WO2021151327A1 - 分诊数据处理方法、装置、设备及介质 - Google Patents

分诊数据处理方法、装置、设备及介质 Download PDF

Info

Publication number
WO2021151327A1
WO2021151327A1 PCT/CN2020/124220 CN2020124220W WO2021151327A1 WO 2021151327 A1 WO2021151327 A1 WO 2021151327A1 CN 2020124220 W CN2020124220 W CN 2020124220W WO 2021151327 A1 WO2021151327 A1 WO 2021151327A1
Authority
WO
WIPO (PCT)
Prior art keywords
triage
term
model
short
patient
Prior art date
Application number
PCT/CN2020/124220
Other languages
English (en)
French (fr)
Inventor
赵惟
徐卓扬
左磊
赵婷婷
孙行智
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021151327A1 publication Critical patent/WO2021151327A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • This application relates to the field of data processing of big data, and in particular to a method, device, equipment, and medium for processing triage data.
  • This application provides a triage data processing method, device, computer equipment, and storage medium, which realize the recognition of the patient's diagnosis data through a short-term triage knowledge model based on decision trees and a long-term triage model based on deep reinforcement learning.
  • the final fusion output triage results, this application is suitable for smart medical and other fields, can further promote the construction of smart cities, can realize the rapid and accurate automatic triage of patients, save patient time, improve the accuracy of medical treatment, and improve The patient experience.
  • a method for processing triage data including:
  • the medical visit data includes a patient identification code
  • the final triage result of the patient is determined and output.
  • a triage data processing device including:
  • the receiving module is used to receive a triage request containing the patient's medical visit data; the medical visit data includes a patient identification code;
  • An acquisition module configured to acquire historical medical visit information associated with the patient identification code, and determine the historical medical visit information and the medical visit data as the patient's data to be triaged;
  • the dividing module is used to input the data to be triaged into a short-term triage knowledge model based on a decision tree, and group the data to be triaged through the short-term triage knowledge model to obtain patient group results and short-term classification Diagnosis result
  • the matching module is used to obtain a long-term triage model based on deep reinforcement learning that matches the results of the patient group;
  • the prediction module is used for predicting the medical visit data through the long-term triage model to obtain the long-term triage result
  • the output module is used to determine and output the final triage result of the patient according to the short-term triage result and the long-term triage result.
  • a computer device includes a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor, and when the processor executes the computer-readable instructions, the following steps are implemented: A triage request containing the patient's medical visit data; the medical visit data includes a patient identification code;
  • the final triage result of the patient is determined and output.
  • One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps:
  • the medical visit data includes a patient identification code
  • the final triage result of the patient is determined and output.
  • the triage data processing method, device, computer equipment, and storage medium provided in this application receive a triage request containing the patient’s visit data; obtain the historical visit information associated with the patient identification code in the visit data, and then The historical visit information and the visit data are determined to be the patient’s data to be triaged; the data to be triaged are input into a short-term triage knowledge model based on a decision tree, and the short-term triage knowledge model is used to compare the
  • the data to be triaged is divided into groups to obtain patient group results and short-term triage results; a long-term triage model based on deep reinforcement learning that matches the patient group results is obtained; the treatment data is obtained through the long-term triage model Make predictions to obtain long-term triage results; according to the short-term triage results and the long-term triage results, the final triage results of the patient are determined and output.
  • the patient’s historical visit information is obtained, and the decision-based The tree's short-term triage knowledge model divides the patient groups and short-term triage results corresponding to the patients, and then matches the patient groups to a long-term triage model based on deep reinforcement learning, predicts the long-term triage results, and integrates the short-term triage results And the long-term triage results determine the final triage results. Therefore, it realizes the combination of the patient’s historical visit information to extract patient characteristics, and the short-term triage knowledge model based on decision trees and the long-term triage model based on deep reinforcement learning are used to analyze the patients. The diagnosis data is identified, and the triage results are finally merged and output, which can realize the automatic triage of patients quickly and accurately, which saves patients' time, improves the accuracy of diagnosis, and improves the patient experience.
  • Fig. 1 is a schematic diagram of an application environment of a triage data processing method in an embodiment of the present application
  • FIG. 2 is a flowchart of a method for processing triage data in an embodiment of the present application
  • step S30 of the triage data processing method in an embodiment of the present application is a flowchart of step S30 of the triage data processing method in an embodiment of the present application
  • step S302 of the triage data processing method in an embodiment of the present application is a flowchart of step S302 of the triage data processing method in an embodiment of the present application
  • FIG. 5 is a flowchart of step S303 of the triage data processing method in an embodiment of the present application.
  • Fig. 6 is a flowchart of step S40 of the triage data processing method in an embodiment of the present application.
  • FIG. 7 is a flowchart of step S403 of the triage data processing method in an embodiment of the present application.
  • FIG. 8 is a flowchart of step S60 of the triage data processing method in an embodiment of the present application.
  • Fig. 9 is a functional block diagram of a triage data processing device in an embodiment of the present application.
  • Fig. 10 is a schematic diagram of a computer device in an embodiment of the present application.
  • the triage data processing method provided in this application can be applied in the application environment as shown in Fig. 1, in which the client (computer equipment) communicates with the server through the network.
  • the client includes, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, cameras, and portable wearable devices.
  • the server can be implemented as an independent server or a server cluster composed of multiple servers.
  • a method for processing triage data is provided, and the technical solution mainly includes the following steps S10-S60:
  • a triage request containing the patient's medical visit data is received; the medical visit data includes a patient identification code.
  • the medical consultation data is the current medical consultation information entered by the patient after logging in and reviewing on the application platform.
  • the medical consultation data can be obtained after the patient enters text on the application platform, or can be obtained by confirming the patient’s voice input on the application platform after converting the patient’s input into text.
  • the medical consultation data includes the patient identification code and the patient Basic information, the patient identification code is a unique code for each patient, and the patient logs in to the application platform through the patient identification code, for example, the patient identification code is the patient’s ID number, social security number, etc., the patient’s basic information Including the patient’s gender, age, work occupation and other basic personal information.
  • S20 Obtain historical medical visit information associated with the patient identification code, and determine the historical medical visit information and the medical visit data as the patient's data to be triaged.
  • the historical medical visit information is information related to the patient's medical visits in the past, including historical diagnosis results, disease history, and medication history , Inspection and inspection results (for example: annual health check) and risk prediction results, etc., and mark the historical visit information and the visit data together as the data to be triaged, wherein the data to be triaged can be
  • the historical medical visit information and the medical visit data are differentiated by adding different weights, and the medical visit data is enhanced, so as to improve the accuracy of subsequent triage.
  • S30 Input the data to be triaged into a short-term triage knowledge model based on a decision tree, and group the data to be triaged through the short-term triage knowledge model to obtain patient group results and short-term triage results.
  • the data to be triaged is input into the short-term triage knowledge model
  • the short-term triage knowledge model is a trained knowledge decision tree model
  • the structure of the short-term triage knowledge model is based on a decision tree
  • the tree structure is constructed, and the short-term triage knowledge model is used to group the data to be triaged, and the group is divided into triage feature extraction of the data to be triaged, and the extracted triage Characteristic decision-making, from the root node down to the next level of internal nodes, until there is a processing process of a leaf node that cannot be further divided, and the leaf node that cannot be further divided is determined as the short-term corresponding to the patient
  • the triage result, the short-term triage result includes the department category and its probability value output by the short-term triage knowledge model, each of the nodes is associated with a group feature, and the nodes include a root node, an internal node, and a leaf Nodes, each time the division of a node indicates that a patient is divided
  • the triage characteristics include group characteristics and symptom characteristics.
  • the group characteristics are characteristics related to the category of the patient group presented by the patient, and the patient group is divided into different groups according to the different characteristics of the patient, such as The characteristics of the group, such as age, frequency of illness, gender, potential risk, etc., divide the patient group into the elderly and more susceptible men, the children who are growing up suddenly, the middle-aged and stable resistant women, etc.; the symptom characteristics
  • the characteristics of the group such as age, frequency of illness, gender, potential risk, etc., divide the patient group into the elderly and more susceptible men, the children who are growing up suddenly, the middle-aged and stable resistant women, etc.
  • the symptom characteristics Features related to medical conditions, such as skin features, orthopedic features, ophthalmological features, otological features, childhood flu features, respiratory features, etc.
  • step S30 before step S30, that is, before inputting the data to be triaged into a short-term triage knowledge model based on a decision tree, the method includes:
  • S301 Obtain a short-term medical visit sample set; wherein the short-term medical visit sample set includes a plurality of short-term medical visit samples, and one short-term medical visit sample is associated with a department label.
  • the short-term consultation sample set is the collection of the collected short-term consultation samples
  • the short-term consultation sample is the collected patient history and the input data of the consultation that has completed triage
  • one of the short-term consultation samples is related to One of the triage labels is associated
  • the department label is the department in which the short-term consultation sample corresponding to it was finally diagnosed in the actual visit, and the department is the various departments included in the hospital.
