WO2022147910A1

WO2022147910A1 - Medical record information verification method and apparatus, and computer device and storage medium

Info

Publication number: WO2022147910A1
Application number: PCT/CN2021/083196
Authority: WO
Inventors: 朱昭苇; 孙行智; 胡岗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2021-01-11
Filing date: 2021-03-26
Publication date: 2022-07-14
Also published as: CN112820367B; CN112820367A

Abstract

A medical record information verification method and apparatus, and a computer device and a storage medium, which relate to the technical field of detection models, and can be applied to the field of smart healthcare, thereby promoting the construction of smart cities. The method comprises: inputting, into a case representation model, case information of medical record text to be verified, so as to obtain a case representation vector; inputting department information into a department representation model, so as to obtain a department representation vector; performing splicing processing on the case representation vector and the department representation vector, so as to obtain a medical record splicing vector; inputting the medical record splicing vector into a case discrimination network model, and determining at least one case determination result corresponding to said medical record text; and matching diagnosis information with each case determination result, and when the diagnosis information successfully matches any case determination result, determining that said medical record text is successfully verified. By means of the method, the efficiency and accuracy of medical record information verification can be improved.

Description

Medical record information verification method, device, computer equipment and storage medium

This application claims the priority of the Chinese patent application filed on January 11, 2021 with the application number 202110032946.6 and the invention titled "Medical Record Information Verification Method, Device, Computer Equipment and Storage Medium", the entire content of which is approved by Reference is incorporated in this application.

technical field

The present application relates to the technical field of detection models, and in particular, to a medical record information verification method, device, computer equipment and storage medium.

Background technique

With the development of science and technology, the medical system has gradually improved. Medical record quality monitoring is one of the effective means to standardize medical behavior. The inventor realized that at present, most of the medical record quality monitoring still adopts manual manual verification. The manual verification method is inefficient, which in turn causes the problem of low quality monitoring accuracy.

Application content

The embodiments of the present application provide a medical record information verification method, device, computer equipment and storage medium, so as to solve the problem of low accuracy of quality monitoring due to incomplete utilization of case information.

A medical record information verification method, comprising:

Obtain the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnostic information;

Inputting the case information into a case representation model to obtain a case representation vector corresponding to the case information; at the same time, inputting the department information into the department representation model to obtain a department representation vector corresponding to the department information;

Perform splicing processing on the case representation vector and the department representation vector to obtain a medical record splicing vector;

Inputting the medical record splicing vector into the case discrimination network model, and determining at least one case judgment result corresponding to the medical record text to be verified;

The diagnostic information is matched with each of the case judgment results, and when the diagnostic information is successfully matched with any one of the case judgment results, it is determined that the text of the medical record to be verified is successfully verified.

A medical record information verification device, comprising:

a medical record text acquisition module, used for acquiring the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnosis information;

The first vector characterization module is used to input the case information into the case characterization model to obtain a case characterization vector corresponding to the case information; meanwhile, input the department information into the department characterization model to obtain the same The department representation vector corresponding to the department information;

a vector splicing module for splicing the case representation vector and the department representation vector to obtain a medical record splicing vector;

A case judgment module, configured to input the medical record splicing vector into a case judgment network model, and determine at least one case judgment result corresponding to the medical record text to be verified;

The case matching module is configured to match the diagnosis information with each of the case judgment results, and when the diagnosis information and any one of the case judgment results are successfully matched, it is determined that the verification of the medical record text to be verified is successful.

A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer-readable instructions:

Obtain the medical record text to be verified; the medical record text to be verified includes case information, department information and diagnostic information associated with the case information;

One or more readable storage media storing computer-readable instructions, wherein the computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform the following steps:

The above-mentioned medical record information verification method, device, computer equipment and storage medium, the method obtains the medical record text to be verified; the medical record text to be verified includes case information, department information and diagnosis information associated with the case information; The case information is input into the case representation model, and a case representation vector corresponding to the case information is obtained; at the same time, the department information is input into the department representation model to obtain a department representation vector corresponding to the department information; The case characterization vector and the department characterization vector are spliced to obtain a medical record splicing vector; the medical record splicing vector is input into the case discrimination network model, and at least one case judgment result corresponding to the medical record text to be verified is determined. ; Match the diagnostic information with each of the case judgment results, and when the diagnostic information and any one of the case judgment results are successfully matched, determine that the text of the medical record to be verified is successfully verified.

By introducing case information and department information, the application learns the correlation between case information and department information through the case representation model and department representation model, so that the case discrimination network model predicts and outputs the case discrimination result based on the case information and department information. It has higher accuracy and improves the efficiency of medical record information verification and monitoring.

The details of one or more embodiments of the application are set forth in the accompanying drawings and the description below, and other features and advantages of the application will become apparent from the description, drawings, and claims.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present application more clearly, the following briefly introduces the drawings that are 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 also be obtained from these drawings without creative labor.

