Inspection Item Push Method, System, Computer Device, and Storage Medium

Abstract

This application relates to artificial intelligence, and provides a method, a system, a computer device, and a storage medium for pushing inspection items. The method for pushing inspection items includes obtaining the target user information and target status parameters of a target user; determining the target inspection item information corresponding to the target user information and the target status parameters based on the association relationship between the user information, status parameters, and inspection item information learned by a preset data push model, and pushing the target inspection item information to the target user. Pushing inspection item information based on the association relationship between the user information, status parameters, and inspection item information in the data push model not only avoids the problem of low push accuracy caused by manual judgment based on the patient's medical treatment data, but also improves the push efficiency of inspection items.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, system, computer device, and storage medium for pushing inspection items. Background Art

[0002] Currently, when a user conducts a project inspection, a doctor needs to combine the patient's past medical treatment data to give the inspection items that need to be tested. Due to the large number of inspection items, the doctor recommends the inspection items that need to be tested to the user (patient) based on the patient's past medical treatment data and their own experience, which not only has a low recommendation accuracy but also a low recommendation efficiency. Summary of the Invention

[0003] This application provides a method, system, computer device, and storage medium for pushing inspection items, aiming to solve the technical problems of low accuracy and efficiency in pushing inspection item data currently.

[0004] In a first aspect, this application provides a method for pushing inspection items, and the method for pushing inspection items includes:

[0005] Obtain the target user information and target status parameters of the target user;

[0006] Based on the association relationship between the user information, status parameters, and inspection item information learned by the preset data push model, determine the target inspection item information corresponding to the target user information and the target status parameters, and push the target inspection item information to the target user.

[0007] In a second aspect, this application also provides a recommendation system for a data push model, and the recommendation system for the data push model includes:

[0008] A data acquisition module, configured to obtain the target user information and target status parameters of the target user;

[0009] An inspection item determination module, configured to determine the target inspection item information corresponding to the target user information and the target status parameters based on the association relationship between the user information, status parameters, and inspection item information learned by the preset data push model.

[0010] In a third aspect, this application also provides a computer device, and the computer device includes:

[0011] A memory and a processor;

[0012] Wherein, the memory is connected to the processor and is used to store programs;

[0013] The processor is configured to implement the steps of the method for pushing inspection items provided in any embodiment of this application by running the programs stored in the memory.

[0014] In a fourth aspect, the present application also provides a computer-readable storage medium storing a computer program, which when executed by a processor causes the processor to implement the steps of any one of the inspection item pushing methods provided in the embodiments of the present application.

[0015] The inspection item pushing method, system, computer device, and storage medium disclosed in the present application obtain the target user information and target status parameters of a target user; based on the association relationship between user information, status parameters, and inspection item information learned by a preset data pushing model, determine the target inspection item information corresponding to the target user information and the target status parameters, and push the target inspection item information to the target user. By pushing inspection item information based on the association relationship between user information, status parameters, and inspection item information in the data pushing model, it not only avoids the problem of low pushing accuracy caused by manual judgment based on the patient's medical treatment data, but also improves the pushing efficiency of inspection and examination items.

[0016] It should be understood that the above general description and subsequent detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 is a flowchart of an inspection item pushing method provided by an embodiment of the present application;

[0019] Figure 2 is a flowchart of learning the association relationship in an inspection item pushing method provided by an embodiment of the present application;

[0020] Figure 3 is a flowchart of determining the association relationship in an inspection item pushing method provided by an embodiment of the present application;

[0021] Figure 4 is a flowchart of vector representation in an inspection item pushing method provided by an embodiment of the present application;

[0022] Figure 5 is a step schematic diagram of another inspection item pushing method provided by an embodiment of the present application;

[0023] Figure 6It is a schematic diagram of the steps of model iterative training in an inspection item push method provided by an embodiment of the present application;

[0024] Figure 7 It is a system block diagram of an inspection item push system provided by an embodiment of the present application;

[0025] Figure 8 It is a schematic block diagram of a computer device provided by an embodiment of the present application.

[0026] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Detailed implementation manners

[0027] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0028] The flowcharts shown in the accompanying drawings are only illustrative, and do not necessarily include all contents and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may change according to the actual situation.

[0029] It should be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application and the appended claims, unless otherwise clearly specified in the context, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0030] It should be understood that in order to facilitate a clear description of the technical solutions in the embodiments of the present application, in the embodiments of the present application, terms such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first callback function and the second callback function are only used to distinguish different callback functions, and do not limit their sequence. Those skilled in the art can understand that the terms "first" and "second" do not limit the quantity and execution order, and the terms "first" and "second" do not necessarily mean different.

[0031] It should also be understood that the term " / and" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the related listed items, and includes these combinations.

[0032] For the inspection items, currently, doctors need to combine the patient's past medical visit data to give the inspection items that need to be tested, and then make a further diagnosis based on the patient's test and inspection results. When diagnosing a patient's disease, doctors need to combine the patient's past medical visit data to give the inspection items that need to be tested, and then make a further diagnosis based on the patient's test and inspection results. Redundancy or insufficiency of inspection items is not conducive to providing better diagnostic services for doctors, nor is it conducive to reducing the misdiagnosis rate and missed diagnosis rate, directly affecting the quality and efficiency of medical diagnosis.