  • S302 Input the short-term consultation samples into a preset knowledge decision tree model; the knowledge decision tree model includes characteristic parameters of each node.
  • the short-term consultation samples are input to a preset knowledge decision tree model
  • the knowledge decision tree model is a model of a decision tree structure
  • the knowledge decision tree model includes a plurality of nodes of the decision tree structure, wherein Each node contains a node feature parameter, which is an attribute parameter of the selected feature in the division classification. For example, if the node feature parameter is set to an attribute parameter greater than 50 years old, the short-term consultation sample is divided , Whether it has the characteristics of more than 50 years old.
  • the method before step S302, that is, before inputting the short-term consultation samples into the preset knowledge decision tree model, the method includes:
  • the data in the clinical guideline knowledge is acquired, and the clinical guideline knowledge is data summarized through the clinical guideline and expert consensus combing and treating rules.
  • the symptom classification, symptom name, symptom phenomenon, and department classification in the clinical guideline knowledge are taken as entities in the diagnosis knowledge map, and the relationship between the entities is constructed according to the clinical guideline knowledge, and the relationship between the entities is constructed according to the knowledge map
  • the top-down model layer (entity-relation-entity) construction method establishes the medical knowledge graph.
  • the established medical knowledge graph is constructed according to a top-down tree structure, and each entity in the medical knowledge graph is converted into a node that has the triage characteristics of the entity, and each entity The relationship between them is converted into the division conditions of decision-making, thereby constructing the knowledge decision tree model.
  • This application realizes that by acquiring clinical guideline knowledge, establishing a medical knowledge graph based on the clinical guideline knowledge; constructing the knowledge decision tree model according to the medical knowledge graph, and being able to construct the clinical guideline knowledge into a medical knowledge graph using the construction method of the knowledge graph , And transform the medical knowledge graph to generate nodes, thereby constructing a knowledge decision tree model, simplifying the construction process of the knowledge decision tree, shortening the construction time, and improving the classification accuracy and reliability of the knowledge decision tree model.
  • S303 Perform group division of the short-term consultation samples through the knowledge decision tree model to obtain sample group results and sample triage results.
  • the knowledge decision tree model divides the short-term consultation samples into groups, and the group division further includes extracting triage features of the short-term consultation samples, and making decisions on the extracted triage features, from The root node is continuously divided down to the next level of internal nodes, until a leaf node that cannot be further divided occurs, and the leaf node that cannot be further divided is determined to be the sample triage result corresponding to the short-term consultation sample ,
  • the sample triage result is each department category and its probability value (also called OR value, that is, the odds ratio corresponding to the logistic regression coefficient) output by the knowledge decision tree model, and each node is associated with a group feature
  • the nodes include root nodes, internal nodes, and leaf nodes. Each time a node is divided, it indicates that a group characteristic is divided for the short-term consultation sample, and the group characteristics associated with all the nodes are summarized. So as to determine the result of the sample population.
  • the step S303 that is, the feature decision of the short-term consultation sample through the knowledge decision tree model, to obtain the sample group result and the sample triage result, includes:
  • S3031 Perform feature division and decision-making on the short-term consultation samples through a decision tree classification method and a recursive partition method to obtain the sample triage results and node path results; the node path results are the results of the knowledge decision tree model The path constituted by the nodes through which the short-term consultation samples make decisions; wherein, each of the nodes is associated with a group feature.
  • the decision tree classification method is a tree structure for classifying the short-term consultation samples, and feature selection is performed on a certain feature of the short-term consultation samples, and the short-term consultation samples are classified according to the result of the feature division. Assign to its next internal node (also called a branch child node); wherein the recursive partitioning method is used in the top-down process, and the recursive partitioning method establishes a model for each node to be divided , Dividing the short-term consultation sample set into different subsets, so that the distribution difference between the subsets is the largest, and the subsets correspond to the features in the decision-making process so as to improve the effectiveness of feature selection.
  • next internal node also called a branch child node
  • the recursive partitioning method establishes a model for each node to be divided , Dividing the short-term consultation sample set into different subsets, so that the distribution difference between the subsets is the largest, and the subsets correspond to the features in the decision-
  • the feature division is that the feature selected from the subset divided by the recursive partitioning method is matched with the feature of the short-term consultation sample, thereby dividing a group feature, and the decision is to allocate according to the result of the feature division
  • the short-term consultation sample will form a top-down path in the knowledge decision tree model to obtain the node path result, and finally
  • the short-term medical consultation sample arrives at a leaf node at the bottom layer, and the leaf node is the sample triage result corresponding to the short-term medical consultation sample.
  • S3032 Acquire group characteristics associated with each node in the node path result, and determine all the acquired group characteristics as the sample group result.
  • the feature selection of the subset of each node in the node path result selects the corresponding group feature, for example: the short-term consultation sample is "60-year-old XXX feels weak legs and feet, dizzy", and performs the feature through the node associated with the age feature
  • the short-term consultation sample is "60-year-old XXX feels weak legs and feet, dizzy"
  • the age characteristics of the elderly are determined as one of the group characteristics of the short-term medical sample , Marked as one of the population characteristics in the sample population results corresponding to the short-term medical sample.
  • the present application realizes the feature division and decision-making of the short-term consultation samples through the decision tree classification method and the recursive partition method to obtain the sample triage result and the node path result; obtain the association with each node in the node path result
  • the population characteristics of all the obtained population characteristics are determined as the sample population results.
  • the sample triage results are obtained through the decision tree method and the recursive partition method, and the population characteristics of all passing nodes are obtained to determine
  • the results of the sample group can be subdivided into obvious subsets in the process of decision tree classification, which improves the accuracy and effectiveness of decision-making.
  • the loss value of the decision tree between the sample triage result and the department label is calculated through the loss function of the knowledge decision tree model, and the loss function is preferably a regularized maximum likelihood function.
  • the convergence condition may be a condition that the value of the decision tree loss value is very small and will not drop after 1000 calculations, that is, the value of the decision tree loss value is very high after 1000 calculations.
  • the convergence condition may also be a condition that the loss value of the decision tree is less than a set threshold, That is, when the loss value of the decision tree is less than the set threshold, the training is stopped, and the knowledge decision tree model after convergence is recorded as a short-term triage knowledge model, so that the loss value of the decision tree does not reach the preset
  • the convergence condition is reached, the characteristic parameters of each node in the knowledge decision tree model are continuously adjusted, and the step of feature decision-making on the short-term medical samples through the knowledge decision tree model is triggered, which can continuously move closer to accurate classification and allow recognition The accuracy rate is getting higher and higher.
  • S40 Acquire a long-term triage model based on deep reinforcement learning that matches the result of the patient group.
  • one of the long-term triage models is associated with a group category, and each of the long-term triage models is based on deep reinforcement learning and is obtained after learning through historical category samples of the group category associated with it.
  • the long-term triage model is more targeted, and the patient group category of the patient is determined according to the patient group results, so as to obtain the long-term triage model corresponding to the group category that matches the patient group category of the patient.
  • the long-term triage model is a deep reinforcement learning (DQN, Deep Q Network) model that combines neural network and reinforcement learning.
  • DQN Deep Q Network
  • the long-term triage model is regarded as an agent, and the treatment plan is regarded as an action.
  • the model learns an optimization strategy through experimentation to maximize long-term rewards, that is, the model chooses an action to act on the environment, and the environment changes after the action is accepted, and at the same time produces a reward (reward or punishment) ) Is fed back to the model as an enhanced signal; the model optimizes the strategy of selecting actions based on the enhanced signal, and the optimization direction is to maximize the long-term expected return; under the optimized strategy, the model then selects the next action according to the current state of the environment.
  • the model learns an optimization strategy through experimentation to maximize long-term rewards, that is, the model chooses an action to act on the environment, and the environment changes after the action is accepted, and at the same time produces a reward (reward or punishment) ) Is fed back to the model as an enhanced signal; the model optimizes the strategy of selecting actions based on the enhanced signal, and the optimization direction is to maximize the long-term expected return; under the optimized strategy, the model then selects the next action according to the current state of the environment.
  • step S40 that is, the obtaining a long-term triage model based on deep reinforcement learning that matches the result of the patient group includes:
  • S401 Input the result of the patient population into a patient population classification model.
  • the patient population results are input into the patient population classification model
  • the patient population classification model is a neural network model trained through a clustering algorithm
  • the patient population classification model realizes the Model based on the characteristics of the group.
  • S402 Perform clustering processing on all the group characteristics through the patient group classification model to obtain a patient group category corresponding to the patient group result.
  • the clustering process is to use the K-means clustering algorithm to perform European calculations on all the population characteristics in the patient population results, and determine the patient population according to the range of the cluster clusters that fall into it.
  • the patient group category can be set according to needs, for example, the patient group category is divided into high-age potential risk groups, middle-age risk groups, and so on.