1 is a schematic diagram of an application environment of a method for verifying medical record information in an embodiment of the present application;

2 is a flowchart of a method for verifying medical record information in an embodiment of the present application;

3 is another flowchart of a method for verifying medical record information in an embodiment of the present application;

4 is a flowchart of step S40 in the method for verifying medical record information in an embodiment of the present application;

5 is a schematic block diagram of a medical record information verification device in an embodiment of the present application;

FIG. 6 is another principle block diagram of the device for verifying medical record information in an embodiment of the present application;

7 is a schematic block diagram of a case judgment module in a medical record information verification device in an embodiment of the present application;

FIG. 8 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 clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

The medical record information verification method provided by the embodiment of the present application can be applied in the application environment shown in FIG. 1 . Specifically, the medical record information verification method is applied in a medical record information verification system. The medical record information verification system includes a client and a server as shown in FIG. Incomplete utilization of information leads to the problem of low accuracy of quality control. Among them, the client, also known as the client, refers to the program corresponding to the server and providing local services for the client. Clients can be installed on, but not limited to, various personal computers, laptops, smartphones, tablets, 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 verifying medical record information is provided, and the method is applied to the server in FIG. 1 as an example for description, including the following steps:

S10: Obtain the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnosis information;

Understandably, the medical record text to be verified refers to the historical medical record text waiting to be verified, and the medical record text to be verified contains case information, such as the patient's basic information (such as name, gender, test date, etc.), symptom information ( Such as chief complaint symptoms, test information, etc.), department information related to case information (such as the respiratory department corresponding to the symptoms of cough and sore throat is department information) and diagnostic information (such as the doctor's judgment for the symptoms of cough and sore throat as throat inflammation for diagnostic information).

S20: Input the case information into a case representation model to obtain a case representation vector corresponding to the case information; at the same time, input the department information into the department representation model to obtain a department representation corresponding to the department information vector;

The case representation model and the department representation model are both constructed based on the convolutional neural network model. The case representation model is used to convert case information into case representation vectors, and the department representation model is used to convert department information into department representation vectors.

Specifically, after obtaining the medical record text to be verified, the case information in the medical record text to be verified is input into the case representation model, and the case information is processed by convolution pooling, etc., to obtain the case representation corresponding to the case information. At the same time, the department information in the medical record text to be verified is input into the department characterization model, and the department information is processed by convolution pooling, etc., to obtain the department characterization vector corresponding to the department information.

Preferably, before inputting the case information into the case representation model, the case information may be preprocessed. For example, assuming that the case information is "I started coughing about three days ago", the case information is trimmed into The shorter sentence pair form of "cough for three days" means that the text length of the case information can be shortened while the important information in the case information is not changed, and then the model vector can be shortened when the case information is input into the case representation model. The conversion time improves the efficiency of medical record information verification; similarly, before inputting the department information into the department representation model, the department information can also be preprocessed. Hospital Respiratory Department”, the department information is cut into a shorter sentence pair form of “Respiratory Department”.

In a specific embodiment, as shown in FIG. 3, before step S20, it further includes:

S01: Obtain a preset medical record sample text set; the preset medical record sample text set includes at least one medical record sample text; the medical record sample text includes case sample information and department sample information corresponding to the case sample information; The medical record sample text is associated with a medical record sample label;

Understandably, the medical record sample text can be obtained by crawling the medical record information text library, and the medical record sample text contains case sample information, such as the basic information of the patient (such as name, gender, test date, etc.), symptom information (such as chief complaint symptoms, test information, etc.), and the department sample information corresponding to the case sample information (for example, the respiratory department corresponding to the symptoms of cough and sore throat is the department information).

Further, a medical record sample text is associated with a medical record sample label, the medical record sample label is determined according to the case sample information and the department sample information, and the medical record sample label includes a positive medical record sample label and a negative medical record sample label; It is understandable that in the medical record sample text If the case sample information and department sample information match each other, the medical record sample label associated with the medical record sample text is the positive medical record sample label; if the case sample information does not match the department sample information in the medical record sample text, the medical record sample text The associated medical record sample labels are negative medical record sample labels. Exemplarily, the case sample information is "cough for 3 days", if the department sample information is "respiratory department", the medical record sample text is the positive medical record sample text, and the medical record sample label is the positive medical record sample label; if the department sample information is " Psychiatry", the medical record sample text is the negative medical record sample text, and the medical record sample label is the negative medical record sample label. Then, the case training model and the department training model of the preset twin representation model in step S02 are trained through different positive medical record sample texts and negative medical record sample texts, so that the case training model and the department training model can achieve better training effects. , you can distinguish whether the case sample information matches the department sample information.

