WO2021151326A1 - Electronic medical record screening method and apparatus based on adversarial network, and device and medium - Google Patents

Abstract

An electronic medical record screening method and apparatus based on an adversarial network, and a device and a medium. The method comprises: a first generator generating simulated misdiagnosis data (S10); a second generator generating simulated missed diagnosis data (S20); acquiring real normal data, and training an initial synthetic discriminator by using the real normal data, the simulated misdiagnosis data and the simulated missed diagnosis data (S30); after the training is completed, determining the trained initial synthetic discriminator to be a synthetic discriminator (S40); and processing, by using the synthetic discriminator, medical records to be subjected to screening to obtain a processing result regarding said medical records (S50). The problem of screening electronic medical records is solved, and medical record screening costs are reduced.

Description

Method, device, equipment and medium for screening electronic medical records based on confrontation network

This application claims the priority of the Chinese patent application filed with the Chinese Patent Office on September 9, 2020, with the application number 202010941842.2, and the invention title "Electronic Medical Record Screening Method, Apparatus, Equipment, and Media Based on Anti-Network", all of which The content is incorporated in this application by reference.

Technical field

This application relates to the field of machine learning, and in particular to a method, device, equipment, and medium for screening electronic medical records based on an adversarial network.

Background technique

The medical record is used to record the treatment information generated by the patient during the diagnosis and treatment process, which is of great significance to the doctor's diagnosis and treatment. With the popularization of computer technology, medical records are gradually becoming electronic, and the formed electronic medical records are stored in the electronic medical record database of the hospital.

However, the inventor found that due to differences in the level and experience of doctors, the quality of recorded medical records is also uneven, which can easily lead to errors in electronic medical record filing and invisible medical risks. Therefore, it is necessary to screen electronic medical records to find out problem medical records in a timely manner. If manual screening is used, it will lead to high labor and time costs, and greatly increase the operating costs of the hospital. Therefore, it is urgent to find a non-manual screening method to screen out problem medical records from electronic medical records.

Application content

Based on this, it is necessary to address the above technical problems and provide a method, device, computer equipment, and storage medium for screening electronic medical records based on a counter-network to solve the screening problem of electronic medical records.

A screening method for electronic medical records based on a confrontation network, including:

Generate simulated misdiagnosis data through the first generator;

Generate simulated missed diagnosis data through the second generator;

Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.

An electronic medical record screening device based on a confrontation network, including:

The first generating module is used to generate simulated misdiagnosis data through the first generator;

The second generating module is used to generate simulated missed diagnosis data through the second generator;

A training module, configured to obtain real normal data, and use the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data to train the initial comprehensive discriminator;

The determining discriminator module is used to determine the trained initial comprehensive discriminator as the comprehensive discriminator after the training is completed;

The screening module is configured to use the comprehensive discriminator to process the medical record to be screened, and obtain the processing result of the medical record to be screened.

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

Generate simulated misdiagnosis data through the first generator;

Generate simulated missed diagnosis data through the second generator;

Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.

One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps:

Generate simulated misdiagnosis data through the first generator;

Generate simulated missed diagnosis data through the second generator;

Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.

The above-mentioned anti-network-based electronic medical record screening method, device, computer equipment and storage medium generate simulated misdiagnosis data through the first generator to generate a large amount of simulated misdiagnosis data close to real, and improve the initial comprehensive discriminator’s ability to discriminate misdiagnosis data . The simulated missed diagnosis data is generated by the second generator to generate a large amount of simulated missed diagnosis data close to the real, and the initial comprehensive discriminator's ability to discriminate the missed diagnosis data is improved. Obtain real normal data, use the real normal data, the simulated misdiagnosis data, and the simulated missed diagnosis data to train the initial comprehensive discriminator. Here, since there is enough data to train the model, the discriminative ability of the model Huge improvements. After the training is completed, the trained initial comprehensive discriminator is determined as the comprehensive discriminator to obtain a discriminator that can screen medical records. The comprehensive discriminator is used to process the medical records to be screened, and the processing results of the medical records to be screened are obtained. Here, the comprehensive discriminator is used to screen whether the medical records are normal, which can improve the accuracy and efficiency of medical record screening, and reduce the medical records The cost of screening. This application can solve the screening problem of electronic medical records. This application can be applied to the smart medical field of smart cities, so as to promote the construction of smart cities.

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

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings that need to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor.

FIG. 1 is a schematic diagram of an application environment of an electronic medical record screening method based on a confrontation network in an embodiment of the present application;

2 is a schematic flowchart of a method for screening electronic medical records based on a confrontation network in an embodiment of the present application;

FIG. 3 is a schematic flowchart of an electronic medical record screening method based on a confrontation network in an embodiment of the present application;

FIG. 4 is a schematic flowchart of an electronic medical record screening method based on a confrontation network in an embodiment of the present application;

FIG. 5 is a schematic flowchart of a method for screening electronic medical records based on a confrontation network in an embodiment of the present application;

FIG. 6 is a schematic flowchart of a method for screening electronic medical records based on a confrontation network in an embodiment of the present application;

FIG. 7 is a schematic flowchart of an electronic medical record screening method based on a confrontation network in an embodiment of the present application;

FIG. 8 is a schematic structural diagram of an electronic medical record screening device based on a confrontation network in an embodiment of the present application;

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

Detailed ways

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

The electronic medical record screening method based on the confrontation network provided in this embodiment can be applied in the application environment as shown in FIG. 1, in which the client communicates with the server. Among them, the client includes, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server can be implemented with an independent server or a server cluster composed of multiple servers.

In one embodiment, as shown in FIG. 2, a method for screening electronic medical records based on a confrontation network is provided. The method is applied to the server in FIG. 1 as an example for description, including the following steps:

S10. Generate simulated misdiagnosis data through the first generator.