  • the Word2vec model calculate the similarity value between the patient group category and the group category associated with each long-term triage model, and determine the long-term triage model associated with the group category corresponding to the largest similarity value as For the long-term triage model matching the patient group category, the complete set of the patient group category may be the same as or different from the complete set of the group category.
  • the degree of matching is measured by the similarity value, It can increase the flexibility of the long-term triage model and improve the accuracy of recognition.
  • This application realizes that by inputting the results of the patient population into a patient population classification model; performing clustering processing on all the population characteristics through the patient population classification model, to obtain the patient population category corresponding to the patient population result;
  • the long-term triage model matching the patient group category in this way, can scientifically select a suitable long-term triage model, improve the accuracy and reliability of triage, and increase the flexibility of the model.
  • the method includes:
  • S4031 Obtain a historical category sample set; the historical category sample set includes a plurality of historical category samples matching the patient group category, and one historical category sample is associated with one historical department label.
  • the historical category sample set is a collection of the historical category samples, and the historical category samples are historically collected medical input data corresponding to patients matching the patient group category, wherein the matching method may be manual
  • the patients are marked with the patient group category, the patients of the same patient group category are determined as patients matching the patient group category, a sample of the historical category is associated with a historical department label, and the historical department label is the actual visit
  • the department in which the historical category sample corresponding to it was last diagnosed, and the department is a variety of departments included in the hospital.
  • the historical category samples are input into the deep reinforcement learning model
  • the deep reinforcement learning model includes the initial parameters
  • the initial parameters can be set according to requirements, for example, the initial parameters are set to zero.
  • the deep reinforcement learning model predicts the historical category samples to obtain a return result.
  • the offline learning strategy method is to use a large number of historical category samples that have been collected for training and learning, so that there will be no unused decision strategies, and the historical category samples are predicted through the deep reinforcement learning model.
  • the prediction is to sort the available medical treatment plans (determining the category of each department) according to the long-term expected return Q value corresponding to each action performed by the deep reinforcement learning model. The larger the Q value, the larger the expected return value of the medical treatment plan. , When the expected return value is the highest, the prediction is completed, and the final treatment plan (predicted department to go to the doctor) is the result of the return.
  • S4034 Determine a network loss value according to the return result and the historical department label associated with the historical category sample.
  • the loss function of the deep reinforcement learning model is used to calculate the network loss value between the return result and the historical department label associated with the historical category sample, and the loss function is preferably a least squares method. function.
  • the network convergence condition may be a condition that the value of the network loss value is small and will not drop after 2000 calculations, that is, the value of the network loss value is very small after 2000 calculations. And when it will no longer drop, stop training, and record the deep reinforcement learning model after convergence as the long-term triage model matching the patient population category; the network convergence condition may also be the network loss The condition that the value is less than the network setting threshold, that is, when the network loss value is less than the network setting threshold, the training is stopped, and the deep reinforcement learning model after convergence is recorded as the long-term The triage model, in this way, when the network loss value does not reach the preset network convergence condition, the initial parameters in the deep reinforcement learning model are continuously adjusted, and the offline learning strategy is triggered.
  • the step of predicting the historical category samples can continuously move closer to accurate results, so that the accuracy of recognition becomes higher and higher. In this way, through more targeted historical category samples and offline learning strategies, the deep reinforcement learning model can be more accurate, and have stronger generalization and robustness.
  • S50 Predict the medical visit data through the long-term triage model to obtain a long-term triage result.
  • the long-term triage model is used to predict the medical visit data, and the prediction also includes a long-term triage obtained after performing a trained action on the medical visit data as a state according to the long-term triage model.
  • Result that is, the treatment plan obtained after performing the action
  • the long-term triage result is the triage result predicted by patients with the same category for a long time after learning
  • the long-term triage result includes the long-term triage model output The category of the department and its Q value.
  • S60 Determine and output a final triage result of the patient according to the short-term triage result and the long-term triage result.
  • each department category and its probability value in the short-term triage result and each department category and its probability value in the long-term triage result calculating according to the set weight parameter
  • the comprehensive score corresponding to each department category is determined, and the department category with the largest comprehensive score is determined as the final triage result of the patient.
  • the final triage result may also include a certain specialty doctor in the department category, etc.
  • the result data related to triage, among which, the difference of the probability value corresponding to each department category is opened by the weight parameter, so that the classification is more accurate.
  • step S60 that is, determining and outputting the final triage result of the patient according to the short-term triage result and the long-term triage result includes:
  • S601 Input the short-term triage result and the long-term triage result into a comprehensive triage model.
  • each department category and its probability value in the short-term triage result and each department category and its probability value in the long-term triage result are input into the comprehensive branch model.
  • S602 Output multiple comprehensive scores associated with each department category one by one through the comprehensive scoring function in the comprehensive triage model, and determine the department category with the largest comprehensive score as the final triage result of the patient.
  • the comprehensive score is calculated by the comprehensive score function, and the department category corresponding to the largest comprehensive score among all comprehensive scores is determined as the final triage result of the patient, and the comprehensive score function for:
  • i is the i-th department category
  • score(i) is the comprehensive score corresponding to the i-th department category
  • OR(i) is the OR value of the short-term triage results corresponding to the i-th department category
  • w1 is the OR value Q(i) is the Q value of the long-term triage results corresponding to the i-th department category
  • w2 is the weight of the Q value.
  • This application realizes that by receiving a triage request containing the patient's medical treatment data; obtaining historical medical consultation information associated with the patient identification code in the medical consultation data, and determining the historical medical consultation information and the medical consultation data as the patient
  • the data to be triaged input the data to be triaged into a short-term triage knowledge model based on a decision tree, and group the data to be triaged through the short-term triage knowledge model to obtain patient group results and short-term Triage results; obtain a long-term triage model based on deep reinforcement learning that matches the results of the patient population; predict the visit data through the long-term triage model to obtain long-term triage results; according to the short-term triage
  • the diagnosis result and the long-term triage result are determined and the final triage result of the patient is determined and output.
  • the historical information of the patient is obtained, and the patient corresponding to the patient is classified through the short-term triage knowledge model based on the decision tree
  • the group and short-term triage results are matched with the patient groups to obtain a long-term triage model based on deep reinforcement learning, and the long-term triage results are predicted, and the short-term triage results and long-term triage results are combined to determine the final triage results.
  • a triage data processing device is provided, and the triage data processing device corresponds to the triage data processing method in the foregoing embodiment one-to-one.
  • the triage data processing device includes a receiving module 11, an acquiring module 12, a dividing module 13, a matching module 14, a predicting module 15 and an output module 15.
  • the detailed description of each functional module is as follows:
  • the receiving module 11 is configured to receive a triage request containing the patient's medical visit data; the medical visit data includes a patient identification code;
  • the obtaining module 12 is configured to obtain historical medical visit information associated with the patient identification code, and determine the historical medical visit information and the medical visit data as the patient's data to be triaged;
  • the dividing module 13 is used to input the data to be triaged into a short-term triage knowledge model based on a decision tree, and divide the data to be triaged by the short-term triage knowledge model to obtain patient group results and short-term Triage results;
  • the matching module 14 is used to obtain a long-term triage model based on deep reinforcement learning that matches the results of the patient group;
  • the prediction module 15 is used for predicting the visit data through the long-term triage model to obtain a long-term triage result
  • the output module 16 is configured to determine and output the final triage result of the patient according to the short-term triage result and the long-term triage result.
  • Each module in the above-mentioned triage data processing device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 10.
  • the computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a readable storage medium and an internal memory.
  • the readable storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer readable instructions in the readable storage medium.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by the processor to realize a triage data processing method.
  • the readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.
  • a computer device including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor. Diagnostic data processing method.
  • one or more readable storage media storing computer readable instructions are provided.
  • the readable storage media provided in this embodiment include non-volatile readable storage media and volatile readable storage. Medium; the readable storage medium stores computer readable instructions, and when the computer readable instructions are executed by one or more processors, the one or more processors implement the triage data processing method in the above-mentioned embodiment.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Data Mining & Analysis (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

