CN111768826A - Electronic health case generation method and device, terminal equipment and storage medium - Google Patents

Electronic health case generation method and device, terminal equipment and storage medium Download PDF

Info

Publication number
CN111768826A
CN111768826A CN202010617099.5A CN202010617099A CN111768826A CN 111768826 A CN111768826 A CN 111768826A CN 202010617099 A CN202010617099 A CN 202010617099A CN 111768826 A CN111768826 A CN 111768826A
Authority
CN
China
Prior art keywords
data
sample
encoder
electronic health
self
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010617099.5A
Other languages
Chinese (zh)
Other versions
CN111768826B (en
Inventor
张国洲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Ping An Smart Healthcare Technology Co ltd
Original Assignee
Ping An International Smart City Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An International Smart City Technology Co Ltd filed Critical Ping An International Smart City Technology Co Ltd
Priority to CN202010617099.5A priority Critical patent/CN111768826B/en
Publication of CN111768826A publication Critical patent/CN111768826A/en
Application granted granted Critical
Publication of CN111768826B publication Critical patent/CN111768826B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Epidemiology (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The application is suitable for the technical field of artificial intelligence, and provides an electronic health case generation method, an electronic health case generation device, terminal equipment and a storage medium, wherein the electronic health case generation method is designed for decoding real sample data and medical data to be converted in a self-encoder mode, so that the real sample data and the medical data to be converted are not required to be converted into a picture format and then decoded, the change of the structure distribution in the real sample data and the medical data to be converted is prevented, the generated electronic health case and the structure distribution of the medical data to be converted are the same, and the accuracy of electronic health case generation is improved. The electronic health case generation method is applied to a smart medical scene, and accordingly construction of a smart city is promoted.

