CN111477310A - Triage data processing method, device, computer equipment and storage medium - Google Patents

Abstract

The invention discloses a triage data processing method, device, computer equipment and storage medium. The method includes: receiving a triage request, obtaining patient information; obtaining symptom information from the patient information through a maximum word matching method; The symptom information is input into the combined prediction model, and the symptom information is predicted and processed by the combined prediction model, and the first symptom department set is obtained; the symptom information is input into the reinforcement learning triage model, and the output containing the first symptom result and the first symptom output after performing the first action is obtained. The first triage result of a state result; when the first state result is the first state, the first symptom result in the first triage result is determined as the final symptom result; the final symptom result is the department where the patient visits. The invention realizes the preprocessing of dimensionality reduction through the probability model and the deep neural network model, and the automatic triage based on the reinforcement learning method, so as to determine the department where the patient seeks the disease, improve the diagnosis accuracy rate and improve the patient experience.

Description

Triage data processing method, device, computer equipment and storage medium

technical field

The invention relates to the field of data processing, in particular to a method, device, computer equipment and storage medium for triage data processing.

Background technique

At present, when patients go to the hospital for treatment, they first need to go to the triage desk for manual triage. During this process, patients need to spend a lot of queuing time, and there are higher requirements for the depth and breadth of professional knowledge of the service personnel of the triage desk. , if the service staff makes a wrong triage for the patient, it is necessary to re-triage the patient, which greatly wastes the patient's time and seriously affects the patient's experience. Reasonable consultation departments, resulting in poor patient experience and low diagnosis accuracy.

SUMMARY OF THE INVENTION

The invention provides a triage data processing method, device, computer equipment and storage medium, which realizes preprocessing of dimension reduction through probability model and deep neural network model, and automatic triage based on reinforcement learning method, which can quickly, Accurately determine the department where the patient needs to seek medical treatment, which improves the accuracy of medical treatment and improves the patient experience.

A triage data processing method, comprising:

Receive a triage request and obtain patient information;

Obtain symptom information from the patient information through a maximum word matching method;

Inputting the symptom information into a combined prediction model, performing prediction processing on the symptom information through the combined prediction model, and obtaining the first symptom department set output by the combined prediction model;

Input the symptom information into the reinforcement learning triage model, and obtain the first triage result output by the reinforcement learning triage model after executing the first action; wherein, the first action is the input of the reinforcement learning triage model to the triage model. After analyzing and processing the symptom information, it is selected from the first action space set, and the first action space set is a preset action space set that is activated after being activated by the first symptom department set; the first action space set is output; A triage result includes the first symptom result and the first state result;

When the first state result is the first state, the first symptom result in the first triage result is determined as the final symptom result; the final symptom result is the department where the patient visits.

A triage data processing device, comprising:

The receiving module is used to receive the triage request and obtain the patient information;

an acquisition module for acquiring symptom information from the patient information through a maximum word matching method;

a prediction module, configured to input the symptom information into a combined prediction model, perform prediction processing on the symptom information through the combined prediction model, and obtain the first symptom department set output by the combined prediction model;

An activation module, configured to input the symptom information into a reinforcement learning triage model, and obtain a first triage result output by the reinforcement learning triage model after performing a first action; wherein the first action is the reinforcement learning The triage model analyzes and processes the input symptom information and selects it from the first action space set, where the first action space set is a preset action space set activated by the first symptom department set output; the first triage result includes a first symptom result and a first state result;

An output module, configured to determine the first symptom result in the first triage result as the final symptom result when the first state result is the first state; the final symptom result is the department where the patient visits a doctor .

A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the triage data processing method when the processor executes the computer program.

A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, implements the steps of the above triage data processing method.

In the triage data processing method, device, computer equipment and storage medium provided by the present invention, accurate symptom information is obtained from the patient information by the maximum word matching method, and the symptom information is input into the symptom prediction probability model completed by training and A combined prediction model composed of a combination of deep convolutional neural network models of departments that have been trained, obtains the first symptom department set, and activates the action space collection through the first symptom department set to output the first action space set, which realizes that the action space set is subjected to dimensionality reduction processing as the first action space set in the reinforcement learning triage model, and then by inputting the symptom information into the reinforcement learning triage model, Obtain the first triage result output after the first action is performed, and if the first state result in the triage result is the first state, determine the first symptom result in the triage result as Final symptom results. It realizes dimensionality reduction preprocessing through probability model and deep neural network model, as well as automatic triage based on reinforcement learning method, which can quickly and accurately determine the departments that patients need to seek medical treatment, which saves patients' time and improves the accuracy of medical treatment. rate and improve the patient experience.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the drawings that are used in the description of the embodiments of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention. , for those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative labor.

