A method, device, electronic device and storage medium for determining similar patients

Abstract

The present application provides a method, device, electronic device and storage medium for determining similar patients, wherein the determination method includes: inputting the feature vector of the target patient into a Gaussian mixture model to obtain the primary patient category to which the target patient belongs; searching for historical patients belonging to the same primary patient category as the target patient to construct a primary patient set; clustering multiple patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set, determining the secondary patient category to which each patient in the primary patient set belongs, and constructing a secondary patient set; calculating the distance between the target patient and any historical patient in the secondary patient set except the target patient, and selecting historical patients similar to the target patient from the secondary patient set based on the distance. The present application selects patients similar to the target patient through two clustering processes, thereby improving the efficiency of searching for similar patients and improving the accuracy of similar patients.

Description

Technical Field

[0001] The present application relates to the technical field of similar patient prediction, and in particular, to a method, device, electronic device and storage medium for determining similar patients. Background Art

[0002] In practice, accurate prediction of a patient's disease development trend can avoid or mitigate the distress caused by the disease to the patient, reduce medical costs, and reduce the use of medical resources. Therefore, prediction of disease development trend is of great significance in disease prevention, early diagnosis, and timely intervention.

[0003] At present, the development trend of the current patient's condition is usually judged based on the condition development of other patients with similar conditions to the current patient. Therefore, it is very necessary to find other patients with similar conditions to the current patient. Generally, the medical data of the current patient is input into the similar patient prediction model, and the output result of the similar patient prediction model is used as the patient with a similar condition to the current patient.

[0004] However, using this similar patient prediction method requires pre-labeling of the medical data generated by historical patients during the treatment process to obtain sample data for training the similar patient prediction model, which will consume a lot of human resources and time resources. Summary of the invention

[0005] In view of this, the purpose of the embodiments of the present application is to provide a method, device, electronic device and storage medium for determining similar patients. Through two clustering processes and calculation of the distance between patients, historical patients similar to the target patient are selected, which saves human and time resources, improves the efficiency of searching for similar historical patients, and improves the accuracy of the similar historical patients found.

[0006] In a first aspect, an embodiment of the present application provides a method for determining similar patients, the method comprising:

[0007] Inputting the feature vector of the target patient into a Gaussian mixture model, and obtaining the primary patient category to which the target patient belongs output by the Gaussian mixture model;

[0008] Finding historical patients who belong to the same primary patient category as the target patient, and determining the target patient and the historical patients who belong to the same primary patient category as the target patient as a primary patient set;

[0009] Clustering a plurality of patients included in the primary patient set according to a feature vector corresponding to each patient in the primary patient set, and determining a secondary patient category to which each patient in the primary patient set belongs;

[0010] Determine the target patient and historical patients belonging to the same secondary patient category as the target patient as a secondary patient set;

[0011] According to the feature vector corresponding to each patient in the secondary patient set, the distance between the target patient and any historical patient in the secondary patient set except the target patient is calculated, and based on the distance, a historical patient similar to the target patient is selected from the secondary patient set.

[0012] In a possible implementation, calculating the distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set includes:

[0013] Determine a plurality of joint distributions corresponding to each historical patient in the secondary patient set except the target patient according to the feature vector corresponding to the target patient and the feature vector corresponding to each historical patient in the secondary patient set except the target patient;

[0014] For each historical patient in the secondary patient set except the target patient, the push distance of the historical patient under each joint distribution corresponding to the historical patient is calculated, and the push distance that meets the preset conditions is determined as the distance between the historical patient and the target patient.

[0015] In a possible implementation, the feature vector corresponding to the target patient is determined in the following manner:

[0016] Generating an intermediate feature matrix corresponding to the target patient according to the intermediate feature vector corresponding to the target patient;

[0017] A feature vector corresponding to the target patient is determined according to the intermediate feature matrix and a preset dimension, wherein the dimension of the feature vector is lower than the dimension of the intermediate feature vector.

[0018] In a possible implementation, the intermediate feature vector corresponding to the target patient is determined in the following manner:

[0019] Acquire an initial feature vector of the target patient;

[0020] If the value of the first element is missing in the initial feature vector of the target patient, searching for a historical patient associated with the target patient from the historical patients according to the values ​​corresponding to each second element except the first element in the initial feature vector of the target patient and the values ​​corresponding to each second element in the initial feature vector of each historical patient, and determining the degree of association between the target patient and each historical patient associated with the target patient;

[0021] According to the numerical value corresponding to the first element in the initial feature vector of each historical patient associated with the target patient and the degree of association corresponding to each historical patient associated with the target patient, the numerical value of the first element in the initial feature vector of the target patient is determined to generate an intermediate feature vector of the target patient.

[0022] In a possible implementation, the initial feature vector of the target patient is determined in the following manner:

[0023] Acquire multiple characteristic data of the target patient, and the data type of each characteristic data; wherein the data types include numerical type, Boolean type, coded type, and categorical type;

[0024] For each characteristic data of the target patient, a conversion rule corresponding to the data type to which the characteristic data belongs is obtained, and the characteristic data is converted based on the conversion rule to generate numerical data corresponding to the characteristic data;

[0025] An initial feature vector of the target patient is generated according to the numerical data corresponding to the target patient.

[0026] In a possible implementation, for each characteristic data of the target patient, obtaining a conversion rule corresponding to a data type to which the characteristic data belongs, performing conversion processing on the characteristic data based on the conversion rule, and generating numerical data corresponding to the characteristic data includes:

[0027] If the data type of the characteristic data is a numerical data type, the characteristic data is determined as the numerical data corresponding to the characteristic data; or,

[0028] If the data type of the characteristic data is Boolean, then based on the correspondence between the characteristic data and the numerical data, the numerical data corresponding to the characteristic data is determined; or,

[0029] If the data type of the characteristic data is a coded type, the disease diagnosis category to which the characteristic data belongs is determined according to the data value of the first specific bit data in the characteristic data, and the numerical data of the characteristic data under the disease diagnosis category is determined according to the data value of the second specific bit data in the characteristic data; or

[0030] If the data type to which the characteristic data belongs is classified, the numerical data corresponding to the characteristic data is determined according to the disease category to which the characteristic data belongs and the corresponding relationship between the disease category and the numerical data.