本申请涉及数据处理领域，提供一种分诊数据处理方法、装置、设备及介质，所述方法包括：接收到包含患者的就诊数据的分诊请求；获取与患者标识码关联的历史就诊信息，将历史就诊信息和就诊数据确定为患者的待分诊数据；通过短期分诊知识模型对待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；获取与患者群体结果相匹配的基于深度强化学习的长期分诊模型；通过长期分诊模型对就诊数据进行预测，得到长期分诊结果；根据短期分诊结果和长期分诊结果，确定患者的最终分诊结果并输出。本申请实现了快速地、准确地对患者进行自动分诊，节省了患者时间，提升了就诊准确率。本申请适用于智慧医疗等领域，可进一步推动智慧城市的建设。

Description

分诊数据处理方法、装置、设备及介质
本申请要求于2020年9月9日提交中国专利局、申请号为202010940847.3，发明名称为“分诊数据处理方法、装置、设备及介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及大数据的数据处理领域，尤其涉及一种分诊数据处理方法、装置、设备及介质。
背景技术
发明人发现目前，患者去医院就诊时，首先需要去分诊台进行人工分诊，在该过程中患者需要消耗大量排队时间，而且对分诊台的服务人员的专业知识深度及广度上有较高的要求，如果服务人员给患者分诊错误，又需要重新进行分诊，大大浪费患者的时间，严重影响患者体验，因此，在现有技术上，患者进行人工分诊过程中耗时长、很难给出合理的就诊科室或者就诊医生，从而导致患者体验差，以及就诊准确率低。
发明内容
本申请提供一种分诊数据处理方法、装置、计算机设备及存储介质，实现了通过基于决策树的短期分诊知识模型和基于深度强化学习的长期分诊模型分别对患者的就诊数据进行识别，最终融合输出分诊结果，本申请适用于智慧医疗等领域，可进一步推动智慧城市的建设，能够实现快速地、准确地对患者进行自动分诊，节省了患者时间，提升了就诊准确率，提升了患者体验。
一种分诊数据处理方法，包括：
接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
一种分诊数据处理装置，包括：
接收模块，用于接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
获取模块，用于获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
划分模块，用于将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
匹配模块，用于获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
预测模块，用于通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
输出模块，用于根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
一个或多个存储有计算机可读指令的可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
本申请提供的分诊数据处理方法、装置、计算机设备及存储介质，通过接收到包含患者的就诊数据的分诊请求；获取与所述就诊数据中的患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出，如此，实现了获取患者的历史就诊信息，通过基于决策树的短期分诊知识模型划分出与患者对应的患者群体和短期分诊结果，再通过患者群体匹配出基于深度强化学习的长期分诊模型，并预测出长期分诊结果，融合短期分诊结果和长期分诊结果确定出最终分诊结果，因此，实现了结合患者历史就诊信息提取患者特征，并通过基于决策树的短期分诊知识模型和基于深度强化学习的长期分诊模型分别对患者的就诊数据进行识别，最终融合输出分诊结果，能够实现快速地、准确地对患者进行自动分诊，节省了患者时间，提升了就诊准确率，提升了患者体验。
本申请的一个或多个实施例的细节在下面的附图和描述中提出，本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例中分诊数据处理方法的应用环境示意图；
图2是本申请一实施例中分诊数据处理方法的流程图；
图3是本申请一实施例中分诊数据处理方法的步骤S30的流程图；
图4是本申请一实施例中分诊数据处理方法的步骤S302的流程图；
图5是本申请一实施例中分诊数据处理方法的步骤S303的流程图；
图6是本申请一实施例中分诊数据处理方法的步骤S40的流程图；
图7是本申请一实施例中分诊数据处理方法的步骤S403的流程图；
图8是本申请一实施例中分诊数据处理方法的步骤S60的流程图；
图9是本申请一实施例中分诊数据处理装置的原理框图；
图10是本申请一实施例中计算机设备的示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请提供的分诊数据处理方法，可应用在如图1的应用环境中，其中，客户端(计算机设备)通过网络与服务器进行通信。其中，客户端(计算机设备)包括但不限于为各种个人计算机、笔记本电脑、智能手机、平板电脑、摄像头和便携式可穿戴设备。服务器可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一实施例中，如图2所示，提供一种分诊数据处理方法，其技术方案主要包括以下步骤S10-S60：
S10，接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码。
可理解地，患者在应用程序平台上确定输入完所述就诊数据之后，触发所述分诊请求，所述就诊数据为患者在所述应用程序平台上登录审核之后输入的当前就诊的信息，所述就诊数据可以通过患者在应用程序平台上进行文本输入后获得，也可以通过患者在应用程序平台对患者输入的语音进行转换成文本后确认获得，所述就诊数据包括所述患者标识码和患者基本信息，所述患者标识码为每个患者的唯一码，患者通过所述患者标识码登录所述应用程序平台，比如患者标识码为患者的身份证号码、社保号码等，所述患者基本信息包括患者的性别、年龄、工作职业等等个人基本信息。
S20，获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据。
可理解地，在应用程序平台上获取与所述患者标识码关联的所有所述历史就诊信息，所述历史就诊信息为患者在过去与就诊相关的信息，包括历史诊断结果、疾病史、用药史、检验检查结果(例如：健康年检检查)和风险预测结果等等，并将所述历史就诊信息和所述就诊数据一起标记为所述待分诊数据，其中，所述待分诊数据可以对所述历史就诊信息和所述就诊数据增加不同的权重进行区分，并且增强就诊数据，以便提高后续的分诊准确性。
S30，将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果。
可理解地，将所述待分诊数据输入至所述短期分诊知识模型，所述短期分诊知识模型为训练完成的知识决策树模型，所述短期分诊知识模型的结构为基于决策树的树结构进行构建，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，所述群体划分为通过对所述待分诊数据进行分诊特征提取，并对提取的分诊特征进行决策，从根节点往下不断往下一层级的内部节点进行划分，直至出现不能再划分的叶节点的处理过程，该不能再划分的叶节点确定为与所述患者对应的所述短期分诊结果，所述短期分诊结果包括所述短期分诊知识模型输出的科室类别及其概率值，每个所述节点都与一个群体特征关联，所述节点包括根节点、内部节点和叶节点，每经过一个节点的划分就表明给患者划分出与患者相 关的一个群体特征，将经过的所有所述节点关联的群体特征进行划分的结果进行汇总，从而确定出所述患者群体结果。
其中，所述分诊特征包括群体特征和症状特征，所述群体特征为患者表现的与患者群体类别相关的特征，所述患者群体为对患者的不同特征进行划分成不同的群体，比如根据患者的群体特征，比如年龄、患病频率、性别、潜在风险等特征，对患者群体划分成高龄多发易感男士人群、少儿成长突发女孩人群、中龄平稳抵御女士人群等等；所述症状特征为与医学上病症相关的特征，例如：皮肤特征、骨科特征、眼科特征、耳科特征、儿童流感特征、呼吸科特征等等。
在一实施例中，如图3所示，所述步骤S30之前，即所述将所述待分诊数据输入基于决策树的短期分诊知识模型中之前，包括：
S301，获取短期就诊样本集；其中，所述短期就诊样本集包括多个短期就诊样本，一个所述短期就诊样本与一个科室标签关联。
可理解地，所述短期就诊样本集为收集的所述短期就诊样本的集合，所述短期就诊样本为收集的患者的历史且已经完成分诊的就诊输入的数据，一个所述短期就诊样本与一个所述分诊标签关联，所述科室标签为在实际就诊中与其对应的所述短期就诊样本最后被确诊的科室，所述科室为医院中包含的各类科室。
S302，将所述短期就诊样本输入预设的知识决策树模型；所述知识决策树模型包括各节点特征参数。
可理解地，将所述短期就诊样本输入至预设的知识决策树模型，所述知识决策树模型为决策树结构的模型，所述知识决策树模型包含有决策树结构的多个节点，其中每个节点都含有节点特征参数，所述节点特征参数为在划分分类中选择特征的属性参数，例如：所述节点特征参数设置为大于50岁的属性参数，则对所述短期就诊样本进行划分，是否具备大于50岁的特征。
在一实施例中，如图4所示，所述步骤S302之前，即所述将所述短期就诊样本输入预设的知识决策树模型之前，包括：
S3021，获取临床指南知识。
可理解地，获取所述临床指南知识中的数据，所述临床指南知识为通过临床指南和专家共识梳理就诊规则汇总的数据。
S3022，根据所述临床指南知识，建立就诊知识图谱。
可理解地，将所述临床指南知识中的症状分类、症状名称、症状现象和科室分类作为所述就诊知识图谱中的实体，根据所述临床指南知识构建各实体之间的关系，按照知识图谱的自顶向下的模式层(实体-关系-实体)构建方法建立所述就诊知识图谱。
S3023，按照就诊知识图谱构建所述知识决策树模型。
可理解地，将建立的所述就诊知识图谱按照自顶而下的树状结构进行构建，将每个所述就诊知识图谱中的实体转换成是否具备该实体的分诊特征的节点，各实体间的关系则转换成决策的划分条件，从而构建出所述知识决策树模型。