1 is a schematic diagram of an application environment of a triage data processing method in an embodiment of the present invention;

2 is a flowchart of a method for processing triage data in an embodiment of the present invention;

3 is a flowchart of a triage data processing method in another embodiment of the present invention;

4 is a flowchart of step S10 of the triage data processing method in an embodiment of the present invention;

5 is a flowchart of step S20 of the triage data processing method in an embodiment of the present invention;

6 is a flowchart of step S30 of the triage data processing method in an embodiment of the present invention;

7 is a flowchart of step S301 of the triage data processing method in an embodiment of the present invention;

8 is a flowchart of step S302 of the triage data processing method in an embodiment of the present invention;

9 is a schematic block diagram of a triage data processing device according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of a computer device in an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The triage data processing method provided by the present invention can be applied in the application environment as shown in FIG. 1 , wherein the client (computer device) 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 one embodiment, as shown in FIG. 2 , a method for processing triage data is provided, and its technical solution mainly includes the following steps S10-S50:

S10, a triage request is received, and patient information is acquired.

Understandably, after the triage request is received, the patient information is obtained, and the triage request is a request triggered after selecting and confirming the patient information that needs to be triaged, and the triggering method can be performed according to requirements. Settings, such as providing a trigger button on the application platform interface that can be triggered by clicking, sliding, etc., automatically triggering after executing a preset program, and so on.

Wherein, the patient information is the information related to the symptoms of the patient input by the patient.

In one embodiment, as shown in FIG. 4 , before the step S10, that is, before the triage request is received and the patient information is obtained, the steps include:

S101, a patient input instruction is received, and patient input information is acquired.

Understandably, receive the patient input instruction, obtain the patient input information, and the patient input instruction is an instruction triggered after the patient input information is input on the display interface of the application program, and the patient input instruction is received. After that, the patient input information is acquired, and the acquisition method can be set as required. For example, the acquisition method can be to acquire the patient input information through the patient input instruction, and obtain the patient input information according to the patient input instruction included in the patient input instruction. A storage path for information captures the patient input information, and the like.

S102: Input the patient input information into a preset preprocessing model, and the preprocessing model recognizes the patient input information to obtain a recognition result; wherein the recognition result includes text, voice and image.

Understandably, the preprocessing model is a preset model for identifying the patient input information, the patient input information is input into the preprocessing model, and the preprocessing model can be based on the patient input information. The format of , determines the recognition result, and the recognition result includes text, speech, and image.

S103: Acquire a conversion model corresponding to the recognition result.

Understandably, a conversion model corresponding to the recognition result is determined according to the recognition result, and the conversion model includes a text conversion model, a speech conversion model, and an image conversion model, that is, if the recognition result is text, the text is obtained. Conversion model, if the recognition result is a voice, then obtain a voice conversion model, if the recognition result is an image, then obtain an image conversion model, and the conversion model is a neural network model that has been trained. Targeted conversion models can improve conversion efficiency and accuracy.

S104: Input the patient input information into the conversion model, and the conversion model performs text conversion on the patient input information, and outputs a conversion result.

Understandably, the patient input information is input into a conversion model corresponding to the recognition result, and the conversion model is used for conversion into the conversion result.

S105: Determine the conversion result as the patient information.

In this way, by performing text, voice and image recognition on the patient input information input by the patient, and corresponding to different conversion models according to different recognition results, the patient information is obtained from the patient input information, providing a variety of The input channel is given to the patient, which improves the patient experience.

S20, obtain symptom information from the patient information through a maximum word matching method.

Understandably, through the maximum word matching method, the patient information is split into multiple single texts, and the multiple single texts are maximalized, that is, the single text and the corresponding single text before and after. Combining, generating pre-text, post-text and full-text, and obtaining the relationship between the single text, the pre-text, the post-text, the full-text and the text in the preset symptom thesaurus. For the text with the highest matching value, the text with the highest matching value is determined as the largest phrase corresponding to the single text, and all the largest phrases are cleared, that is, the largest phrase with zero matching value is removed, and the clearing process is performed. All the largest phrases after zero processing are determined as the symptom information.

In one embodiment, as shown in FIG. 5 , in the step S20, that is, the maximum word matching method is used to obtain symptom information from the patient information, including:

S201: Acquire a preset symptom thesaurus; the symptom thesaurus includes a plurality of symptom words.

Understandably, the symptom thesaurus is a warehouse that stores all symptom words, that is, the symptom thesaurus contains a plurality of symptom words, and the symptom words can be added to the symptom thesaurus according to requirements, and the symptoms The thesaurus can be added over time, and the symptom word is a word that gives a symptom a name. For example, the symptom word database can filter out the symptom words of common symptoms in the public data of symcat.

S202, splitting the patient information into multiple single texts.

Understandably, the single text can be a single word or a single phrase, for example, if the patient information is "coughing from yesterday to today", it can be divided into "yesterday", "to", "today", "always". ","cough".

S203: Obtain the start position and the end position of the single text, combine the previous single text at the start position with the single text to generate pre-text, and combine the last single text at the end position with the single text The text is combined to generate the post text, and the previous single text at the start position, the single text and the next single text at the end position are combined to generate the full text.

Understandably, the starting position can be set according to requirements, for example, the starting position can be that the single text can reach the single text after the first digit from the left to the right in the patient information starts. Then the number of digits is the start position of the single text; the end position can also be set according to requirements, for example, the end position can be obtained by adding the number of digits of the single text after the start position; the The previous single text at the start position is combined with the single text to generate pre-text, the next single text at the end position is combined with the single text to generate post-text, and the previous single text at the start position is combined. A single text, the single text and the last single text at the end position are combined to generate a full text, for example: the patient information is epigastric pain for three days, and the split multiple single texts are upper, abdominal pain, Three days, with abdominal pain as a single text, the pre-text text is epigastric pain, the back text is abdominal pain three days, and the full text is epigastric pain three days.