Description

Electronic health case generation method and device, terminal equipment and storage medium
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a method and an apparatus for generating an electronic health case, a terminal device, and a storage medium.
Background
Electronic Health Record (EHR) is patient-centric, electronically recorded information about a person's health, such as hospital records, disease information, allergy history, and the like. These records provide great help for doctors to understand the patient's condition, and today, with machine learning and the rapid development of artificial intelligence, AI medical algorithms (disease prediction, intelligent diagnosis, etc.) are also heavily dependent on a large amount of electronic health medical data.
Due to the reasons that information cannot be shared among different medical institutions, medical data is secret, cases for a certain disease are few and the like, a large amount of effective data is difficult to obtain, and a large amount of electronic health case data cannot be directly used by researchers. Therefore, the method for generating EHR data is of great significance in the application of AI medical field.
In the existing electronic health case data generation process, original data are converted into a picture format according to an ICD-9 disease coding list, then a picture with corresponding data distribution is generated by using a generative countermeasure network (GAN), and finally the picture is decoded into available electronic health case data. However, the original discrete data needs to be converted into the picture format, so that the structural distribution of the original data is changed, the generated electronic health case data is inconsistent with the original data, and the accuracy of the generation of the electronic health case data is reduced.
Disclosure of Invention
In view of this, embodiments of the present application provide an electronic health case generation method, an apparatus, a terminal device, and a storage medium, so as to solve the problem in the prior art that the electronic health case generation method needs to convert original discrete data into a picture format first, which changes the structural distribution of the original data, and thus the electronic health case data generation accuracy is poor.
A first aspect of an embodiment of the present application provides an electronic health case generation method, including:
acquiring real sample data, and training a self-encoder according to the real sample data;
acquiring preset noise data, and inputting the preset noise data into a generative countermeasure network for data generation to obtain false data;
controlling the trained self-encoder to decode the real sample data and the false data to obtain sample decoded data and false decoded data;
performing weight parameter optimization on the generative confrontation network according to the sample decoding data and the false decoding data until the generative confrontation network meets a preset end condition, and merging the trained self-encoder into the generative confrontation network after parameter optimization;
inputting the medical data to be converted into the combined generative confrontation network, and instructing the generative confrontation network to generate data according to the medical data to be converted to obtain the electronic health case.
Further, training an auto-encoder according to the real sample data includes:
converting the disease information in the real sample data into sample coded data according to a disease code list, and performing vector conversion on the sample coded data to obtain sample vector data;
sampling the sample vector data to obtain a sampling vector, and controlling an encoder in the self-encoder to encode the sampling vector to obtain an encoding result;
controlling a decoder in the self-encoder to decode the encoding result to obtain a decoding result, and performing loss calculation according to the decoding result to obtain a loss value;
and respectively updating parameters of the encoder and the decoder according to the loss values.
Further, after the parameter updating is performed on the encoder and the decoder according to the loss values, the method further includes:
if the self-encoder meets preset optimization conditions, stopping training of the self-encoder;
and if the self-encoder does not meet the preset optimization condition, sampling the sample vector data again until the self-encoder meets the preset optimization condition.
Further, the controlling the trained auto-encoder to decode the real sample data and the dummy data to obtain sample decoded data and dummy decoded data includes:
respectively sampling real sample data and false data according to a preset sampling value to obtain real sample data and false sample data;
controlling the decoder to decode the real sampling data and the false sampling data respectively to obtain a sample decoding result and a false decoding result;
specifically, the performing weight parameter optimization on the generative countermeasure network according to the sample decoded data and the dummy decoded data includes:
carrying out weight parameter optimization on a discriminator in the generative countermeasure network according to the sample decoding data and the false decoding data;
and optimizing weight parameters of the generator in the generative countermeasure network according to the sample decoding data and the false decoding data.
Further, a first optimization formula adopted for optimizing weight parameters of the discriminator in the generative countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure BDA0002564158830000031
Figure BDA0002564158830000032
θd＝clip(θd，-c，c)；
wherein α is learning rate, RMSProp is optimization algorithm, clip is interception function, c is interception fixed value in the interception function, ziFor the dummy decoded data, xiDecoding data for said sample, m being said predetermined sampling value, D (x)i) Probability, G (z), of decoding data for said sample as said true sample datai) The first optimization formula is used to calculate a loss function value in the discriminator for data generated after the dummy decoded data is input to the generator
Figure BDA0002564158830000033
Value of the loss function
Figure BDA0002564158830000034
Weight θ for the discriminatordAnd updating the numerical value.
Further, a second optimization formula adopted for optimizing the weight parameters of the generator in the generator countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure BDA0002564158830000035
Figure BDA0002564158830000041
wherein α is learning rate, RMSProp is optimization algorithm, ziFor the dummy decoded data, xiDecoding data for said samples, m being said predetermined sample value, said second optimization formula being used to calculate a value of a loss function in said generator
Figure BDA0002564158830000042
Value of the loss function
Figure BDA0002564158830000043
Weight θ for the generatorgAnd updating the numerical value.
Further, the method further comprises:
uploading the electronic health case into a blockchain.
A second aspect of an embodiment of the present application provides an electronic health case generation apparatus, including:
the self-encoder training unit is used for acquiring real sample data and training the self-encoder according to the real sample data;
