WO2021164388A1

WO2021164388A1 - Triage fusion model training method, triage method, apparatus, device, and medium

Info

Publication number: WO2021164388A1
Application number: PCT/CN2020/135343
Authority: WO
Inventors: 唐蕊; 李彦轩; 朱昭苇; 孙行智
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-09-25
Filing date: 2020-12-10
Publication date: 2021-08-26
Also published as: CN112101550A; CN112101550B

Abstract

The present application relates to the field of big data processing. Provided is a triage fusion model training method, a triage method, an apparatus, a device and a medium. The method comprises: obtaining a treatment sample set; inputting said treatment samples into a multi-fusion neural network model containing initial parameters; performing prediction with respect to the treatment samples and obtaining at least two triage results; performing standardization conversion on each triage result to obtain standardized results; performing weight fusion on all standardized results to obtain sample triage results; obtaining, by means of loss modeling in the multi-fusion neural network model, a total loss value; if the total loss value has not reached a preset convergence condition, iteratively refreshing the initial parameters of the multi-fusion neural network model until convergence, then recording the post-convergence multi-fusion neural network model as a triage fusion model. The present method improves performance and accuracy of multi-fusion neural network model identification. The present application is applicable to the fields of smart medical care, etc., and can further promote the development of smart cities.

Description

Triage fusion model training method, triage method, device, equipment and medium

This application requires the priority of a Chinese patent application filed with the Chinese Patent Office on September 25, 2020 with the application number 202011023857.7 and the invention title of "Triage Fusion Model Training Method, Triage Method, Apparatus, Equipment, and Media". The entire content is incorporated into this application by reference.

Technical field

This application relates to the field of big data data processing, and in particular to a triage fusion model training method, triage method, device, equipment, and medium.

Background technique

With the advancement and development of medicine, hospitals have become more specialized in setting up departments. The problem that comes with it is that users choose departments to bring certain difficulties. In order to solve this problem, major hospitals have added guidance links, including guidance. Clinicians and self-directed diagnosis services mainly help patients recommend diagnosis and treatment departments.

At present, the inventor found that when a patient goes to the hospital for treatment, he first needs to go to the triage table for manual triage. In this process, the patient needs to spend a lot of time in queuing, and the depth and breadth of the professional knowledge of the guides at the triage table is limited. Higher requirements, if the guides give the patient a wrong diagnosis and need to perform the triage again, it will greatly waste the patient’s time and seriously affect the patient’s experience. Therefore, in the prior art, the manual triage of the patient takes a long time. , It is difficult to give a reasonable medical department or doctor, resulting in poor patient experience and low medical accuracy.

Summary of the invention

This application provides a triage fusion model training, a triage method, a device, a computer device, and a storage medium, which enables accurate recommendation of topic data to users, improves the accuracy of topic recommendation, and avoids showing disliked topic data to users Users have improved user experience satisfaction and the effectiveness of topic recommendations. This application is suitable for smart medical and other fields, which can further promote the construction of smart cities.

A training method for triage fusion model, including:

Obtaining a medical consultation sample set; the medical consultation sample set includes a plurality of medical consultation samples, and each medical consultation sample is associated with a triage label;

Inputting the consultation samples into a multi-fusion neural network model containing initial parameters;

Predicting the consultation samples by using the multi-fusion neural network model to obtain at least two triage results;

Perform standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results;

Perform weight fusion on all the standardized results to obtain sample triage results;

Performing a loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

When the total loss value does not reach the preset convergence condition, iteratively update the initial parameters of the multi-fusion neural network model, until the total loss value reaches the preset convergence condition, the subsequent convergence The multi-fusion neural network model is recorded as a triage fusion model.

A method of triage, including:

Receive the patient's triage request, and obtain the patient's medical information in the triage request;

Input the patient's medical information into the triage fusion model trained by the above-mentioned triage fusion model training method, and obtain the final triage result output by the triage fusion model.

A triage fusion model training device, including:

The obtaining module is used to obtain a medical examination sample set; the medical examination sample set includes a plurality of medical examination samples, and each of the medical examination samples is associated with a triage label;

An input module, used to input the consultation samples into a multi-fusion neural network model containing initial parameters;

A prediction module, used for predicting the medical sample through the multi-fusion neural network model, and obtaining at least two triage results;

The standardization module is used to perform standardized conversion of each of the triage results to obtain a standardized result corresponding to each of the triage results;

The weight module is used to perform weight fusion of all the standardized results to obtain sample triage results;

A loss module, configured to perform loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

The iterative module is used to iteratively update the initial parameters of the multi-fusion neural network model when the total loss value does not reach the preset convergence condition, until the total loss value reaches the preset convergence condition, The multi-fusion neural network model after convergence is recorded as a triage fusion model.

A triage device, including:

The receiving module is used to receive the patient's triage request, and obtain the patient's medical treatment information in the triage request;

The triage module is used to input the patient information into the triage fusion model trained by the above-mentioned triage fusion model training method, and obtain the final triage result output by the triage fusion model.

A computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, and the processor implements the following steps when the processor executes the computer-readable instructions:

A computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, and the processor further implements the following steps when the processor executes the computer-readable instructions:

The patient visit information is input into the triage fusion model trained by the triage fusion model training method, and the final triage result output by the triage fusion model is obtained.