S204: Obtain the matching value of the single text, the pre-text, the post-text and the full text with the text in the symptom lexicon, and determine the text with the highest matching value as The largest phrase corresponding to the single text.

Understandably, the matching value is the number of words that match the same text, search and match the single text with the text in the symptom lexicon, obtain a matching value corresponding to the single text, and use the The pre-text is searched and matched with the text in the symptom lexicon to obtain a matching value corresponding to the pre-text, and the post-text is searched and matched with the text in the symptom lexicon to obtain For the matching value corresponding to the post text, search and match the full text with the text in the symptom lexicon to obtain the matching value corresponding to the full text, for example: in the above example, the corresponding The matching value is 2, the matching value corresponding to epigastric pain is 3, the matching value corresponding to three days of abdominal pain is 0, and the matching value corresponding to three days of abdominal pain is 0.

S205: Perform zero-clearing processing on all the largest phrases corresponding to each of the single texts in the patient information, and determine all the largest phrases after zero-clearing processing as the symptom information.

Understandably, among all the largest phrases corresponding to each of the single texts, the largest phrases with a matching value of zero are removed, and all the removed largest phrases are determined as the symptom information.

In this way, the maximum word matching method can ensure the accuracy of symptom acquisition, and improve the correctness and accuracy of symptom acquisition.

S30: Input the symptom information into a combined prediction model, perform prediction processing on the symptom information by using the combined prediction model, and obtain a first symptom department set output by the combined prediction model.

Understandably, the combined prediction model is a model composed of a combination of a trained symptom prediction probability model and a trained department deep convolutional neural network model, and the symptom prediction probability model is to input the first symptom sample into the Bayesian model. The probability model is trained and obtained, the department deep convolutional neural network model is obtained by inputting the second symptom sample into the deep neural network model for training, and the first symptom department set is the predicted probability distribution output by the symptom prediction probability model The result is obtained by splicing and normalizing the department prediction distribution results output by the department deep convolutional neural network model. The predicted probability distribution results are the distribution map of the probability values corresponding to all the symptom words, and the department prediction The distribution result is the distribution map of the probability values corresponding to all departments. The department is the department name set according to the needs, such as surgery, internal medicine, orthopedics, etc. The predicted probability distribution is a set of data, and the department predicts the distribution result. It is also a set of data, and the first symptom department set is a set of arrays after the prediction probability distribution and the department prediction distribution result are processed.

In one embodiment, as shown in FIG. 6 , in step S30, that is, inputting the symptom information into a combination prediction model, performing prediction processing on the symptom information through the combination prediction model, and obtaining the combination The first symptom department set output by the prediction model, including:

S301: Input the symptom information into a trained symptom prediction probability model, predict the symptom information by using the symptom prediction probability model, and obtain a prediction probability distribution result output by the symptom prediction probability model; wherein the prediction The probability distribution result characterizes a matching probability distribution of symptoms in the symptom set associated with the symptom information.

Understandably, the symptom prediction probability model is a trained Bayesian probability model, and by inputting the symptom information into the symptom prediction probability model for prior distribution processing, the prediction probability distribution result can be obtained, and the A predicted probability distribution result is a matching probability distribution of symptoms in the symptom set associated with the symptom information, the predicted probability distribution result is an array of percentage values ranging from 0% to 100%, the symptom The set is the total set of all symptom words.

In one embodiment, as shown in FIG. 7 , before the step S301, that is, before the symptom information is input into the trained symptom prediction probability model, it includes:

S3011: Obtain a first symptom sample; wherein each of the first symptom samples is associated with a symptom category label.

The first symptom sample is collected symptom information, and each first symptom sample is manually determined and associated with a symptom category label, that is, each first symptom sample is labeled, and the symptom category Labels are symptom words in the symptom set.

S3012: Input the first symptom sample into a Bayesian probability model including a first initial parameter.

Understandably, the Bayesian probability model is a model based on a neural network structure formed by a Bayesian algorithm, and the Bayesian probability model includes the first initial parameter, wherein the first initial parameter may be Set according to requirements, for example, set the first initial parameter to a default value by default, or a random parameter value, and so on.

S3013: Perform prior distribution processing on the first symptom sample by using the Bayesian probability model.

Understandably, the prior distribution processing is the probability distribution processing in the Bayesian algorithm, and the prior distribution processing is performed on the first symptom sample.

S3014: Obtain the distribution result output by the Bayesian probability model, and determine a first loss value according to the degree of matching between the distribution result and the symptom category label.

Understandably, according to the probability distribution processed by the prior distribution according to the Bayesian probability model, the probability distribution is a distribution diagram of the probability values corresponding to all the symptom words in the symptom set, and the obtained The distribution result of the Bayesian probability model, the distribution result is the distribution map of the probability values corresponding to all the symptom words, and the symptom words corresponding to all the probability values in the distribution result are obtained. The corresponding probability value is compared with the symptom category label of the first symptom sample to determine the corresponding loss value, that is, the first loss value is calculated through the loss function of the Bayesian probability model. S3015, when the first loss value reaches a preset first convergence condition, record the Bayesian probability model after convergence as a trained symptom prediction probability model.