the data generation unit is used for acquiring preset noise data and inputting the preset noise data into a generation type countermeasure network for data generation to obtain false data;
the data decoding unit is used for controlling the trained self-encoder to decode the real sample data and the false data to obtain sample decoding data and false decoding data;
a parameter optimization unit, configured to perform weight parameter optimization on the generative countermeasure network according to the sample decoded data and the dummy decoded data until the generative countermeasure network meets a preset end condition, and merge the trained self-encoder into the generative countermeasure network after parameter optimization;
and the medical record generation unit is used for inputting the medical data to be converted into the combined generative confrontation network and indicating the generative confrontation network to generate data according to the medical data to be converted to obtain the electronic health case.
A fourth aspect of the embodiments of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the terminal device, where the processor implements the steps of the electronic health case generation method provided by the first aspect when executing the computer program.
A fifth aspect of embodiments of the present application provides a storage medium storing a computer program that, when executed by a processor, implements the steps of the electronic health case generation method provided by the first aspect.
The electronic health case generation method, the electronic health case generation device, the terminal equipment and the storage medium have the following beneficial effects:
according to the electronic health case generation method provided by the embodiment of the application, the real sample data and the medical data to be converted are decoded in a self-encoder mode, so that the real sample data and the medical data to be converted do not need to be decoded after being converted into the picture format, the structural distribution in the real sample data and the medical data to be converted is prevented from being changed, the generated electronic health case and the medical data to be converted are identical in structural distribution, and the generation accuracy of the electronic health case is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a flowchart illustrating an implementation of an electronic health case generation method according to an embodiment of the present disclosure;
fig. 2 is a flowchart illustrating an implementation of an electronic health case generation method according to another embodiment of the present application;
FIG. 3 is a flow chart of an implementation of an electronic health case generation method according to yet another embodiment of the present application;
FIG. 4 is a schematic diagram of a network structure of the generator provided in the embodiment of FIG. 3;
fig. 5 is a block diagram of an electronic health case generation apparatus according to an embodiment of the present application;
fig. 6 is a block diagram of a terminal device according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The electronic health case generation method according to the embodiment of the present application may be executed by a control device or a terminal (hereinafter, referred to as a "mobile terminal").
The electronic health case generation method is applied to a smart medical scene, and by means of the design of decoding medical data to be converted in a self-encoder mode, the medical data to be converted are not required to be converted into a picture format and then decoded, so that structural distribution change in the medical data to be converted is prevented, the generated electronic health case is identical to the structural distribution of the medical data to be converted, the accuracy of electronic health case generation is improved, and the construction of a smart city is promoted.
Referring to fig. 1, fig. 1 shows a flowchart of an implementation of an electronic health case generation method according to an embodiment of the present application, including:
step S10, acquiring real sample data, and training a self-encoder according to the real sample data;
the real sample data is preset real electronic medical data, the self-encoder is an artificial neural network capable of learning input data to efficiently represent through unsupervised learning, and the self-encoder trained according to the real sample data is used for performing dimension reduction (dimensional reduction) on the input data to obtain low-dimensional discrete data.
Specifically, the self-encoder includes an encoder (encoder) for compressing the features of the input data into a hidden layer representation, which may be represented by an encoding function h ═ f (x), where h is the hidden layer, and a decoder (decoder) for reconstructing the input features of the hidden layer representation, which may be represented by a decoding function r ═ g (h), so that the whole self-encoder may be described by a function g (f (x)) r, where r is the output data and x is the input data, and the self-encoder is trained according to the real sample data, so that the hidden layer h carries the useful features in the real sample data.
Step S20, acquiring preset noise data, inputting the preset noise data into a generative countermeasure network for data generation to obtain false data;
the generative countermeasure network comprises a generator and a discriminator, wherein the generator is used for generating data according to the data input into the generative countermeasure network to generate the false data G (z), the discriminator is used for judging whether the false data generated by the generator is real data or not, the output of the discriminator is D (x), D (x) represents the probability that x is the real data, if the probability is 1, 100% of the data is real data, and the output is 0, x is not the real data.
Specifically, in this step, the preset noise data is input into a generator in a generative countermeasure network to generate the dummy data, and the dummy data is used for network training of the generative countermeasure network.
Step S30, controlling the trained self-encoder to decode the real sample data and the false data to obtain sample decoded data and false decoded data;
the trained self-encoder is controlled to decode the real sample data and the false data, so that the real sample data and the false data are subjected to feature extraction, and the extracted features are subjected to dimension reduction to obtain low-dimension sample decoded data and false decoded data, thereby effectively facilitating subsequent network training of the generative countermeasure network.
Step S40, carrying out weight parameter optimization on the generative confrontation network according to the sample decoding data and the false decoding data until the generative confrontation network meets a preset end condition, and merging the trained self-encoder into the generative confrontation network after parameter optimization;