[0031] In a possible implementation manner, the determination method further includes:

[0032] Predicting the medical decision of the target patient based on the medical decisions of historical patients similar to the target patient; wherein the medical decision includes one or more of the following decisions: consultation decision, triage decision, diagnosis decision and treatment decision.

[0033] In a second aspect, an embodiment of the present application provides a similar patient determination device, the determination device comprising:

[0034] A first acquisition module is used to input the feature vector of the target patient into a Gaussian mixture model to obtain the primary patient category to which the target patient belongs output by the Gaussian mixture model;

[0035] A first determination module is used to find historical patients belonging to the same primary patient category as the target patient, and determine the target patient and the historical patients belonging to the same primary patient category as the target patient as a primary patient set;

[0036] a second determination module, configured to cluster the multiple patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set, and determine the secondary patient category to which each patient in the primary patient set belongs;

[0037] A third determination module is used to determine the target patient and historical patients belonging to the same secondary patient category as the target patient as a secondary patient set;

[0038] A selection module is used to calculate the distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set, and based on the distance, select a historical patient similar to the target patient from the secondary patient set.

[0039] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of the method for determining similar patients as described in any one of the first aspects.

[0040] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for determining similar patients described in any one of the first aspects are executed.

[0041] The embodiments of the present application provide a method, device, electronic device and storage medium for determining similar patients. The method comprises the following steps: inputting a feature vector of a target patient into a Gaussian mixture model to obtain a primary patient category to which the target patient belongs output by the Gaussian mixture model; searching for historical patients belonging to the same primary patient category as the target patient, and determining the target patient and the historical patients belonging to the same primary patient category as the target patient as a primary patient set; clustering a plurality of patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set to determine a secondary patient category to which each patient in the primary patient set belongs; determining the target patient and the historical patients belonging to the same secondary patient category as the target patient as a secondary patient set; calculating a distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set, and selecting a historical patient similar to the target patient from the secondary patient set based on the distance. The embodiment of the present application selects historical patients similar to the target patient through two clustering processes and calculation of the distance between patients, which saves human and time resources, improves the efficiency of searching for similar historical patients, and improves the accuracy of searching for similar historical patients.

[0042] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A flow chart of a method for determining similar patients provided in an embodiment of the present application is shown;

[0045] Figure 2 A flow chart showing another method for determining similar patients provided in an embodiment of the present application is shown;

[0046] Figure 3 A flow chart showing another method for determining similar patients provided in an embodiment of the present application is shown;

[0047] Figure 4 A flow chart showing another method for determining similar patients provided in an embodiment of the present application is shown;

[0048] Figure 5 A flow chart showing another method for determining similar patients provided in an embodiment of the present application is shown;

[0049] Figure 6 A schematic diagram of the structure of a similar patient determination device provided in an embodiment of the present application is shown;

[0050] Figure 7 A schematic diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0051] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0052] At present, the development trend of the current patient's condition is usually judged based on the development of other patients with similar conditions to the current patient. Therefore, it is very necessary to find other patients with similar conditions to the current patient. Generally, the medical data of the current patient is input into the similar patient prediction model, and the output result of the similar patient prediction model is used as the patient with similar condition to the current patient. However, using this similar patient prediction method, it is necessary to mark the medical data generated by historical patients during the treatment process in advance to obtain sample data for training the similar patient prediction model, which will consume a lot of human resources and time resources.

[0053] Based on the above problems, the embodiments of the present application provide a method, device, electronic device and storage medium for determining similar patients, which inputs a feature vector of a target patient into a Gaussian mixture model to obtain the primary patient category to which the target patient belongs output by the Gaussian mixture model; searches for historical patients belonging to the same primary patient category as the target patient, and determines the target patient and the historical patients belonging to the same primary patient category as the target patient as a primary patient set; clusters multiple patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set, and determines the secondary patient category to which each patient in the primary patient set belongs; determines the target patient and the historical patients belonging to the same secondary patient category as the target patient as a secondary patient set; calculates the distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set, and selects historical patients similar to the target patient from the secondary patient set based on the distance. The embodiment of the present application selects historical patients similar to the target patient through two clustering processes and calculation of the distance between patients, which saves human and time resources, improves the efficiency of searching for similar historical patients, and improves the accuracy of searching for similar historical patients.

[0054] The defects existing in the above solutions are the results obtained by the inventor after practice and careful research. Therefore, the process of discovering the above problems and the solutions proposed by this application for the above problems below should be the contributions made by the inventor to this application during the application process.

[0055] The technical solutions in the present application will be clearly and completely described below in conjunction with the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0056] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0057] To facilitate understanding of this embodiment, a method for determining similar patients disclosed in an embodiment of the present application is first introduced in detail.

[0058] See also Figure 1 As shown, Figure 1 A flowchart of a method for determining similar patients provided in an embodiment of the present application, the method comprising the following steps:

[0059] S101. Input a feature vector of a target patient into a Gaussian mixture model, and obtain the primary patient category to which the target patient belongs output by the Gaussian mixture model.

[0060] In this step, the target patient is the patient whose medical decision is to be determined. Based on the medical decisions of multiple historical patients, the medical decision of the target patient is predicted. The Gaussian mixture model is used to determine the primary patient category to which the target patient belongs. The Gaussian mixture model is a linear combination of K1 Gaussian distributions. It assumes that all patients can be divided into K1 categories, that is, K1 primary patient categories, and each category of patients obeys a Gaussian distribution. The training process of the Gaussian mixture model is to estimate the probability density of each Gaussian distribution and the weight of each Gaussian distribution. After the Gaussian mixture model is trained, the feature vector of the target patient is input into the Gaussian mixture model, and the target patient is substituted into the K1 Gaussian distributions to obtain the probability that the target patient belongs to each primary patient category. The primary patient category with the largest probability is taken as the primary patient category to which the target patient finally belongs.