The preset first convergence condition may be a condition that the loss value is small after 500 calculations and will not decrease again, that is, the loss value is small after 500 calculations And when it does not drop any more, stop training, and record the Bayesian probability model after convergence as the symptom prediction probability model after training; the preset first convergence condition can also be that the loss value is less than the set value. The condition of setting a threshold value, that is, when the loss value is less than the set threshold value, stop training, and record the Bayesian probability model after convergence as the symptom prediction probability model after training.

In this way, by inputting the first symptom sample to train the Bayesian probability model, the accuracy and reliability of the distribution result can be improved.

In one embodiment, after step S3014, that is, after obtaining the distribution result output by the Bayesian probability model, and determining the loss value according to the degree of matching between the distribution result and the symptom category label, the method includes:

S3016, when the first loss value does not reach the preset first convergence condition, iteratively update the first initial parameter of the Bayesian probability model until the first loss value reaches the preset first When the convergence conditions are met, the Bayesian probability model after convergence is recorded as the symptom prediction probability model after training.

In this way, when the first loss value does not reach the preset first convergence condition, the first initial parameter of the iterative Bayesian probability model is continuously updated, so as to continuously move closer to the accurate distribution result and make the distribution result more accurate. rate is getting higher.

S302, input the symptom information into the trained department deep convolutional neural network model, perform text feature extraction on the symptom information through the department deep convolutional neural network model, and obtain the department deep convolutional neural network model output The department prediction distribution result; wherein the department prediction distribution result represents the matching probability distribution of the departments related to the symptom information in the department set.

Understandably, the department deep convolutional neural network model is a trained deep neural network model, and by inputting the symptom information into the deep convolutional neural network model for text feature extraction, the predicted distribution result of the department can be obtained. , the department prediction distribution result is the matching probability distribution of the departments related to the symptom information in the department set, the department set is the total set of departments in all hospitals, and the department prediction distribution result is a group consisting of An array of percentage values in the range 0% to 100%.

In one embodiment, as shown in FIG. 8 , before the step S302, that is, before the input of the symptom information into the trained department deep convolutional neural network model, the method includes:

S3021: Obtain a second symptom sample; wherein each of the second symptom samples is associated with a department label.

The second symptom sample is collected symptom information, and each second symptom sample is manually determined and associated with a department label, that is, each second symptom sample is labeled, and the department label is All the corresponding names of the departments mentioned.

S3022: Input the second symptom sample into a deep neural network model including second initial parameters.

Understandably, the deep neural network model is a model based on a multi-classified neural network structure, and the network structure of the deep neural network model can be set according to requirements, for example, the neural network structure can be selected as VGG, GoogLeNet, etc. , the deep neural network model includes the second initial parameter, where the second initial parameter can be set according to requirements, for example, setting the second initial parameter to a default preset value, or a random parameter values, etc.

S3023, extracting text features in the symptom samples by using the deep neural network model.

Understandably, the text feature is to convert text data into a feature vector, and a commonly used method for extracting text features is the bag-of-words extraction, and the feature vector corresponding to the text is obtained according to the frequency of occurrence.

S3024: Obtain the recognition result output by the deep neural network model according to the text feature, and determine a second loss value according to the matching degree between the recognition result and the department label.

Understandably, according to the text features extracted by the deep neural network model, the recognition result of the deep neural network model is obtained, and the recognition result is the distribution map of the probability values corresponding to all the departments, and the For all probability values in the identification results, the probability values corresponding to all the departments are compared with the department labels of the second symptom samples, and the corresponding loss values are determined, that is, through the deep neural network model. A loss function calculates the second loss value.

S3025, when the second loss value reaches a preset second convergence condition, record the deep neural network model after convergence as a trained department deep convolutional neural network model.

The preset second convergence condition may be a condition that the second loss value is small and will not decrease after 5,000 calculations, that is, after the second loss is calculated 5,000 times When the value is very small and will not drop any more, stop training, and record the deep neural network model after convergence as the department deep convolutional neural network model after training; the preset second convergence condition can also be The condition that the second loss value is less than the set threshold, that is, when the second loss value is less than the set threshold, the training is stopped, and the deep neural network model after convergence is recorded as the department deep convolutional neural network after training. network model.

In this way, by training the initial neural network model according to the document type label value in the training image sample, it is possible to perform recognition by extracting texture features and improve the accuracy and reliability of the recognition result.

In one embodiment, after step S3024, it includes:

S3026, when the second loss value does not reach a preset second convergence condition, iteratively update a second initial parameter of the deep neural network model until the second loss value reaches the preset second convergence condition When conditions are met, the deep neural network model after convergence is recorded as the department deep convolutional neural network model after training.

In this way, when the second loss value does not reach the preset second convergence condition, the second initial parameter of the iterative deep neural network model is continuously updated, which can continuously move closer to the accurate recognition result, so that the accuracy of the recognition result becomes more and more accurate. higher.