the method comprises the steps of carrying out weight parameter optimization design on a generative confrontation network according to sample decoding data and false decoding data to achieve the effect of network training of the generative confrontation network, stopping the training of the generative confrontation network when the generative confrontation network is judged to meet a preset ending condition, wherein the preset ending condition can be set according to requirements, for example, the preset ending condition can be used for judging whether the generative confrontation network reaches a preset iteration number or judging whether the generative confrontation network converges to judge whether the network training of the generative confrontation network stops.
Specifically, in the network training process of the generative countermeasure network, the generator is intended to generate true data as much as possible to cheat the discriminator, the discriminator is intended to distinguish the data generated by the generator from the true data as much as possible, and a dynamic "game process" is formed between the generator and the discriminator, so that the final game result is that the generator generates a picture g (z) enough to "falsely and falsely" and it is difficult for the discriminator to determine whether the data generated by the generator is true, so that D (g (z)) is 0.5, and when D (g (z)) is 0.5, the generative countermeasure network is determined to converge.
Step S50, inputting the medical data to be converted into the combined generative confrontation network, and instructing the generative confrontation network to generate data according to the medical data to be converted to obtain an electronic health case;
when the generative confrontation network converges, the authenticity of the electronic health case generated based on the medical data to be converted is higher, and the accuracy of generating the electronic health case is improved.
The real sample data and the medical data to be converted are decoded in a self-encoder mode, so that the real sample data and the medical data to be converted do not need to be decoded after being converted into the picture format, the change of the structure distribution in the real sample data and the medical data to be converted is prevented, the generated electronic health case is identical to the structure distribution of the medical data to be converted, and the accuracy of the generation of the electronic health case is improved.
Referring to fig. 2, fig. 2 is a flowchart illustrating an implementation of an electronic health case generation method according to another embodiment of the present application. With respect to the embodiment corresponding to fig. 1, the electronic health case generation method provided in this embodiment is further detailed in step S10 in the embodiment corresponding to fig. 1, and step S10 includes:
s11, acquiring real sample data, converting the disease information in the real sample data into sample coded data according to a disease code list, and performing vector conversion on the sample coded data to obtain sample vector data;
the disease code list is an ICD-9 table, and the corresponding relationship between different ICD9 codes and corresponding disease names is stored in the disease code list, for example, 0703(ICD9 code) -acute icteric viral hepatitis e (disease name), 0091(ICD9 code) -acute bacterial enteritis (disease name).
Specifically, in this step, the disease information in the real sample data is converted into sample coded data according to the disease code list, and the sample coded data is converted into a vector X ∈ V with a fixed | C | dimensioncThe V iscIs sample vector data.
S12, sampling the sample vector data to obtain a sampling vector, and controlling an encoder in the self-encoder to encode the sampling vector to obtain an encoding result;
wherein, from VcS pieces of data x in the sample vector data are randomly sampled and input into the encoder to be encoded, so as to obtain an encoding result enc (x), where the value of s may be set as required, for example, s may be set to a value of 1000, 2000, or 3000.
S13, controlling a decoder in the self-encoder to decode the encoding result to obtain a decoding result, and performing loss calculation according to the decoding result to obtain a loss value;
wherein, the decoder is controlled to decode the coding result Enc (x) to obtain a decoding result Dec (Enc (x)), and the loss function calculation is performed on the decoding result Dec (Enc (x)) to obtain a loss value
Figure BDA0002564158830000091
Optionally, in this step, the loss function used may be:
Figure BDA0002564158830000092
wherein in the above formula, s is VcThe value of the sample in the sample vector data,
Figure BDA0002564158830000093
for calculated loss values
Figure BDA0002564158830000094
xiIs the decoding result Dec (enc (x)).
S14, updating parameters of the encoder and the decoder according to the loss values respectively;
based on the loss value, parameters of the encoder and the decoder can be updated in a gradient descent method (BGD) or a random gradient descent method (SGD) mode, so that the effect of optimizing the parameters of the encoder and the decoder is achieved.
S15, if the self-encoder meets the preset optimization condition, stopping the training of the self-encoder;
when the self-encoder is judged to meet the preset optimization condition, the parameter optimization of the self-encoder is stopped, and the preset optimization condition can be set according to requirements, for example, the preset optimization condition can be used for judging whether the self-encoder reaches a preset iteration number or not, or judging whether the self-encoder converges or not so as to judge whether the training of the self-encoder is stopped or not.
Specifically, in this step, when the loss value is judged
Figure BDA0002564158830000095
And if the loss threshold is smaller than the loss threshold, the self-encoder is judged to be converged so as to stop the optimization training of the self-encoder.
S16, if the self-encoder does not meet the preset optimization condition, sampling the sample vector data again until the self-encoder meets the preset optimization condition;
in this embodiment, the trained self-encoder can perform dimension reduction (dimensional reduction) on input data according to the real sample data to obtain low-dimensional discrete data, a hidden layer of the self-encoder carries useful features in the real sample data, and based on a decoder in the trained self-encoder, data generated by a generator in a generative countermeasure network can be discretized into an electronic health case without converting medical data to be converted into a picture format and then decoding, and the generated electronic health case better conforms to the structural distribution of the medical data to be converted.
Referring to fig. 3, fig. 3 is a flowchart illustrating an implementation of an electronic health case generation method according to another embodiment of the present application. With respect to the embodiment corresponding to fig. 1, the electronic health case generation method provided in this embodiment is further detailed in step S30 in the embodiment corresponding to fig. 1, and step S30 includes:
s31, respectively sampling real sample data and false data according to preset sampling values to obtain real sample data and false sample data;
v corresponding to real sample data according to preset sampling value mcSample vector data and dummy data pzSampling is carried out so as to correspondingly obtain real sampling data x and false sampling data z.