本申请实现了通过获取临床指南知识，根据所述临床指南知识，建立就诊知识图谱；按照就诊知识图谱构建所述知识决策树模型，能够运用知识图谱的构建方式将临床指南知识构建出就诊知识图谱，并将该就诊知识图谱转换生成各节点，从而构建出知识决策树模型，简化了知识决策树的构建过程，缩短了构建时间，并提高了知识决策树模型的分类准确率性可靠性。
S303，通过所述知识决策树模型对所述短期就诊样本进行群体划分，得到样本群体结果和样本分诊结果。
可理解地，所述知识决策树模型对所述短期就诊样本进行群体划分，所述群体划分还包括通过对所述短期就诊样本进行分诊特征提取，并对提取的分诊特征进行决策，从根节 点往下不断往下一层级的内部节点进行划分，直至出现不能再划分的叶节点的处理过程，该不能再划分的叶节点确定为与所述短期就诊样本对应的所述样本分诊结果，所述样本分诊结果为所述知识决策树模型输出的各科室类别及其概率值(也称为OR值，即逻辑回归系数对应的比数比)，每个节点都与一个群体特征关联，所述节点包括根节点、内部节点和叶节点，每经过一个节点的划分就表明给所述短期就诊样本划分出一个群体特征，将经过的所有节点关联的群体特征进行划分的结果进行汇总，从而确定出所述样本群体结果。
在一实施例中，如图5所示，所述步骤S303中，即所述通过所述知识决策树模型对所述短期就诊样本进行特征决策，得到样本群体结果和样本分诊结果，包括：
S3031，通过决策树分类方法和递归分区方法，对所述短期就诊样本进行特征划分及决策，得到所述样本分诊结果和节点路径结果；所述节点路径结果为所述知识决策树模型对所述短期就诊样本进行决策经过的节点构成的路径；其中，每个所述节点都与一个群体特征关联。
可理解地，所述决策树分类方法为对所述短期就诊样本进行分类的树形结构，对所述短期就诊样本的某一特征进行特征选择，根据特征划分的结果，将所述短期就诊样本进行分配到其下一个内部节点(也可称为分支子节点)；其中在自顶向下的过程中采用所述递归分区方法，所述递归分区方法为在每个待划分的节点建立一个模型，将所述短期就诊样本集划分出不同的子集，使得各子集间的分布差异是最大的，子集对应决策过程中的特征从而能够提高特征选择的有效性。
其中，所述特征划分为通过所述递归分区方法划分出的子集所选取的特征与所述短期就诊样本的特征进行匹配，从而划分出一个群体特征，所述决策为根据特征划分的结果分配到相应的下一个内部节点的过程，经过特征划分及决策的处理，所述短期就诊样本会在所述知识决策树模型中构成一条自顶向下的路径，得到所述节点路径结果，最后所述短期就诊样本抵达最底层的一个叶节点，该叶节点为与所述短期就诊样本对应的所述样本分诊结果。
S3032，获取与所述节点路径结果中的各节点关联的群体特征，将获取的所有所述群体特征确定为所述样本群体结果。
可理解地，获取所述节点路径结果中的各节点的子集的特征选择对应的群体特征，例如：短期就诊样本为“60岁XXX感觉腿脚无力，头晕”，经过年龄特征关联的节点进行特征划分时，将其60岁的特征与节点中的各子集对应的年龄特征进行匹配，划分出与高龄的年龄特征相匹配，则该高龄的年龄特征确定为该短期就诊样本的其中一个群体特征，标记为与该短期就诊样本对应的所述样本群体结果中的其中一个群体特征。
本申请实现了通过决策树分类方法和递归分区方法，对所述短期就诊样本进行特征划分及决策，得到所述样本分诊结果和节点路径结果；获取与所述节点路径结果中的各节点关联的群体特征，将获取的所有所述群体特征确定为所述样本群体结果，如此，实现了通过决策树方法和递归分区方法，得到样本分诊结果，并获取所有经过节点的群体特征，确定出样本群体结果，能够在决策树分类的过程中经一部细分出各明显的子集，提高了决策的准确性和有效性。
S304，根据所述样本分诊结果和所述科室标签，得到决策树损失值。
可理解地，通过所述知识决策树模型的损失函数，计算所述样本分诊结果和所述科室标签之间的所述决策树损失值，所述损失函数优选为正则化的极大似然函数。
S305，在所述决策树损失值未达到预设的决策树收敛条件时，调整所述知识决策树模型中的各节点特征参数，并触发通过所述知识决策树模型对所述短期就诊样本进行特征决策的步骤，直至所述决策树损失值达到所述决策树收敛条件时，将收敛之后的所述知识决策树模型记录为短期分诊知识模型。
可理解地，所述收敛条件可以为所述决策树损失值经过了1000次计算后值为很小且 不会再下降的条件，即在所述决策树损失值经过1000次计算后值为很小且不会再下降时，停止训练，并将收敛之后的所述知识决策树模型记录为短期分诊知识模型；所述收敛条件也可以为所述决策树损失值小于设定阈值的条件，即在所述决策树损失值小于设定阈值时，停止训练，并将收敛之后的所述知识决策树模型记录为短期分诊知识模型，如此，在所述决策树损失值未达到预设的收敛条件时，不断调整所述知识决策树模型中的各节点特征参数，并触发通过所述知识决策树模型对所述短期就诊样本进行特征决策的步骤，可以不断向准确的分类靠拢，让识别的准确率越来越高。
S40，获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型。
可理解地，一个所述长期分诊模型都与一种群体类别关联，各所述长期分诊模型都是基于深度强化学习并通过与其关联的群体类别的历史类别样本进行学习后获得，能够让长期分诊模型更具针对性，根据所述患者群体结果确定出所述患者的患者群体类别，从而获取与所述患者的患者群体类别匹配的群体类别对应的所述长期分诊模型，所述长期分诊模型为结合神经网络和强化学习的深度强化学习(DQN，Deep Q Network)模型，将所述长期分诊模型作为智能体(Agent)，就诊方案看做动作(Action)，患者的就诊信息作为状态(State)，模型通过试验的方式来学习一个优化策略，以最大化长期回报，即模型选择一个动作作用于环境，环境接受该动作后状态发生变化，同时产生一个回报(奖励或惩罚)作为强化信号反馈给模型；模型根据该强化信号优化选择动作的策略，优化方向是使长期预期回报最高；在优化后的策略下，模型再根据环境当前的状态选择下一个动作。
在一实施例中，如图6所示，所述步骤S40中，即所述获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型，包括：
S401，将所述患者群体结果输入患者群体分类模型。
可理解地，将所述患者群体结果输入所述患者群体份分类模型，所述患者群体分类模型为通过聚类算法进行训练完成的神经网络模型，所述患者群体分类模型实现了对收集的患者的群体特征进行分类的模型。
S402，通过所述患者群体分类模型对所有所述群体特征进行聚类处理，得到与所述患者群体结果对应的患者群体类别。
可理解地，所述聚类处理为运用K-means聚类算法对所述患者群体结果中的所有所述群体特征进行欧式计算，根据落入的聚类簇的范围，确定出所述患者群体类别，所述患者群体类别可以根据需求设定，比如患者群体类别分为高龄潜在风险人群、中龄风险人群等等。
S403，获取与所述患者群体类别匹配的所述长期分诊模型。
可理解地，通过Word2vec模型，计算该患者群体类别与各长期分诊模型关联的群体类别之间的相似度值，将与最大的相似度值对应的群体类别关联的长期分诊模型确定为与该患者群体类别匹配的所述长期分诊模型，所述患者群体类别的全集可以与所述群体类别的全集相同，也可以与所述群体类别的全集不同，通过相似度值的衡量匹配程度，可以增加长期分诊模型的灵活性，并提高了识别的准确性。
本申请实现了通过将所述患者群体结果输入患者群体分类模型；通过所述患者群体分类模型对所有所述群体特征进行聚类处理，得到与所述患者群体结果对应的患者群体类别；获取与所述患者群体类别匹配的所述长期分诊模型，如此，能够科学地选择合适的长期分诊模型，提高了分诊的准确性和可靠性，并增加了模型的灵活性。
在一实施例中，如图7所示，所述步骤S403之前，即所述获取与所述患者群体类别匹配的所述长期分诊模型之前，包括：
S4031，获取历史类别样本集；所述历史类别样本集包含多个与所述患者群体类别匹配的历史类别样本，一个所述历史类别样本与一个历史科室标签关联。
可理解地，所述历史类别样本集为所述历史类别样本的集合，所述历史类别样本为与 所述患者群体类别匹配的患者对应的历史收集的就诊输入的数据，其中匹配方式可以为人工对患者进行患者群体类别的标记，将相同患者群体类别的患者确定为与所述患者群体类别匹配的患者，一个所述历史类别样本与一个历史科室标签关联，所述历史科室标签为在实际就诊中与其对应的所述历史类别样本最后被确诊的科室，所述科室为医院中包含的各类科室。
S4032，将所述历史类别样本输入含有初始参数的深度强化学习模型。
可理解地，将所述历史类别样本输入所述深度强化学习模型中，所述深度强化学习模型包括所述初始参数，所述初始参数可以根据需求设定，比如初始参数设置为零。
S4033，通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测，得到回报结果。
可理解地，所述离线学习策略方式为利用已经收集的大量历史类别样本进行训练学习，从而不会出现未曾采用过的决策策略，通过所述深度强化学习模型对所述历史类别样本进行预测，所述预测为根据所述深度强化学习模型执行的各个动作对应的长期预期回报Q值对可选就诊方案(确定出各科室类别)进行排序，Q值越大表示此就诊方案预期回报值越大，在预期回报值最高时，则预测完毕，最终得出的就诊方案(预测出就诊的科室)即为所述回报结果。
S4034，根据所述回报结果和与所述历史类别样本关联的历史科室标签，确定网络损失值。
可理解地，通过所述深度强化学习模型的损失函数，计算所述回报结果和与所述历史类别样本关联的历史科室标签之间的所述网络损失值，所述损失函数优选为最小二乘法函数。
S4035，在所述网络损失值未达到预设的网络收敛条件时，调整所述深度强化学习模型中的初始参数，并触发通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测的步骤，直至所述网络损失值达到所述网络收敛条件时，将收敛之后的所述深度强化学习模型记录为与所述患者群体类别匹配的所述长期分诊模型。
可理解地，所述网络收敛条件可以为所述网络损失值经过了2000次计算后值为很小且不会再下降的条件，即在所述网络损失值经过2000次计算后值为很小且不会再下降时，停止训练，并将收敛之后的所述深度强化学习模型记录为与所述患者群体类别匹配的所述长期分诊模型；所述网络收敛条件也可以为所述网络损失值小于网络设定阈值的条件，即在所述网络损失值小于网络设定阈值时，停止训练，并将收敛之后的所述深度强化学习模型记录为与所述患者群体类别匹配的所述长期分诊模型，如此，在所述网络损失值未达到预设的网络收敛条件时，不断调整所述深度强化学习模型中的初始参数，并触发通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测的步骤，可以不断向准确的结果靠拢，让识别的准确率越来越高。如此，通过更具针对性的历史类别样本，以及通过离线学习策略方式能够让深度强化学习模型的准确率更高，而且有更强的泛化度及鲁棒性。