S303, splicing and normalizing the predicted probability distribution result and the predicted distribution result of the department to obtain the first symptom department set.

Understandably, splicing the predicted probability distribution result with the predicted distribution result of the department, that is, splicing the array corresponding to the predicted probability distribution result with the array corresponding to the predicted distribution result of the department will obtain the result of the action space. The splicing array of the same dimension is collected, and the splicing array is normalized, that is, the percentage value in the splicing array that is greater than or equal to the preset percentage threshold is normalized to 1, and the percentage value in the splicing array is normalized to 1. The percentage value smaller than the preset percentage threshold is normalized to 0, and the normalized splicing array is determined as the first symptom department set.

In this way, the symptom information is predicted by the symptom prediction probability model, the predicted probability distribution result of the matching probability distribution of the symptoms related to the symptom information is obtained, and the deep convolutional neural network model of the department is used Predict the symptom information, obtain the department prediction distribution result of the matching probability distribution of the department related to the symptom information, and then splicing and normalizing the predicted probability distribution result and the department prediction distribution result After processing, the first symptom department set is obtained, which realizes reasonable prediction of the symptom information, and provides an activation factor for the action space total set.

S40: Input the symptom information into a reinforcement learning triage model, and obtain a first triage result output by the reinforcement learning triage model after executing a first action; wherein the first action is the reinforcement learning triage model After analyzing and processing the input symptom information, it is selected from the first action space set, and the first action space set is output after the preset action space set is activated by the first symptom department set; The first triage result includes a first symptom result and a first state result.

Understandably, the reinforcement learning triage model can be set as a learning method according to requirements. Preferably, the reinforcement learning triage model is set to the DQN learning method, and the symptom information is used as the Agent of the reinforcement learning triage model. (agent), the reinforcement learning triage model guides the input symptom information, selects a first action from the first action space set, and the first action is to execute the symptom information The action space collection is a collection containing all actions, and the action space collection is a set of arrays, wherein the action space collection is the same as the array dimension of the first symptom department collection, and the After the action space set is activated by the first symptom department set, the first action space set is output, and the activation process is to match the action space set and the first symptom department set according to the activation principle. Obtain, the activation principle mode is to match and retain or delete the array contents corresponding to two arrays one-to-one, that is, the array contents of the action space total set are judged according to the corresponding value in the array of the first symptom department set. Retain or delete, if the value of the first symptom department set is 1, then the contents of the array in the action space total set corresponding to the value of the symptom department set will be retained, if the first symptom department set If the value is 0, the contents of the array in the action space set corresponding to the value of the symptom department set are deleted, and the first action space set finally obtained is the preset action space set Obtained after being activated by the first symptom department set, the first action space set provides a set of all guided actions, and the reinforcement learning triage model outputs the output after performing the first action on the symptom information. The first triage result.

In this way, the first action space set is activated by the first symptom department set to output the first action space set, so that the dimensionality reduction process of the action space set is realized as the reinforcement learning triage model The first action space set in can improve the learning effect of the reinforcement learning triage model, and improve the accuracy of the reinforcement learning triage model.

S50, when the first state result is the first state, determine the first symptom result in the first triage result as the final symptom result; the final symptom result is the department where the patient seeks a doctor.

Understandably, the first state can be set as a terminal state, that is, the final state is reached after the interactive dialogue with the reinforcement learning triage model, and when the result of the first state is the first state, the The first symptom result is determined as the final symptom result, and the symptom result is the department to which the patient needs to see a doctor.

In the present invention, accurate symptom information is obtained from the patient information through the maximum word matching method, and the symptom information is input into the combined prediction model (a symptom prediction probability model completed by training and a department deep convolutional neural network model completed by training). The model composed of combination), obtain the first symptom department set, activate the action space set through the first symptom department set, output the first action space set, and realize the action space set After dimensionality reduction processing is performed, it is used as the first action space set in the reinforcement learning triage model (the learning effect of the reinforcement learning triage model can be improved, and the accuracy of the reinforcement learning triage model can be improved), Then, by inputting the symptom information into the reinforcement learning triage model, obtain the first triage output after executing the first action (selected from the first action space set after analyzing and processing the symptom information). As a result, if the first state result in the triage result is the first state (termination state), the first symptom result in the triage result is determined as the final symptom result (department where the patient visits).

In this way, the present invention realizes the preprocessing of dimensionality reduction through the probability model and the deep neural network model, and the automatic triage based on the reinforcement learning method, which can quickly and accurately determine the department where the patient needs to seek medical treatment, and saves the patient's time. Improve the accuracy of diagnosis and improve the patient experience.

In one embodiment, the first triage result further includes a first reward result; as shown in FIG. 3 , after the step S40, that is, the acquired result output after the reinforcement learning triage model performs the first action. After the first triage results, including:

S60, when the first state result is the second state, use the first symptom result in the first triage result as the next symptom information, and at the same time use the first reward result and the next Symptom information association.