S32, controlling the decoder to decode the real sampling data and the false sampling data respectively to obtain a sample decoding result and a false decoding result;
wherein the control decoder decodes the real sample data x and the dummy sample data z to obtain D (x) correspondinglyi) A sample decoding result;
specifically, the dummy sampling data z is input into the generator to obtain the generated data g (z), and the generated data g (z) is input into the decoder for decoding to obtain the dummy decoding result Dec (g (z)).
Specifically, the performing weight parameter optimization on the generative countermeasure network according to the sample decoded data and the dummy decoded data includes:
carrying out weight parameter optimization on a discriminator in the generative countermeasure network according to the sample decoding data and the false decoding data;
optimizing weight parameters of a generator in the generative countermeasure network according to the sample decoding data and the false decoding data;
in this embodiment, a first optimization formula adopted for performing weight parameter optimization on the discriminator in the generative countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure BDA0002564158830000101
Figure BDA0002564158830000102
θd＝clip(θd，-c，c)；
in the above formula, α is the learning rate, RMSProp is the optimization algorithm, clip is the intercept function, c is the intercept fixed value in the intercept function, ziFor the dummy decoded data, xiDecoding data for said sample, m being said predetermined sampling value, D (x)i) Probability, G (z), of decoding data for said sample as said true sample datai) The first optimization formula is used to calculate a loss function value in the discriminator for data generated after the dummy decoded data is input to the generator
Figure BDA0002564158830000111
Value of the loss function
Figure BDA0002564158830000112
Weight θ for the discriminatordAnd updating the numerical value.
In addition, a second optimization formula adopted for optimizing the weight parameters of the generator in the generative countermeasure network according to the sample decoding data and the dummy decoding data is as follows:
Figure BDA0002564158830000113
Figure BDA0002564158830000114
wherein α is learning rate, RMSProp is optimization algorithm, ziFor the dummy decoded data, xiDecoding data for said samples, m being said predetermined sample value, said second optimization formula being used to calculate a value of a loss function in said generator
Figure BDA0002564158830000115
Value of the loss function
Figure BDA0002564158830000116
Weight θ for the generatorgAnd updating the numerical value.
Referring to fig. 4, optionally, in this embodiment, the generator may be built in a manner of a compression and Excitation network (SENET) -Residual network (ResNets) module, where the generator sequentially includes a Residual network, a compression layer (szeeze), a weight Excitation layer (Excitation), and a weight weighting layer.
Specifically, in this embodiment, the compression layer uses Global Average capacitance (GAP for short), the generator performs feature compression on the input noise data through the compression layer, each two-dimensional feature channel is changed into a real number, the real number has a Global receptive field, and the output dimension is matched with the number of input feature channels to represent the Global distribution of responses on the feature channels, and the layer close to the input can also obtain the Global receptive field.
The weight excitation layer comprises a Bottleneck layer (Bottleneck) structure formed by two Fully Connected layers (Fully Connected layers) to model the correlation between channels, and outputs the weight with the same number as that of the input feature, firstly, the feature dimension is reduced to 1/16 of the input feature, and then the feature dimension is raised back to the original dimension through one Fully Connected layer after being activated by a ReLu activation function. The benefit of this approach over using a full link layer directly is: 1) more nonlinearity is provided, and complex correlation among channels can be better fitted; 2) the parameter amount and the calculation amount are greatly reduced.
In the embodiment, the weight weighting layer adopts a Sigmoid function, obtains normalized weights between 0 and 1 through one Sigmoid function, and finally weights the normalized weights to the characteristics of each channel by adopting a scaling operation (Scale operation).
In the embodiment, parameter optimization is performed on the discriminator and the generator by adopting a Wasserstein GAN algorithm, so that the stability of the generative confrontation network training is effectively improved, the unicity of generated data of the generator is reduced, the diversity of virtual samples is improved, the generator is built in a mode of compressing and exciting a network-residual error network module, the calculation accuracy of the generative confrontation network is effectively improved, and the SENET design is introduced into the generator, so that the data generation effect of the generative confrontation network is improved.
In all embodiments of the present application, the corresponding electronic health case is generated based on the medical data to be converted, and specifically, the electronic health case is obtained by performing data generation on the medical data to be converted, such as by using a generative countermeasure network. Uploading the electronic health case to the blockchain can ensure the safety and the fair transparency of the user. The user device may download the electronic health case from the blockchain to verify whether the electronic health case is tampered with. The blockchain referred to in this example is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm, and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.
Referring to fig. 5, fig. 5 is a block diagram of an electronic health case generation apparatus 100 according to an embodiment of the present application. The electronic health case generation apparatus 100 in this embodiment includes units for performing the steps in the embodiments corresponding to fig. 1 to 3. Please refer to fig. 1 to 3 and fig. 1 to 3 for the corresponding embodiments. For convenience of explanation, only the portions related to the present embodiment are shown. Referring to fig. 5, the electronic health case generation apparatus 100 includes: the self-encoder training unit 10, the data generating unit 11, the data decoding unit 12, the parameter optimizing unit 13 and the medical record generating unit 14, wherein:
the self-encoder training unit 10 is used for acquiring real sample data and training a self-encoder according to the real sample data;
the data generating unit 11 is configured to acquire preset noise data, input the preset noise data into a generating countermeasure network, and generate data to obtain dummy data;
a data decoding unit 12, configured to control the trained self-encoder to decode the real sample data and the dummy data, so as to obtain sample decoded data and dummy decoded data;
a parameter optimization unit 13, configured to perform weight parameter optimization on the generative countermeasure network according to the sample decoded data and the dummy decoded data until the generative countermeasure network meets a preset end condition, and merge the trained self-encoder into the generative countermeasure network after parameter optimization;