S50，通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果。
可理解地，通过所述长期分诊模型对所述就诊数据进行预测，所述预测还包括根据所述长期分诊模型针对所述就诊数据作为状态执行训练后的动作后得出的长期分诊结果(也即执行动作后得到的就诊方案)，所述长期分诊结果为具有长期相同类别的患者进行学习后预测出的分诊结果，所述长期分诊结果包括所述长期分诊模型输出的科室类别及其Q值。
S60，根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
可理解地，通过将所述短期分诊结果中的各科室类别及其概率值和所述长期分诊结果中的各科室类别及其概率值进行加权处理，根据设定的权重参数进行计算得出各科室类别 对应的综合分值，将综合分值最大的科室类别确定为所述患者的最终分诊结果，所述最终分诊结果还可以包含科室类别中某一特长的就诊医生等其他与分诊相关的结果数据，其中，通过权重参数拉开各科室类别对应的概率值的差距，让分类更加准确。
在一实施例中，如图8所示，所述步骤S60中，即所述根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出，包括：
S601，将所述短期分诊结果和所述长期分诊结果输入综合分诊模型。
可理解地，将所述短期分诊结果中的各科室类别及其概率值和所述长期分诊结果中的各科室类别及其概率值输入所述综合分支模型中。
S602，通过所述综合分诊模型中的综合评分函数，输出与各科室类别一一关联的多个综合分值，将综合分值最大的科室类别确定为所述患者的最终分诊结果。
可理解地，通过所述综合评分函数计算得出所述综合分值，将所有综合分值中最大的综合分值对应的科室类别确定为所述患者的最终分诊结果，所述综合评分函数为：
score(i)＝OR(i)×w1+Q(i)×w2
其中，i为第i个科室类别，score(i)为第i个科室类别对应的综合分值，OR(i)为第i个科室类别对应的短期分诊结果的OR值，w1为OR值的权重；Q(i)为第i个科室类别对应的长期分诊结果的Q值；w2为Q值的权重。
本申请实现了通过接收到包含患者的就诊数据的分诊请求；获取与所述就诊数据中的患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出，如此，实现了获取患者的历史就诊信息，通过基于决策树的短期分诊知识模型划分出与患者对应的患者群体和短期分诊结果，再通过患者群体匹配出基于深度强化学习的长期分诊模型，并预测出长期分诊结果，融合短期分诊结果和长期分诊结果确定出最终分诊结果，因此，实现了结合患者历史就诊信息提取患者特征，并通过基于决策树的短期分诊知识模型和基于深度强化学习的长期分诊模型分别对患者的就诊数据进行识别，最终融合输出分诊结果，能够实现快速地、准确地对患者进行自动分诊，节省了患者时间，提升了就诊准确率，提升了患者体验。
在一实施例中，提供一种分诊数据处理装置，该分诊数据处理装置与上述实施例中分诊数据处理方法一一对应。如图9所示，该分诊数据处理装置包括接收模块11、获取模块12、划分模块13、匹配模块14、预测模块15和输出模块15。各功能模块详细说明如下：
接收模块11，用于接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
获取模块12，用于获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
划分模块13，用于将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
匹配模块14，用于获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
预测模块15，用于通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
输出模块16，用于根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
关于分诊数据处理装置的具体限定可以参见上文中对于分诊数据处理方法的限定，在 此不再赘述。上述分诊数据处理装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图10所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括可读存储介质、内存储器。该可读存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为可读存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种分诊数据处理方法。本实施例所提供的可读存储介质包括非易失性可读存储介质和易失性可读存储介质。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机可读指令，处理器执行计算机可读指令时实现上述实施例中分诊数据处理方法。
在一个实施例中，提供了一个或多个存储有计算机可读指令的可读存储介质，本实施例所提供的可读存储介质包括非易失性可读存储介质和易失性可读存储介质；该可读存储介质上存储有计算机可读指令，该计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器实现上述实施例中分诊数据处理方法。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质或易失性可读存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种分诊数据处理方法，其中，包括：
    接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
    获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
    将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
    获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
    通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
    根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
  2. 如权利要求1所述的分诊数据处理方法，其中，所述将所述待分诊数据输入基于决策树的短期分诊知识模型中之前，包括：
    获取短期就诊样本集；其中，所述短期就诊样本集包括多个短期就诊样本，一个所述短期就诊样本与一个科室标签关联；
    将所述短期就诊样本输入预设的知识决策树模型；所述知识决策树模型包括各节点特征参数；
    通过所述知识决策树模型对所述短期就诊样本进行群体划分，得到样本群体结果和样本分诊结果；
    根据所述样本分诊结果和所述科室标签，得到决策树损失值；
    在所述决策树损失值未达到预设的决策树收敛条件时，调整所述知识决策树模型中的各节点特征参数，并触发通过所述知识决策树模型对所述短期就诊样本进行特征决策的步骤，直至所述决策树损失值达到所述决策树收敛条件时，将收敛之后的所述知识决策树模型记录为短期分诊知识模型。
  3. 如权利要求2所述的分诊数据处理方法，其中，所述将所述短期就诊样本输入预设的知识决策树模型之前，包括：
    获取临床指南知识；
    根据所述临床指南知识，建立就诊知识图谱；
    按照就诊知识图谱构建所述知识决策树模型。
  4. 如权利要求2所述的分诊数据处理方法，其中，所述通过所述知识决策树模型对所述短期就诊样本进行特征决策，得到样本群体结果和样本分诊结果，包括：
    通过决策树分类方法和递归分区方法，对所述短期就诊样本进行特征划分及决策，得到所述样本分诊结果和节点路径结果；所述节点路径结果为所述知识决策树模型对所述短期就诊样本进行决策经过的节点构成的路径；其中，每个所述节点都与一个群体特征关联；
    获取与所述节点路径结果中的各节点关联的群体特征，将获取的所有所述群体特征确定为所述样本群体结果。
  5. 如权利要求1所述的分诊数据处理方法，其中，所述获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型，包括：
    将所述患者群体结果输入患者群体分类模型；
    通过所述患者群体分类模型对所有所述群体特征进行聚类处理，得到与所述患者群体结果对应的患者群体类别；
    获取与所述患者群体类别匹配的所述长期分诊模型。
  6. 如权利要求5所述的分诊数据处理方法，其中，所述获取与所述患者群体类别匹配的所述长期分诊模型之前，包括：
    获取历史类别样本集；所述历史类别样本集包含多个与所述患者群体类别匹配的历史 类别样本，一个所述历史类别样本与一个历史科室标签关联；
    将所述历史类别样本输入含有初始参数的深度强化学习模型；
    通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测，得到回报结果；
    根据所述回报结果和与所述历史类别样本关联的历史科室标签，确定网络损失值；
    在所述网络损失值未达到预设的网络收敛条件时，调整所述深度强化学习模型中的初始参数，并触发通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测的步骤，直至所述网络损失值达到所述网络收敛条件时，将收敛之后的所述深度强化学习模型记录为与所述患者群体类别匹配的所述长期分诊模型。
  7. 如权利要求6所述的分诊数据处理方法，其中，所述根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出，包括：
    将所述短期分诊结果和所述长期分诊结果输入综合分诊模型；