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:

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 further execute the following steps:

The triage fusion model training method, device, computer equipment, and storage medium provided in this application are obtained by obtaining a consultation sample set; the consultation sample set includes a plurality of consultation samples, and each of the consultation samples is associated with a triage label; The medical sample input contains a multi-fusion neural network model containing initial parameters; the medical sample is predicted by the multi-fusion neural network model to obtain at least two triage results; each of the triage results is standardized and converted, Obtain a standardized result corresponding to each of the triage results; perform weight fusion on all the standardized results to obtain a sample triage result; use the loss model in the multi-fusion neural network model to compare the sample triage result and The triage label performs loss analysis to obtain a total loss value; when the total loss value does not reach the preset convergence condition, iteratively update the initial parameters of the multi-fusion neural network model until the total loss value reaches the total loss value. When the preset convergence conditions are described, the multi-fusion neural network model after convergence is recorded as a triage fusion model. Therefore, this application provides a triage fusion model training method, and at least two models are predicted by the multi-fusion neural network model. Triage results (multi-fusion neural network model includes at least two models, one model corresponds to one triage result), standardized conversion of each triage result to obtain each standardized result, and weight fusion of all standardized results to obtain sample triage result , Perform loss analysis through the loss model to obtain the total loss value, and iteratively update the multi-fusion neural network model according to the total loss value until convergence, which realizes the standardization of the triage results output by the different models in the multi-fusion neural network model, so that It has size-related comparability, breaking the limitation of independence between models, and through weight fusion and loss analysis, it can make the training of multi-fusion neural network models more efficient and accurate, and improve the multi-fusion neural network model. Performance and accuracy of recognition.

The triage method, device, computer equipment, and storage medium provided in this application obtain the patient's medical information in the triage request by receiving the patient's triage request; and input the patient's medical information into the above-mentioned triage fusion model training Methods The trained triage fusion model is used to obtain the final triage results output by the triage fusion model. In this way, this application uses the triage fusion model training method to train the completed triage fusion model to predict the patient's visit information, and obtain the final The triage result, the final triage result provides an accurate basis for patients to make an appointment, improves the accuracy and efficiency of triage, improves user satisfaction, and enhances the effectiveness of triage.

The details of one or more embodiments of the present application are presented in the following drawings and description, and other features and advantages of the present application will become apparent from the description, drawings and claims.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings that need to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only 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 labor.

FIG. 1 is a schematic diagram of an application environment of a triage fusion model training method or a triage method in an embodiment of the present application;

2 is a flowchart of a method for training a triage fusion model in an embodiment of the present application;

Fig. 3 is a flowchart of step S40 of the method for training a triage fusion model in an embodiment of the present application;

4 is a flowchart of step S50 of the method for training a triage fusion model in an embodiment of the present application;

FIG. 5 is a flowchart of step S60 of the method for training a triage fusion model in an embodiment of the present application;

Fig. 6 is a flowchart of a triage method in an embodiment of the present application;

Figure 7 is a functional block diagram of a triage fusion model training device in an embodiment of the present application;

Fig. 8 is a functional block diagram of a triage device in an embodiment of the present application;

Fig. 9 is a schematic diagram of a computer device in an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

The triage fusion model training method provided by 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. Among them, the client (computer equipment) 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.

In an embodiment, as shown in FIG. 2, a method for training a triage fusion model is provided, and its technical solution mainly includes the following steps S10-S70:

S10: Obtain a medical visit sample set; the medical visit sample set includes multiple medical visit samples, and each medical visit sample is associated with a triage label.

Understandably, the medical consultation sample set is a collection of the medical consultation samples collected, and the medical consultation sample is data of medical consultation information that has been input from the collected history of the patient and has completed triage. The triage label is associated, and the triage label is a department after the corresponding visit sample was finally triaged in the actual visit, and the department is a variety of departments included in the hospital.

S20: Input the patient visit samples into a multi-fusion neural network model containing initial parameters.

Understandably, the multi-fusion neural network model is a neural network model that merges at least two models. The multi-fusion neural network model can select the fused model according to requirements. For example, the selected fusion model can be an LSTM model, a depth Convolutional neural network model (DCNN model), recurrent neural network model (RNN model), deep residual network model (DRN model) or reinforcement learning model, etc., the initial parameters include each of the multiple fusion neural network models The parameters of the model, the parameters of each model can be directly transferred from the parameters of the model of the same type as each of the models in other fields by means of transfer learning, that is, the model of the same type as each of the models in other fields The parameters are used as the parameters of the model.

S30: Predict the medical samples through the multi-fusion neural network model, and obtain at least two triage results.

Understandably, the multi-fusion neural network model includes a neural network model of at least two models, and the multi-fusion neural network model predicts the consultation sample, that is, each of the models in the multi-fusion neural network model The predictions are made on the samples to be visited respectively, and the prediction is that each of the models is processed according to their respective algorithms to predict the predicted value of each triage category, so that each of the models predicts the respective triage result, namely One of the models corresponds to one of the triage results, indicating that the multi-fusion neural network model outputs at least two of the triage results, and the triage results include all triage categories and the predicted value of each triage category , The predicted value predicts the possibility of the triage category, and the complete set of triage categories is the same as the complete set of triage labels. For example, if the triage label has 100 categories, the triage category is also There are 100 categories in one-to-one correspondence with the triage labels, and there are 100 triage categories and 100 predicted values corresponding to the triage categories in the triage result.