In this embodiment, the first generator refers to a simulation generator obtained after training by a Generative Adversarial Network (GAN), and is used to generate simulated misdiagnosis data. The first generator can generate a large amount of simulated misdiagnosis data close to the real, ensuring that the initial comprehensive discriminator has enough misdiagnosis training data to improve the ability to discriminate misdiagnosed data. Simulated misdiagnosis data is a type of medical record data, which refers to medical records that have misdiagnosis problems.

S20: Generate simulated missed diagnosis data through the second generator.

Similarly, the second generator is also trained by a Generative Adversarial Network (GAN) to obtain a simulation generator for generating simulated missed diagnosis data. The second generator can generate a large amount of simulated missed diagnosis data that is close to real, ensuring that the initial comprehensive discriminator has enough missed diagnosis training data to improve the ability to discriminate the missed diagnosis data. The simulated missed diagnosis data is a type of medical record data, which refers to the medical records that have the problem of missed diagnosis.

S30. Obtain real normal data, and use the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data to train an initial comprehensive discriminator.

The true normal data refers to the true medical record data without any missed diagnosis or misdiagnosis. Here, the initial comprehensive discriminator is a three-class classifier. When training the initial comprehensive discriminator, using real normal data, simulated misdiagnosis data and simulated missed diagnosis data for training can greatly improve the initial comprehensive discriminator's ability to discriminate these three types of data, and then accurately distinguish the types of medical records. Specifically, during training, the discriminant data of the initial comprehensive discriminator can be returned to the first generator and the second generator to improve the correlation between the first generator, the second generator and the initial comprehensive discriminator (based on the loss Function) to further improve the discrimination ability of the initial comprehensive discriminator.

S40. After the training is completed, determine the trained initial comprehensive discriminator as a comprehensive discriminator.

When the training is completed, the discriminant data of the initial comprehensive discriminator converges. At this time, the trained initial comprehensive discriminator can be determined as the comprehensive discriminator. The comprehensive discriminator can be used to discriminate the type of medical record data. The initial comprehensive discriminator combines the simulated data of the first generator and the second generator (including simulated missed diagnosis data and simulated misdiagnosis data), and the final comprehensive discriminator has good discrimination capabilities and can accurately distinguish the types of electronic medical records.

S50. Use the comprehensive discriminator to process the medical record to be screened, and obtain a processing result of the medical record to be screened.

In this embodiment, the medical record to be screened refers to the medical record that needs to be screened. The comprehensive discriminator is used to process the medical records to be screened, and the processing results of the medical records to be screened can be obtained. There are three treatment results: missed diagnosis, misdiagnosis and normal medical records. Among them, missed medical records and misdiagnosed medical records are abnormal medical records (also called problem medical records). Since the comprehensive discriminator provided in this embodiment has good discrimination capabilities, it can distinguish between normal medical records and abnormal medical records with high accuracy, greatly reducing the processing time and cost of medical record screening, and reducing hospital operating costs.

In steps S10-S50, the simulated misdiagnosis data is generated by the first generator to generate a large amount of simulated misdiagnosis data close to the real, and the initial comprehensive discriminator's ability to discriminate the misdiagnosis data is improved. The simulated missed diagnosis data is generated by the second generator to generate a large amount of simulated missed diagnosis data close to the real, and the initial comprehensive discriminator's ability to discriminate the missed diagnosis data is improved. Obtain real normal data, use the real normal data, the simulated misdiagnosis data, and the simulated missed diagnosis data to train the initial comprehensive discriminator. Here, since there is enough data to train the model, the discriminative ability of the model Huge improvements. After the training is completed, the trained initial comprehensive discriminator is determined as the comprehensive discriminator to obtain a discriminator that can screen medical records. The comprehensive discriminator is used to process the medical records to be screened, and the processing results of the medical records to be screened are obtained. Here, the comprehensive discriminator is used to screen whether the medical records are normal, which can improve the accuracy and efficiency of medical record screening, and reduce the medical records The cost of screening.

Optionally, as shown in FIG. 3, step S10, that is, before generating simulated misdiagnosis data by the first generator, further includes:

S101. In the first confrontation neural network, the first initial generator receives the first random noise, and generates the first simulation data;

S102. The first discriminator receives the true misdiagnosis data with a tag of 1, and generates first real discrimination data; the first discriminator receives the first simulation data with a tag of 0, and generates first simulation discrimination data;

S103. Calculate the first discrimination loss value of the first discriminator according to the first true discrimination data and the first simulation discrimination data; calculate the first discrimination loss value of the first initial generator according to the first simulation discrimination data -Generate loss value;

S104. Update the first discriminator according to the first discrimination loss value, and update the first initial generator according to the first generation loss value;

S105. Repeat the step of updating the first discriminator and the step of updating the first initial generator until the first preset termination condition is satisfied, and the first preset termination condition is that the first simulation judgment data is in The first preset range;

S106. Determine the first initial generator that meets the first preset termination condition as the first generator.

In this embodiment, the first confrontation neural network includes a first initial generator and a first discriminator. The first random noise can be generated by a random algorithm. In the initial stage of training, when the first initial generator processes the first random noise, the generated first simulation data is quite different from the real misdiagnosis data, and it is easy to be identified by the first discriminator. At this time, the generator will obtain a larger first generation loss value, and adjust the calculation parameters in the first initial generator according to the first generation loss value, so that the data generated by the first initial generator gradually approaches the true misdiagnosis data. At the same time, if the first discriminator makes a misjudgment when discriminating the first analog data, it will also obtain a larger first discrimination loss value, and adjust the first discriminator according to the first discrimination loss value Calculating parameters in, so that it has a stronger ability to distinguish between the first simulated data and the real misdiagnosed data.

Repeating the step of updating the first discriminator refers to repeating the steps related to the first discriminator in steps S101-S104. Repeating the step of updating the first generator refers to repeating the steps related to the first initial generator in steps S101-S104. The update steps of the first discriminator and the first initial generator are performed simultaneously.