    通过所述综合分诊模型中的综合评分函数，输出与各科室类别一一关联的多个综合分值，将综合分值最大的科室类别确定为所述患者的最终分诊结果。
  8. 一种分诊数据处理装置，其中，包括：
    接收模块，用于接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
    获取模块，用于获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
    划分模块，用于将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
    匹配模块，用于获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
    预测模块，用于通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
    输出模块，用于根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
  9. 一种计算机设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，其中，所述处理器执行所述计算机可读指令时实现如下步骤：接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
    获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
    将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
    获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
    通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
    根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
  10. 如权利要求9所述的计算机设备，其中，所述将所述待分诊数据输入基于决策树的短期分诊知识模型中之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取短期就诊样本集；其中，所述短期就诊样本集包括多个短期就诊样本，一个所述短期就诊样本与一个科室标签关联；
    将所述短期就诊样本输入预设的知识决策树模型；所述知识决策树模型包括各节点特征参数；
    通过所述知识决策树模型对所述短期就诊样本进行群体划分，得到样本群体结果和样本分诊结果；
    根据所述样本分诊结果和所述科室标签，得到决策树损失值；
    在所述决策树损失值未达到预设的决策树收敛条件时，调整所述知识决策树模型中的 各节点特征参数，并触发通过所述知识决策树模型对所述短期就诊样本进行特征决策的步骤，直至所述决策树损失值达到所述决策树收敛条件时，将收敛之后的所述知识决策树模型记录为短期分诊知识模型。
  11. 如权利要求10所述的计算机设备，其中，所述将所述短期就诊样本输入预设的知识决策树模型之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取临床指南知识；
    根据所述临床指南知识，建立就诊知识图谱；
    按照就诊知识图谱构建所述知识决策树模型。
  12. 如权利要求10所述的计算机设备，其中，所述通过所述知识决策树模型对所述短期就诊样本进行特征决策，得到样本群体结果和样本分诊结果，包括：
    通过决策树分类方法和递归分区方法，对所述短期就诊样本进行特征划分及决策，得到所述样本分诊结果和节点路径结果；所述节点路径结果为所述知识决策树模型对所述短期就诊样本进行决策经过的节点构成的路径；其中，每个所述节点都与一个群体特征关联；
    获取与所述节点路径结果中的各节点关联的群体特征，将获取的所有所述群体特征确定为所述样本群体结果。
  13. 如权利要求9所述的计算机设备，其中，所述获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型，包括：
    将所述患者群体结果输入患者群体分类模型；
    通过所述患者群体分类模型对所有所述群体特征进行聚类处理，得到与所述患者群体结果对应的患者群体类别；
    获取与所述患者群体类别匹配的所述长期分诊模型。
  14. 如权利要求13所述的计算机设备，其中，所述获取与所述患者群体类别匹配的所述长期分诊模型之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取历史类别样本集；所述历史类别样本集包含多个与所述患者群体类别匹配的历史类别样本，一个所述历史类别样本与一个历史科室标签关联；
    将所述历史类别样本输入含有初始参数的深度强化学习模型；
    通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测，得到回报结果；
    根据所述回报结果和与所述历史类别样本关联的历史科室标签，确定网络损失值；
    在所述网络损失值未达到预设的网络收敛条件时，调整所述深度强化学习模型中的初始参数，并触发通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测的步骤，直至所述网络损失值达到所述网络收敛条件时，将收敛之后的所述深度强化学习模型记录为与所述患者群体类别匹配的所述长期分诊模型。
  15. 一个或多个存储有计算机可读指令的可读存储介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
    接收到包含患者的就诊数据的分诊请求；所述就诊数据包括患者标识码；
    获取与所述患者标识码关联的历史就诊信息，将所述历史就诊信息和所述就诊数据确定为所述患者的待分诊数据；
    将所述待分诊数据输入基于决策树的短期分诊知识模型中，通过所述短期分诊知识模型对所述待分诊数据进行群体划分，得到患者群体结果和短期分诊结果；
    获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型；
    通过所述长期分诊模型对所述就诊数据进行预测，得到长期分诊结果；
    根据所述短期分诊结果和所述长期分诊结果，确定所述患者的最终分诊结果并输出。
  16. 如权利要求15所述的可读存储介质，其中，所述将所述待分诊数据输入基于决策树的短期分诊知识模型中之前，所述计算机可读指令被一个或多个处理器执行时，使得所 述一个或多个处理器还执行如下步骤：
    获取短期就诊样本集；其中，所述短期就诊样本集包括多个短期就诊样本，一个所述短期就诊样本与一个科室标签关联；
    将所述短期就诊样本输入预设的知识决策树模型；所述知识决策树模型包括各节点特征参数；
    通过所述知识决策树模型对所述短期就诊样本进行群体划分，得到样本群体结果和样本分诊结果；
    根据所述样本分诊结果和所述科室标签，得到决策树损失值；
    在所述决策树损失值未达到预设的决策树收敛条件时，调整所述知识决策树模型中的各节点特征参数，并触发通过所述知识决策树模型对所述短期就诊样本进行特征决策的步骤，直至所述决策树损失值达到所述决策树收敛条件时，将收敛之后的所述知识决策树模型记录为短期分诊知识模型。
  17. 如权利要求16所述的可读存储介质，其中，所述将所述短期就诊样本输入预设的知识决策树模型之前，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    获取临床指南知识；
    根据所述临床指南知识，建立就诊知识图谱；
    按照就诊知识图谱构建所述知识决策树模型。
  18. 如权利要求16所述的可读存储介质，其中，所述通过所述知识决策树模型对所述短期就诊样本进行特征决策，得到样本群体结果和样本分诊结果，包括：
    通过决策树分类方法和递归分区方法，对所述短期就诊样本进行特征划分及决策，得到所述样本分诊结果和节点路径结果；所述节点路径结果为所述知识决策树模型对所述短期就诊样本进行决策经过的节点构成的路径；其中，每个所述节点都与一个群体特征关联；
    获取与所述节点路径结果中的各节点关联的群体特征，将获取的所有所述群体特征确定为所述样本群体结果。
  19. 如权利要求15所述的可读存储介质，其中，所述获取与所述患者群体结果相匹配的基于深度强化学习的长期分诊模型，包括：
    将所述患者群体结果输入患者群体分类模型；
    通过所述患者群体分类模型对所有所述群体特征进行聚类处理，得到与所述患者群体结果对应的患者群体类别；
    获取与所述患者群体类别匹配的所述长期分诊模型。
  20. 如权利要求19所述的可读存储介质，其中，所述获取与所述患者群体类别匹配的所述长期分诊模型之前，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器还执行如下步骤：
    获取历史类别样本集；所述历史类别样本集包含多个与所述患者群体类别匹配的历史类别样本，一个所述历史类别样本与一个历史科室标签关联；
    将所述历史类别样本输入含有初始参数的深度强化学习模型；
    通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测，得到回报结果；
    根据所述回报结果和与所述历史类别样本关联的历史科室标签，确定网络损失值；
    在所述网络损失值未达到预设的网络收敛条件时，调整所述深度强化学习模型中的初始参数，并触发通过离线学习策略方式，所述深度强化学习模型对所述历史类别样本进行预测的步骤，直至所述网络损失值达到所述网络收敛条件时，将收敛之后的所述深度强化学习模型记录为与所述患者群体类别匹配的所述长期分诊模型。
PCT/CN2020/124220 2020-09-09 2020-10-28 分诊数据处理方法、装置、设备及介质 WO2021151327A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010940847.3 2020-09-09
CN202010940847.3A CN112017789B (zh) 2020-09-09 2020-09-09 分诊数据处理方法、装置、设备及介质