S02: Input the medical record sample text into a preset twin representation model, and perform vector representation on the case sample information through a case training model that includes a first initial parameter in the preset twin representation model, to obtain a case sample vector At the same time, vector characterization is carried out on the sample information of the department through the department training model including the second initial parameter in the preset twin characterization model to obtain the department sample vector;

Understandably, the preset twin representation model is used to learn the representation of case sample information and department sample information. The preset twin representation model includes a case training model and a department training model. Both the case training model and the department training model are based on convolution. Generated by neural network model building.

Further, after obtaining the preset medical record sample text set, the medical record sample text is input into the preset twin representation model, and the case is trained by the case training model including the first initial parameter in the preset twin representation model. The sample information is represented by a vector, that is, the case sample information is subjected to convolution pooling and other processing to obtain a case sample vector; at the same time, the department is trained by the department training model that includes the second initial parameter in the preset twin representation model. The sample information is represented by a vector, that is, the sample information of the department is processed by convolution and pooling, and the sample vector of the department is obtained.

Further, if only the department sample information is used for model training, that is, the case sample information and department sample information are not used for model training, the model cannot learn because the department sample information name is too short and does not have rich semantic information. The ability to distinguish the sample information of each department, so in this embodiment, the model training is performed through the case sample information and the department sample information, so that the department training model can also learn the department information representation that contains the semantic information rich in the case sample information.

S03: Perform splicing processing on the case sample vector and the department sample vector to obtain a sample splicing vector, and input the sample splicing vector into an initial regression model to determine the label prediction probability corresponding to the medical record sample text;

Specifically, when the medical record sample text is input into a preset twin representation model, the case sample information is represented by a vector through a case training model that includes the first initial parameter in the preset twin representation model, and a case is obtained. sample vector; at the same time, vector representation is performed on the department sample information through the department training model including the second initial parameter in the preset twin representation model, and after the department sample vector is obtained, the department sample vector is spliced to the case At the back end of the sample vector, the sample splicing vector is obtained, and the sample splicing vector is input into the initial regression model to determine the label prediction probability corresponding to the sample splicing vector, that is, to determine whether the department sample vector matches the case sample vector.

S04: Determine the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability;

Specifically, after splicing the case sample vector and the department sample vector, a sample splicing vector is obtained, and the sample splicing vector is input into the initial regression model, and the label prediction corresponding to the medical record sample text is determined. After the probability, according to the medical record sample label and the label prediction probability, the predicted loss value is determined by the cross-entropy loss function; the cross-entropy loss function is:

Loss=w1*y*log(p)+w0*(1-y)*log(1-p)

Wherein, Loss is the prediction loss value; w1 and w0 are the weights of the preset twin representation model; y is the label of the medical record sample; p is the label prediction probability.

Understandably, it is pointed out in step S01 that the medical record sample text includes positive medical record sample text and negative medical record sample text. When the medical record sample text is a positive medical record sample text, the associated medical record sample label is a positive medical record sample label. The label value of the label is 1; when the medical record sample text is the negative medical record sample text, the associated medical record sample label is the negative medical record sample label, and the label value of the negative medical record sample label is 0; therefore, when input to the preset twin representation model The medical record sample text is positive medical record sample text. According to the above cross-entropy loss function, y is 1, and p represents the probability that the predicted department sample information matches the case sample information; when the medical record sample text input to the preset twin representation model is negative For the medical record sample text, according to the above cross-entropy loss function, y is 0, and 1-p represents the probability that the predicted department sample information does not match the case sample information.

Further, w1 and w0 in the above-mentioned cross-entropy loss function are weight values. It is understandable that w1 is used to predict the positive medical record sample text into the negative medical record sample text (that is, the department sample information is matched with the case sample information, predicting The prediction loss function of the department sample information does not match the case sample information) has a larger loss of rotation, w0 is to predict the negative medical record sample text into the positive medical record sample text (that is, the department sample information does not match the case sample information, the prediction is The prediction loss function of the department sample information and the case sample information match) is smaller and the loss is reversed, so that the recall rate of the preset twin representation model can be improved, the generalization ability of the preset twin representation model can be improved, and the result obtained in step S20 can be prevented. Case representation vectors and department representation vectors are filtered out of too much important information.

S05: When the predicted loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the case training model and the second initial parameter of the department training model, until the predicted loss value reaches the predetermined value. When the preset convergence conditions are met, the case training model after convergence is recorded as the case characterization model, and the department training model after convergence is recorded as the department characterization model.

Understandably, the convergence condition can be the condition that the predicted loss value is less than the set threshold, that is, when the predicted loss value is less than the set threshold, the training is stopped; the convergence condition can also be that the predicted loss value after 10,000 calculations is The condition is very small and will not decrease again, that is, when the predicted loss value is small and will not decrease after 10,000 calculations, stop training, and record the case training model after convergence as the case representation model, The department training model after convergence is recorded as the department characterization model.