and the medical record generating unit 14 is used for inputting the medical data to be converted into the combined generative confrontation network, and instructing the generative confrontation network to generate data according to the medical data to be converted, so as to obtain the electronic health case.
As an embodiment of the present application, the self-encoder training unit 10 is further configured to: converting the disease information in the real sample data into sample coded data according to a disease code list, and performing vector conversion on the sample coded data to obtain sample vector data;
sampling the sample vector data to obtain a sampling vector, and controlling an encoder in the self-encoder to encode the sampling vector to obtain an encoding result;
controlling a decoder in the self-encoder to decode the encoding result to obtain a decoding result, and performing loss calculation according to the decoding result to obtain a loss value;
and respectively updating parameters of the encoder and the decoder according to the loss values.
If the self-encoder meets preset optimization conditions, stopping training of the self-encoder;
as an embodiment of the present application, the self-encoder training unit 10 is further configured to: and if the self-encoder does not meet the preset optimization condition, sampling the sample vector data again until the self-encoder meets the preset optimization condition.
As an embodiment of the present application, the data decoding unit 12 is further configured to: respectively sampling real sample data and false data according to a preset sampling value to obtain real sample data and false sample data;
and controlling the decoder to decode the real sampling data and the false sampling data respectively to obtain a sample decoding result and a false decoding result.
As an embodiment of the present application, the parameter optimization unit 13 is further configured to:
carrying out weight parameter optimization on a discriminator in the generative countermeasure network according to the sample decoding data and the false decoding data;
and optimizing weight parameters of the generator in the generative countermeasure network according to the sample decoding data and the false decoding data.
As an embodiment of the present application, in the parameter optimization unit 13, a first optimization formula adopted for performing weight parameter optimization on the discriminator in the generative countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure BDA0002564158830000141
Figure BDA0002564158830000142
θd＝clip(θd，-c，c)；
wherein α is learning rate, RMSProp is optimization algorithm, clip is interception function, c is interception fixed value in the interception function, ziFor the dummy decoded data, xiDecoding data for said sample, m being said predetermined sampling value, D (x)i) Decoding data for the sample as the trueProbability of real sample data, G (z)i) The first optimization formula is used to calculate a loss function value in the discriminator for data generated after the dummy decoded data is input to the generator
Figure BDA0002564158830000143
Value of the loss function
Figure BDA0002564158830000144
Weight θ for the discriminatordAnd updating the numerical value.
As an embodiment of the present application, in the parameter optimization unit 13, a second optimization formula adopted by the weight parameter optimization of the generator in the generative countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure BDA0002564158830000145
Figure BDA0002564158830000146
wherein α is learning rate, RMSProp is optimization algorithm, ziFor the dummy decoded data, xiDecoding data for said samples, m being said predetermined sample value, said second optimization formula being used to calculate a value of a loss function in said generator
Figure BDA0002564158830000147
Value of the loss function
Figure BDA0002564158830000148
Weight θ for the generatorgAnd updating the numerical value.
As can be seen from the above, according to the electronic health case generation device 100 provided by this embodiment, by adopting the design of decoding the real sample data and the medical data to be converted in the manner of the self-encoder, the real sample data and the medical data to be converted do not need to be decoded after being converted into the picture format, and thus, the change of the structure distribution in the real sample data and the medical data to be converted is prevented, the generated electronic health case and the medical data to be converted have the same structure distribution, and the accuracy of generating the electronic health case is improved.
Fig. 6 is a block diagram of a terminal device 2 according to another embodiment of the present application. As shown in fig. 6, the terminal device 2 of this embodiment includes: a processor 20, a memory 21 and a computer program 22, such as a program of an electronic health case generation method, stored in said memory 21 and executable on said processor 20. The processor 20, when executing the computer program 22, implements the steps in the various embodiments of the electronic health case generating method described above, such as S10-S50 shown in fig. 1, or S11-S16 and S31-S32 shown in fig. 2 and 3. Alternatively, when the processor 20 executes the computer program 22, the functions of the units in the embodiment corresponding to fig. 5, for example, the functions of the units 10 to 14 shown in fig. 5, are implemented, for which reference is specifically made to the relevant description in the embodiment corresponding to fig. 6, which is not repeated herein.
Illustratively, the computer program 22 may be divided into one or more units, which are stored in the memory 21 and executed by the processor 20 to accomplish the present application. The one or more units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution of the computer program 22 in the terminal device 2. For example, the computer program 22 may be divided into a structured analysis unit 10, a program adding unit 11 and a data merging unit 12, and a synchronization display unit 13, each of which functions as described above.
The terminal device may include, but is not limited to, a processor 20, a memory 21. Those skilled in the art will appreciate that fig. 6 is merely an example of a terminal device 2 and does not constitute a limitation of terminal device 2 and may include more or fewer components than shown, or some components may be combined, or different components, e.g., the terminal device may also include input-output devices, network access devices, buses, etc.
The Processor 20 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory 21 may be an internal storage unit of the terminal device 2, such as a hard disk or a memory of the terminal device 2. The memory 21 may also be an external storage device of the terminal device 2, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the terminal device 2. Further, the memory 21 may also include both an internal storage unit and an external storage device of the terminal device 2. The memory 21 is used for storing the computer program and other programs and data required by the terminal device. The memory 21 may also be used to temporarily store data that has been output or is to be output.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.