Publications (1)

Publication Number Publication Date
WO2021151327A1 true WO2021151327A1 (zh) 2021-08-05

Family

ID=73522185

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/124220 WO2021151327A1 (zh) 2020-09-09 2020-10-28 分诊数据处理方法、装置、设备及介质

Country Status (2)

Country Link
CN (1) CN112017789B (zh)
WO (1) WO2021151327A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114496231A (zh) * 2022-02-16 2022-05-13 平安科技（深圳）有限公司 基于知识图谱的体质识别方法、装置、设备和存储介质
CN117012374A (zh) * 2023-10-07 2023-11-07 之江实验室 一种融合事件图谱和深度强化学习的医疗随访系统及方法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112652386A (zh) * 2020-12-25 2021-04-13 平安科技（深圳）有限公司 分诊数据处理方法、装置、计算机设备及存储介质
CN116959654A (zh) * 2021-04-21 2023-10-27 广州医科大学附属第一医院 建立基于诊断时效的covid-19分诊系统的方法、该系统及分诊方法
CN113642854A (zh) * 2021-07-23 2021-11-12 重庆中烟工业有限责任公司 烟支单支克重预测方法、装置及计算机可读存储介质
CN113488125A (zh) * 2021-07-27 2021-10-08 心医国际数字医疗系统(大连)有限公司 信息处理方法、装置、电子设备及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104809125A (zh) * 2014-01-24 2015-07-29 腾讯科技（深圳）有限公司 一种网页类别的识别方法和装置
US20190189268A1 (en) * 2017-12-15 2019-06-20 International Business Machines Corporation Differential diagnosis mechanisms based on cognitive evaluation of medical images and patient data
CN111477310A (zh) * 2020-03-04 2020-07-31 平安国际智慧城市科技股份有限公司 分诊数据处理方法、装置、计算机设备及存储介质
CN111564206A (zh) * 2019-02-13 2020-08-21 东软医疗系统股份有限公司 一种分诊方法、装置、设备及介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109559822A (zh) * 2018-11-12 2019-04-02 平安科技（深圳）有限公司 智能初诊方法、装置、计算机设备及存储介质
CN110751996B (zh) * 2019-09-10 2020-12-15 浙江大学 基于递归分区计算的高血压用药推荐模型及其构建方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104809125A (zh) * 2014-01-24 2015-07-29 腾讯科技（深圳）有限公司 一种网页类别的识别方法和装置
US20190189268A1 (en) * 2017-12-15 2019-06-20 International Business Machines Corporation Differential diagnosis mechanisms based on cognitive evaluation of medical images and patient data
CN111564206A (zh) * 2019-02-13 2020-08-21 东软医疗系统股份有限公司 一种分诊方法、装置、设备及介质
CN111477310A (zh) * 2020-03-04 2020-07-31 平安国际智慧城市科技股份有限公司 分诊数据处理方法、装置、计算机设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114496231A (zh) * 2022-02-16 2022-05-13 平安科技（深圳）有限公司 基于知识图谱的体质识别方法、装置、设备和存储介质
CN114496231B (zh) * 2022-02-16 2024-03-26 平安科技（深圳）有限公司 基于知识图谱的体质识别方法、装置、设备和存储介质
CN117012374A (zh) * 2023-10-07 2023-11-07 之江实验室 一种融合事件图谱和深度强化学习的医疗随访系统及方法
CN117012374B (zh) * 2023-10-07 2024-01-26 之江实验室 一种融合事件图谱和深度强化学习的医疗随访系统及方法

Also Published As

Publication number Publication date
CN112017789B (zh) 2023-08-01
CN112017789A (zh) 2020-12-01

Similar Documents

Publication Publication Date Title
WO2021151327A1 (zh) 分诊数据处理方法、装置、设备及介质
US11810671B2 (en) System and method for providing health information
CN108986908B (zh) 问诊数据处理方法、装置、计算机设备和存储介质
WO2020220545A1 (zh) 基于长短期记忆模型的疾病预测方法、装置和计算机设备
ȚĂRANU Data mining in healthcare: decision making and precision.
JP2021532499A (ja) 機械学習に基づく医療データ分類方法、装置、コンピュータデバイス及び記憶媒体
CN110111886A (zh) 一种基于XGBoost疾病预测的智能问诊系统及方法
US20090287503A1 (en) Analysis of individual and group healthcare data in order to provide real time healthcare recommendations
CN110929752B (zh) 基于知识驱动和数据驱动的分群方法及相关设备
JP2016520228A (ja) 医療システムにおけるコンテキスト意識予測
CN113345577B (zh) 诊疗辅助信息的生成方法、模型训练方法、装置、设备以及存储介质
CN112016295A (zh) 症状数据处理方法、装置、计算机设备及存储介质
CN112216361A (zh) 基于人工智能的随访计划单生成方法、装置、终端及介质
CN111370102A (zh) 科室导诊方法、装置以及设备
WO2021151356A1 (zh) 分诊数据处理方法、装置、计算机设备及存储介质
US11710572B2 (en) Experience engine-method and apparatus of learning from similar patients
CN112447270A (zh) 一种用药推荐方法、装置、设备及存储介质
CN114628001B (zh) 基于神经网络的处方推荐方法、系统、设备及存储介质
CN112071431B (zh) 基于深度学习和知识图谱的临床路径自动生成方法及系统
WO2023178970A1 (zh) 医疗数据处理方法、装置、设备及存储介质
Santos et al. Enabling ubiquitous data mining in intensive care-features selection and data pre-processing
Cheong et al. Towards Gender Fairness for Mental Health Prediction.
CN116313086A (zh) 一种亚健康预测模型构建方法、装置、设备及存储介质
KR20190104713A (ko) 자가 적응형 의료 데이터 분석 장치 및 방법
CN113919963A (zh) 基于人工智能的自动理赔方法、装置、设备及介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20916739

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205N DATED 04/05/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20916739

Country of ref document: EP

Kind code of ref document: A1