Further, after determining the predicted loss value of the preset twin representation model according to the medical record sample label corresponding to the case sample text and the label prediction probability, when the predicted loss value does not reach the preset convergence condition, according to the The predicted loss value adjusts the first initial parameter of the case training model and the second initial parameter of the department training model, and re-inputs the case sample text into the preset twin representation model after adjusting the first and second initial parameters , so that when the predicted loss value corresponding to the medical record sample text reaches the preset convergence condition, select another medical record sample text in the preset medical record sample text set, and execute the above steps S01 to S04, and obtain the corresponding medical record sample text. Predict the loss value, and when the predicted loss value does not reach the preset convergence condition, adjust the first initial parameter of the case training model and the second initial parameter of the department training model again according to the predicted loss value, so that the medical record sample text The corresponding prediction loss value reaches the preset convergence condition.

In this way, after the preset twin representation model is trained through all the medical record sample texts in the preset medical record sample text set, the output results of the preset twin representation model can continue to be closer to the accurate results, so that the recognition accuracy is getting higher and higher. Until the predicted loss values corresponding to all the medical record sample texts reach the preset convergence condition, the case training model after convergence is recorded as the case representation model, and the department training model after convergence is recorded as the department Representation model.

S30: Perform splicing processing on the case representation vector and the department representation vector to obtain a medical record splicing vector;

Specifically, the case information is input into the case representation model, and the case representation vector corresponding to the case information is obtained; at the same time, the department information is input into the department representation model, and the corresponding department information is obtained. After the department representation vector, the department representation vector is spliced to the back end of the case representation vector to obtain the medical record splicing vector.

S40: Input the medical record splicing vector into the case discrimination network model, and determine at least one case judgment result corresponding to the medical record text to be verified;

Understandably, the case discrimination network model is used to determine the case determination result corresponding to the medical record to be verified according to the medical record splicing vector (ie, case information and department information). After splicing the case characterization vector and the department characterization vector to obtain the medical record splicing vector, the medical record splicing vector is input into the case discrimination network model, so as to diagnose and predict the medical records to be verified according to the medical record splicing vector, and then determine At least one case judgment result corresponding to the medical record text to be verified. Understandably, for case information and department information, one or more different case judgment results may be included.

Further, after splicing the case representation vector and the department representation vector to obtain the medical record splicing vector, the medical record splicing vector is input into the case discrimination network model, and the medical record splicing vector is subjected to convolution pooling. After classification and other processing, at least one case judgment result corresponding to the medical record text to be verified is obtained, and one case judgment result is also associated with a judgment probability, that is, according to the case information and department information in the medical record text to be verified, it can be determined. The probability that the corresponding diagnostic information is a case judgment result is the judgment probability.

In one embodiment, as shown in FIG. 4 , step S40 includes:

S401: Perform convolution pooling on the medical record splicing vector through a preset convolutional neural network in the case discrimination network model to obtain a medical record output vector;

Specifically, after splicing the case representation vector and the department representation vector to obtain the medical record splicing vector, the medical record splicing vector is subjected to convolution pooling processing through the preset convolutional neural network in the case discrimination network model, Get the medical record output vector. Optionally, the preset convolutional neural network may be a TextCNN network (text classification convolutional neural network).

Further, before the medical record splicing vector is input into the case discrimination network model, it also includes:

obtaining the third initial parameter of the case characterization model, and the fourth initial parameter of the department characterization model;

Among them, the third initial parameter refers to the parameters of the case representation model obtained after the training of the case training model in steps S01-S05 is completed. It is understandable that the model parameters of the case training model are the first initial parameters, and the case is obtained after the training is completed. The model parameters characterizing the model are updated to the third initial parameters. Similarly, the fourth initial parameter refers to the parameters of the department representation model obtained after the training of the department training model in steps S01-S05 is completed. It is understandable that the model parameters of the department training model are the second initial parameters, which are obtained after the training is completed. The model parameters of the department characterization model are updated to the fourth initial parameters.

The average value of the third initial parameter and the fourth initial parameter is recorded as the discriminative initial parameter of the preset convolutional neural network.

Understandably, compared with random initialization parameters, using the average value of the third initial parameter and the fourth initial parameter as the initial discriminant parameter of the preset convolutional neural network can give the preset convolutional neural network a better value. The initial parameter distribution space, on the other hand, speeds up the training of the case discrimination network model. Further, before the medical record splicing vector is input into the case discrimination network model, the case discrimination network model can be trained by a preset training sample (such as the positive medical record sample text in step S01), so that the case discrimination network model can be trained. Learn the case sample information and department sample information in the positive medical record sample text, and predict more accurate case judgment results based on the case sample information and department sample information; understandably, the positive diagnosis information contained in the positive medical record sample text can be regarded as In order to be correct information, the case judgment results output by the case discrimination network model are close to or even the same as the positive diagnosis information.

S402: Perform case classification on the medical record output vector through a preset classification network in the case discrimination network model, and determine a case judgment result corresponding to the medical record text to be verified.

