity recognition models to identify the entities of the keywords, improving the accuracy of the entity recognition of the keywords, and helping to improve the classification accuracy of the target medical data. After obtaining the entities of the keywords, the target medical data is classified based on the matrix rank constraint and the entity to obtain multiple categories of target medical data, thereby predicting the disease category for each category of target medical data, obtaining the disease category prediction result, and improving the accuracy of the disease category prediction. Real-time prediction of the outbreak trend of epidemics helps the CDC to grasp the possible risk of epidemic outbreaks in the first place, take corresponding preventive measures in advance, and indirectly improve the effect and reliability of epidemic monitoring.

[0239] Embodiment 3

[0240] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above disease prediction method embodiment are implemented, such as Figure 1 S11-S14 shown:

[0241] S11, receiving medical records reported by medical institutions, and extracting target medical data from the medical records;

[0242] S12, identifying credible keywords and untrustworthy keywords in the target medical data, using a preset first entity recognition model to identify entities of the credible keywords, and using a preset second entity recognition model to identify entities of the untrustworthy keywords;

[0243] S13, classifying the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data;

[0244] S14, predicting the disease category for each type of the target medical data to obtain a disease category prediction result.

[0245] Alternatively, when the computer program is executed by a processor, the functions of each module / unit in the above-mentioned device embodiment are realized, for example Figure 2 Modules 201-204:

[0246] The receiving module 201 is used to receive medical records reported by medical institutions and extract target medical data from the medical records;

[0247] The determination module 202 is used to identify the trusted keywords and untrusted keywords in the target medical data, use a preset first entity recognition model to identify the entities of the trusted keywords, and use a preset second entity recognition model to identify the entities of the untrusted keywords;

[0248] The output module 203 is used to classify the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data;

[0249] The prediction module 204 is used to predict the disease category of each type of the target medical data to obtain a disease category prediction result.

[0250] Embodiment 4

[0251] See also Figure 3 FIG. 2 is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. In a preferred embodiment of the present invention, the electronic device 3 includes a memory 31 , at least one processor 32 , at least one communication bus 33 and a transceiver 34 .

[0252] Those skilled in the art should understand that Figure 3 The structure of the electronic device shown does not constitute a limitation of the embodiments of the present invention, and may be a bus structure or a star structure. The electronic device 3 may also include more or less other hardware or software than shown in the figure, or a different component arrangement.

[0253] In some embodiments, the electronic device 3 is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors, and embedded devices. The electronic device 3 may also include client devices, which include but are not limited to any electronic product that can interact with a client through a keyboard, mouse, remote control, touchpad, or voice-controlled device, such as a personal computer, tablet computer, smart phone, digital camera, etc.

[0254] It should be noted that the electronic device 3 is only an example, and other existing or future electronic products that are suitable for the present invention should also be included in the protection scope of the present invention and included here by reference.

[0255] In some embodiments, the memory 31 stores a computer program, and when the computer program is executed by the at least one processor 32, all or part of the steps in the disease prediction method as described above are implemented. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0256] Furthermore, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0257] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0258] In some embodiments, the at least one processor 32 is the control core (ControlUnit) of the electronic device 3, and uses various interfaces and lines to connect various components of the entire electronic device 3, and executes various functions and processes data of the electronic device 3 by running or executing programs or modules stored in the memory 31, and calling data stored in the memory 31. For example, when the at least one processor 32 executes the computer program stored in the memory, it implements all or part of the steps of the disease prediction method described in the embodiment of the present invention; or implements all or part of the functions of the disease prediction device. The at least one processor 32 can be composed of an integrated circuit, for example, it can be composed of a single packaged integrated circuit, or it can be composed of multiple integrated circuits with the same function or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0259] In some embodiments, the at least one communication bus 33 is configured to implement connection and communication between the memory 31 and the at least one processor 32, etc.

[0260] Although not shown, the electronic device 3 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 32 through a power management device, so that the power management device can manage charging, discharging, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 3 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0261] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, electronic device, or network device, etc.) or a processor to execute a part of the method described in each embodiment of the present invention.

[0262] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0263] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0264] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0265] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "including" does not exclude other units or, and the singular does not exclude the plural. Multiple units or devices stated in the specification may also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0266] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A disease prediction method, characterized in that: The method comprises: Receiving medical records reported by medical institutions, and extracting target medical data from the medical records; Identify credible keywords and untrustworthy keywords in the target medical data, use a preset first entity recognition model to identify entities of the credible keywords, and use a preset second entity recognition model to identify entities of the untrustworthy keywords; Classifying the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data; Perform disease category prediction on each type of target medical data to obtain disease category prediction results, including: inputting each type of target medical data into a first disease prediction architecture in a preset disease probability prediction model to obtain a first predicted disease, inputting the first predicted disease into a selective connection layer in the disease probability prediction model, obtaining a second disease prediction architecture selected by the selective connection layer, inputting the first predicted disease and designated medical data into the designated second disease prediction architecture to obtain a second predicted disease, mapping the designated medical data, the first predicted disease and the second predicted disease into a disease classification vector, and determining the disease category of each type of target medical data according to a preset standard classification vector and the disease classification vector.