[0061] Among them, the feature vector of the target patient is used to characterize the target patient's gender, age, disease diagnosis, test indicators and other medical and health data. The first-level patient category is the preset K1 patient categories.

[0062] S102, searching for historical patients who belong to the same primary patient category as the target patient, and determining the target patient and the historical patients who belong to the same primary patient category as the target patient as a primary patient set.

[0063] In this step, the medical and health data of historical patients are obtained from the hospital's HIS, CIS, LIS, PACS, EMR and other information systems, and the feature vectors of historical patients are determined based on the medical and health data. The feature vectors of each historical patient are respectively input into the Gaussian mixture model to obtain the first-level patient category to which each historical patient belongs, and then multiple historical patients under each first-level patient category are obtained. After determining the first-level patient category to which the target patient belongs, historical patients who belong to the same first-level patient category as the target patient are searched, and a first-level patient set is formed based on the historical patients and the target patients found, and the patients included in the first-level patient set correspond to the same first-level patient category.

[0064] S103. Clustering multiple patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set, and determining the secondary patient category to which each patient in the primary patient set belongs.

[0065] In this step, the first-level patient category is a generalized patient category. Under each first-level patient category, the first-level patient category is subdivided into multiple second-level patient categories. The multiple patients included in the first-level patient set are clustered based on the K-means algorithm. Specifically, K2 second-level patient categories are preset. Based on the feature vector corresponding to each patient in the first-level patient set, the K-means algorithm is used to cluster the multiple patients included in the first-level patient set to obtain K2 subclasses, each of which includes at least one patient. Each subclass corresponds to a specific second-level patient category, and the second-level patient category corresponding to the subclass is determined as the second-level patient category corresponding to each patient in the subclass.

[0066] Specifically, the basic idea of ​​the K-means algorithm is to cluster objects with K2 points in space as the center (K2 is the number of subclasses), cluster the objects closest to each of the K2 points, and update the values ​​of each cluster center one by one through iteration until the best clustering result is obtained. Assuming that the multiple patients included in the primary patient set are clustered into K2 subclasses, the specific steps of the K-means algorithm include:

[0067] Step 1: Appropriately select the initial centers of K2 subclasses, usually randomly;

[0068] Step 2: In each iteration, for any patient in the first-level patient set, calculate the Euclidean distance from each of the K2 centers, and assign the patient to the subclass with the center with the shortest distance;

[0069] Step 3: Use the mean method to update the center values ​​of the K2 subclasses;

[0070] Step 4: For all K2 cluster centers, repeat steps 2 and 3. When the moving distance of the center value of the subclass is less than the preset threshold, the iteration ends, clustering is completed, and multiple subclasses are obtained.

[0071] Here, the above preset threshold can be set according to the actual needs of the user. If the user has high requirements for clustering accuracy, the preset threshold can be reduced. If the user has high requirements for computing speed, the preset threshold can be increased.

[0072] S104. Determine the target patient and historical patients who belong to the same secondary patient category as the target patient as a secondary patient set.

[0073] In this step, based on the clustering results of the K-means algorithm, the secondary patient category to which the target patient belongs is found, and the historical patients who belong to the same secondary patient category as the target patient are found. Based on the target patient and the historical patients who belong to the same secondary patient category as the target patient, a secondary patient set is constructed.

[0074] S105. Calculate the distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set, and select a historical patient similar to the target patient from the secondary patient set based on the distance.

[0075] In this step, for each historical patient in the secondary patient set, the distance between the target patient and the historical patient is calculated based on the feature vector of the target patient and the feature vector of the historical patient, and the distance between each historical patient and the target patient is obtained, and the historical patients whose distances meet the preset distance requirements are determined as historical patients similar to the target patient.

[0076] Optionally, historical patients whose distances are greater than a preset distance threshold are determined as historical patients similar to the target patient; or, the distances of historical patients are sorted, and historical patients ranked in a specific position are determined as historical patients similar to the target patient, for example, historical patients ranked in the top five are determined as historical patients similar to the target patient.

[0077] The embodiment of the present application provides a method for determining similar patients. The method determines the primary patient category to which the target patient belongs based on a Gaussian mixture model, obtains a primary patient set including the target patient and historical patients belonging to the same primary patient category as the target patient, clusters multiple patients included in the primary patient set based on the K-means algorithm, obtains a secondary patient set including the target patient and historical patients belonging to the same secondary patient category as the target patient, calculates the distance between the target patient and each historical patient in the secondary patient set, and selects historical patients similar to the target patient from the secondary patient set based on the calculated distance. Through two clustering processes and calculation of the distance between patients, historical patients similar to the target patient are selected without marking a large number of historical patients, saving human resources and time resources, improving the efficiency of searching for similar historical patients, and improving the accuracy of similar historical patients found.

[0078] Furthermore, in a method for determining similar patients provided in an embodiment of the present application, the method further includes:

[0079] Predicting the medical decision of the target patient based on the medical decisions of historical patients similar to the target patient; wherein the medical decision includes one or more of the following decisions: consultation decision, triage decision, diagnosis decision and treatment decision.

[0080] In an embodiment of the present application, the medical decision of each historical patient similar to the target patient is obtained, and the medical decision to be used for the target patient is predicted based on the medical decision of each historical patient similar to the target patient, wherein the medical decision includes one or more of the following decisions: consultation decision, triage decision, diagnosis decision and treatment decision. Specifically, the corresponding type of medical decision can be selected according to the type of disease and severity of the disease of the target patient.

[0081] For further information, see Figure 2 As shown, Figure 2 A flowchart of another method for determining similar patients provided in an embodiment of the present application, wherein the distance between the target patient and any historical patient other than the target patient in the secondary patient set is calculated based on the feature vector corresponding to each patient in the secondary patient set, including:

[0082] S201. Determine multiple joint distributions corresponding to each historical patient in the secondary patient set except the target patient based on the feature vector corresponding to the target patient and the feature vector corresponding to each historical patient in the secondary patient set except the target patient.