Specifically, after the medical record splicing vector is subjected to convolution pooling processing by the preset convolutional neural network in the case discrimination network model to obtain the medical record output vector, the predetermined classification network in the case discrimination network model is used to perform convolution and pooling processing. The medical record output vector is used for case classification, and a case judgment result corresponding to the medical record text to be verified is determined. Optionally, the preset classification network is the softmax layer in the case discrimination network model.

S50: Match the diagnostic information with each of the case judgment results, and when the diagnostic information and any one of the case judgment results are successfully matched, determine that the text of the medical record to be verified is successfully verified.

Specifically, after inputting the medical record splicing vector into the case discrimination network model, after determining at least one case determination result corresponding to the medical record text to be verified, the diagnostic information is matched with each case determination result, exemplarily , for example, by determining the similarity between the diagnosis information and the judgment results of each case, or by performing character matching between the diagnosis information and the judgment results of each case through regular expressions, and then when the diagnosis information and any case judgment result are successfully matched , if the similarity between the diagnostic information and the case judgment result is greater than the preset similarity threshold (such as 95%), or the character matching between the diagnostic information and the case judgment result reaches more than 95%, it is determined that the text of the medical record to be verified is verified. Success, that is, it is determined that the diagnostic information in the medical record text to be verified is correct.

Further, in step S40, it is pointed out that a case judgment result is also associated with a judgment probability, so in inputting the medical record splicing vector into the case judgment network model, at least one case judgment corresponding to the medical record text to be verified is determined. After the results, insert the judgment results of each case into the case judgment sequence in descending order of judgment probability; from the first case judgment result in the case judgment sequence, compare the judgment results of each case with the diagnostic information; When it is successfully matched with any case judgment result, the case judgment result is recorded as the judgment result to be confirmed; when the judgment result to be confirmed is not the case judgment result in the first position in the case judgment sequence, that is, the judgment result to be confirmed The judgment probability corresponding to the result is not the largest, and then all the case judgment results in the case judgment sequence before the judgment result to be confirmed are sent to the preset receiver, so that the preset receiver can judge whether the medical record to be verified is verified. success. The preset recipient may be a medical record manager or a medical record inspector.

In a specific embodiment, after step S50, after matching the diagnosis information with each of the case judgment results, the method further includes:

When the diagnostic information does not match all the case judgment results, it is determined that the text verification of the medical record to be verified fails, and a risk of misjudgment exists in the diagnostic information.

Understandably, after the diagnosis information is matched with the judgment results of each of the cases, if the diagnosis information does not match the judgment results of all the cases, the representative diagnosis information may not match the case information and department information, and then it is determined that the diagnosis information does not match the case information and the department information. The text verification of the medical record to be verified fails, and a risk of misjudgment of the diagnostic information is prompted, so as to wait for the preset recipient to manually verify the medical record to be verified.

In this embodiment, by introducing case information and department information, the correlation between case information and department information is learned through the case representation model and department representation model, so that the case discrimination network model can predict and output the case information and department information. The results of the case judgment have higher accuracy, and improve the efficiency of medical record information verification and monitoring.

It should be understood that the size of the sequence numbers of the steps in the above embodiments does not mean the sequence of execution, and the execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

In one embodiment, a medical record information verification apparatus is provided, and the medical record information verification apparatus corresponds one-to-one with the medical record information verification method in the above embodiment. As shown in FIG. 5 , the medical record information verification device includes a medical record text acquisition module 10 , a first vector representation module 20 , a vector splicing module 30 , a case judgment module 40 and a case matching module 50 . The detailed description of each functional module is as follows:

The medical record text acquisition module 10 is used to acquire the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnosis information;

The first vector representation module 20 is used for inputting the case information into the case representation model to obtain a case representation vector corresponding to the case information; meanwhile, inputting the department information into the department representation model to obtain a case representation vector corresponding to the case information. Describe the department representation vector corresponding to the department information;

The vector splicing module 30 is used for splicing the case representation vector and the department representation vector to obtain a medical record splicing vector;

The case judgment module 40 is used to input the medical record splicing vector into the case judgment network model, and determine at least one case judgment result corresponding to the medical record text to be verified;

The case matching module 50 is used to match the diagnosis information with each of the case judgment results, and when the diagnosis information is successfully matched with any of the case judgment results, it is determined that the verification of the medical record text to be verified is successful .

Preferably, as shown in Figure 6, the medical record information verification device further includes:

A medical record sample text set acquisition module 01, configured to acquire a preset medical record sample text set; the preset medical record sample text set includes at least one medical record sample text; the medical record sample text includes case sample information and corresponds to the case sample information Department sample information; a medical record sample text associated with a medical record sample label;

The second vector characterization module 02 is configured to input the text of the medical record sample into a preset twin representation model, and perform vectorization on the case sample information through the case training model including the first initial parameter in the preset twin representation model Characterization, to obtain a case sample vector; at the same time, through the department training model including the second initial parameter in the preset twin representation model, the department sample information is vectorized to obtain the department sample vector;

Label prediction module 03, configured to perform splicing processing on the case sample vector and the department sample vector to obtain a sample splicing vector, and input the sample splicing vector into the initial regression model, and determine that it corresponds to the medical record sample text The label prediction probability of ;

A predicted loss value determination module 04, configured to determine the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability;

A parameter update module 05, configured to update and iterate the first initial parameters of the case training model and the second initial parameters of the department training model when the predicted loss value does not reach a preset convergence condition, until the When the predicted loss value reaches the preset convergence condition, the case training model after convergence is recorded as the case characterization model, and the department training model after convergence is recorded as the department characterization model.