Claims (10)

1. An electronic health case generation method, comprising:
acquiring real sample data, and training a self-encoder according to the real sample data;
acquiring preset noise data, and inputting the preset noise data into a generative countermeasure network for data generation to obtain false data;
controlling the trained self-encoder to decode the real sample data and the false data to obtain sample decoded data and false decoded data;
performing weight parameter optimization on the generative confrontation network according to the sample decoding data and the false decoding data until the generative confrontation network meets a preset end condition, and merging the trained self-encoder into the generative confrontation network after parameter optimization;
inputting the medical data to be converted into the combined generative confrontation network, and instructing the generative confrontation network to generate data according to the medical data to be converted to obtain the electronic health case.
2. The electronic health case generation method of claim 1, wherein training an auto-encoder according to the real sample data comprises:
converting the disease information in the real sample data into sample coded data according to a disease code list, and performing vector conversion on the sample coded data to obtain sample vector data;
sampling the sample vector data to obtain a sampling vector, and controlling an encoder in the self-encoder to encode the sampling vector to obtain an encoding result;
controlling a decoder in the self-encoder to decode the encoding result to obtain a decoding result, and performing loss calculation according to the decoding result to obtain a loss value;
and respectively updating parameters of the encoder and the decoder according to the loss values.
3. The electronic health case generation method according to claim 2, further comprising, after the parameter updating the encoder and the decoder according to the loss values, respectively:
if the self-encoder meets preset optimization conditions, stopping training of the self-encoder;
and if the self-encoder does not meet the preset optimization condition, sampling the sample vector data again until the self-encoder meets the preset optimization condition.
4. The electronic health case generation method according to claim 1, wherein the self-encoder after the control training decodes the real sample data and the dummy data to obtain sample decoded data and dummy decoded data, and comprises:
respectively sampling real sample data and false data according to a preset sampling value to obtain real sample data and false sample data;
controlling the decoder to decode the real sampling data and the false sampling data respectively to obtain a sample decoding result and a false decoding result;
specifically, the performing weight parameter optimization on the generative countermeasure network according to the sample decoded data and the dummy decoded data includes:
carrying out weight parameter optimization on a discriminator in the generative countermeasure network according to the sample decoding data and the false decoding data;
and optimizing weight parameters of the generator in the generative countermeasure network according to the sample decoding data and the false decoding data.
5. The electronic health case generation method of claim 4, wherein a first optimization formula for optimizing weight parameters of discriminators in the generative countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure FDA0002564158820000021
Figure FDA0002564158820000022
θd＝clip(θd，-c，c)；
wherein α is learning rate, RMSProp is optimization algorithm, clip is interception function, c is interception fixed value in the interception function, ziFor the dummy decoded data, xiDecoding data for said sample, m being said predetermined sampling value, D (x)i) Probability, G (z), of decoding data for said sample as said true sample datai) The first optimization formula is used to calculate a loss function value in the discriminator for data generated after the dummy decoded data is input to the generator
Figure FDA0002564158820000023
Value of the loss function
Figure FDA0002564158820000024
Weight θ for the discriminatordAnd updating the numerical value.
6. The electronic health case generation method of claim 4, wherein a second optimization formula for optimizing the weight parameters of the generators in the generative countermeasure network according to the sample decoded data and the dummy decoded data is as follows:
Figure FDA0002564158820000031
Figure FDA0002564158820000032
wherein α is learning rate, RMSProp is optimization algorithm, ziFor the dummy decoded data, xiDecoding data for said sample, m being said predetermined sample value, said second optimization formula being used to calculateLoss function value in said generator
Figure FDA0002564158820000033
Value of the loss function
Figure FDA0002564158820000034
Weight θ for the generatorgAnd updating the numerical value.
7. The electronic health case generation method of claim 1, further comprising:
uploading the electronic health case into a blockchain.
8. An electronic health case generation device, comprising:
the self-encoder training unit is used for acquiring real sample data and training the self-encoder according to the real sample data;
the data generation unit is used for acquiring preset noise data and inputting the preset noise data into a generation type countermeasure network for data generation to obtain false data;
the data decoding unit is used for controlling the trained self-encoder to decode the real sample data and the false data to obtain sample decoding data and false decoding data;
a parameter optimization unit, configured to perform weight parameter optimization on the generative countermeasure network according to the sample decoded data and the dummy decoded data until the generative countermeasure network meets a preset end condition, and merge the trained self-encoder into the generative countermeasure network after parameter optimization;
and the medical record generation unit is used for inputting the medical data to be converted into the combined generative confrontation network and indicating the generative confrontation network to generate data according to the medical data to be converted to obtain the electronic health case.
9. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 7 when executing the computer program.
10. A storage medium storing a computer program, characterized in that the computer program realizes the steps of the method according to any one of claims 1 to 7 when executed by a processor.
CN202010617099.5A 2020-06-30 2020-06-30 Electronic health case generation method, device, terminal equipment and storage medium Active CN111768826B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010617099.5A CN111768826B (en) 2020-06-30 2020-06-30 Electronic health case generation method, device, terminal equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010617099.5A CN111768826B (en) 2020-06-30 2020-06-30 Electronic health case generation method, device, terminal equipment and storage medium