When the first simulation judgment data is in the first preset range, it can be considered that the first adversarial neural network meets the first preset termination condition. For example, when the first discrimination data of the first simulation data of the first discriminator is 0.5, it is difficult for the first discriminator to determine whether the first simulation data output by the first initial generator is true. That is, the first simulation data generated by the first generator is very similar to the real misdiagnosis data. At this time, the first adversarial neural network has reached convergence.

In steps S101-S106, in the first counter neural network, the first initial generator receives the first random noise and generates first simulation data, where the first initial generator continuously generates new first simulation data . The first discriminator receives the true misdiagnosis data with a label of 1, and generates first real discrimination data; the first discriminator receives the first simulation data with a label of 0, and generates first simulation discrimination data, where, The first discriminator simultaneously discriminates the real misdiagnosed data and the first simulated data, which can improve the ability of the first discriminator to discriminate the misdiagnosed data. Calculate the first discrimination loss value of the first discriminator according to the first real discrimination data and the first simulation discrimination data; calculate the first generation of the first initial generator according to the first simulation discrimination data Loss value, calculate the loss value that can be used to update the model parameters. The first discriminator is updated according to the first discriminating loss value, and the first initial generator is updated according to the first generation loss value. Here, the parameters of the respective models are gradually updated through the loss value to improve the model’s performance Accuracy. Repeat the step of updating the first discriminator and the step of updating the first initial generator until the first preset termination condition is satisfied, and the first preset termination condition is that the first simulation judgment data is in the first The preset range is used to complete the training of the model. The first initial generator that satisfies the first preset termination condition is determined as the first generator, so as to obtain a first generator that can be used to generate first simulation data.

Optionally, as shown in FIG. 4, in step S103, the calculation of the first discrimination loss value of the first discriminator according to the first real discrimination data and the first simulation discrimination data includes:

S1031. Process the first real discrimination data and the first simulation discrimination data by using a first discrimination loss function to generate the first discrimination loss value, and the first discrimination loss function is:

Wherein, x _r1 represents the true misdiagnosis data, D ₁ (x) represents the first true discrimination data, E is the expected calculation symbol, x _f1 represents the first simulation data, and z ₁ represents the first random noise , G ₁ (z ₁ ) represents the first simulation data, D ₁ (G ₁ (z ₁ )) represents the first discrimination loss value,

Represents the first discriminant loss value;

In step S103, the calculating the first generation loss value of the first initial generator according to the first simulation discrimination data includes:

S1032. Process the first simulation discrimination data by using a first generation loss function to generate the first generation loss value, where the first generation loss function is:

Wherein, D ₁ (x _f1 ) is the first simulation discrimination data, D ₃ (x _f1 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the first simulation data, α is a hyperparameter,

Is the first generation loss value.

Here, G ₁ refers to the first initial generator, and D ₁ refers to the first discriminator. The first initial generator is used to generate the first simulation data, and the first discriminator is used to discriminate the first simulation data and generate the first simulation discrimination data; it is also used to discriminate the real misdiagnosis data and generate the first real discrimination data. The training of the first initial generator and the first discriminator is a process of confrontation. α is a hyperparameter, which can be set before model training.

When calculating the first generation loss value, a loss item (α·logD ₃ (x _f1 )) including the _{discriminant data of D 3 (initial comprehensive discriminator) can be added.} The addition of α·logD ₃ (x _f1 ) can make the distribution of the misdiagnosis data generated by the first generator meet the requirements of the actual scene.

Optionally, as shown in FIG. 5, step S20, that is, before generating the simulated missed diagnosis data by the second generator, further includes:

S201. In the second confrontation neural network, the second initial generator receives the second random noise, and generates second simulation data;

S202. The second discriminator receives the real missed diagnosis data with a label of 1, and generates second real discrimination data; the second discriminator receives the second simulated data with a label of 0, and generates second simulated discrimination data;

S203. Calculate the second discrimination loss value of the second discriminator according to the second real discrimination data and the second simulation discrimination data; calculate the second discrimination loss value of the second initial generator according to the second simulation discrimination data 2. Generate loss value;

S204. Update the second discriminator according to the second discrimination loss value, and update the second initial generator according to the second generation loss value;

S205. Repeat the step of updating the second discriminator and the step of updating the second initial generator until a second preset termination condition is met, and the second preset termination condition is that the second simulation judgment data is in Second preset range;

S206. Determine the second initial generator that meets a second preset termination condition as the second generator.

In this embodiment, the second adversarial neural network includes a second initial generator and a second discriminator. The second random noise can be generated by a random algorithm. In the initial stage of training, when the second initial generator processes the second random noise, the generated second simulated data is quite different from the real missed diagnosis data, and it is easy to be identified by the second discriminator. At this time, the generator will obtain two larger second generation loss values, and adjust the calculation parameters in the second initial generator according to the second generation loss values, so that the data generated by the second initial generator gradually approaches the real Missed diagnosis data. At the same time, if the second discriminator makes a misjudgment when discriminating the second analog data, it will also obtain two larger second discrimination loss values, and adjust the second discrimination according to the second discrimination loss value The calculation parameters in the device make it more capable of distinguishing between the second simulated data and the real missed diagnosis data.

Repeating the step of updating the second discriminator refers to repeating the steps related to the second discriminator in steps S201-S204. Repeating the step of updating the second generator refers to repeating the steps related to the second initial generator in steps S201-S204. The update steps of the second discriminator and the second initial generator are performed simultaneously.

When the second simulation judgment data is in the second preset range, it can be considered that the second adversarial neural network meets the second preset termination condition. For example, when the second discriminator's second analog discrimination data for the second analog data is 0.5, it is difficult for the second discriminator to determine whether the second analog data output by the second initial generator is true. That is, the second simulation data generated by the second generator is very similar to the real missed diagnosis data. At this time, the second counter neural network has reached convergence.