Understandably, the second state may be a non-terminal state, that is, the final state has not been reached after the interactive dialogue with the reinforcement learning triage model. The input of the interactive dialogue of the reinforcement learning triage model, that is, the first symptom result is used as the next symptom information, and the first triage result also includes the first reward result, and the first reward result is the reinforcement learning The reward value given to the first action by the triage model after performing the first action, the reward value can be the value given each time, or the value that is continuously accumulated according to the number of interactive dialogues, through the reward value The reinforcement learning triage model can be provided to move closer to the accurate diagnosis department (in the direction of maximizing reward value), and associate the first reward result with the next symptom information.

S70: Input the next symptom information into the combined prediction model, perform prediction processing on the symptom information by using the combined prediction model, and obtain a second symptom department set output by the combined prediction model.

Understandably, by inputting the next symptom information into the combined prediction model, the second symptom department set is obtained, the second symptom department set is also an array, and the second symptom department set is the same as The first symptom department sets have the same dimension.

S80: Input the next symptom information and the first reward result associated with the next symptom information into the reinforcement learning triage model, and obtain the first reward output after the reinforcement learning triage model performs the second action. Two triage results; wherein, the second action is that the reinforcement learning triage model analyzes and processes the input first symptom information and the first reward result and selects it from the second action space set , the second action space set is output after the action space total set is activated by the second symptom department set; the second triage result includes a second symptom result and a second state result.

Understandably, the reinforcement learning triage model analyzes the inputted next symptom information and the first reward result associated with the next symptom information, so as to guide the next symptom information, A second action is selected from the second action space set, and the second action is an action to be performed on the symptom information, wherein the array dimension of the action space set and the second symptom department set In the same way, the action space set is activated and processed by the second symptom department set, and the second action space set is output. The second action space set provides a set of all guided actions, and the reinforcement learning triage The model outputs the second triage result after performing the second action on the symptom information.

S90, when the second state result is the first state, determine the second symptom result in the second triage result as the final symptom result; the final symptom result is the department where the patient seeks a doctor.

In this way, the reinforcement learning triage model learns from the interactive dialogue and guides the direction of the optimal strategy to move closer, and when in the second state (non-terminal state), the reinforcement learning triage model continuously learns and interacts with dialogue, Until it is in the first state (termination state), the final symptom result is output, and the accuracy of medical treatment is improved.

In one embodiment, a triage data processing apparatus is provided, and the triage data processing apparatus is in one-to-one correspondence with the triage data processing method in the above embodiment. As shown in FIG. 9 , the triage data processing apparatus includes a receiving module 11 , an acquisition module 12 , a prediction module 13 , an activation module 14 and an output module 15 .

The detailed description of each functional module is as follows:

The receiving module 11 is used to receive a triage request and obtain patient information;

an acquisition module 12 for acquiring symptom information in the patient information through a maximum word matching method;

A prediction module 13, configured to input the symptom information into a combined prediction model, perform prediction processing on the symptom information through the combined prediction model, and obtain the first symptom department set output by the combined prediction model;

The activation module 14 is configured to input the symptom information into the reinforcement learning triage model, and obtain the first triage result output after the reinforcement learning triage model performs the first action; wherein, the first action is the reinforcement The learning triage model analyzes and processes the input symptom information and selects it from the first action space set, and the first action space set is a preset action space set activated by the first symptom department set and then output; the first triage result includes a first symptom result and a first state result;

The output module 15 is configured to, when the first state result is the first state, determine the first symptom result in the first triage result as the final symptom result; the final symptom result is the result of the patient's consultation department.

In one embodiment, the activation module 14 includes:

The reward unit is configured to use the first symptom result in the first triage result as the next symptom information when the first state result is the second state, and at the same time combine the first reward result with all associated with the next symptom information;

an input unit, configured to input the next symptom information into the combined prediction model, perform prediction processing on the symptom information through the combined prediction model, and obtain a second symptom department set output by the combined prediction model;

An activation unit, configured to input the next symptom information and the first reward result associated with the next symptom information into the reinforcement learning triage model, and obtain after the reinforcement learning triage model performs the second action The output of the second triage result; wherein, the second action is that the reinforcement learning triage model analyzes and processes the input first symptom information and the first reward result from the second action space Selected from the set, the second action space set is output after the action space total set is activated by the second symptom department set; the second triage result includes the second symptom result and the second state result;

A determination unit, configured to determine the second symptom result in the second triage result as the final symptom result when the second state result is the first state; the final symptom result is the department where the patient visits a doctor .

In one embodiment, the receiving module 11 includes:

a receiving unit, configured to receive the patient input instruction and obtain the patient input information;

a recognition unit, configured to input the patient input information into a preset preprocessing model, and the preprocessing model recognizes the patient input information to obtain a recognition result; wherein the recognition result includes text, voice and image;

a selection unit for obtaining a conversion model corresponding to the recognition result;

a conversion unit, configured to input the patient input information into the conversion model, and the conversion model performs text conversion on the patient input information, and outputs a conversion result;

A result output unit, configured to determine the conversion result as the patient information.