In one embodiment, the multi-fusion neural network model includes an LSTM model, a deep convolutional neural network model, and a reinforcement learning model, and the LSTM model (Long short term memory network, long and short term memory network model) is based on the LSTM algorithm. The model for predicting the results of triage of the consultation samples, and the deep convolutional neural network model is to compare the consultation samples with the network structure in the deep convolutional neural network model (such as VGG16, GoogleNet, ResNet, etc.) Perform convolution, pooling, and fully connected models for predicting triage results. The reinforcement learning model is the expected return Q value obtained after performing various actions on the sample of the visit and predicts the triage based on the expected return Q value The resulting model.

S40: Perform standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results.

Understandably, since the triage results of different models are not comparable in size, it is necessary to standardize the predicted value of the probability in the triage results output by the different models. To make each of the triage results have size-related comparability, and the standardization is converted to perform the triage in descending order of the predicted value of each of the triage categories in the triage results Sort the results, obtain the triage categories with the preset extraction number in the sequence from the sorted triage results, and determine the obtained triage categories of the triage categories with the preset extraction number. For the sorting result corresponding to each of the triage results, standardization technology is used to perform standardized assignment to each triage category in each of the sorting results, that is, to associate each triage category with the triage category in the sorting result Comparable values corresponding to the sequences in, so as to obtain a standardized result corresponding to each of the sorting results one-to-one, that is, the standardized result corresponding to each of the triage results.

In an embodiment, as shown in FIG. 3, in step S40, that is, performing standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results includes:

S401: Sort the triage categories in each of the triage results from large to small, and obtain the pre-set number of triage categories with the first sequence after the sorting, and determine each of the obtained triage results as The sorting result corresponding to each of the triage results; the triage result includes the triage category.

Understandably, the triage results are sorted according to the predicted value of each of the triage categories in the triage results in descending order, and the sequence is obtained from the sorted triage results Triage category of the previous preset extraction number, the preset extraction number is the number of the first sequence extracted from the triage result, the preset extraction number can be set according to requirements, the preset extraction number It is less than the number of the complete set of triage categories. Preferably, the preset extraction number is set to a multiple of 10, such as 10, 20, 30, and so on.

S402: Perform standardized assignment to each triage category in each sorting result through standardization technology.

Understandably, the standardization technique is to determine the comparable value assigned to each triage category according to the sequence digits of each triage category in the ranking result through a standardization function, and the standardization function is:

Wherein, h is the preset extraction quantity, j is the number of sequence digits of the triage category whose sequence is the jth position in the sorting result, such as the first position, the second position, etc., and Y _j is the In the ranking result, the sequence is the comparable value of the j-th triage category.

Wherein, the standardized assignment value will give a comparable value to each triage category in each of the ranking results, and the comparable value is a value calculated using the standardized function according to the number of sequence digits of the triage category in the ranking result , Associating each of the triage categories in the ranking result with the corresponding comparable value.

S403: Determine each of the sorting results after the assignment as a standardized result corresponding to each of the sorting results in a one-to-one manner.

Understandably, each of the ranking results assigned by the standardization technology is determined as the corresponding standardized result, and the standardized result includes the triage category, the predicted value corresponding to the triage category, and The comparable value associated with the triage category is such that the possibility of the triage category of different models is comparable in size.

This application realizes that by sorting the triage categories in each of the triage results from large to small, and obtaining the pre-set number of triage categories with the first sequence after the sorting, the obtained triage categories The result is determined as the sorting result corresponding to each of the triage results; through standardization technology, each triage category in each of the sorting results is standardized and assigned; each of the sorting results after the assignment is determined to be the same as each of the sorting results. The sorting results correspond to the standardized results one-to-one. In this way, by standardizing the triage results of different models, the triage categories in the sorting results of different models are comparable in size, so that the subsequent multi-models can be compared. Provide assistance during the fusion process to further improve the performance of the multi-fusion neural network model, and improve the accuracy and efficiency of triage prediction.

S50: Perform weight fusion on all the standardized results to obtain sample triage results.

Understandably, the process of the weight fusion is as follows: firstly, each of the standardized results is converted into a triage vector of a preset dimension to obtain a triage vector corresponding to each of the standardized results. The difference between the triage result and the triage label is determined, and the accuracy rate corresponding to each of the triage results is determined, that is, the accuracy rate corresponding to each of the standardized results is obtained; secondly, according to all the results The accuracy rate generates a weight corresponding to the triage vector corresponding to the standardized result corresponding to each of the accuracy rates; finally, the sample triage result is obtained according to each of the triage vectors and each of the weights.

Wherein, the preset dimension is the number of complete sets of triage categories, and the preset dimension is greater than the preset extraction quantity.

In one embodiment, as shown in FIG. 4, in the step S50, that is, performing weight fusion of all the standardized results to obtain the sample triage results, including:

S501: Transform each of the standardized results into a triage vector of a preset dimension to obtain a triage vector corresponding to each of the standardized results.