In steps S201-S206, in the second counter neural network, the second initial generator receives the second random noise and generates second simulation data, where the second initial generator continuously generates new second simulation data . The second discriminator receives the true missed diagnosis data with a label of 1, and generates second real discrimination data; the second discriminator receives the second simulated data with a label of 0, and generates second simulated discrimination data, where, The second discriminator simultaneously discriminates the real missed diagnosis data and the second simulated data, which can improve the ability of the second discriminator to discriminate the missed data. Calculate the second discrimination loss value of the second discriminator according to the second real discrimination data and the second simulation discrimination data; calculate the second generation of the second initial generator according to the second simulation discrimination data Loss value, calculate the loss value that can be used to update the model parameters. The second discriminator is updated according to the second discriminant loss value, and the second initial generator is updated according to the second generation loss value. Here, the parameters of the respective models are gradually updated by the loss value to improve the performance of the model. Accuracy. Repeat the step of updating the second discriminator and the step of updating the second initial generator until a second preset termination condition is met, and the second preset termination condition is that the second simulation judgment data is in the second The preset range is used to complete the training of the model. The second initial generator that meets the second preset termination condition is determined as the second generator to obtain a second generator that can be used to generate second simulation data.

Optionally, as shown in FIG. 6, in step S203, the calculation of the second discrimination loss value of the second discriminator according to the second real discrimination data and the second simulation discrimination data includes:

S2031. Process the second real discrimination data and the second simulation discrimination data by using a second discrimination loss function to generate the second discrimination loss value, and the second discrimination loss function is:

Wherein, x _r2 represents the real missed diagnosis data, D ₂ (x) represents the second true discrimination data, E is the expected calculation symbol, x _f2 represents the second simulation data, and z ₂ represents the second random noise , G ₂ (z ₂ ) represents the second simulation data, D ₂ (G ₂ (z ₂ )) represents the second discrimination loss value,

Represents the second discriminant loss value;

In step S203, the calculating the second generation loss value of the second initial generator according to the second simulation discrimination data includes:

S2032. Process the second simulation discrimination data by using a second generation loss function to generate the second generation loss value, where the second generation loss function is:

Wherein, D ₂ (x _f2 ) is the second simulation discrimination data, D ₃ (x _f2 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the second simulation data, β is a hyperparameter,

Is the second generation loss value.

Here, G ₂ refers to the second initial generator, and D ₂ refers to the second discriminator. The second initial generator is used to generate second simulation data, and the second discriminator is used to discriminate the second simulation data and generate the second simulation discrimination data; it is also used to discriminate the true missed diagnosis data and generate the second real discrimination data. The training of the second initial generator and the second discriminator are two processes of confrontation. β is a hyperparameter, which can be set before model training.

When calculating the second generation loss value, a loss item (β·logD ₃ (x _f2 )) including the _{discriminant data of D 3 (initial comprehensive discriminator) can be added.} The addition of β·logD ₃ (x _f2 ) can make the distribution of the missed diagnosis data generated by the second generator meet the requirements of the actual scene.

Optionally, as shown in FIG. 7, step S30, that is, acquiring real normal data, using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data to train the initial comprehensive discriminator, includes :

S301. Use the initial comprehensive discriminator to discriminate the real normal data, the simulated misdiagnosis data, and the simulated missed diagnosis data, and generate comprehensive discrimination data, the comprehensive discrimination data including a missed diagnosis rate, a misdiagnosis rate, and a normal rate;

S302: Calculate a comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosis rate, and the normal rate;

S303. Update the initial comprehensive discriminator according to the comprehensive discrimination loss value;

S304. Repeat the steps of updating the initial comprehensive discriminator until the comprehensive discriminating loss value meets the preset convergence condition.

In this embodiment, the missed diagnosis rate in the comprehensive discrimination data refers to the ratio of the number of missed diagnosed medical records determined by the initial comprehensive discriminator to the total number of discriminated medical records. For example, if the total number of discriminated medical records is 100, and the number of missed diagnoses determined by the initial comprehensive discriminator is 4, the missed diagnosis rate in the comprehensive discriminator data is 4%. Similarly, the misdiagnosis rate in the comprehensive discrimination data refers to the ratio of the number of misdiagnosed medical records determined by the initial comprehensive discriminator to the total number of discriminated medical records; the normal rate in the comprehensive discrimination data refers to the normal medical records judged by the initial comprehensive discriminator The ratio of the number to the total number of discriminative medical records.

When calculating the comprehensive discrimination loss value, in addition to the initial comprehensive discriminator to generate comprehensive discrimination data, it is also necessary to combine the true missed diagnosis rate, misdiagnosis rate and normal rate.

The step of repeatedly updating the initial comprehensive discriminator refers to repeatedly performing steps S301-S303.

The preset convergence condition can mean that the comprehensive judgment loss value approaches a certain value.

Optionally, step S302, that is, calculating the comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosed rate, and the normal rate includes:

The missed diagnosis rate, the misdiagnosis rate, and the normal rate are processed by a comprehensive loss function to generate the comprehensive discrimination loss value, and the comprehensive loss function is:

Among them, when k=1, y ₁ is the true missed diagnosis rate,

Is the missed diagnosis rate in the comprehensive discrimination data; when k=2, y ₂ is the true misdiagnosis rate,

Is the misdiagnosis rate in the comprehensive discrimination data; when k=3, y ₃ is the true normal rate,

Is the normal rate in the comprehensive discrimination data;

Is the comprehensive discrimination loss value.

Here, the comprehensive discrimination loss value can be calculated by the comprehensive loss function. When the initial comprehensive discriminator converges, the comprehensive discriminant loss value approaches a certain value.

It should be understood that the size of the sequence number of each step in the foregoing embodiment does not mean the order of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

In one embodiment, an electronic medical record screening device based on a confrontation network is provided, and the electronic medical record screening device based on the confrontation network corresponds to the electronic medical record screening method based on the confrontation network in the above-mentioned embodiment in a one-to-one correspondence. As shown in FIG. 8, the electronic medical record screening device based on the confrontation network includes a first generation module 10, a second generation module 20, a training module 30, a determination discriminator module 40 and a screening module 50. The detailed description of each functional module is as follows:

The first generating module 10 is configured to generate simulated misdiagnosis data through the first generator;

The second generating module 20 is used to generate simulated missed diagnosis data through the second generator;

The training module 30 is configured to obtain real normal data, and use the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data to train the initial comprehensive discriminator;

The determining discriminator module 40 is used to determine the trained initial comprehensive discriminator as a comprehensive discriminator after the training is completed;

The screening module 50 is configured to use the comprehensive discriminator to process the medical record to be screened, and obtain the processing result of the medical record to be screened.