2. The disease prediction method according to claim 1, characterized in that: The identifying of credible keywords and untrustworthy keywords in the target medical data includes: Performing word segmentation processing on the target medical data to obtain multiple keywords; Traversing the multiple keywords, and obtaining a keyword from the multiple keywords as a target keyword in each traversal; Obtain the first character and the last character of the target keyword; Acquire a first neighbor character located before the first character in the target medical data, and acquire a second neighbor character located after the last character in the target medical data; Determine whether the first character and the first neighbor character can form a word, and determine whether the last character and the second neighbor character can form a word; When the first character and the first neighboring characters cannot form a word, and the last character and the second neighboring characters cannot form a word, determining that the target keyword is a credible keyword; When the first character and the first neighboring character can form a word, and / or the last character and the second neighboring character can form a word, the target keyword is determined to be an untrustworthy keyword.

3. The disease prediction method according to claim 1, characterized in that: The target medical data is classified based on the matrix rank constraint and the entity to obtain multiple types of target medical data including: splicing multiple entities corresponding to the target medical data into an entity vector; Performing alignment processing on the entity vector to obtain an aligned entity vector; Selecting a plurality of target medical data pairs from the target medical data; Acquire the shared neighbor target medical data of each target medical data pair, and generate a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data; Calculate a Laplace matrix based on the similarity matrix; Finding the undirected graph connected branch structure of the Laplace matrix based on the Laplace matrix rank constraint theory; Output multiple categories of target medical data according to the undirected graph connected branch structure.

4. The disease prediction method according to claim 3, characterized in that: The acquiring of the shared neighbor target medical data of each target medical data pair and generating a similarity matrix according to the aligned entity vectors corresponding to the shared neighbor target medical data comprises: Acquire the first target medical data and the second target medical data in each of the target medical data pairs; Finding K nearest neighbor target medical data of the first target medical data and K nearest neighbor target medical data of the second target medical data; Acquire shared nearest neighbor target medical data from two of the K nearest neighbor target medical data; Calculate the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data; A similarity matrix is ​​generated according to the similarities of the plurality of target medical data pairs.

5. The disease prediction method according to claim 4, characterized in that: The calculating the similarity of the corresponding target medical data pair according to the aligned entity vector of the first target medical data, the aligned entity vector of the second target medical data and the aligned entity vector corresponding to the shared neighbor target medical data comprises: An average aligned entity vector is calculated based on the aligned entity vectors corresponding to the shared neighbor target medical data; Calculate a first distance according to the aligned entity vector of the first target medical data and the average aligned entity vector; Calculate a second distance according to the aligned entity vector of the second target medical data and the average aligned entity vector; Calculating a third distance according to the aligned entity vector of the first target medical data and the aligned entity vector of the second target medical data; The similarity is calculated based on the first distance, the second distance and the third distance using a preset similarity calculation model.

6. The disease prediction method according to any one of claims 1 to 4, characterized in that: The extracting target medical data from the medical record comprises: Acquire a plurality of initial medical data from the medical record, wherein the initial medical data includes patient description information and medical description information; Sampling the multiple initial medical data to obtain multiple medical samples having the same data distribution as the initial medical data; Determining a medical description value in the plurality of medical samples such that a ratio of the number of initial medical data including the medical description value to the number of the plurality of medical samples is greater than a first preset threshold; Acquire a patient description value corresponding to the medical description value from the plurality of medical samples, so that a ratio of the number of initial medical data including the patient description value to the number of the plurality of medical samples is greater than a second preset threshold; Searching the plurality of initial medical data for initial medical data that includes the medical description value but does not include the patient description value; The found initial medical data is used as the target medical data.

7. A disease prediction device, the device being used to implement the disease prediction method according to any one of claims 1 to 6, characterized in that: The device comprises: A receiving module, used for receiving medical records reported by medical institutions and extracting target medical data from the medical records; An identification module, used to identify credible keywords and untrustworthy keywords in the target medical data, using a preset first entity recognition model to identify entities of the credible keywords, and using a preset second entity recognition model to identify entities of the untrustworthy keywords; A classification module, used for classifying the target medical data based on the matrix rank constraint and the entity to obtain multiple categories of target medical data; The prediction module is used to predict the disease category of each type of target medical data to obtain a disease category prediction result.

8. An electronic device, characterized in that: The electronic device comprises a processor and a memory, and the processor is used to implement the disease prediction method according to any one of claims 1 to 6 when executing a computer program stored in the memory.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the disease prediction method according to any one of claims 1 to 6 is implemented.

Images