[0033] Exemplarily, when a patient visits a doctor for a diagnosis, the doctor will make a comprehensive judgment based on the patient's basic personal information, including gender, age, etc., and also need to combine the patient's medical history, current medical history, etc. to give the test and inspection items to assist the doctor's diagnosis process. Since there are differences in the requirements for the inspection data that doctors need to master when further understanding the disease, there are also differences in the test and inspection items given, and there are redundant or insufficient situations in the items to be inspected.

[0034] To this end, the embodiments of the present application provide an inspection item push method, system, computer device, and storage medium. Among them, based on the trained data push model, it can recommend the inspection items that the patient needs to test based on the patient's past medical visit data, assist the doctor's diagnosis process, thereby helping the doctor provide better diagnostic services, reducing the probability of misdiagnosis and missed diagnosis, and further improving the quality and efficiency of medical care.

[0035] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0036] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an inspection item push method provided by an embodiment of the present application. This method can be applied to a computer device and is used to push inspection item information to a target user.

[0037] As Figure 1 shown, this inspection item push method includes steps S10 to S20.

[0038] S10. Obtain the target user information and target status parameters of the target user;

[0039] S20. Based on the association relationship between the user information, status parameters, and inspection item information learned by the preset data push model, determine the target inspection item information corresponding to the target user information and the target status parameters, and push the target inspection item information to the target user.

[0040] Exemplarily, in the embodiments of the present application, when the target user is a patient seeking medical treatment, the target user information and the target status parameter are the medical treatment data of the patient seeking medical treatment. Specifically, the medical treatment data is specifically the current medical treatment data of the patient seeking medical treatment, and the current medical treatment data includes user information and medical treatment results, that is: the target user information is the user information of the patient seeking medical treatment, and the target status parameter is the medical treatment result information in the current medical treatment data of the patient seeking medical treatment.

[0041] In the push application stage of the inspection and examination items for patients seeking medical treatment, for the medical treatment data of patients seeking medical treatment, doctors or other medical staff can input the user information of the patients seeking medical treatment and the medical treatment results diagnosed by doctors according to the current symptoms of the patients into the computer device.

[0042] In some embodiments, the user information of each patient seeking medical treatment includes, but is not limited to, data such as the name, age, and gender of each patient, which are used to characterize the basic information of the patient. The medical treatment results of each patient seeking medical treatment include, but are not limited to, disease name information and disease symptom information.

[0043] This embodiment provides a method for pushing inspection items. The method obtains the target user information and the target status parameter of the target user; based on the association relationship between the user information, the status parameter, and the inspection item information learned by the preset data push model, determines the target inspection item information corresponding to the target user information and the target status parameter, and pushes the target inspection item information to the target user. Based on the association relationship between the user information, the status parameter, and the inspection item information in the data push model, pushing the inspection item information not only avoids the problem of low push accuracy caused by manual judgment based on the medical treatment data of patients, but also improves the push efficiency of the inspection and examination items.

[0044] In some embodiments, as Figure 2 shown, before determining the target inspection item information corresponding to the target user information and the target status parameter based on the association relationship between the user information, the status parameter, and the inspection item information learned by the preset data push model, the association relationship includes steps S101 to S103:

[0045] Step S101, obtain sample data, where the sample data includes the sample user information of the sample user, the historical status parameter corresponding to the sample user information, and the historical inspection item information corresponding to the sample user information;

[0046] Step S102, based on the sample data, train a preset initial push model to determine the probability that the user information, the status parameter, and the inspection item information appear together, as the association relationship between the user information, the status parameter, and the inspection item information;

[0047] Step S103: Determine the data push model based on the trained initial push model.

[0048] In an embodiment of the present application, before using the data push model to determine the target inspection item information corresponding to the target user information and the target status parameter, a preset initial push model is trained with sample data. Based on the relationship among the sample user information in the sample data, the historical status parameter corresponding to the sample user information, and the historical inspection item information corresponding to the sample user information, the preset initial push model learns the co-occurrence probability, so that after the training of the initial push model is completed, a data push model is obtained.

[0049] In some embodiments of the present application, the training of the preset initial push model based on the sample data further includes:

[0050] Sort the sample data based on the time stamp order corresponding to the sample data, and train the initial push model based on the sorted sample data.

[0051] In an embodiment of the present application, after obtaining the sample data, the sample data needs to be processed. The sample data includes sample user information, the historical status parameter corresponding to the sample user information, and the historical inspection item information corresponding to the sample user information. The sample user information, historical status parameter, and historical inspection item information are several historical medical visit collection data of the sample user. The several historical medical visit collection data are corresponding to time stamps, and the historical medical visit collection data are arranged in the order of time stamps. The historical medical visit collection data includes the user information and medical visit results of the sample user and one or more corresponding inspection item information.

[0052] In the present application, the sample data is used as the data for modeling the data push model. Each sample data includes multiple medical visit data of a sample user sorted in the order of medical visit time. The medical visit data of the sample user includes the basic information of the sample user, the diagnosis result, and the list of inspection items performed during this medical visit. Among them, the basic information of the sample user is the sample user information in the sample data, the diagnosis result is the historical status parameter in the sample data, and the list of inspection items is the historical inspection item information in the sample data.