Understandably, the triage vector is converted into a vector of the same dimension according to the predicted value corresponding to each triage category in each of the standardized results, that is, the preset is initialized for each standardized result A vector array of dimensions, the vector array reserves an element position for all triage categories, and fills the element position corresponding to the triage category in the vector array according to the predicted value corresponding to each triage category in the standardized result , The remaining element positions are filled with zeros, so that the filled vector array is determined as the process of the triage vector corresponding to the normalized result.

S502: Determine an accuracy rate corresponding to each of the triage results according to each of the triage results and the triage tags.

Understandably, the triage result output by each model is compared with the triage label to obtain the accuracy rate of each model, that is, in the ranking result corresponding to the triage result When there is a triage category that is the same as the triage label, all the predicted values of the ranking result are summed to obtain the accuracy rate of the model corresponding to the triage result. When the sorting result does not have the same triage category as the triage label, the prediction value corresponding to the triage category that is the same as the triage label in the triage result is determined to be the same as that of the triage label. State the accuracy of the model corresponding to the triage results.

S503: Generate a weight corresponding to each of the triage vectors according to all the accuracy rates.

Understandably, all the accuracy rates are normalized, and after the normalization, the weight corresponding to each accuracy rate is calculated, that is, all the accuracy rates are summed to obtain the total accuracy rate. The ratio of the accuracy rate corresponding to the triage vector to the total accuracy rate is determined as the weight corresponding to the triage vector.

S504: Obtain the sample triage result according to each of the triage vectors and each of the weights.

Understandably, the triage vector and the weight corresponding to the triage vector are multiplied, that is, each of the predicted values in the triage vector is multiplied by the weight to obtain the updated weight The weight prediction value in the triage vector is added to the weight prediction values of the same triage category in all the updated weight triage vectors, that is, the weight prediction values in all the updated weight triage vectors are the same The weight prediction values of the triage category are summarized to obtain a vector array of the preset dimension, the summarized vector array is determined as the sample triage result, and the sample triage result is The weight prediction value is determined as a probability value.

This application realizes that the triage vector corresponding to each of the standardized results is obtained by converting each of the standardized results into a triage vector of a preset dimension; according to each of the triage results and the triage label, it is determined The accuracy rate corresponding to each of the triage results; according to all the accuracy rates, a weight corresponding to each of the triage vectors is generated; according to each of the triage vectors and each of the weights, the sample triage is obtained As a result, in this way, by constructing the weights of the multi-fusion neural network model, fusing each model, and fusing to generate a sample triage result of a preset dimension, it is possible to scientifically and accurately fuse multiple models in the multi-fusion neural network model. Improve the accuracy and reliability of triage.

S60: Perform a loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value.

Understandably, the sample triage results are sorted according to the predicted value in the sample triage results in descending order, and the sorted sample triage results in the sequence of the first preset number The triage category, the obtained sample triage result is determined as the filtered triage result; the sample triage result includes the triage category and the probability value corresponding to the triage category, and gives the filter score Each of the triage categories in the diagnosis result is assigned an order-related value, wherein, if there is the triage category that is the same as the triage label, the total probability value is substituted for the triage that is the same as the triage label. The probability value corresponding to the diagnosis category, the sequence correlation value and the probability value (including the probability value after replacement) corresponding to each of the triage categories in the filtered triage result are input into the In the loss model, loss analysis is performed through the loss model to obtain the total loss value.

Wherein, the preset number can be set according to requirements, the preset number can be the same as the preset extraction number, or it can be different from the preset extraction number, and the sum of the probabilities is all the probabilities The sum of values.

In an embodiment, as shown in FIG. 5, in the step S60, the loss model in the multi-fusion neural network model is used to perform a loss analysis on the sample triage result and the triage label , Get the total loss value, including:

S601: Sort the sample triage results in descending order, and obtain a preset number of triage categories in the sequence of the sorted sample triage results, and triage the obtained sample The result is determined as a filtered triage result; the sample triage result includes the triage category and the probability value corresponding to the triage category.

Understandably, the sample triage results are sorted according to the probability value of each of the triage categories in the sample triage results in descending order, from the sorted sample triage results The preset number of triage categories with the pre-sequence obtained in the triage category, the pre-set number is the number of the pre-sequence extracted from the sample triage results, the preset number can be set according to requirements, the preset number It may be the same as the preset extraction quantity, or may be different from the preset extraction quantity, for example, the preset quantity is 20, 30, etc., and the obtained sample triage result is determined as the filtering Triage results.

S602. Assign an order correlation value to each of the triage categories in the filtered triage result.

Understandably, the sequence correlation value ln(k-i+1) is assigned to each of the triage categories in the filtered triage results through the sequence correlation function, where k is the preset number, i The sequence in the filtered triage result is the i-th position.

S603: If the triage category that is the same as the triage label exists, replace the probability value corresponding to the triage category that is the same as the triage label with the total probability value, and the total probability value is all The sum of the probability values.

Understandably, when there is the triage category that is the same as the triage label among all the triage categories in the filtered triage result, the total probability value is substituted for the triage category corresponding to the triage category. The probability value, when there is no triage category that is the same as the triage label in all triage categories in the filtered triage result, there is no need to process the probability value in the filtered triage result.

S604. Input the sequence correlation value and the probability value corresponding to each of the triage categories in the filtered triage result into the loss model, and perform loss analysis through the loss model to obtain the total Loss value.