Preferably, the predicted loss value determination module includes:

A predicted loss value determination unit, configured to determine the predicted loss value through a cross-entropy loss function according to the medical record sample label and the label prediction probability; the cross-entropy loss function is:

Loss=w1*y*log(p)+w0*(1-y)*log(1-p)

Preferably, the vector splicing module 30 includes:

The vector splicing unit is configured to obtain the medical record splicing vector after splicing the department representation vector to the back end of the case representation vector.

Preferably, as shown in FIG. 7 , the case judgment module 40 includes:

A convolution pooling unit 401, configured to perform convolution pooling on the medical record splicing vector through a preset convolutional neural network in the case discrimination network model to obtain a medical record output vector;

The case classification unit 402 is configured to perform case classification on the medical record output vector through a preset classification network in the case discrimination network model, and determine a case judgment result corresponding to the medical record text to be verified.

Preferably, the medical record information verification device further includes:

an initial parameter acquisition module for acquiring the third initial parameter of the case characterization model and the fourth initial parameter of the department characterization model;

The initial parameter recording module is configured to record the mean value of the third initial parameter and the fourth initial parameter as the initial parameter for discrimination of the case discrimination network model.

A verification failure prompting module is configured to determine that the text verification of the medical record to be verified fails when the diagnostic information does not match the judgment results of all the cases, and prompt that the diagnostic information has a risk of misjudgment.

For specific limitations on the medical record information verification device, reference may be made to the above limitations on the medical record information verification method, which will not be repeated here. Each module in the above-mentioned medical record information verification device may 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. 8 . 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 readable storage medium, 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 execution of the operating system and computer-readable instructions in the readable storage medium. The database of the computer device is used to store the data used in the medical record information verification method in the above embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer-readable instructions, when executed by the processor, implement a method for verifying medical record information. The readable storage medium provided in this embodiment includes a non-volatile readable storage medium and a volatile readable storage medium.

In one embodiment, there is provided a computer apparatus comprising a memory, a processor, and computer readable instructions stored in the memory and executable on the processor, wherein the processor executes the computer The following steps are implemented when readable instructions:

In one embodiment, one or more readable storage media are provided having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by one or more processors, cause the one or more processing The device performs the following steps:

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 computer-readable instructions, and the computer-readable instructions can be stored in a non-volatile computer. In a readable storage medium or a volatile computer-readable storage medium, the computer-readable instructions, when executed, may include the processes of the foregoing method embodiments. Wherein, any reference to memory, storage, database or other medium used in the various embodiments provided in this application 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 for illustration. In practical applications, the above-mentioned functions can be allocated to different functional units, 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 application, but not to limit them; although the present application has been described in detail with reference to the above-mentioned embodiments, those of ordinary skill in the art should understand that: it can still be used for the above-mentioned 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 in the embodiments of the application, and should be included in the within the scope of protection of this application.

Claims

A medical record information verification method, comprising:

Obtain the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnostic information;

Inputting the case information into a case representation model to obtain a case representation vector corresponding to the case information; at the same time, inputting the department information into the department representation model to obtain a department representation vector corresponding to the department information;

Perform splicing processing on the case representation vector and the department representation vector to obtain a medical record splicing vector;

Inputting the medical record splicing vector into the case discrimination network model, and determining at least one case judgment result corresponding to the medical record text to be verified;

The diagnostic information is matched with each of the case judgment results, and when the diagnostic information is successfully matched with any one of the case judgment results, it is determined that the text of the medical record to be verified is successfully verified.
The method for verifying medical record information according to claim 1, wherein before the inputting the case information into the case representation model and obtaining the case representation vector corresponding to the case information, the method comprises:

Obtain a preset medical record sample text set; the preset medical record sample text set includes at least one medical record sample text; the medical record sample text includes case sample information and department sample information corresponding to the case sample information; one of the medical record sample information The text is associated with a medical record sample label;

Inputting the medical record sample text into a preset twin representation model, and performing vector representation on the case sample information through a case training model that includes the first initial parameter in the preset twin representation model to obtain a case sample vector; , performing vector representation on the department sample information through the department training model including the second initial parameter in the preset twin characterization model, to obtain a department sample vector;

Perform splicing processing on the case sample vector and the department sample vector to obtain a sample splicing vector, and input the sample splicing vector into an initial regression model to determine the label prediction probability corresponding to the medical record sample text;