[0053] In some embodiments of the present invention, during the learning phase of the association relationship among user information, status parameters, and examination item information, the basic information and diagnosis results of the sample user can also be used as the medical visit sample data for this visit, and the examination item information for this visit can be used as the true examination item information label corresponding to the medical visit sample data, and input into the data push model established by applying deep learning technology to learn the relationship among the user information, medical visit results, and examination item information of the sample user.

[0054] In the embodiments of the present application, each sample user's medical visit data includes the patient's basic information, diagnosis results, and the list of test and examination items conducted during this visit. Among them, the patient's basic information includes age and gender; the diagnosis results include the main diagnosis code and the secondary diagnosis code.

[0055] Each piece of sample data for pre-training the data push model is composed of the multiple medical visit data of a patient arranged in chronological order.

[0056] Exemplarily, the k medical visit data of patient i is V i =(v i,1 , v i,2 , v i,2 , …, v i,k ), where the j-th medical visit data of patient i contains the basic information of the patient for this visit (i.e., age p i,j,age and gender p i,j,sex ), diagnosis results (i.e., main diagnosis d i,j,x and secondary diagnosis d i,j,2 ), and the test and examination list c i,j =(c i,j,1 , c i,j,2 , c i,j,3 , …, c i,j,m ) (for example: m test and examination items). That is, the j-th medical visit data of patient i can be expressed as:

[0057] v i,j =(p i,j , d i,j , c i,j )=(p i,j,age , p i,j,sex , d i,j,1 , d i,j,2 , c i,j,1 , c i,j,2 , c i,j,3 , …, c i,j,m ).

[0058] By constructing the above sample data structure, based on the medical visit sample data and the corresponding real examination item information tags, the data push model is trained to learn the probabilities that occur among the user information, medical visit results, and examination item information.

[0059] In some embodiments, based on the sample data, a preset initial push model is trained to determine the probability correlation relationship among the user information, status parameters, and examination item information. As Figure 3 shown, the method for determining this correlation relationship specifically includes steps S201 to S202.

[0060] S201: Based on the name embedding layer, type embedding layer, and medical visit order embedding layer of the initial push model, map the sample user information, historical status parameters, and historical examination item information into target vector data;

[0061] S202: Based on the main body module of the initial push model and the target vector data, determine the probability correlation relationship among the user information, status parameters, and examination item information.

[0062] In the embodiments of the present application, when learning the relationship among the sample user information, historical status parameters, and historical examination item information of a sample user, that is, when learning the relationship among the user information, medical visit results, and examination item information of a sample user, the sample data input into the data push model is first vectorially represented, and the user information, medical visit results, and examination item information of the sample user in the sample data are converted into vectors that can be recognized by the data push model.

[0063] Specifically, as shown in Figure 4 below, based on the name embedding layer, type embedding layer, and medical visit order embedding layer of the initial push model, mapping the sample user information, historical status parameters, and historical examination item information into target vector data includes steps S301 to S302:

[0064] Step S301: Based on the name embedding layer, perform vector representation on the name data in the sample user information, historical status parameters, and historical examination item information to obtain name vector data;

[0065] Step S302: Based on the type embedding layer, perform vector representation on the types of the sample user information, historical status parameters, and historical examination item information to obtain type vector data;

[0066] Step S303: Based on the medical visit order embedding layer, perform vector representation on the sample user information, historical status parameters, and historical examination item information in the order of medical visit time to obtain medical visit order vector data;

[0067] Step S304: Sum the name vector data, type vector data, and visit order vector data to generate the target vector data.

[0068] Specifically, in the embodiments of the present application, the data push model consists of an input module, a main body module, and an output model. Among them, the input module is represented by three embedding layers, and the three embedding layers are respectively the name embedding layer E name , type embedding layer E type , and visit order embedding layer E seq .

[0069] In some embodiments, each embedding layer of the input module may consist of a fixed number of n tokens. Exemplarily, each embedding layer consists of 512 fixed tokens, and each token is represented by a vector.

[0070] Map the sample data of the sample user, that is, the visit data, to the name embedding layer, type embedding layer, and visit order embedding layer for representation, where:

[0071] (1) Name embedding layer E name : Represent the names in the basic information, diagnosis results, and inspection item list of each input visit data as vectors. When representing as vectors, first perform initialization and then update during the training process of the model.

[0072] Exemplarily, the name embedding layer represents the age, gender, and inspection items in each data as vectors. Among them, the age is used to segment the age, and each segment is represented as a vector; the gender is used to represent three vectors corresponding to male, female, and unknown; each inspection item corresponds to a vector representation.

[0073] In some embodiments, each name in each data corresponds to each token of the input module in order for vector representation.

[0074] Specifically, discretize the age, and each age group corresponds to a vector representation.

[0075] For example: Divide the age into the following 6 age groups based on medical knowledge, and each age group is represented by a vector. The dimension of each vector can be 128 dimensions, and each vector is initialized to obtain the vector representation of each age group after initialization:

[0076] Age group [0,1): For example, it corresponds to vector a1 with a dimension of 128 dimensions;

[0077] Age group [1,6): For example, it corresponds to vector a2 with a dimension of 128 dimensions;

[0078] Age range [6, 14): For example, it corresponds to vector a3 with a dimension of 128;

[0079] Age range [14, 45): For example, it corresponds to vector a4 with a dimension of 128;

[0080] Age range [45, 65): For example, it corresponds to vector a5 with a dimension of 128;

[0081] Age range [65, +∞): For example, it corresponds to vector a6 with a dimension of 128.