Understandably, the sequence correlation value and the probability value corresponding to each of the triage categories are input into the loss function in the loss model, and the total loss value is calculated by the loss function.

In one embodiment, in step S604, that is, inputting the sequence correlation value and the probability value corresponding to each of the triage categories in the filtered triage result into the loss model, Perform loss analysis through the loss model to obtain the total loss value, including:

S6041. Input the sequence correlation value and the probability value corresponding to each of the triage categories into a loss function in the loss model, and calculate the total loss value through the loss function; the loss function for:

L=-(p ₁ ×ln(k)+…+p _i ×ln(k-i+1)+…+p _k ×ln(1))

in,

L is the total loss value;

k is the preset number;

p ₁ is the probability value corresponding to the triage category whose sequence is the first in the filtered triage result;

ln(k) is the sequence correlation value corresponding to the triage category whose sequence is the first in the filtered triage result;

p _i is the probability value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

ln(k-i+1) is the sequence correlation value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

p _k is the probability value corresponding to the triage category whose sequence is the kth in the filtered triage result;

ln(1) is the sequence correlation value corresponding to the triage category whose sequence is the kth in the filtered triage result.

S70: When the total loss value does not reach the preset convergence condition, iteratively update the initial parameters of the multi-fusion neural network model, until the total loss value reaches the preset convergence condition, the subsequent convergence The multi-fusion neural network model is recorded as a triage fusion model.

Understandably, the convergence condition may be a condition that the value of the total loss value is small and will not drop after 3000 calculations, that is, the value of the total loss value is very small after 3000 calculations. When it will no longer drop, stop training, and record the multi-fusion neural network model after convergence as a triage fusion model; the convergence condition can also be the condition that the total loss value is less than the set threshold, that is, the convergence condition is less than the set threshold. When the total loss value is less than the set threshold, stop training, and the multi-fusion neural network model after convergence is recorded as a triage fusion model. In this way, when the total loss value does not reach the preset convergence condition, it is continuously adjusted The initial parameters in the multi-fusion neural network model and trigger the prediction of the medical sample through the multi-fusion neural network model, and the steps of obtaining at least two triage results can continuously move closer to accurate results, allowing identification The accuracy rate is getting higher and higher. In this way, by improving the loss function of the multi-fusion neural network model, it can optimize the sample triage results of the multi-fusion neural network model, and this improvement improves the performance of the multi-fusion neural network model.

This application realizes the acquisition of a medical consultation sample set; the medical consultation sample set includes a plurality of medical consultation samples, and each of the medical consultation samples is associated with a triage label; the medical consultation sample is input into a multi-fusion neural network model containing initial parameters; The multi-fusion neural network model is used to predict the medical samples to obtain at least two triage results; perform standardized conversion of each triage result to obtain a standardized result corresponding to each triage result; The standardized results are weighted and fused to obtain the sample triage results; the loss model in the multi-fusion neural network model is used to perform loss analysis on the sample triage results and the triage labels to obtain the total loss value; When the total loss value does not reach the preset convergence condition, the initial parameters of the multi-fusion neural network model are iteratively updated, until the total loss value reaches the preset convergence condition, the convergence of the multiple The fusion neural network model is recorded as a triage fusion model. Therefore, this application provides a triage fusion model training method, which predicts at least two triage results through a multi-fusion neural network model (the multi-fusion neural network model includes at least two models, A model corresponds to a triage result), the standardized conversion of each triage result is performed to obtain each standardized result, and all the standardized results are weighted and fused to obtain the sample triage result, and the loss analysis is performed through the loss model to obtain the total loss value, according to the total loss The value is iteratively updated the multi-fusion neural network model until it converges, which realizes the standardization of the triage results output by the different models in the multi-fusion neural network model, so that it has size-related comparability and breaks the mutual relationship between the models. The independent limitation, through weight fusion and loss analysis, can make the training of multi-fusion neural network models more efficient and accurate, and improve the performance and accuracy of multi-fusion neural network model recognition.

The triage method provided in this application can be applied in the application environment as shown in Figure 1, where the client (computer equipment) communicates with the server via the network. Among them, the client (computer equipment) 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.

In an embodiment, as shown in FIG. 6, a triage method is provided, and the technical solution mainly includes the following steps S100-S200:

S100: Receive a triage request from a patient, and obtain the patient's medical treatment information in the triage request.

Understandably, after the patient determines on the application platform to input the patient's medical consultation information, the triage request is triggered, and the patient's medical consultation information is the current medical consultation input after the patient logs in on the application platform. Information, the patient visit information can be obtained by the patient after text input on the application platform, or can be obtained by confirming the patient after the patient converts the voice input by the patient into text on the application platform.

S200: Input the patient visit information into the triage fusion model trained by the above triage fusion model training method, and obtain the final triage result output by the triage fusion model.

Understandably, the patient visit information is input to the triage fusion model trained and trained as in the above-mentioned triage fusion model training method, and the sample triage results output from the above-mentioned triage fusion model training method The triage category corresponding to the highest probability value is determined as the final triage result, and the final triage result provides an accurate basis for the patient to make an appointment, and facilitates the patient to select an accurate department to make an appointment.