[0083] S202. For each historical patient in the secondary patient set except the target patient, calculate the bulldozing distance of the historical patient under each joint distribution corresponding to the historical patient, and determine the bulldozing distance that meets the preset conditions as the distance between the historical patient and the target patient.

[0084] Combining step 201 and step 202, the distance between the target patient and each historical patient in the secondary patient set is the Earth-Mover (EM) distance, also known as the Wasserstein distance. Specifically, the Earth-Mover distance between patients is calculated using the following formula:

[0085]

[0086] Among them, P1 is the target patient, P2 is any historical patient in the secondary patient set, Π(P1,P2) is the set of all possible joint distributions of P1 and P2. For each joint distribution γ in the set of joint distributions, (x,y)~γ can be sampled to obtain the target feature vector x of P1 and the target feature vector y of P2, and the ||xy|| between x and y is calculated, and then the expected value E of ||xy|| of x and y under each joint distribution γ is calculated. (x,y)～γ [||xy||], that is, the push-up distance between P1 and P2 under each joint distribution. The lower bound of this expected value in all possible joint distributions is the push-up distance between P1 and P2.

[0087] The push-up distance between the target patient and each historical patient in the secondary patient set is calculated. If the calculated push-up distance is short, the similarity between the patients is large. If the calculated push-up distance is long, the similarity between the patients is small.

[0088] For further information, see Figure 3 As shown, Figure 3 This is a flow chart of another method for determining similar patients provided in an embodiment of the present application, wherein the feature vector corresponding to the target patient is determined in the following manner:

[0089] S301. Generate an intermediate feature matrix corresponding to the target patient according to the intermediate feature vector corresponding to the target patient.

[0090] S302. Determine a feature vector corresponding to the target patient according to the intermediate feature matrix and a preset dimension, wherein the dimension of the feature vector is lower than the dimension of the intermediate feature vector.

[0091] In combination with step 301 and step 302, the target patient corresponds to a variety of feature data, and the various feature data of the target patient are converted into numerical data to obtain the initial feature vector of the target patient, and the missing values ​​in the initial feature vector are interpolated to obtain the intermediate feature vector of the target patient, and the high-dimensional intermediate feature vector is reduced in dimension to obtain a low-dimensional feature vector. Accordingly, the initial feature vector, intermediate feature vector and feature vector of each historical patient are obtained.

[0092] Let the total number of the target patient and historical patients be n. Each patient corresponds to an intermediate feature vector of d dimensions. The feature mapping maps the original set of d dimensions to a new set of k dimensions (k < d). Here, the k dimensions are preset dimensions, and it reduces the feature dimensions by combining medical features rather than deleting medical features. Using the idea of dimensionality reduction, multiple medical features are transformed into a few comprehensive medical features (i.e., principal components). Each principal component can reflect most of the information of the feature data, and the information contained is non-repetitive. Through a linear transformation, the original d-dimensional medical features are transformed into k dimensions, where each dimension in the k dimensions is linearly independent.

[0093] Specifically, the following method is used to perform dimensionality reduction on the intermediate feature vectors of d dimensions corresponding to the target patient and historical patients respectively:

[0094] Step 1: Take the intermediate feature vectors of d dimensions corresponding to n patients as row vectors to obtain an intermediate feature matrix X0;

[0095] Step 2: Zero-mean each column of X0 (representing a medical feature), that is, subtract the mean of the column to obtain a matrix X1;

[0096] Step 3: Calculate the covariance matrix C = 1 / n * X1 * X1 T ;

[0097] Step 4: Calculate the eigenvalues and corresponding eigenvectors of the covariance matrix C;

[0098] Step 5: Arrange the eigenvectors in a matrix row by row from top to bottom according to the corresponding eigenvalue magnitudes, and take the first k rows to form a matrix W;

[0099] Step 6: X2 = W * X1, X2 is the feature matrix after being reduced to k dimensions. X2 is a matrix of n rows and k columns. According to the values corresponding to each row in the target feature matrix X2, determine the feature vectors corresponding to the target patient and historical patients respectively.

[0100] Further, as shown in Figure 4 shown, Figure 4 is a flowchart of another method for determining similar patients provided by an embodiment of the present application. The intermediate feature vector corresponding to the target patient is determined by the following method:

[0101] S401: Obtain the initial feature vector of the target patient.

[0102] In an embodiment of the present application, when collecting the medical and health data of the target patient, due to the different medical conditions of different patients, the types of medical and health data collected are different, and there are cases where the types of medical and health data are missing. Correspondingly, the values ​​of some elements in the initial feature vector corresponding to the target patient may be missing, and the cluster filling method is used to interpolate the values ​​of the missing elements. The cluster filling method is the K-means clustering method. First, the K3 historical patients closest to the target patient are determined based on Euclidean distance or correlation analysis, and the weighted average of the values ​​of the element in the initial feature vector of these K3 historical patients is used to interpolate the value of the element in the initial feature vector of the target patient. The same mean interpolation method belongs to single value interpolation. The difference is that it uses a hierarchical clustering model to predict the type of missing elements, and then interpolates with the mean of the type.

[0103] Specifically, the initial feature vector of the target patient and the intermediate feature vector of each historical patient are obtained, wherein each historical patient corresponds to an initial feature vector, and the cluster filling method in the embodiment of the present application is used in advance to interpolate the values ​​of the missing elements in the initial feature vector to obtain the intermediate feature vector of each historical patient.

[0104] S402. If the value of the first element is missing in the initial feature vector of the target patient, search for historical patients associated with the target patient from the historical patients based on the values ​​corresponding to each second element except the first element in the initial feature vector of the target patient and the values ​​corresponding to each second element in the initial feature vector of each historical patient, and determine the degree of association between the target patient and each historical patient associated with the target patient.