[0082] Among them, the method of initializing each vector is to sample from a Gaussian distribution, and the initialized vector representation is used for model training. During the model training process, the vectors corresponding to each age range will be learned and updated.

[0083] Exemplarily, the gender is discretized, and each gender feature corresponds to a vector, and the dimension of each vector can be 128. The vector representation corresponding to each gender feature is:

[0084] Female gender: For example, it corresponds to vector s1 with a dimension of 128;

[0085] Male gender: For example, it corresponds to vector s2 with a dimension of 128;

[0086] Unknown gender: For example, it corresponds to vector s3 with a dimension of 128.

[0087] The vectors corresponding to each gender feature are initialized by sampling from a Gaussian distribution to obtain the vector representation of each initialized gender value. The initialized vector representation is used for model training. During the model training process, the vectors corresponding to each gender value will be learned and updated.

[0088] Exemplarily, the test items are discretized, and each test item corresponds to a vector, and the dimension of each vector can be 128. The vector representation corresponding to each test item is:

[0089] White blood cells: For example, it corresponds to vector c1 with a dimension of 128;

[0090] Glycated hemoglobin: For example, it corresponds to vector c2 with a dimension of 128;

[0091] Red blood cells: For example, it corresponds to vector c3 with a dimension of 128.

[0092] Initialize the vectors corresponding to each inspection item by sampling from a Gaussian distribution to obtain the vector representation of each initialized inspection. Apply the initialized vector representation for model training. During the model training process, the vectors corresponding to each inspection will be learned and updated.

[0093] (2) Type Embedding Layer E type : Represent the data type of each input medical visit data as a vector.

[0094] The said Type Embedding Layer E type The method of performing vector representation is similar to the above-mentioned Name Embedding Layer E name For the vector representations of age, gender, and inspection items, each data type corresponds to a vector, and the dimension of each vector can be 128. Initialize each vector by sampling from a Gaussian distribution to obtain the initialized vector representation of each data type. Apply the initialized vector representation for model training. During the model training process, the vectors corresponding to each data type will be learned and updated.

[0095] In the embodiment of the present application, the said Type Embedding Layer E type Includes three data types, namely the type representing the patient's basic information, the type representing the diagnosis code, and the type representing the inspection items, and perform vector representations for these three types respectively.

[0096] Exemplarily, the vector representations of the three data types of the said Type Embedding Layer E type are as follows:

[0097] The type representing the patient's basic information: For example, the corresponding vector is t1, with a dimension of 128;

[0098] The type representing the inspection items: For example, the corresponding vector is t2, with a dimension of 128;

[0099] The type representing the diagnosis code: For example, the corresponding vector is t3, with a dimension of 128.

[0100] (3) Medical Visit Order Embedding Layer E seq : Since each piece of data is the medical visit data of a patient, therefore, sort the medical visit data of this patient in chronological order, and then assign a vector representation of a specific order to each order.

[0101] Different from the above-mentioned Name Embedding Layer E's name vector representations of age, gender, and inspection items, when the Medical Visit Order Embedding Layer E seq represents the vector representation of the sequence order, the vector representation corresponding to each position is fixed, that is: the vectors corresponding to each position in all sequences are the same.

[0102] For example, the vector corresponding to position 1 is p1, with a dimension of 128; the vector corresponding to position 2 is p2, with a dimension of 128; and so on.

[0103] Specifically, the vector representing the sequence is obtained from the sine function, and the vector representation of the corresponding position is obtained by corresponding to different frequencies through the absolute position and relative position of each element in the sequence (i.e., i = 1, 2,...). During the model training process, the position vector remains unchanged and does not change with model training.

[0104] When converting the medical visit data into vector-form data, each piece of data (i.e., the multiple medical visit data of a patient) is mapped to the vector representations of the name embedding layer E name , type embedding layer E type , and medical visit order embedding layer E seq . The vector sum of the name embedding layer, type embedding layer, and medical visit order embedding layer obtains the vector-form data of the input embedding layer.

[0105] In some embodiments, the input embedding layer is: E input = E name + E type + E seq , that is, the vectors of the corresponding tokens of each embedding layer are added together;

[0106] Among them, E input represents the input embedding layer, E name represents the name embedding layer, E type represents the type embedding layer, and E seq represents the medical visit order embedding layer.

[0107] In some embodiments, when learning the correlation relationship between user information, medical visit results, and examination item information in the vector-form data based on the main module of the data push model, that is, when learning the correlation relationship between user information, status parameters, and examination item information, the input embedding layer E input is input into the main module of the model. Among them, the input embedding layer E input is composed of n m-dimensional vectors; preferably, the input embedding layer E input is composed of 512 m-dimensional vectors.

[0108] In the embodiments of the present application, based on the deep learning network structure of the main module of the data push model, the relationship between the input sample data is learned.

[0109] Exemplarily, the main module of the data push model may be a k-layer bidirectional encoder representation deep learning network structure based on transformers. Each layer consists of n transformers (corresponding to each token), and the n transformers in each layer are bidirectionally connected to each transformer in the previous layer. The relationship between user information, medical treatment results, and examination item information in the input data is learned through the main module.

[0110] In an embodiment of the present application, the input of the main module of the data push model is an input embedding layer, and the input embedding layer is E input = E name + E type + E seq , where E input represents the input embedding layer, E name represents the name embedding layer, E type represents the type embedding layer, and E seq represents the medical treatment order embedding layer.