This application realizes that by receiving the patient's triage request, obtain the patient's visit information in the triage request; input the patient's visit information into the triage fusion model trained as the above-mentioned triage fusion model training method, and obtain the The final triage result output by the triage fusion model. In this case, the triage fusion model trained by the triage fusion model training method predicts the patient's visit information and obtains the final triage result. The final triage result is presented to the patient Appointment provides an accurate basis, improves the accuracy and efficiency of triage, improves user satisfaction, and enhances the effectiveness of triage.

In one embodiment, a triage fusion model training device is provided, and the triage fusion model training device corresponds to the triage fusion model training method in the foregoing embodiment in a one-to-one correspondence. As shown in FIG. 7, the triage fusion model training device includes an acquisition module 11, an input module 12, a prediction module 13, a standardization module 14, a weight module 15, a loss module 16 and an iteration module 17. The detailed description of each functional module is as follows:

The obtaining module 11 is configured to obtain a medical consultation sample set; the medical consultation sample set includes a plurality of medical consultation samples, and each of the medical consultation samples is associated with a triage label;

The input module 12 is used to input the consultation samples into a multi-fusion neural network model containing initial parameters;

The prediction module 13 is used for predicting the medical sample through the multi-fusion neural network model to obtain at least two triage results;

The standardization module 14 is used to perform standardized conversion of each of the triage results to obtain a standardized result corresponding to each of the triage results;

The weight module 15 is used to perform weight fusion of all the standardized results to obtain the sample triage results;

The loss module 16 is configured to perform loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

The iterative module 17 is configured to iteratively update the initial parameters of the multi-fusion neural network model when the total loss value does not reach the preset convergence condition, until the total loss value reaches the preset convergence condition, The multi-fusion neural network model after convergence is recorded as a triage fusion model.

Regarding the specific limitation of the triage fusion model training device, please refer to the above limitation on the triage fusion model training method, which will not be repeated here. Each module in the above-mentioned triage fusion model training 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.

In one embodiment, a triage device is provided, and the triage device corresponds to the triage method in the foregoing embodiment one-to-one. As shown in FIG. 8, the triage device includes a receiving module 101 and a triage module 102. The detailed description of each functional module is as follows:

The receiving module 101 is configured to receive a patient's triage request, and obtain the patient's medical treatment information in the triage request;

The triage module 102 is configured to input the patient's visit information into the triage fusion model trained by the above-mentioned triage fusion model training method, and obtain the final triage result output by the triage fusion model.

For the specific limitation of the triage device, please refer to the above limitation on the triage method, which will not be repeated here. Each module in the above-mentioned triage 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.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 9. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, 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 fusion model training method, or triage method. The readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and capable of running on the processor. Diagnosis fusion model training method, or when the processor executes computer-readable instructions, the triage method in the above embodiment is implemented.

In one embodiment, 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 fusion model training method in the above-mentioned embodiment, or When the computer-readable instructions are executed by the processor, the triage method in the above-mentioned embodiment is implemented.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a non-volatile computer. In a readable storage medium or a volatile readable storage medium, when the computer readable instruction is executed, it may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. 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. As an illustration and not a limitation, 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.

Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the division of the above functional units and modules is used as an example. In practical applications, the above functions can be allocated to different functional units and modules as needed. Module completion, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still implement the foregoing The technical solutions recorded in the examples are modified, or some of the technical features are equivalently replaced; these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the application, and should be included in Within the scope of protection of this application.

Claims

A training method for triage fusion model, which includes:

Obtaining a medical consultation sample set; the medical consultation sample set includes a plurality of medical consultation samples, and each medical consultation sample is associated with a triage label;

Inputting the consultation samples into a multi-fusion neural network model containing initial parameters;

Predicting the consultation samples by using the multi-fusion neural network model to obtain at least two triage results;

Perform standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results;

Perform weight fusion on all the standardized results to obtain sample triage results;

Performing a loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

When the total loss value does not reach the preset convergence condition, iteratively update the initial parameters of the multi-fusion neural network model, until the total loss value reaches the preset convergence condition, the subsequent convergence The multi-fusion neural network model is recorded as a triage fusion model.
3. The triage fusion model training method according to claim 1, wherein said performing standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results comprises:

Sort the triage categories in each of the triage results from large to small, and obtain the pre-set number of triage categories with the first sequence after the sorting, and determine the obtained triage results to be the same as each The sorting result corresponding to the triage result; the triage result includes the triage category;

Using standardization technology, standardize the assignment of each triage category in each of the sorting results;

Each of the sorting results after the assignment is determined as a standardized result corresponding to each of the sorting results one-to-one.
The method for training a triage fusion model according to claim 1, wherein said performing weight fusion of all the standardized results to obtain sample triage results comprises:

Transform each of the standardized results into a triage vector of a preset dimension to obtain a triage vector corresponding to each of the standardized results;

According to each of the triage results and the triage label, determine the accuracy rate corresponding to each of the triage results;

According to all the accuracy rates, generate a weight corresponding to each of the triage vectors;

According to each of the triage vectors and each of the weights, the sample triage results are obtained.
The triage fusion model training method according to claim 1, wherein the loss model in the multi-fusion neural network model is used to perform a loss analysis on the sample triage result and the triage label to obtain a total Loss value, including:

Sort the sample triage results in descending order, and obtain a preset number of triage categories in the sequence of the sorted sample triage results, and determine the obtained sample triage results To filter the triage results; the sample triage results include the triage category and the probability value corresponding to the triage category;