[0105] In this step, the initial feature vector of the target patient includes d elements, and the element with a missing value in the d elements is called the first element. The first element can be multiple elements or one element. For each first element in the initial feature vector of the target patient, the elements other than the first element in the d elements are collectively referred to as the second element. For example, if the initial feature vector is (x, y, z), if "y" is the first element, then "x" and "z" are the second elements. Based on the value of each second element in the initial feature vector of the target patient and the value of each second element in the initial feature vector of the historical patient, the difference between the target patient and the historical patient under each second element is calculated. If the difference under any second element is less than a preset threshold, the historical patient is determined as a historical patient associated with the target patient. According to the average value of the difference between the target patient and each historical patient associated with the target patient under each second element, the correlation between the target patient and each historical patient associated with the target patient is determined. Here, the larger the average value, the smaller the correlation.

[0106] S403. Determine the value of the first element in the initial feature vector of the target patient according to the value corresponding to the first element in the initial feature vector of each historical patient associated with the target patient and the degree of association corresponding to each historical patient associated with the target patient, and generate an intermediate feature vector of the target patient.

[0107] In this step, according to the value corresponding to the first element in the initial feature vector of each historical patient associated with the target patient, and the correlation degree corresponding to each historical patient associated with the target patient, the sum of the product of the first element and the correlation degree is calculated, and the calculated sum is used as the value of the first element in the initial feature vector of the target patient. For example, the historical patients associated with the target patient are historical patient 1, historical patient 2, and historical patient 3. The value of the first element of historical patient 1 is 1, and the corresponding correlation degree is 0.5; the value of the first element of historical patient 2 is 2, and the corresponding correlation degree is 0.2; the value of the first element of historical patient 3 is 3, and the corresponding correlation degree is 0.3. Then the sum of the product of the first element and the correlation degree is 1×0.5+2×0.2+3×0.3=1.8, and the value of the first element in the initial feature vector of the target patient is 1.8.

[0108] After completing the numerical interpolation of each first element in the initial feature vector of the target patient, the intermediate feature vector of the target patient is obtained.

[0109] For further information, see Figure 5 As shown, Figure 5 This is a flow chart of another method for determining similar patients provided in an embodiment of the present application, wherein the initial feature vector of the target patient is determined in the following manner:

[0110] S501. Acquire multiple characteristic data of the target patient, and the data type to which each characteristic data belongs; wherein the data types include numerical type, Boolean type, coded type, and categorical type.

[0111] In this step, characteristic data of the target patient is obtained from multiple data sources, and characteristic data of multiple historical patients is obtained. The characteristic data here refers to the patient's medical health data. For example, multiple characteristic data of the patient are obtained from hospital information systems such as HIS, CIS, LIS, PACS, EMR, etc. After obtaining multiple characteristic data from multiple data sources, it is necessary to count the multiple characteristic data corresponding to each patient. Specifically, according to the ID of each patient, the characteristic data from different data sources but with the same patient ID are integrated together in the order of preset medical characteristics to create a wide table. In this way, the integration of the characteristic data is completed. The characteristic data of each row in the wide table is the characteristic data of the same patient.

[0112] Feature data include multiple data types, specifically, data types include numeric, Boolean, coded, and classified. The data type of each feature data corresponding to the target patient is obtained, and correspondingly, the data type of each feature data corresponding to the historical patient is obtained.

[0113] S502: For each characteristic data of the target patient, obtain a conversion rule corresponding to the data type to which the characteristic data belongs, convert the characteristic data based on the conversion rule, and generate numerical data corresponding to the characteristic data.

[0114] In this step, since the characteristic data includes multiple data types, in order to facilitate calculation, the characteristic data of multiple data types need to be converted into unified numerical data. Characteristic data of different data types correspond to different conversion rules. For each characteristic data of the target patient, the conversion rule corresponding to the data type to which the characteristic data belongs is obtained, and the characteristic data is converted based on the corresponding conversion rule to generate numerical data corresponding to the characteristic data.

[0115] Accordingly, for each characteristic data of the historical patients, the conversion rule corresponding to the data type to which the characteristic data belongs is obtained, and the characteristic data is converted based on the corresponding conversion rule to generate the numerical data corresponding to the characteristic data. The characteristic data of the historical patients can be converted before the characteristic data of the target patient is converted.

[0116] S503: Generate an initial feature vector of the target patient according to the numerical data corresponding to the target patient.

[0117] In this step, after determining the numerical data corresponding to each characteristic data of the target patient, multiple numerical data corresponding to the target patient are obtained, and the multiple numerical data are combined according to a preset medical characteristic order to obtain an initial characteristic vector of the target patient.

[0118] Correspondingly, after determining the numerical data corresponding to each characteristic data of the historical patient, multiple numerical data corresponding to the historical patient are obtained, and the multiple numerical data are combined according to a preset medical characteristic order to obtain an initial characteristic vector of the historical patient.

[0119] Furthermore, in a method for determining similar patients provided in an embodiment of the present application, for each type of characteristic data of the target patient, a conversion rule corresponding to the data type to which the characteristic data belongs is obtained, and the characteristic data is converted based on the conversion rule. In the process of generating numerical data corresponding to the characteristic data, there are multiple conversion rules for the characteristic data. The data types to which the characteristic data belongs are different, and the conversion rules corresponding to the characteristic data are different. The conversion rules for numerical, Boolean, coded, and classified characteristic data are explained respectively.

[0120] Conversion rule 1: If the data type of the feature data is a numerical type, the feature data is determined to be numerical data corresponding to the feature data.

[0121] The data obtained from laboratory tests are generally numerical data, such as red blood cell count, hematocrit measurement, and platelet count. If the data type of the feature data is numerical, the feature data will not be processed and will be numerical data.

[0122] Conversion rule 2: If the data type of the characteristic data is Boolean, based on the correspondence between the characteristic data and the numerical data, determine the numerical data corresponding to the characteristic data.