[0111] Exemplarily, the input embedding layer E input can be a 512-dimensional vector group (v1, v2, v3,..., v512), where each vector is, for example, 128-dimensional. For example, the v1 vector is represented as (w1, w2,..., w128). Each vector, that is, the token representation, represents age, gender, and examination items.

[0112] Exemplarily, the main module may be a 12-layer bidirectional encoder representation deep learning network structure based on transformers. Each layer consists of 512 transformers, each transformer corresponding to each token. The 512 transformers in each layer are connected to each transformer in the previous layer, that is, bidirectionally connected. Through the complex and delicate model structure of the data push model, the relationship between the input age, gender, and examination items is learned.

[0113] In some embodiments, when learning the association relationship between user information, status parameters, and examination item information, the pre-training task of the data push model is set as the mask task, and this task is unsupervised.

[0114] Specifically, randomly mask t% of the vectors in the input embedding layer of the model. t can be set to 15%. Predict these masked vectors in the last layer of the main module of the model, where each token in the input embedding layer corresponds one-to-one with each token in the last layer of the main module.

[0115] Through the mask task, the vector representation of the input embedding layer of the model and the parameters in the main module of the model are learned to pre-train the model, and the association relationship between user information, status parameters, and examination item information is learned.

[0116] Among them, the vector representation of the input embedding layer of the model, that is, the name embedding layer E name , Type embedding layer E type , and the visit sequence embedding layer E seq The vector representation of each part in .

[0117] In some embodiments, the network structure of the main module of the data push model is used to learn the association between the input data. For example, 512 m-dimensional vectors, i.e., tokens, are input, and each token represents the patient's age, gender, and test items. When the main module of the data push model learns the relationship between the input data, an unsupervised learning method is used, that is, the input and output results are the same.

[0118] In the embodiment of the present application, the input embedding layer is input into the model main body module, and the output of the model main body module is set as the name embedding layer E corresponding to 512 data in the pre-training task. name During the model training process, the main module of the data push model learns the relationship between the 512 tokens and the probability of the user information, status parameters and inspection item information appearing together, that is, the probability of the basic information, diagnosis results and inspections appearing together. The model parameters are trained by massive sample data, so that the model can learn age, gender and inspection tasks to better learn network parameters.

[0119] It should be noted that the mask task is to construct a supervised task by hiding some output tokens and predicting the hidden tokens through all output tokens and the network. The actual input and output are known, and the relationship between age, gender, and test examination of the pre-trained data push model is obtained through the mask task.

[0120] In some embodiments, after the input module and the main module of the pre-trained model, the data push model is trained. A d-layer fully connected network is connected at the first token of the last layer of the main module. Assuming that there are a total of j inspection items, the last layer of the fully connected network is j-dimensional, and a sigmoid layer is connected thereafter for multi-classification tasks. The output module of the model is composed of the above network structure.

[0121] It should be noted that during training, each sample data is organized into the following format: each sample data contains f visits, the first f-1 visits data are input into the data push model, and the output of the data push model is the test and examination list of the fth visit.

[0122] Specifically, the output of the model is a j-dimensional vector, where each dimension takes a value of 0 or 1. A value of 1 indicates the presence of the test item corresponding to that dimension, and a value of 0 indicates its absence. The training of the test data push model is supervised.

[0123] When applying the trained data push model, for each patient, the patient's visit data is input into the data push model in sequence, and the model outputs a recommended list of tests and examinations that should be performed for the current visiting patient.

[0124] In some embodiments, the target user information and target status parameters of the target user are obtained, that is, the visit data of the patient is obtained. The visit data includes the current visit data of the visiting patient, and the current visit data includes user information and visit results. The visit data of the patient also includes the historical visit data of the target user, and the historical visit data includes the user information, visit results, and examination items of the target user's historical visits.

[0125] In this application, the historical visit data, that is: the relevant user information, relevant status parameters, and relevant examination item information of the target user; in the embodiments of this application, the user information of the target user's historical visit is the relevant user information of the target user, the visit result in the historical visit data is the relevant status parameter of the target user, and the examination item in the historical visit data is the relevant examination item information of the target user.

[0126] In some embodiments, as Figure 5 shown, determining the data push model based on the trained initial push model specifically includes steps S111 to S112.

[0127] S111. Obtain the relevant user information, relevant status parameters, and relevant examination item information of the target user;

[0128] S112. Based on the relevant user information, relevant status parameters, and relevant examination item information, perform iterative training on the trained initial push model to obtain an updated initial push model as the data push model.

[0129] In the embodiments of this application, as Figure 6 shown, performing iterative training on the trained initial push model based on the relevant user information, relevant status parameters, and relevant examination item information includes steps S1101 to S1102:

[0130] Step S1101. Iteratively update the sample data based on the relevant user information, relevant status parameters, and relevant examination item information, and mark the true examination item labels corresponding to the relevant examination item information;

[0131] Step S1102: Update the iterative sample data and the true inspection item labels, and train the initial push model to obtain a corrected initial push model. When the historical medical data of the target user is obtained, based on the completion of the training of the data push model, the historical medical data of the target user can be used as the sample data, and the inspection items corresponding to the actual medical treatment of the current target user can be used as the true inspection item labels. The model is trained with the historical medical data of the current target user. The user information, medical treatment results, and inspection items are all personalized based on the medical treatment data of the target user itself, and a more personalized data push model can be obtained. The current medical treatment data of the target user is processed by the corrected data push model, and the predicted result is closer to the actual situation of the target user.