Assigning sequence related values to each of the triage categories in the filtered triage results;

If there is the triage category that is the same as the triage label, the total probability value is substituted for the probability value corresponding to the triage category that is the same as the triage label, and the total probability value is all the Sum of probability values;

Input the sequence correlation value and the probability value corresponding to each of the triage categories in the filtered triage result into the loss model, and perform loss analysis through the loss model to obtain the total loss value .
The triage fusion model training method according to claim 4, wherein said inputting said sequence correlation value and said probability value corresponding to each of said triage categories in said filtered triage result into said loss In the model, the loss analysis is performed through the loss model to obtain the total loss value, including:

The sequence correlation value and the probability value corresponding to each of the triage categories are input into the loss function in the loss model, and the total loss value is calculated by the loss function; the loss function is:

L=-(p 1 ×ln(k)+…+p i ×ln(k-i+1)+…+p k ×ln(1))

in,

L is the total loss value;

k is the preset number;

p 1 is the probability value corresponding to the triage category whose sequence is the first in the filtered triage result;

ln(k) is the sequence correlation value corresponding to the triage category whose sequence is the first in the filtered triage result;

p i is the probability value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

ln(k-i+1) is the sequence correlation value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

p k is the probability value corresponding to the triage category whose sequence is the kth in the filtered triage result;

ln(1) is the sequence correlation value corresponding to the triage category whose sequence is the kth in the filtered triage result.
A method of triage, which includes:

Receive the patient's triage request, and obtain the patient's medical information in the triage request;

The patient visit information is input into the triage fusion model trained by the triage fusion model training method according to any one of claims 1 to 5, and the final triage result output by the triage fusion model is obtained.
A triage fusion model training device, which includes:

An obtaining module, configured to obtain a medical examination sample set; the medical examination sample set includes a plurality of medical examination samples, and each of the medical examination samples is associated with a triage label;

An input module, used to input the consultation samples into a multi-fusion neural network model containing initial parameters;

A prediction module, used for predicting the medical sample through the multi-fusion neural network model, and obtaining at least two triage results;

The standardization module is used to perform standardized conversion of each of the triage results to obtain a standardized result corresponding to each of the triage results;

The weight module is used to perform weight fusion of all the standardized results to obtain sample triage results;

A loss module, configured to perform loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

The iterative module is used to iteratively update the initial parameters of the multi-fusion neural network model when the total loss value does not reach the preset convergence condition, until the total loss value reaches the preset convergence condition, The multi-fusion neural network model after convergence is recorded as a triage fusion model.
A triage device, which includes:

The receiving module is used to receive the patient's triage request, and obtain the patient's medical treatment information in the triage request;

The triage module is used to input the patient's visit information into the triage fusion model trained by the triage fusion model training method according to any one of claims 1 to 5, and obtain the final triage output by the triage fusion model result.
A computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, wherein the processor implements the following steps when the processor executes the computer-readable instructions: Obtaining a medical consultation sample set; the medical consultation sample set includes a plurality of medical consultation samples, and each medical consultation sample is associated with a triage label;

Inputting the consultation samples into a multi-fusion neural network model containing initial parameters;

Predicting the consultation samples by using the multi-fusion neural network model to obtain at least two triage results;

Perform standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results;

Perform weight fusion on all the standardized results to obtain sample triage results;

Performing a loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

When the total loss value does not reach the preset convergence condition, iteratively update the initial parameters of the multi-fusion neural network model, until the total loss value reaches the preset convergence condition, the subsequent convergence The multi-fusion neural network model is recorded as a triage fusion model.
9. The computer device according to claim 9, wherein said performing standardized conversion on each of said triage results to obtain a standardized result corresponding to each of said triage results comprises:

Sort the triage categories in each of the triage results from large to small, and obtain the pre-set number of triage categories with the first sequence after the sorting, and determine the obtained triage results to be the same as each The sorting result corresponding to the triage result; the triage result includes the triage category;

Using standardization technology, standardize the assignment of each triage category in each of the sorting results;

Each of the sorting results after the assignment is determined as a standardized result corresponding to each of the sorting results one-to-one.
8. The computer device according to claim 9, wherein the weighted fusion of all the standardized results to obtain a sample triage result comprises:

Transform each of the standardized results into a triage vector of a preset dimension to obtain a triage vector corresponding to each of the standardized results;

According to each of the triage results and the triage label, determine the accuracy rate corresponding to each of the triage results;

According to all the accuracy rates, generate a weight corresponding to each of the triage vectors;

According to each of the triage vectors and each of the weights, the sample triage results are obtained.
The computer device according to claim 9, wherein the loss analysis is performed on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value, including :

Sort the sample triage results in descending order, and obtain a preset number of triage categories in the sequence of the sorted sample triage results, and determine the obtained sample triage results To filter the triage results; the sample triage results include the triage category and the probability value corresponding to the triage category;

Assigning sequence related values to each of the triage categories in the filtered triage results;

If there is the triage category that is the same as the triage label, the total probability value is substituted for the probability value corresponding to the triage category that is the same as the triage label, and the total probability value is all the Sum of probability values;