[0123] There are two types of data corresponding to Boolean data, namely 0 and 1. Boolean data is either 0 or 1. For example, the results of the nucleic acid test for the new coronavirus are generally expressed as negative or positive. In order to unify the calculation, they are mapped to 0 and 1. You can directly use 0 or 1 as numerical data, or you can determine the numerical data corresponding to the feature data based on the preset correspondence between the feature data and the numerical data. For example, "0" corresponds to the numerical data "0.2", and "1" corresponds to the numerical data "0.5".

[0124] Conversion rule three: If the data type of the feature data is a coded type, the disease diagnosis category to which the feature data belongs is determined based on the data value of the first specific bit data in the feature data, and the numerical data of the feature data under the disease diagnosis category is determined based on the data value of the second specific bit data in the feature data.

[0125] There are a lot of coded feature data in clinical data, such as disease classification and coding (ICD-10). The coding system is a reflection of expert knowledge, which contains a lot of semantic relationships. The first 4 digits of ICD-10 code are mixed letters and numbers, where the first digit is an English letter and the last 3 digits are Arabic numerals. The first 3 digits of the code are the category code, and the 4th digit is the sub-classification of the first 3 digits.

[0126] For the code “E11.0”, the first three digits “E11” represent “type 2 diabetes”, and together with the fourth digit “0” represent “type 2 diabetes with coma”.

[0127] According to the ICD-10 coding rules, when the first letters of two disease diagnosis codes are the same, the distance between the second and third digits in the disease diagnosis codes reflects the difference between the two diseases to a certain extent; when the first letters of the two disease diagnosis codes are different, it is difficult to measure the distance between the two diseases using the information of the codes themselves. Therefore, the conversion rules corresponding to the coded feature data are:

[0128] (1) First, all ICD-10 diagnosis codes are grouped according to whether their first letters are the same, resulting in m groups. These m different letters are used as features to create m new features, i.e., m disease diagnosis categories.

[0129] (2) If the first letter of the disease code is the same as the created feature, the second digit of the code is used as the tens digit and the third digit is used as the ones digit, and the two digits are converted into a two-digit decimal integer, and this decimal integer is used as the numerical data under the disease diagnosis category. Here, the first digit is the first specific bit in the feature data, and the second and third digits are the second specific bits in the feature data.

[0130] (3) If the first letter of the disease code is different from the created feature, the corresponding feature value is 100.

[0131] For example, there are currently five disease diagnosis codes: A05.7, B17.8, H25.3, H28.5, and H43.4. The above disease diagnosis codes are converted according to the conversion rules corresponding to the coding types. The conversion results of the disease diagnosis codes are shown in Table 1.

[0132] Disease diagnosis code A B H A05.7 5 100 100 B17.8 100 17 100 H25.3 100 100 25 H28.5 100 100 28 H43.4 100 100 43

[0133] Table 1. Transformation results of disease diagnosis codes

[0134] Conversion rule 4: If the data type of the characteristic data is classified, the numerical data corresponding to the characteristic data is determined according to the disease category to which the characteristic data belongs and the correspondence between the disease category and the numerical data.

[0135] One-Hot encoding is used to convert the categorical feature data into numerical data. For example, if the "pathological type" in the feature data set is a categorical type, the possible values ​​of this feature data include: "large cell carcinoma", "small cell carcinoma", "neuroendocrine carcinoma", "adenocarcinoma", "squamous cell carcinoma", and "squamous cell carcinoma". In order to encode these 6 possible values, 6 new features are created, namely "large cell carcinoma", "small cell carcinoma", "neuroendocrine carcinoma", "adenocarcinoma", "squamous cell carcinoma", and "squamous cell carcinoma". If the "pathological type" of a case is a new feature, then the new feature is set to 1, and the new features other than the new feature are all set to 0. Therefore, for each categorical feature data, only one of the 6 new features has a value of 1, and the values ​​of the other new features are all 0.

[0136] Specifically, One-Hot encoding is used to encode pathological type features, and the conversion results of pathological types are shown in Table 2.

[0137] Pathological type Large cell carcinoma Small cell carcinoma Neuroendocrine cancer Adenocarcinoma Squamous cell carcinoma Squamous adenocarcinoma Large cell carcinoma 1 0 0 0 0 0 Small cell carcinoma 0 1 0 0 0 0 Neuroendocrine cancer 0 0 1 0 0 0 Adenocarcinoma 0 0 0 1 0 0 Squamous cell carcinoma 0 0 0 0 1 0 Squamous adenocarcinoma 0 0 0 0 0 1

[0138] Table 2. Conversion results of pathological types

[0139] Based on the same inventive concept, an embodiment of the present application also provides a similar patient determination device corresponding to the similar patient determination method. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the similar patient determination method in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0140] See also Figure 6 As shown, Figure 6 A schematic diagram of a similar patient determination device provided in one embodiment of the present application, the determination device comprising:

[0141] A first acquisition module 601 is used to input the feature vector of the target patient into a Gaussian mixture model to obtain the primary patient category to which the target patient belongs output by the Gaussian mixture model;

[0142] A first determination module 602 is used to find historical patients belonging to the same primary patient category as the target patient, and determine the target patient and the historical patients belonging to the same primary patient category as the target patient as a primary patient set;

[0143] A second determination module 603 is used to cluster the multiple patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set, and determine the secondary patient category to which each patient in the primary patient set belongs;

[0144] A third determination module 604 is used to determine the target patient and historical patients belonging to the same secondary patient category as the target patient as a secondary patient set;

[0145] The selection module 605 is used to calculate the distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set, and select a historical patient similar to the target patient from the secondary patient set based on the distance.

[0146] In a possible implementation, the selection module 605, when calculating the distance between the target patient and any historical patient other than the target patient in the secondary patient set according to the feature vector corresponding to each patient in the secondary patient set, includes:

[0147] Determine a plurality of joint distributions corresponding to each historical patient in the secondary patient set except the target patient according to the feature vector corresponding to the target patient and the feature vector corresponding to each historical patient in the secondary patient set except the target patient;

[0148] For each historical patient in the secondary patient set except the target patient, the push distance of the historical patient under each joint distribution corresponding to the historical patient is calculated, and the push distance that meets the preset conditions is determined as the distance between the historical patient and the target patient.