Optionally, the first generation module 10 includes:

Generating a first simulation data unit, used in the first counter neural network, the first initial generator receives the first random noise, and generates the first simulation data;

The first discriminating unit is used for the first discriminator to receive the true misdiagnosis data with a label of 1, and generate first real discriminating data; the first discriminator receives the first simulation data with the label of 0, and generates a first simulation Discriminate data;

The first loss value calculating unit is configured to calculate the first discriminant loss value of the first discriminator according to the first true discriminant data and the first simulation discriminant data; calculate the first discriminant loss value of the first discriminator according to the first simulation discriminant data The first generation loss value of the first initial generator;

A first update unit, configured to update the first discriminator according to the first discrimination loss value, and update the first initial generator according to the first generation loss value;

The first iterative update unit is configured to repeat the step of updating the first discriminator and the step of updating the first initial generator until the first preset termination condition is satisfied, and the first preset termination condition is the The first simulation judgment data is in the first preset range;

A determining first generator unit is configured to determine the first initial generator meeting a first preset termination condition as the first generator.

Optionally, the unit for calculating the first loss value includes:

A unit for calculating a first discriminant loss value is used to process the first true discriminant data and the first simulated discriminant data through a first discriminant loss function to generate the first discriminant loss value, and the first discriminant loss function is :

Wherein, x _r1 represents the true misdiagnosis data, D ₁ (x) represents the first true discrimination data, E is the expected calculation symbol, x _f1 represents the first simulation data, and z ₁ represents the first random noise , G ₁ (z ₁ ) represents the first simulation data, D ₁ (G ₁ (z ₁ )) represents the first discrimination loss value,

Represents the first discriminant loss value;

A unit for calculating a first generation loss value is configured to process the first simulation discrimination data through a first generation loss function to generate the first generation loss value, and the first generation loss function is:

Wherein, D ₁ (x _f1 ) is the first simulation discrimination data, D ₃ (x _f1 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the first simulation data, α is a hyperparameter,

Is the first generation loss value.

Optionally, the second generation module 20 includes:

Generating a second simulation data unit, used in the second counter neural network, the second initial generator receives the second random noise, and generates the second simulation data;

The second discriminating unit is used for the second discriminator to receive the true missed diagnosis data with the tag of 1, and to generate second real discriminating data; the second discriminator receives the second simulation data with the tag of 0, and generates the second simulation Discriminate data;

The second loss value calculation unit is configured to calculate the second discriminant loss value of the second discriminator according to the second true discriminant data and the second simulation discriminant data; calculate the second discriminant loss value of the second discriminator according to the second simulation discriminant data The second generation loss value of the second initial generator;

A second update unit, configured to update the second discriminator according to the second discrimination loss value, and update the second initial generator according to the second generation loss value;

The second iterative update unit is configured to repeat the step of updating the second discriminator and the step of updating the second initial generator until a second preset termination condition is satisfied, and the second preset termination condition is the The second simulation discrimination data is in the second preset range;

A determining second generator unit is configured to determine the second initial generator meeting a second preset termination condition as the second generator.

Optionally, the unit for calculating the second loss value includes:

A unit for calculating a second discriminant loss value is used to process the second true discriminant data and the second simulated discriminant data through a second discriminant loss function to generate the second discriminant loss value, and the second discriminant loss function is :

Wherein, x _r2 represents the real missed diagnosis data, D ₂ (x) represents the second true discrimination data, E is the expected calculation symbol, x _f2 represents the second simulation data, and z ₂ represents the second random noise , G ₂ (z ₂ ) represents the second simulation data, D ₂ (G ₂ (z ₂ )) represents the second discrimination loss value,

Represents the second discriminant loss value;

The second generation loss value calculation unit is configured to process the second simulation discrimination data through a second generation loss function to generate the second generation loss value, and the second generation loss function is:

Wherein, D ₂ (x _f2 ) is the second simulation discrimination data, D ₃ (x _f2 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the second simulation data, β is a hyperparameter,

Is the second generation loss value.

Optionally, the training module 30 includes:

Generating a comprehensive discrimination data unit, configured to use the initial comprehensive discriminator to discriminate the true normal data, the simulated misdiagnosis data and the simulated missed diagnosis data, and generate comprehensive discrimination data, the comprehensive discrimination data including missed diagnosis rate, Misdiagnosis rate and normal rate;

Generating a comprehensive discrimination loss value unit for calculating a comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosis rate, and the normal rate;

Updating the initial comprehensive discriminator unit, configured to update the initial comprehensive discriminator according to the comprehensive discrimination loss value;

The iterative update of the initial comprehensive discriminator unit is used to repeat the steps of updating the initial comprehensive discriminator until the comprehensive discrimination loss value meets the preset convergence condition.

Optionally, a unit for generating a comprehensive discrimination loss value is further configured to process the missed diagnosis rate, the misdiagnosis rate, and the normal rate through a comprehensive loss function to generate the comprehensive discrimination loss value, and the comprehensive loss function is:

Among them, when k=1, y ₁ is the true missed diagnosis rate,

Is the missed diagnosis rate in the comprehensive discrimination data; when k=2, y ₂ is the true misdiagnosis rate,

Is the misdiagnosis rate in the comprehensive discrimination data; when k=3, y ₃ is the true normal rate,

Is the normal rate in the comprehensive discrimination data;

Is the comprehensive discrimination loss value.