Input the sequence correlation value and the probability value corresponding to each of the triage categories in the filtered triage result into the loss model, and perform loss analysis through the loss model to obtain the total loss value .
The computer device according to claim 12, wherein said inputting said sequence correlation value and said probability value corresponding to each of said triage categories in said filtered triage result into said loss model, by The loss model performs loss analysis to obtain the total loss value, including:

The sequence correlation value and the probability value corresponding to each of the triage categories are input into the loss function in the loss model, and the total loss value is calculated by the loss function; the loss function is:

L=-(p 1 ×ln(k)+…+p i ×ln(k-i+1)+…+p k ×ln(1))

in,

L is the total loss value;

k is the preset number;

p 1 is the probability value corresponding to the triage category whose sequence is the first in the filtered triage result;

ln(k) is the sequence correlation value corresponding to the triage category whose sequence is the first in the filtered triage result;

p i is the probability value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

ln(k-i+1) is the sequence correlation value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

p k is the probability value corresponding to the triage category whose sequence is the kth in the filtered triage result;

ln(1) is the sequence correlation value corresponding to the triage category whose sequence is the kth in the filtered 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, wherein the processor further implements the following steps when executing the computer readable instructions :

Receive the patient's triage request, and obtain the patient's medical information in the triage request;

The patient visit information is input into the triage fusion model trained by the triage fusion model training method, and the final triage result output by the triage fusion model is obtained.
One or more readable storage media storing computer readable instructions, where when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps:

Obtaining a medical consultation sample set; the medical consultation sample set includes a plurality of medical consultation samples, and each medical consultation sample is associated with a triage label;

Inputting the consultation samples into a multi-fusion neural network model containing initial parameters;

Predicting the consultation samples by using the multi-fusion neural network model to obtain at least two triage results;

Perform standardized conversion on each of the triage results to obtain a standardized result corresponding to each of the triage results;

Perform weight fusion on all the standardized results to obtain sample triage results;

Performing a loss analysis on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value;

When the total loss value does not reach the preset convergence condition, iteratively update the initial parameters of the multi-fusion neural network model, until the total loss value reaches the preset convergence condition, the subsequent convergence The multi-fusion neural network model is recorded as a triage fusion model.
15. The readable storage medium according to claim 15, wherein the standardized conversion of each of the triage results to obtain a standardized result corresponding to each of the triage results comprises:

Sort the triage categories in each of the triage results from large to small, and obtain the pre-set number of triage categories with the first sequence after the sorting, and determine the obtained triage results to be the same as each The sorting result corresponding to the triage result; the triage result includes the triage category;

Using standardization technology, standardize the assignment of each triage category in each of the sorting results;

Each of the sorting results after the assignment is determined as a standardized result corresponding to each of the sorting results one-to-one.
15. The readable storage medium according to claim 15, wherein said performing weight fusion of all the standardized results to obtain sample triage results comprises:

Transform each of the standardized results into a triage vector of a preset dimension to obtain a triage vector corresponding to each of the standardized results;

According to each of the triage results and the triage label, determine the accuracy rate corresponding to each of the triage results;

According to all the accuracy rates, generate a weight corresponding to each of the triage vectors;

According to each of the triage vectors and each of the weights, the sample triage results are obtained.
The readable storage medium of claim 15, wherein the loss analysis is performed on the sample triage result and the triage label through the loss model in the multi-fusion neural network model to obtain a total loss value ,include:

Sort the sample triage results in descending order, and obtain a preset number of triage categories in the sequence of the sorted sample triage results, and determine the obtained sample triage results To filter the triage results; the sample triage results include the triage category and the probability value corresponding to the triage category;

Assigning sequence related values to each of the triage categories in the filtered triage results;

If there is the triage category that is the same as the triage label, the total probability value is substituted for the probability value corresponding to the triage category that is the same as the triage label, and the total probability value is all the Sum of probability values;

Input the sequence correlation value and the probability value corresponding to each of the triage categories in the filtered triage result into the loss model, and perform loss analysis through the loss model to obtain the total loss value .
The readable storage medium of claim 18, wherein the sequence correlation value and the probability value corresponding to each of the triage categories in the filtered triage result are input into the loss model , Perform loss analysis through the loss model to obtain the total loss value, including:

The sequence correlation value and the probability value corresponding to each of the triage categories are input into the loss function in the loss model, and the total loss value is calculated by the loss function; the loss function is:

L=-(p 1 ×ln(k)+…+p i ×ln(k-i+1)+…+p k ×ln(1))

in,

L is the total loss value;

k is the preset number;

p 1 is the probability value corresponding to the triage category whose sequence is the first in the filtered triage result;

ln(k) is the sequence correlation value corresponding to the triage category whose sequence is the first in the filtered triage result;

p i is the probability value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

ln(k-i+1) is the sequence correlation value corresponding to the triage category whose sequence is the i-th in the filtered triage result;

p k is the probability value corresponding to the triage category whose sequence is the kth in the filtered triage result;

ln(1) is the sequence correlation value corresponding to the triage category whose sequence is the kth in the filtered triage result.
One or more readable storage media storing computer readable instructions, where when the computer readable instructions are executed by one or more processors, the one or more processors further execute the following steps:

Receive the patient's triage request, and obtain the patient's medical information in the triage request;

The patient visit information is input into the triage fusion model trained by the triage fusion model training method, and the final triage result output by the triage fusion model is obtained.