[0132] Please refer to Figure 7 , Figure 7 which is a schematic system block diagram of an inspection item push system 400 provided by an embodiment of the present application. As Figure 7 shown, the inspection item push system 400 includes:

[0133] A data acquisition module 401, configured to acquire the target user information and target status parameters of the target user;

[0134] An inspection item determination module 402, configured to determine the associated relationship of the target inspection item information corresponding to the target user information and target status parameters based on the associated relationship between the user information, status parameters, and inspection item information learned by the preset data push model.

[0135] In some embodiments, when the data acquisition module 401 acquires the target user information and target status parameters of the target user, the target user information and target status parameters may be the medical treatment data of the patient. The medical treatment data includes the user information and medical treatment results of the current medical treatment data of the patient, and the user information of the medical treatment patient and the medical treatment results diagnosed by the doctor according to the current symptoms of the patient are input into the computer device by the doctor or other medical staff.

[0136] In some embodiments, the user information of each medical treatment patient includes, but is not limited to, data such as the name, age, and gender of each patient for characterizing the basic information of the patient. The medical treatment results of each medical treatment patient include, but are not limited to, disease name information and disease symptom information.

[0137] In some embodiments, when the data acquisition module 401 acquires the target user information and target status parameters of the target user, the target user information and target status parameters are the medical treatment data of the patient seeking medical treatment. The medical treatment data includes not only the current medical treatment data of the target user, but also the historical medical treatment data of the target user. Among them, the current medical treatment data includes user information and treatment results, and the historical medical treatment data includes the user information, treatment results, and examination items of the target user's historical medical treatment.

[0138] In some embodiments, the examination item determination module 402 processes the acquired medical treatment data of the target user based on the data push model, and according to the correlation relationship learned by the data push model among the user information, treatment results, and examination item information, that is, learns the correlation relationship among the user information, status parameters, and examination item information, and filters out the target examination item information suitable for the target user.

[0139] In some embodiments, in the learning stage of the correlation relationship among the user information, status parameters, and examination item information, the examination item determination module 402 uses the basic information and diagnosis results of the patient in the sample data as the medical treatment sample data for this medical treatment, and uses the examination item information for this medical treatment as the true examination item information label corresponding to the medical treatment sample data, and inputs them into the data push model established by applying deep learning technology to learn the relationship among the user information, treatment results, and examination item information.

[0140] In some embodiments, the establishment of the correlation relationship among the user information, treatment results, and examination item information specifically includes:

[0141] Based on the embedding layer module of the data push model, convert the medical treatment data into vector-form data; based on the main body module of the data push model, learn the correlation relationship among the user information, treatment results, and examination item information in the vector-form data.

[0142] In some embodiments, when the medical treatment data input to the pre-trained data push model includes the current medical treatment data and historical medical treatment data of the target user, the examination item determination module 402 for examination item push further includes:

[0143] Based on the historical medical treatment data of the target user, iteratively update the sample data and the corresponding true examination item labels, and train the data push model to obtain a trained and corrected data push model; based on the corrected data push model, process the current medical treatment data of the target user, obtain the examination item information corresponding to the current medical treatment data of the target user, and push the examination item information to the target user.

[0144] Please refer to Figure 8 ,Figure 8 It is a schematic block diagram of a computer device provided by an embodiment of the present application. As Figure 8 shown, the computer device 500 includes one or more processors 501 and a memory 502. The processor 501 and the memory 502 are connected through a bus, and the bus is, for example, an I2C (Inter-integrated Circuit) bus.

[0145] Among them, one or more processors 501 work alone or jointly to execute the steps of the inspection item pushing method provided by the above embodiment.

[0146] Specifically, the processor 501 can be a microcontroller unit (MCU), a central processing unit (CPU), a digital signal processor (DSP), etc.

[0147] Specifically, the memory 502 can be a Flash chip, a read-only memory (ROM), a magnetic disk, an optical disc, a USB flash drive, a mobile hard disk, etc.

[0148] Among them, the processor 501 is used to run a computer program stored in the memory 502 and implement the steps of the inspection item pushing method provided by the above embodiment when executing the computer program.

[0149] Exemplarily, the processor 501 is used to run a computer program stored in the memory 502 and, when executing the computer program, implement the following steps:

[0150] Obtain the target user information and target status parameters of the target user;

[0151] Based on the association relationship between the user information, status parameters, and inspection item information learned by the preset data push model, determine the target inspection item information corresponding to the target user information and the target status parameters, and push the target inspection item information association relationship to the target user.

[0152] In some embodiments, before determining the target inspection item information corresponding to the target user information and the target status parameter based on the association relationship among the user information, status parameter, and inspection item information learned by the preset data push model, it includes: obtaining sample data, where the sample data includes the sample user information of the sample user, the historical status parameter corresponding to the sample user information, and the historical inspection item information corresponding to the sample user information; training a preset initial push model based on the sample data to determine the probability of the co-occurrence of the user information, status parameter, and inspection item information as the association relationship among the user information, status parameter, and inspection item information; and determining the data push model based on the trained initial push model.