Regarding the specific limitation of the electronic medical record screening device based on the confrontation network, please refer to the above limitation of the electronic medical record screening method based on the confrontation network, which will not be repeated here. The various modules in the above-mentioned anti-network-based electronic medical record screening device can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 9. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer equipment is used to store the data involved in the above-mentioned electronic medical record screening method. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer-readable instructions are executed by the processor, a method for screening electronic medical records based on a counter-network is realized.

In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored on the memory and capable of running on the processor, and the processor implements the following steps when the processor executes the computer-readable instructions:

Generate simulated misdiagnosis data through the first generator;

Generate simulated missed diagnosis data through the second generator;

Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.

In one embodiment, one or more computer-readable storage media storing computer-readable instructions are provided. The readable storage media provided in this embodiment include non-volatile readable storage media and volatile readable storage media. Storage medium. The readable storage medium stores computer readable instructions, and when the computer readable instructions are executed by one or more processors, the following steps are implemented:

Generate simulated misdiagnosis data through the first generator;

Generate simulated missed diagnosis data through the second generator;

Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a non-volatile memory. In a readable storage medium or a volatile readable storage medium, when the computer-readable instructions are executed, they may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

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

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

Claims (20)

1. A screening method for electronic medical records based on a confrontation network, which includes:

   Generate simulated misdiagnosis data through the first generator;

   Generate simulated missed diagnosis data through the second generator;

   Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

   After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

   The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.
2. The method for screening electronic medical records based on a confrontation network according to claim 1, wherein before said generating simulated misdiagnosis data by the first generator, the method further comprises:

   In the first confrontation neural network, the first initial generator receives the first random noise and generates the first simulation data;

   The first discriminator receives the true misdiagnosis data with a tag of 1, and generates first real discrimination data; the first discriminator receives the first simulation data with a tag of 0, and generates first simulation discrimination data;

   Calculate the first discrimination loss value of the first discriminator according to the first real discrimination data and the first simulation discrimination data; calculate the first generation of the first initial generator according to the first simulation discrimination data Loss value

   Update the first discriminator according to the first discriminant loss value, and update the first initial generator according to the first generation loss value;

   Repeat the step of updating the first discriminator and the step of updating the first initial generator until the first preset termination condition is satisfied, and the first preset termination condition is that the first simulation judgment data is in the first Preset range

   The first initial generator that meets the first preset termination condition is determined as the first generator.
3. The method for screening electronic medical records based on a confrontation network according to claim 2, wherein the first discrimination loss value of the first discriminator is calculated according to the first real discrimination data and the first simulation discrimination data ,include:

   The first real discrimination data and the first simulation discrimination data are processed by a first discrimination loss function to generate the first discrimination loss value, and the first discrimination loss function is:

   

   Wherein, x _r1 represents the true misdiagnosis data, D ₁ (x) represents the first true discrimination data, E is the expected calculation symbol, x _f1 represents the first simulation data, and z ₁ represents the first random noise , G ₁ (z ₁ ) represents the first simulation data, D ₁ (G ₁ (z ₁ )) represents the first discrimination loss value,

   

   Represents the first discriminant loss value;

   The calculating the first generation loss value of the first initial generator according to the first simulation discrimination data includes:

   The first simulation discrimination data is processed by a first generation loss function to generate the first generation loss value, and the first generation loss function is:

   

   Wherein, D ₁ (x _f1 ) is the first simulation discrimination data, D ₃ (x _f1 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the first simulation data, α is a hyperparameter,

   

   Is the first generation loss value.
4. The method for screening electronic medical records based on a confrontation network according to claim 1, wherein before said generating simulated missed diagnosis data by the second generator, the method further comprises:

   In the second confrontation neural network, the second initial generator receives the second random noise and generates the second simulation data;

   The second discriminator receives the real missed diagnosis data with a label of 1, and generates second real discrimination data; the second discriminator receives the second simulated data with a label of 0, and generates second simulated discrimination data;

   Calculate the second discrimination loss value of the second discriminator according to the second real discrimination data and the second simulation discrimination data; calculate the second generation of the second initial generator according to the second simulation discrimination data Loss value

   Update the second discriminator according to the second discriminant loss value, and update the second initial generator according to the second generation loss value;

   Repeat the step of updating the second discriminator and the step of updating the second initial generator until a second preset termination condition is met, and the second preset termination condition is that the second simulation judgment data is in the second Preset range

   The second initial generator that meets the second preset termination condition is determined as the second generator.
5. The method for screening electronic medical records based on a confrontation network according to claim 4, wherein the second discriminant loss value of the second discriminator is calculated according to the second real discriminant data and the second simulated discriminant data ,include:

   The second real discrimination data and the second simulation discrimination data are processed by a second discrimination loss function to generate the second discrimination loss value, and the second discrimination loss function is:

   

   Wherein, x _r2 represents the real missed diagnosis data, D ₂ (x) represents the second true discrimination data, E is the expected calculation symbol, x _f2 represents the second simulation data, and z ₂ represents the second random noise , G ₂ (z ₂ ) represents the second simulation data, D ₂ (G ₂ (z ₂ )) represents the second discrimination loss value,

   

   Represents the second discriminant loss value;

   The calculating the second generation loss value of the second initial generator according to the second simulation discrimination data includes:

   The second simulation discrimination data is processed by a second generation loss function to generate the second generation loss value, and the second generation loss function is:

   

   Wherein, D ₂ (x _f2 ) is the second simulation discrimination data, D ₃ (x _f2 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the second simulation data, β is a hyperparameter,

   

   Is the second generation loss value.
6. The method for screening electronic medical records based on a confrontation network according to claim 1, wherein said acquiring real normal data uses said real normal data, said simulated misdiagnosis data and said simulated misdiagnosis data to make said initial comprehensive judgment Training, including:

   Use the initial comprehensive discriminator to discriminate the real normal data, the simulated misdiagnosis data, and the simulated missed diagnosis data to generate comprehensive discrimination data, the comprehensive discrimination data including missed diagnosis rate, misdiagnosis rate, and normal rate;

   Calculating the comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosis rate, and the normal rate;

   Updating the initial comprehensive discriminator according to the comprehensive discrimination loss value;