In one embodiment, the obtaining module 12 includes:

a first obtaining unit, configured to obtain a preset symptom thesaurus; the symptom thesaurus includes a plurality of symptom words;

a splitting unit for splitting the patient information into multiple single texts;

The second obtaining unit is configured to obtain the start position and the end position of the single text, combine the previous single text at the start position with the single text to generate the pre-text, and obtain the last single text at the end position. The text is combined with the single text to generate post-text, and the previous single text at the start position, the single text and the next single text at the end position are combined to generate a full text;

The matching unit is used to obtain the matching value of the text in the single text, the pre-text, the post-text and the full text with the text in the symptom lexicon, and select the one with the highest matching value. The text is determined as the largest phrase corresponding to the single text;

A clearing unit, configured to perform clearing processing on all the largest phrases corresponding to each of the single texts in the patient information, and determine all the largest phrases after clearing processing as the symptom information.

In one embodiment, the prediction module 13 includes:

a first model unit, configured to input the symptom information into a trained symptom prediction probability model, predict the symptom information by using the symptom prediction probability model, and obtain a prediction probability distribution result output by the symptom prediction probability model; Wherein, the predicted probability distribution result represents a matching probability distribution of symptoms in the symptom set related to the symptom information;

The second model unit is used to input the symptom information into the department deep convolutional neural network model that has been trained, and perform text feature extraction on the symptom information through the department deep convolutional neural network model, and obtain the department deep volume The department prediction distribution result output by the cumulative neural network model; wherein, the department prediction distribution result represents the matching probability distribution of the department related to the symptom information in the department set;

A splicing unit, configured to splicing and normalizing the predicted probability distribution result and the predicted distribution result of the department to obtain the first symptom department set.

In one embodiment, the first model unit includes:

a first obtaining subunit, used to obtain a first symptom sample; wherein each of the first symptom samples is associated with a symptom category label;

a first input subunit, configured to input the first symptom sample into a Bayesian probability model including a first initial parameter;

a first processing subunit, configured to perform prior distribution processing on the first symptom sample by using the Bayesian probability model;

a first output subunit, configured to obtain a distribution result output by the Bayesian probability model, and determine a first loss value according to the degree of matching between the distribution result and the symptom category label;

A first convergence subunit, configured to record the Bayesian probability model after convergence as a trained symptom prediction probability model when the first loss value reaches a preset first convergence condition.

In one embodiment, the second model unit includes:

a second acquisition subunit, configured to acquire a second symptom sample; wherein each of the second symptom samples is associated with a department label;

a second input subunit, configured to input the second symptom sample into the deep neural network model including the second initial parameter;

a second processing subunit, configured to extract text features in the symptom samples through the deep neural network model;

The second output subunit is used to obtain the recognition result output by the deep neural network model according to the text feature, and determine the second loss value according to the degree of matching between the recognition result and the department label;

The second convergence subunit is configured to record the deep neural network model after convergence as a trained department deep convolutional neural network model when the second loss value reaches a preset second convergence condition.

For the specific limitation of the triage data processing apparatus, reference may be made to the above limitation on the triage data processing method, which will not be repeated here. Each module in the above triage data processing device can be implemented in whole or in part by software, hardware and combinations thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

In one embodiment, a computer device is provided, and the computer device may be a server, and its internal structure diagram may be as shown in FIG. 10 . The computer device includes a processor, memory, a network interface, and a database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The nonvolatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program implements a triage data processing method when executed by the processor.

In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the triage data processing method in the above embodiment when the processor executes the computer program .

In one embodiment, a computer-readable storage medium is provided, and a computer program is stored thereon, and when the computer program is executed by a processor, the method for processing triage data in the foregoing embodiment is implemented.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage In the medium, when the computer program is executed, it may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other medium used in the various embodiments provided by the present invention may include non-volatile and/or volatile memory. Nonvolatile 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. By way of illustration and not limitation, RAM is available in various forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Road (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 simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. Module completion, that is, dividing the internal structure of the device 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 invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it is still possible to implement the foregoing implementations. The technical solutions described in the examples are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the within the protection scope of the present invention.

Claims (10)