[0153] In some embodiments, training the preset initial push model based on the sample data to determine the probability of the co-occurrence of the user information, status parameter, and inspection item information includes: mapping the sample user information, historical status parameter, and historical inspection item information into target vector data based on the name embedding layer, type embedding layer, and visit order embedding layer of the initial push model; and determining the probability of the co-occurrence of the user information, status parameter, and inspection item information based on the main body module of the initial push model and the target vector data.

[0154] In some embodiments, mapping the sample user information, historical status parameter, and historical inspection item information into target vector data based on the name embedding layer, type embedding layer, and visit order embedding layer of the initial push model includes: performing vector representation on the name data in the sample user information, historical status parameter, and historical inspection item information based on the name embedding layer to obtain name vector data; performing vector representation on the types of the sample user information, historical status parameter, and historical inspection item information based on the type embedding layer to obtain type vector data; performing vector representation on the sample user information, historical status parameter, and historical inspection item information in the order of visit time based on the visit order embedding layer to obtain visit order vector data; and summing the name vector data, type vector data, and visit order vector data to generate the target vector data.

[0155] In some embodiments, training the preset initial push model based on the sample data further includes: sorting the sample data based on the time stamp order corresponding to the sample data, and training the association relationship of the initial push model based on the sorted sample data.

[0156] In some embodiments, determining the data push model based on the trained initial push model includes: obtaining relevant user information, relevant status parameters, and relevant inspection item information of the target user; and iteratively training the trained initial push model based on the relevant user information, relevant status parameters, and relevant inspection item information to obtain an updated initial push model as the association relationship among the user information, status parameters, and inspection item information of the data push model.

[0157] In some embodiments, the iteratively training the trained initial push model based on the relevant user information, relevant status parameters, and relevant inspection item information includes: iteratively updating the sample data based on the relevant user information, relevant status parameters, and relevant inspection item information, and marking the true inspection item labels corresponding to the relevant inspection item information; and training the initial push model according to the iteratively updated sample data and the true inspection item labels to obtain a corrected initial push model association relationship.

[0158] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, which when executed by a processor causes the processor to implement the steps of the inspection item push method provided in the above embodiments:

[0159] Obtaining target user information and target status parameters of a target user;

[0160] Based on the association relationship among the user information, status parameters, and inspection item information learned by a preset data push model, determining target inspection item information corresponding to the target user information and the target status parameters, and pushing the target inspection item information to the target user.

[0161] In some embodiments, before determining the target inspection item information corresponding to the target user information and the target status parameters based on the association relationship among the user information, status parameters, and inspection item information learned by a preset data push model, it includes: obtaining sample data, where the sample data includes sample user information of a sample user, historical status parameters corresponding to the sample user information, and historical inspection item information corresponding to the sample user information; training a preset initial push model based on the sample data to determine the probability of co-occurrence of the user information, status parameters, and inspection item information as the association relationship among the user information, status parameters, and inspection item information; and determining the data push model based on the trained initial push model.

[0162] In some embodiments, training the preset initial push model based on the sample data to determine the co-occurrence probability of the user information, status parameters, and examination item information includes: mapping the sample user information, historical status parameters, and historical examination item information into target vector data based on the name embedding layer, type embedding layer, and visit order embedding layer of the initial push model; and determining the co-occurrence probability of the user information, status parameters, and examination item information based on the main module of the initial push model and the target vector data.

[0163] In some embodiments, mapping the sample user information, historical status parameters, and historical examination item information into target vector data based on the name embedding layer, type embedding layer, and visit order embedding layer of the initial push model includes: performing vector representation on the name data in the sample user information, historical status parameters, and historical examination item information based on the name embedding layer to obtain name vector data; performing vector representation on the types of the sample user information, historical status parameters, and historical examination item information based on the type embedding layer to obtain type vector data; performing vector representation on the sample user information, historical status parameters, and historical examination item information in the order of visit time based on the visit order embedding layer to obtain visit order vector data; and summing the name vector data, type vector data, and visit order vector data to generate the target vector data.

[0164] In some embodiments, training the preset initial push model based on the sample data further includes: sorting the sample data based on the time stamp order corresponding to the sample data, and training the initial push model based on the sorted sample data.

[0165] In some embodiments, determining the data push model based on the trained initial push model includes: obtaining the relevant user information, relevant status parameters, and relevant examination item information of the target user; and performing iterative training on the trained initial push model based on the relevant user information, relevant status parameters, and relevant examination item information to obtain an updated initial push model as the data push model.

[0166] In some embodiments, performing iterative training on the trained initial push model based on the relevant user information, relevant status parameters, and relevant examination item information includes: iteratively updating the sample data based on the relevant user information, relevant status parameters, and relevant examination item information, and marking the true examination item label corresponding to the relevant examination item information; and training the initial push model based on the iteratively updated sample data and the true examination item label to obtain a corrected initial push model association relationship.

[0167] Among them, the computer-readable storage medium may be an internal storage unit of the computer device described in any of the foregoing embodiments, such as the hard disk or memory of the terminal device. The computer-readable storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the terminal device.