[0149] In a possible implementation manner, the device for determining similar patients further includes:

[0150] A first generating module, used for generating an intermediate feature matrix corresponding to the target patient according to the intermediate feature vector corresponding to the target patient;

[0151] The fourth determination module is used to determine the feature vector corresponding to the target patient according to the intermediate feature matrix and a preset dimension, wherein the dimension of the feature vector is lower than the dimension of the intermediate feature vector.

[0152] In a possible implementation manner, the device for determining similar patients further includes:

[0153] A second acquisition module, used to acquire an initial feature vector of the target patient;

[0154] a fifth determination module, configured to search for a historical patient associated with the target patient from the historical patients according to the numerical values ​​respectively corresponding to each second element except the first element in the initial feature vector of the target patient and the numerical values ​​respectively corresponding to each second element in the initial feature vector of each historical patient, and determine the degree of association between the target patient and each historical patient associated with the target patient, if the numerical value of the first element is missing in the initial feature vector of the target patient;

[0155] The sixth determination module is used to determine the value of the first element in the initial feature vector of the target patient based on the value corresponding to the first element in the initial feature vector of each historical patient associated with the target patient and the degree of association corresponding to each historical patient associated with the target patient, so as to generate an intermediate feature vector of the target patient.

[0156] In a possible implementation manner, the device for determining similar patients further includes:

[0157] The third acquisition module is used to acquire multiple characteristic data of the target patient and the data type of each characteristic data; wherein the data type includes numerical type, Boolean type, coding type, and classification type;

[0158] The second generating module is used to obtain, for each characteristic data of the target patient, a conversion rule corresponding to the data type to which the characteristic data belongs, and convert the characteristic data based on the conversion rule to generate numerical data corresponding to the characteristic data;

[0159] The third generating module is used to generate an initial feature vector of the target patient according to the numerical data corresponding to the target patient.

[0160] In a possible implementation, the second generating module, for each characteristic data of the target patient, obtains a conversion rule corresponding to a data type to which the characteristic data belongs, and converts the characteristic data based on the conversion rule to generate numerical data corresponding to the characteristic data, including:

[0161] If the data type of the characteristic data is a numerical data type, the characteristic data is determined as the numerical data corresponding to the characteristic data; or,

[0162] If the data type of the characteristic data is Boolean, then based on the correspondence between the characteristic data and the numerical data, the numerical data corresponding to the characteristic data is determined; or,

[0163] If the data type of the characteristic data is a coded type, the disease diagnosis category to which the characteristic data belongs is determined according to the data value of the first specific bit data in the characteristic data, and the numerical data of the characteristic data under the disease diagnosis category is determined according to the data value of the second specific bit data in the characteristic data; or

[0164] If the data type to which the characteristic data belongs is classified, the numerical data corresponding to the characteristic data is determined according to the disease category to which the characteristic data belongs and the corresponding relationship between the disease category and the numerical data.

[0165] In a possible implementation manner, the device for determining similar patients further includes:

[0166] A prediction module is used to predict the medical decision of the target patient based on the medical decisions of historical patients similar to the target patient; wherein the medical decision includes one or more of the following decisions: consultation decision, triage decision, diagnosis decision and treatment decision.

[0167] The embodiment of the present application provides a device for determining similar patients. The device determines the primary patient category to which a target patient belongs based on a Gaussian mixture model, obtains a primary patient set including the target patient and historical patients belonging to the same primary patient category as the target patient, clusters multiple patients included in the primary patient set based on the K-means algorithm, obtains a secondary patient set including the target patient and historical patients belonging to the same secondary patient category as the target patient, calculates the distance between the target patient and each historical patient in the secondary patient set, and selects historical patients similar to the target patient from the secondary patient set based on the calculated distance. Through two clustering processes and calculation of the distance between patients, historical patients similar to the target patient are selected, without the need to mark a large number of historical patients, saving human resources and time resources, improving the efficiency of searching for similar historical patients, and improving the accuracy of similar historical patients found.

[0168] See also Figure 7 As shown, Figure 7 A schematic diagram of an electronic device provided in an embodiment of the present application, the electronic device 700 includes: a processor 701, a memory 702 and a bus 703, the memory 702 stores machine-readable instructions executable by the processor 701, when the electronic device is running, the processor 701 communicates with the memory 702 via the bus 703, and the processor 701 executes the machine-readable instructions to perform the steps of the method for determining similar patients as described above.

[0169] Specifically, the memory 702 and the processor 701 can be general-purpose memories and processors, which are not specifically limited here. When the processor 701 runs the computer program stored in the memory 702, the method for determining similar patients can be executed.

[0170] Corresponding to the above-mentioned method for determining similar patients, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method for determining similar patients are executed.

[0171] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules 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 communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0172] 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 modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0173] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0174] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that can be executed by a processor. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform 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 read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0175] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for determining similar patients, characterized in that: The determination method comprises: Inputting the feature vector of the target patient into a Gaussian mixture model, and obtaining the primary patient category to which the target patient belongs output by the Gaussian mixture model; Finding historical patients who belong to the same primary patient category as the target patient, and determining the target patient and the historical patients who belong to the same primary patient category as the target patient as a primary patient set; Performing K-means clustering on a plurality of patients included in the primary patient set according to a feature vector corresponding to each patient in the primary patient set, and determining a secondary patient category to which each patient in the primary patient set belongs; Determine the target patient and historical patients belonging to the same secondary patient category as the target patient as a secondary patient set; Calculate the distance between the target patient and any historical patient in the secondary patient set except the target patient according to the feature vector corresponding to each patient in the secondary patient set, and select a historical patient similar to the target patient from the secondary patient set based on the distance; The Gaussian mixture model is a linear combination of several Gaussian distributions, and the step of inputting the feature vector of the target patient into the Gaussian mixture model includes: Inputting the feature vector of the target patient into several Gaussian distributions to obtain the probability that the target patient belongs to each primary patient category; The primary patient category with the highest probability is taken as the primary patient category to which the target patient finally belongs; The performing K-means clustering on the multiple patients included in the primary patient set comprises: Based on the feature vector corresponding to each patient in the primary patient set, a K-means algorithm is used to cluster the multiple patients included in the primary patient set into a plurality of subclasses, wherein each subclass includes at least one patient, and each subclass corresponds to a specific secondary patient category; The secondary patient category corresponding to the subcategory is determined as the secondary patient category corresponding to each patient within the subcategory.