Determine the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability;

When the predicted loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the case training model and the second initial parameter of the department training model until the predicted loss value reaches the predicted loss value. When the convergence condition is set, the case training model after convergence is recorded as the case characterization model, and the department training model after convergence is recorded as the department characterization model.
The method for verifying medical record information according to claim 2, wherein the determining the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability comprises:

According to the medical record sample label and the label prediction probability, the predicted loss value is determined by a cross-entropy loss function; the cross-entropy loss function is:

Loss=w1*y*log(p)+w0*(1-y)*log(1-p)

Wherein, Loss is the prediction loss value; w1 and w0 are the weights of the preset twin representation model; y is the label of the medical record sample; p is the label prediction probability.
The medical record information verification method according to claim 1, wherein the splicing process is performed on the case representation vector and the department representation vector to obtain a medical record splicing vector, comprising:

After splicing the department representation vector to the back end of the case representation vector, the medical record splicing vector is obtained.
The medical record information verification method according to claim 1, wherein the inputting the medical record splicing vector into the input case discrimination network model, and determining the case judgment result corresponding to the medical record text to be verified, comprising:

The medical record splicing vector is subjected to convolution pooling processing through the preset convolutional neural network in the case discrimination network model to obtain a medical record output vector;

Case classification is performed on the medical record output vector through a preset classification network in the case discrimination network model, and a case judgment result corresponding to the medical record text to be verified is determined.
The medical record information verification method according to claim 5, wherein, before the inputting the medical record splicing vector into the case discrimination network model, the method comprises:

obtaining the third initial parameter of the case characterization model, and the fourth initial parameter of the department characterization model;

The average value of the third initial parameter and the fourth initial parameter is recorded as the discriminative initial parameter of the preset convolutional neural network.
The method for verifying medical record information according to claim 1, wherein after the matching of the diagnosis information with each of the case judgment results, the method further comprises:

When the diagnostic information does not match all the case judgment results, it is determined that the text verification of the medical record to be verified fails, and a risk of misjudgment exists in the diagnostic information.
A medical record information verification device, comprising:

a medical record text acquisition module, used for acquiring the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnosis information;

The first vector characterization module is used to input the case information into the case characterization model to obtain a case characterization vector corresponding to the case information; meanwhile, input the department information into the department characterization model to obtain the same The department representation vector corresponding to the department information;

a vector splicing module for splicing the case representation vector and the department representation vector to obtain a medical record splicing vector;

A case judgment module, configured to input the medical record splicing vector into a case judgment network model, and determine at least one case judgment result corresponding to the medical record text to be verified;

The case matching module is configured to match the diagnosis information with each of the case judgment results, and when the diagnosis information and any one of the case judgment results are successfully matched, it is determined that the verification of the medical record text to be verified is successful.
A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer-readable instructions:

Obtain the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnostic information;

Inputting the case information into a case representation model to obtain a case representation vector corresponding to the case information; at the same time, inputting the department information into the department representation model to obtain a department representation vector corresponding to the department information;

Perform splicing processing on the case representation vector and the department representation vector to obtain a medical record splicing vector;

Inputting the medical record splicing vector into the case discrimination network model, and determining at least one case judgment result corresponding to the medical record text to be verified;

The diagnostic information is matched with each of the case judgment results, and when the diagnostic information is successfully matched with any one of the case judgment results, it is determined that the text of the medical record to be verified is successfully verified.
The computer device of claim 9, wherein the processor executes the computer readable instructions before the case information is input into a case representation model to obtain a case representation vector corresponding to the case information Also implement the following steps:

Obtain a preset medical record sample text set; the preset medical record sample text set includes at least one medical record sample text; the medical record sample text includes case sample information and department sample information corresponding to the case sample information; one of the medical record sample information The text is associated with a medical record sample label;

Inputting the medical record sample text into a preset twin representation model, and performing vector representation on the case sample information through a case training model that includes the first initial parameter in the preset twin representation model to obtain a case sample vector; , performing vector representation on the department sample information through the department training model including the second initial parameter in the preset twin characterization model, to obtain a department sample vector;

Perform splicing processing on the case sample vector and the department sample vector to obtain a sample splicing vector, and input the sample splicing vector into an initial regression model to determine the label prediction probability corresponding to the medical record sample text;

Determine the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability;

When the predicted loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the case training model and the second initial parameter of the department training model until the predicted loss value reaches the predicted loss value. When the convergence condition is set, the case training model after convergence is recorded as the case characterization model, and the department training model after convergence is recorded as the department characterization model.
The computer device according to claim 10, wherein the determining the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability comprises:

According to the medical record sample label and the label prediction probability, the predicted loss value is determined by a cross-entropy loss function; the cross-entropy loss function is:

Loss=w1*y*log(p)+w0*(1-y)*log(1-p)