[0168] As described above, the above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for pushing inspection items, characterized in that, the method for pushing inspection items includes: Obtaining the target user information and target status parameters of the target user; Based on the association relationship between user information, status parameters, and inspection item information learned by a preset data push model, determining the target inspection item information corresponding to the target user information and the target status parameters, and pushing the target inspection item information to the target user; wherein, the sample data pre-trained for the data push model is composed of multiple medical visit data of the user arranged in chronological order; wherein, the data push model is composed of an input module, a main module, and an output model. The input module is represented by three embedding layers, and the three embedding layers are respectively a name embedding layer, a type embedding layer, and a medical visit order embedding layer; Performing vector representation on the basic information, diagnosis result, and name in the inspection item list of each input medical visit data as the vector of the name embedding layer; Performing vector representation on the data type of each input medical visit data as the vector of the type embedding layer; Sorting the medical visit data in chronological order, and specifying a specific order for each order for vector representation as the vector of the medical visit order embedding layer; Based on the vector summation of the name embedding layer, the type embedding layer, and the medical visit order embedding layer, obtaining the vector of the input embedding layer; Inputting the vector of the input embedding layer into the main module to learn the association relationship between the user information, status parameters, and inspection item information based on the main module.

2. The method for pushing inspection items according to claim 1, characterized in that, before determining the target inspection item information corresponding to the target user information and the target status parameters based on the association relationship between user information, status parameters, and inspection item information learned by a preset data push model, it includes: Obtaining sample data, where the sample data includes the sample user information of the sample user, the historical status parameters corresponding to the sample user information, and the historical inspection item information corresponding to the sample user information; Based on the sample data, training a preset initial push model to determine the probability of the co-occurrence of the user information, status parameters, and inspection item information as the association relationship between the user information, status parameters, and inspection item information; Based on the trained initial push model, determining the data push model.

3. The method for pushing inspection items according to claim 2, characterized in that, the training of the preset initial push model based on the sample data to determine the probability of the co-occurrence of the user information, status parameters, and inspection item information includes: Based on the name embedding layer, type embedding layer, and medical visit order embedding layer of the initial push model, mapping the sample user information, historical status parameters, and historical inspection item information into target vector data; Based on the main module of the initial push model and the target vector data, determining the probability of the co-occurrence of the user information, status parameters, and inspection item information.

4. The method for pushing inspection items according to claim 3, characterized in that, The name embedding layer, type embedding layer, and visit order embedding layer based on the initial push model map the sample user information, historical status parameters, and historical examination item information into target vector data, including: Performing vector representation on the name data in the sample user information, historical status parameters, and historical examination item information based on the name embedding layer to obtain name vector data; Performing vector representation on the types of the sample user information, historical status parameters, and historical examination item information based on the type embedding layer to obtain type vector data; Performing vector representation on the sample user information, historical status parameters, and historical examination item information in the order of visit time based on the visit order embedding layer to obtain visit order vector data; Summing the name vector data, type vector data, and visit order vector data to generate the target vector data.

5. The examination item push method according to claim 2, wherein, The training of the preset initial push model based on the sample data further includes: Sorting the sample data based on the time stamp order corresponding to the sample data, and training the initial push model based on the sorted sample data.

6. The examination item push method according to claim 2, wherein, Determining the data push model based on the trained initial push model includes: Obtaining the relevant user information, relevant status parameters, and relevant examination item information of the target user; Performing iterative training on the trained initial push model based on the relevant user information, relevant status parameters, and relevant examination item information to obtain the updated initial push model as the data push model.

7. The examination item push method according to claim 6, wherein, The iterative training of the trained initial push model based on the relevant user information, relevant status parameters, and relevant examination item information includes: Iteratively updating the sample data based on the relevant user information, relevant status parameters, and relevant examination item information, and marking the true examination item labels corresponding to the relevant examination item information; Training the initial push model according to the iteratively updated sample data and true examination item labels to obtain the corrected initial push model.

8. An examination item push system, wherein, The examination item push system is used to execute the examination item push method according to any one of claims 1-7, and the push system includes: A data acquisition module for acquiring the target user information and target status parameters of the target user; An examination item determination module for determining the target examination item information corresponding to the target user information and target status parameters based on the association relationship between the user information, status parameters, and examination item information learned by the preset data push model; Among them, the sample data for pre-training the data push model is composed of the multiple visit data of the user arranged in chronological order, and the correlation relationship among the user information, status parameters, and examination item information is the probability that the user information, status parameters, and examination item information appear together; Among them, the data push model is composed of an input module, a main body module, and an output model. The input module is represented by three embedding layers, and the three embedding layers are respectively a name embedding layer, a type embedding layer, and a visit order embedding layer; Perform vector representation on the basic information, diagnosis result, and name in the examination item list of each input visit data as the vector of the name embedding layer; Perform vector representation on the data type of each input visit data as the vector of the type embedding layer; Sort the visit data in chronological order, and specify a specific order for each order for vector representation as the vector of the visit order embedding layer; Based on the summation of the vectors of the name embedding layer, type embedding layer, and visit order embedding layer, obtain the vector of the input embedding layer; Input the vector of the input embedding layer into the main body module to learn the correlation relationship among the user information, status parameters, and examination item information based on the main body module.

9. A computer device, characterized in that, the computer device includes: a memory and a processor; Among them, the memory is connected to the processor for storing programs; The processor is used to implement the steps of the examination item push method according to any one of claims 1-7 by running the program stored in the memory.

10. A computer-readable storage medium, characterized in that, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to implement the steps of the examination item push method according to any one of claims 1-7.

Images