   The steps of updating the initial comprehensive discriminator are repeated until the comprehensive discriminating loss value meets the preset convergence condition.
7. The method for screening electronic medical records based on the adversarial network according to claim 6, wherein the calculation of the comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosis rate, and the normal rate includes :

   The missed diagnosis rate, the misdiagnosis rate, and the normal rate are processed by a comprehensive loss function to generate the comprehensive discrimination loss value, and the comprehensive loss function is:

   

   Among them, when k=1, y ₁ is the true missed diagnosis rate,

   

   Is the missed diagnosis rate in the comprehensive discrimination data; when k=2, y ₂ is the true misdiagnosis rate,

   

   Is the misdiagnosis rate in the comprehensive discrimination data; when k=3, y ₃ is the true normal rate,

   

   Is the normal rate in the comprehensive discrimination data;

   

   Is the comprehensive discrimination loss value.
8. An electronic medical record screening device based on a confrontation network, which includes:

   The first generating module is used to generate simulated misdiagnosis data through the first generator;

   The second generating module is used to generate simulated missed diagnosis data through the second generator;

   A training module, configured to obtain real normal data, and use the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data to train the initial comprehensive discriminator;

   The determining discriminator module is used to determine the trained initial comprehensive discriminator as the comprehensive discriminator after the training is completed;

   The screening module is configured to use the comprehensive discriminator to process the medical record to be screened, and obtain the processing result of the medical record to be screened.
9. A computer device includes a memory, a processor, and computer-readable instructions that are stored in the memory and can run on the processor, wherein the processor implements the following steps when the processor executes the computer-readable instructions:

   Generate simulated misdiagnosis data through the first generator;

   Generate simulated missed diagnosis data through the second generator;

   Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

   After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

   The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.
10. 9. The computer device according to claim 9, wherein before said generating the simulated misdiagnosis data by the first generator, the method further comprises:

    In the first confrontation neural network, the first initial generator receives the first random noise and generates the first simulation data;

    The first discriminator receives the true misdiagnosis data with a tag of 1, and generates first real discrimination data; the first discriminator receives the first simulation data with a tag of 0, and generates first simulation discrimination data;

    Calculate the first discrimination loss value of the first discriminator according to the first real discrimination data and the first simulation discrimination data; calculate the first generation of the first initial generator according to the first simulation discrimination data Loss value

    Update the first discriminator according to the first discriminant loss value, and update the first initial generator according to the first generation loss value;

    Repeat the step of updating the first discriminator and the step of updating the first initial generator until the first preset termination condition is satisfied, and the first preset termination condition is that the first simulation judgment data is in the first Preset range

    The first initial generator that meets the first preset termination condition is determined as the first generator.
11. 11. The computer device according to claim 10, wherein said calculating a first discriminant loss value of said first discriminator based on said first real discriminant data and said first simulated discriminant data comprises:

    The first real discrimination data and the first simulation discrimination data are processed by a first discrimination loss function to generate the first discrimination loss value, and the first discrimination loss function is:

    

    Wherein, x _r1 represents the true misdiagnosis data, D ₁ (x) represents the first true discrimination data, E is the expected calculation symbol, x _f1 represents the first simulation data, and z ₁ represents the first random noise , G ₁ (z ₁ ) represents the first simulation data, D ₁ (G ₁ (z ₁ )) represents the first discrimination loss value,

    

    Represents the first discriminant loss value;

    The calculating the first generation loss value of the first initial generator according to the first simulation discrimination data includes:

    The first simulation discrimination data is processed by a first generation loss function to generate the first generation loss value, and the first generation loss function is:

    

    Wherein, D ₁ (x _f1 ) is the first simulation discrimination data, D ₃ (x _f1 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the first simulation data, α is a hyperparameter,

    

    Is the first generation loss value.
12. 9. The computer device according to claim 9, wherein before said generating the simulated missed diagnosis data by the second generator, it further comprises:

    In the second confrontation neural network, the second initial generator receives the second random noise and generates the second simulation data;

    The second discriminator receives the real missed diagnosis data with a label of 1, and generates second real discrimination data; the second discriminator receives the second simulated data with a label of 0, and generates second simulated discrimination data;

    Calculate the second discrimination loss value of the second discriminator according to the second real discrimination data and the second simulation discrimination data; calculate the second generation of the second initial generator according to the second simulation discrimination data Loss value

    Update the second discriminator according to the second discriminant loss value, and update the second initial generator according to the second generation loss value;

    Repeat the step of updating the second discriminator and the step of updating the second initial generator until a second preset termination condition is met, and the second preset termination condition is that the second simulation judgment data is in the second Preset range

    The second initial generator that meets the second preset termination condition is determined as the second generator.
13. The computer device according to claim 12, wherein the calculating the second discriminant loss value of the second discriminator based on the second real discriminant data and the second simulation discriminant data comprises:

    The second real discrimination data and the second simulation discrimination data are processed by a second discrimination loss function to generate the second discrimination loss value, and the second discrimination loss function is:

    

    Wherein, x _r2 represents the true missed diagnosis data, D _D (x) represents the second true discrimination data, E is the expected calculation symbol, x _f2 represents the second simulation data, and z ₂ represents the second random noise , G ₂ (z ₂ ) represents the second simulation data, D ₂ (G ₂ (z ₂ )) represents the second discrimination loss value,

    

    Represents the second discriminant loss value;

    The calculating the second generation loss value of the second initial generator according to the second simulation discrimination data includes:

    The second simulation discrimination data is processed by a second generation loss function to generate the second generation loss value, and the second generation loss function is:

    

    Wherein, D ₂ (x _f2 ) is the second simulation discrimination data, D ₃ (x _f2 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the second simulation data, β is a hyperparameter,

    

    Is the second generation loss value.
14. 9. The computer device according to claim 9, wherein said acquiring real normal data and using said real normal data, said simulated misdiagnosis data and said simulated misdiagnosis data to train said initial comprehensive discriminator comprises:

    Use the initial comprehensive discriminator to discriminate the real normal data, the simulated misdiagnosis data, and the simulated missed diagnosis data to generate comprehensive discrimination data, the comprehensive discrimination data including missed diagnosis rate, misdiagnosis rate, and normal rate;

    Calculating the comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosis rate, and the normal rate;

    Updating the initial comprehensive discriminator according to the comprehensive discrimination loss value;

    The steps of updating the initial comprehensive discriminator are repeated until the comprehensive discriminating loss value meets the preset convergence condition.
15. One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps:

    Generate simulated misdiagnosis data through the first generator;

    Generate simulated missed diagnosis data through the second generator;

    Acquiring real normal data, and training the initial comprehensive discriminator using the real normal data, the simulated misdiagnosis data, and the simulated misdiagnosis data;

    After the training is completed, determine the trained initial comprehensive discriminator as the comprehensive discriminator;

    The comprehensive discriminator is used to process the medical record to be screened, and the processing result of the medical record to be screened is obtained.
16. 15. The readable storage medium according to claim 15, wherein before generating the simulated misdiagnosis data by the first generator, the method further comprises:

    In the first confrontation neural network, the first initial generator receives the first random noise and generates the first simulation data;

    The first discriminator receives the true misdiagnosis data with a tag of 1, and generates first real discrimination data; the first discriminator receives the first simulation data with a tag of 0, and generates first simulation discrimination data;

    Calculate the first discrimination loss value of the first discriminator according to the first real discrimination data and the first simulation discrimination data; calculate the first generation of the first initial generator according to the first simulation discrimination data Loss value

    Update the first discriminator according to the first discriminant loss value, and update the first initial generator according to the first generation loss value;

    Repeat the step of updating the first discriminator and the step of updating the first initial generator until the first preset termination condition is satisfied, and the first preset termination condition is that the first simulation judgment data is in the first Preset range

    The first initial generator that meets the first preset termination condition is determined as the first generator.
17. 15. The readable storage medium of claim 16, wherein the calculating the first discriminant loss value of the first discriminator based on the first real discriminant data and the first simulated discriminant data comprises:

    The first real discrimination data and the first simulation discrimination data are processed by a first discrimination loss function to generate the first discrimination loss value, and the first discrimination loss function is:

    

    Wherein, x _r1 represents the true misdiagnosis data, D ₁ (x) represents the first true discrimination data, E is the expected calculation symbol, x _f1 represents the first simulation data, and z ₁ represents the first random noise , G ₁ (z ₁ ) represents the first simulation data, D ₁ (G ₁ (z ₁ )) represents the first discrimination loss value,

    

    Represents the first discriminant loss value;

    The calculating the first generation loss value of the first initial generator according to the first simulation discrimination data includes:

    The first simulation discrimination data is processed by a first generation loss function to generate the first generation loss value, and the first generation loss function is:

    

    Wherein, D ₁ (x _f1 ) is the first simulation discrimination data, D ₃ (x _f1 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the first simulation data, α is a hyperparameter,

    

    Is the first generation loss value.
18. 15. The readable storage medium according to claim 15, wherein before generating the simulated missed diagnosis data by the second generator, the method further comprises:

    In the second confrontation neural network, the second initial generator receives the second random noise and generates the second simulation data;

    The second discriminator receives the real missed diagnosis data with a label of 1, and generates second real discrimination data; the second discriminator receives the second simulated data with a label of 0, and generates second simulated discrimination data;

    Calculate the second discrimination loss value of the second discriminator according to the second real discrimination data and the second simulation discrimination data; calculate the second generation of the second initial generator according to the second simulation discrimination data Loss value

    Update the second discriminator according to the second discriminant loss value, and update the second initial generator according to the second generation loss value;

    Repeat the step of updating the second discriminator and the step of updating the second initial generator until a second preset termination condition is met, and the second preset termination condition is that the second simulation judgment data is in the second Preset range

    The second initial generator that meets the second preset termination condition is determined as the second generator.
19. 17. The readable storage medium of claim 18, wherein the calculating the second discriminant loss value of the second discriminator based on the second real discriminant data and the second simulated discriminant data comprises:

    The second real discrimination data and the second simulation discrimination data are processed by a second discrimination loss function to generate the second discrimination loss value, and the second discrimination loss function is:

    

    Wherein, x _r2 represents the real missed diagnosis data, D ₂ (x) represents the second true discrimination data, E is the expected calculation symbol, x _f2 represents the second simulation data, and z ₂ represents the second random noise , G ₂ (z ₂ ) represents the second simulation data, D ₂ (G ₂ (z ₂ )) represents the second discrimination loss value,

    

    Represents the second discriminant loss value;

    The calculating the second generation loss value of the second initial generator according to the second simulation discrimination data includes:

    The second simulation discrimination data is processed by a second generation loss function to generate the second generation loss value, and the second generation loss function is:

    

    Wherein, D ₂ (x _f2 ) is the second simulation discrimination data, D ₃ (x _f2 ) is the discrimination data generated after the initial comprehensive discriminator discriminates the second simulation data, β is a hyperparameter,

    

    Is the second generation loss value.
20. The readable storage medium according to claim 15, wherein said acquiring real normal data and using said real normal data, said simulated misdiagnosis data and said simulated misdiagnosis data to train said initial comprehensive discriminator comprises :

    Use the initial comprehensive discriminator to discriminate the real normal data, the simulated misdiagnosis data, and the simulated missed diagnosis data to generate comprehensive discrimination data, the comprehensive discrimination data including missed diagnosis rate, misdiagnosis rate, and normal rate;

    Calculating the comprehensive discrimination loss value of the initial comprehensive discriminator according to the missed diagnosis rate, the misdiagnosis rate, and the normal rate;

    Updating the initial comprehensive discriminator according to the comprehensive discrimination loss value;

    The steps of updating the initial comprehensive discriminator are repeated until the comprehensive discriminating loss value meets the preset convergence condition.

Images