2. The method for determining similar patients according to claim 1, characterized in that: The calculating, according to the feature vector corresponding to each patient in the secondary patient set, the distance between the target patient and any historical patient in the secondary patient set except the target patient, comprises: Determine a plurality of joint distributions corresponding to each historical patient in the secondary patient set except the target patient according to the feature vector corresponding to the target patient and the feature vector corresponding to each historical patient in the secondary patient set except the target patient; For each historical patient in the secondary patient set except the target patient, the push distance of the historical patient under each joint distribution corresponding to the historical patient is calculated, and the push distance that meets the preset conditions is determined as the distance between the historical patient and the target patient.

3. The method for determining similar patients according to claim 1, characterized in that: The feature vector corresponding to the target patient is determined in the following manner: Generating an intermediate feature matrix corresponding to the target patient according to the intermediate feature vector corresponding to the target patient; A feature vector corresponding to the target patient is determined according to the intermediate feature matrix and a preset dimension, wherein the dimension of the feature vector is lower than the dimension of the intermediate feature vector.

4. The method for determining similar patients according to claim 3, characterized in that: The intermediate feature vector corresponding to the target patient is determined in the following manner: Acquire an initial feature vector of the target patient; If the value of the first element is missing in the initial feature vector of the target patient, searching for a historical patient associated with the target patient from the historical patients according to the values ​​corresponding to each second element except the first element in the initial feature vector of the target patient and the values ​​corresponding to each second element in the initial feature vector of each historical patient, and determining the degree of association between the target patient and each historical patient associated with the target patient; According to the numerical value corresponding to the first element in the initial feature vector of each historical patient associated with the target patient and the degree of association corresponding to each historical patient associated with the target patient, the numerical value of the first element in the initial feature vector of the target patient is determined to generate an intermediate feature vector of the target patient.

5. The method for determining similar patients according to claim 4, characterized in that: The initial feature vector of the target patient is determined by: Acquire multiple characteristic data of the target patient, and the data type of each characteristic data; wherein the data types include numerical type, Boolean type, coded type, and categorical type; For each characteristic data of the target patient, a conversion rule corresponding to the data type to which the characteristic data belongs is obtained, and the characteristic data is converted based on the conversion rule to generate numerical data corresponding to the characteristic data; An initial feature vector of the target patient is generated according to the numerical data corresponding to the target patient.

6. The method for determining similar patients according to claim 5, characterized in that: For each characteristic data of the target patient, obtaining a conversion rule corresponding to the data type to which the characteristic data belongs, performing conversion processing on the characteristic data based on the conversion rule, and generating numerical data corresponding to the characteristic data, including: If the data type of the characteristic data is a numerical data type, the characteristic data is determined as the numerical data corresponding to the characteristic data; or, If the data type of the characteristic data is Boolean, then based on the correspondence between the characteristic data and the numerical data, the numerical data corresponding to the characteristic data is determined; or, If the data type of the characteristic data is a coded type, the disease diagnosis category to which the characteristic data belongs is determined according to the data value of the first specific bit data in the characteristic data, and the numerical data of the characteristic data under the disease diagnosis category is determined according to the data value of the second specific bit data in the characteristic data; or, If the data type to which the characteristic data belongs is classified, the numerical data corresponding to the characteristic data is determined according to the disease category to which the characteristic data belongs and the corresponding relationship between the disease category and the numerical data.

7. The method for determining similar patients according to claim 1, characterized in that: The determination method further comprises: Predicting the medical decision of the target patient based on the medical decisions of historical patients similar to the target patient; wherein the medical decision includes one or more of the following decisions: consultation decision, triage decision, diagnosis decision and treatment decision.

8. A device for determining similar patients, characterized in that: The determining device comprises: A first acquisition module is used to input the feature vector of the target patient into a Gaussian mixture model to obtain the primary patient category to which the target patient belongs output by the Gaussian mixture model; A first determination module is used to find historical patients belonging to the same primary patient category as the target patient, and determine the target patient and the historical patients belonging to the same primary patient category as the target patient as a primary patient set; A second determination module is used to perform K-means clustering on multiple patients included in the primary patient set according to the feature vector corresponding to each patient in the primary patient set, and determine the secondary patient category to which each patient in the primary patient set belongs; A third determination module is used to determine the target patient and historical patients belonging to the same secondary patient category as the target patient as a secondary patient set; A selection module, configured to calculate the distance between the target patient and any historical patient other than the target patient in the secondary patient set according to the feature vector corresponding to each patient in the secondary patient set, and select a historical patient similar to the target patient from the secondary patient set based on the distance; The Gaussian mixture model is a linear combination of several Gaussian distributions, and the step of inputting the feature vector of the target patient into the Gaussian mixture model includes: Inputting the feature vector of the target patient into several Gaussian distributions to obtain the probability that the target patient belongs to each primary patient category; The primary patient category with the highest probability is taken as the primary patient category to which the target patient finally belongs; The performing K-means clustering on the multiple patients included in the primary patient set comprises: Based on the feature vector corresponding to each patient in the primary patient set, a K-means algorithm is used to cluster the multiple patients included in the primary patient set into a plurality of subclasses, wherein each subclass includes at least one patient, and each subclass corresponds to a specific secondary patient category; The secondary patient category corresponding to the subcategory is determined as the secondary patient category corresponding to each patient within the subcategory.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the method for determining similar patients as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the method for determining similar patients as claimed in any one of claims 1 to 7.

Images