Wherein, Loss is the prediction loss value; w1 and w0 are the weights of the preset twin representation model; y is the label of the medical record sample; p is the label prediction probability.
The computer device according to claim 9, wherein the splicing process is performed on the case representation vector and the department representation vector to obtain a medical record splicing vector, comprising:

After splicing the department representation vector to the back end of the case representation vector, the medical record splicing vector is obtained.
The computer device according to claim 9, wherein the inputting the medical record splicing vector into the input case discrimination network model, and determining the case judgment result corresponding to the medical record text to be verified, comprising:

The medical record splicing vector is subjected to convolution pooling processing through the preset convolutional neural network in the case discrimination network model to obtain a medical record output vector;

Case classification is performed on the medical record output vector through a preset classification network in the case discrimination network model, and a case judgment result corresponding to the medical record text to be verified is determined.
The computer device according to claim 13, wherein, before the medical record splicing vector is input into the case discrimination network model, the processor further implements the following steps when executing the computer-readable instructions:

obtaining the third initial parameter of the case characterization model, and the fourth initial parameter of the department characterization model;

The average value of the third initial parameter and the fourth initial parameter is recorded as the discriminative initial parameter of the preset convolutional neural network.
One or more readable storage media storing computer-readable instructions, wherein the computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform the following steps:

Obtain the medical record text to be verified; the medical record text to be verified includes case information, department information associated with the case information, and diagnostic information;

Inputting the case information into a case representation model to obtain a case representation vector corresponding to the case information; at the same time, inputting the department information into the department representation model to obtain a department representation vector corresponding to the department information;

Perform splicing processing on the case representation vector and the department representation vector to obtain a medical record splicing vector;

Inputting the medical record splicing vector into the case discrimination network model, and determining at least one case judgment result corresponding to the medical record text to be verified;

The diagnostic information is matched with each of the case judgment results, and when the diagnostic information is successfully matched with any one of the case judgment results, it is determined that the text of the medical record to be verified is successfully verified.
16. The readable storage medium of claim 15, wherein before the case information is input into the case representation model and the case representation vector corresponding to the case information is obtained, the computer-readable instructions are executed by one or more When the multiple processors are executed, the one or more processors are caused to further perform the following steps:

Obtain a preset medical record sample text set; the preset medical record sample text set includes at least one medical record sample text; the medical record sample text includes case sample information and department sample information corresponding to the case sample information; one of the medical record sample information The text is associated with a medical record sample label;

Inputting the medical record sample text into a preset twin representation model, and performing vector representation on the case sample information through a case training model that includes the first initial parameter in the preset twin representation model to obtain a case sample vector; , performing vector representation on the department sample information through the department training model including the second initial parameter in the preset twin characterization model, to obtain a department sample vector;

Perform splicing processing on the case sample vector and the department sample vector to obtain a sample splicing vector, and input the sample splicing vector into an initial regression model to determine the label prediction probability corresponding to the medical record sample text;

Determine the predicted loss value of the preset twin representation model according to the medical record sample label and the label prediction probability;

When the predicted loss value does not reach the preset convergence condition, update and iterate the first initial parameter of the case training model and the second initial parameter of the department training model until the predicted loss value reaches the predicted loss value. When the convergence condition is set, the case training model after convergence is recorded as the case characterization model, and the department training model after convergence is recorded as the department characterization model.
The readable storage medium according to claim 16, wherein the determining the prediction loss value of the preset twin representation model according to the medical record sample label and the label prediction probability comprises:

According to the medical record sample label and the label prediction probability, the predicted loss value is determined by a cross-entropy loss function; the cross-entropy loss function is:

Loss=w1*y*log(p)+w0*(1-y)*log(1-p)

Wherein, Loss is the prediction loss value; w1 and w0 are the weights of the preset twin representation model; y is the label of the medical record sample; p is the label prediction probability.
The readable storage medium according to claim 15, wherein the splicing process on the case representation vector and the department representation vector to obtain a medical record splicing vector, comprising:

After splicing the department representation vector to the back end of the case representation vector, the medical record splicing vector is obtained.
The readable storage medium according to claim 15, wherein the inputting the medical record splicing vector into the input case discrimination network model, and determining the case judgment result corresponding to the medical record text to be verified comprises:

The medical record splicing vector is subjected to convolution pooling processing through the preset convolutional neural network in the case discrimination network model to obtain a medical record output vector;

Case classification is performed on the medical record output vector through a preset classification network in the case discrimination network model, and a case judgment result corresponding to the medical record text to be verified is determined.
19. The readable storage medium of claim 19, wherein, before the input of the medical record splicing vector into the case discrimination network model, the computer-readable instructions, when executed by one or more processors, cause The one or more processors also perform the following steps:

obtaining the third initial parameter of the case characterization model, and the fourth initial parameter of the department characterization model;

The average value of the third initial parameter and the fourth initial parameter is recorded as the discriminative initial parameter of the preset convolutional neural network.