1. a triage data processing method, is characterized in that, comprises: Receive a triage request and obtain patient information; Obtain symptom information from the patient information through a maximum word matching method; Inputting the symptom information into a combined prediction model, performing prediction processing on the symptom information through the combined prediction model, and obtaining the first symptom department set output by the combined prediction model; Input the symptom information into the reinforcement learning triage model, and obtain the first triage result output by the reinforcement learning triage model after executing the first action; wherein, the first action is the input of the reinforcement learning triage model to the triage model. After analyzing and processing the symptom information, it is selected from the first action space set, and the first action space set is a preset action space set that is activated after being activated by the first symptom department set; the first action space set is output; A triage result includes the first symptom result and the first state result; When the first state result is the first state, the first symptom result in the first triage result is determined as the final symptom result; the final symptom result is the department where the patient visits. 2. The triage data processing method according to claim 1, wherein the first triage result further comprises a first reward result; That is, after obtaining the first triage result output after the reinforcement learning triage model performs the first action, it includes: When the first state result is the second state, the first symptom result in the first triage result is used as the next symptom information, and the first reward result and the next symptom information are used simultaneously. association; Inputting the next symptom information into the combined prediction model, performing prediction processing on the symptom information through the combined prediction model, and obtaining the second symptom department set output by the combined prediction model; Input the next symptom information and the first reward result associated with the next symptom information into the reinforcement learning triage model, and obtain the second score output after the reinforcement learning triage model performs the second action. diagnosis result; wherein, the second action is that the reinforcement learning triage model analyzes and processes the input first symptom information and the first reward result and selects it from the second action space set. The second action space set is output after the action space total set is activated by the second symptom department set; the second triage result includes a second symptom result and a second state result; When the second state result is the first state, the second symptom result in the second triage result is determined as the final symptom result; the final symptom result is the department where the patient visits. 3. triage data processing method as claimed in claim 1, is characterized in that, described receiving triage request, before obtaining patient information, comprising: Receive patient input instructions, and obtain patient input information; Inputting the patient input information into a preset preprocessing model, and the preprocessing model recognizes the patient input information to obtain a recognition result; wherein the recognition result includes text, voice and image; obtaining a conversion model corresponding to the recognition result; Inputting the patient input information into the conversion model, and the conversion model performs text conversion on the patient input information, and outputs a conversion result; The conversion result is determined as the patient information. 4. The triage data processing method according to claim 1, characterized in that, obtaining symptom information in the patient information by a maximum word matching method, comprising: obtaining a preset symptom thesaurus; the symptom thesaurus includes a plurality of symptom words; splitting the patient information into multiple single texts; Obtain the start position and the end position of the single text, combine the previous single text at the start position with the single text to generate a pre-text, and perform the last single text at the end position with the single text. Combining to generate post text, combining the previous single text at the start position, the single text and the last single text at the end position to generate a full text; Obtain the matching value of the single text, the pre-text, the post-text and the full text with the text in the symptom lexicon, and determine the text with the highest matching value as the text with the highest matching value. The largest phrase corresponding to the single text; Perform zero clearing processing on all the largest phrases corresponding to each of the single texts in the patient information, and determine all the largest phrases after zero clearing as the symptom information. 5 . The triage data processing method according to claim 1 , wherein the symptom information is input into a combined prediction model, the symptom information is predicted and processed by the combined prediction model, and the combination is obtained. 6 . The first symptom department set output by the prediction model, including: Input the symptom information into the trained symptom prediction probability model, predict the symptom information through the symptom prediction probability model, and obtain the prediction probability distribution result output by the symptom prediction probability model; wherein, the prediction probability distribution the result characterizes the matching probability distribution of symptoms in the symptom set related to the symptom information; Input the symptom information into the trained department deep convolutional neural network model, perform text feature extraction on the symptom information through the department deep convolutional neural network model, and obtain the department output by the department deep convolutional neural network model a prediction distribution result; wherein the department prediction distribution result represents the matching probability distribution of the departments related to the symptom information in the department set; The predicted probability distribution result and the department predicted distribution result are spliced and normalized to obtain the first symptom department set. 6. The triage data processing method according to claim 5, characterized in that, before the described symptom information is input into the trained symptom prediction probability model, the method comprises: obtaining a first symptom sample; wherein each of the first symptom samples is associated with a symptom category label; inputting the first symptom sample into a Bayesian probability model including a first initial parameter; Perform prior distribution processing on the first symptom sample by using the Bayesian probability model; obtaining a distribution result output by the Bayesian probability model, and determining a first loss value according to the degree of matching between the distribution result and the symptom category label; When the first loss value reaches a preset first convergence condition, the Bayesian probability model after convergence is recorded as a trained symptom prediction probability model. 7. The triage data processing method according to claim 5, characterized in that, before the described symptom information is input into the department deep convolutional neural network model that has been trained, the method comprises: obtaining second symptom samples; wherein each of the second symptom samples is associated with a department label; inputting the second symptom sample into a deep neural network model including second initial parameters; Extract text features in the symptom samples through the deep neural network model; Obtain the recognition result output by the deep neural network model according to the text feature, and determine the second loss value according to the degree of matching between the recognition result and the department label; When the second loss value reaches a preset second convergence condition, the deep neural network model after convergence is recorded as a trained department deep convolutional neural network model. 8. A triage data processing device, comprising: The receiving module is used to receive the triage request and obtain the patient information; an acquisition module for acquiring symptom information from the patient information through a maximum word matching method; a prediction module, configured to input the symptom information into a combined prediction model, perform prediction processing on the symptom information through the combined prediction model, and obtain the first symptom department set output by the combined prediction model; An activation module, configured to input the symptom information into a reinforcement learning triage model, and obtain a first triage result output by the reinforcement learning triage model after performing a first action; wherein the first action is the reinforcement learning The triage model analyzes and processes the input symptom information and selects it from the first action space set, where the first action space set is a preset action space set activated by the first symptom department set output; the first triage result includes a first symptom result and a first state result; An output module, configured to determine the first symptom result in the first triage result as the final symptom result when the first state result is the first state; the final symptom result is the department where the patient visits a doctor . 9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the computer program as claimed in the claims The triage data processing method according to any one of 1 to 7. 10. A computer-readable storage medium storing a computer program, wherein the computer program realizes the triage data according to any one of claims 1 to 7 when the computer program is executed by a processor Approach.

Images