Publications (2)

Publication Number Publication Date
CN111768826A true CN111768826A (en) 2020-10-13
CN111768826B CN111768826B (en) 2023-06-27

Family

ID=72724367

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010617099.5A Active CN111768826B (en) 2020-06-30 2020-06-30 Electronic health case generation method, device, terminal equipment and storage medium

Country Status (1)

Country Link
CN (1) CN111768826B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114756894A (en) * 2022-06-15 2022-07-15 北京中科开迪软件有限公司 Electronic image archive protection method and system based on optical disk library

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108550118A (en) * 2018-03-22 2018-09-18 深圳大学 Fuzzy processing method, device, equipment and the storage medium of motion blur image
US20190046068A1 (en) * 2017-08-10 2019-02-14 Siemens Healthcare Gmbh Protocol independent image processing with adversarial networks
CN110032931A (en) * 2019-03-01 2019-07-19 阿里巴巴集团控股有限公司 Generate confrontation network training, reticulate pattern minimizing technology, device and electronic equipment
WO2019157435A1 (en) * 2018-02-12 2019-08-15 Rensselaer Polytechnic Institute Ct big data from simulation, emulation and transfer learning
US20190333623A1 (en) * 2018-04-30 2019-10-31 Elekta, Inc. Radiotherapy treatment plan modeling using generative adversarial networks
US20200134415A1 (en) * 2018-10-30 2020-04-30 Huawei Technologies Co., Ltd. Autoencoder-Based Generative Adversarial Networks for Text Generation

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190046068A1 (en) * 2017-08-10 2019-02-14 Siemens Healthcare Gmbh Protocol independent image processing with adversarial networks
WO2019157435A1 (en) * 2018-02-12 2019-08-15 Rensselaer Polytechnic Institute Ct big data from simulation, emulation and transfer learning
CN108550118A (en) * 2018-03-22 2018-09-18 深圳大学 Fuzzy processing method, device, equipment and the storage medium of motion blur image
US20190333623A1 (en) * 2018-04-30 2019-10-31 Elekta, Inc. Radiotherapy treatment plan modeling using generative adversarial networks
US20200134415A1 (en) * 2018-10-30 2020-04-30 Huawei Technologies Co., Ltd. Autoencoder-Based Generative Adversarial Networks for Text Generation
CN110032931A (en) * 2019-03-01 2019-07-19 阿里巴巴集团控股有限公司 Generate confrontation network training, reticulate pattern minimizing technology, device and electronic equipment

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114756894A (en) * 2022-06-15 2022-07-15 北京中科开迪软件有限公司 Electronic image archive protection method and system based on optical disk library
CN114756894B (en) * 2022-06-15 2022-08-23 北京中科开迪软件有限公司 Electronic image archive protection method and system based on optical disk library

Also Published As

Publication number Publication date
CN111768826B (en) 2023-06-27

Similar Documents

Publication Publication Date Title
CN112259246B (en) Disease prediction method integrating medical concept hierarchy structure and related equipment
CN112418292B (en) Image quality evaluation method, device, computer equipment and storage medium
CN109478222A (en) The method of the data exchanged between protection service user and ISP
JP2019075108A (en) Information processing method and device and information detection method and device
CN109698017B (en) Medical record data generation method and device
CN111340638A (en) Abnormal medical insurance document identification method and device, computer equipment and storage medium
CN112257578A (en) Face key point detection method and device, electronic equipment and storage medium
CN112435755A (en) Disease analysis method, disease analysis device, electronic device, and storage medium
CN111768826B (en) Electronic health case generation method, device, terminal equipment and storage medium
CN112529767B (en) Image data processing method, device, computer equipment and storage medium
EP3893159A1 (en) Training a convolutional neural network
CN117557689A (en) Image processing method, device, electronic equipment and storage medium
CN116821113A (en) Time sequence data missing value processing method and device, computer equipment and storage medium
CN116095183A (en) Data compression method and related equipment
CN116313148A (en) Drug sensitivity prediction method, device, terminal equipment and medium
CN115914488A (en) Medical image identity confusion sharing method and system, terminal equipment and storage medium
CN113239814B (en) Facial expression recognition method, device, equipment and medium based on optical flow reconstruction
CN115588492A (en) Diagnosis guiding method and system based on artificial intelligence
CN113554719B (en) Image encoding method, decoding method, storage medium and terminal equipment
KR20230022005A (en) Device and method for extracting compound information
Akhtar et al. Significance of ROI coding using MAXSHIFT scaling applied on MRI images in teleradiology-telemedicine
US20230353765A1 (en) Lossless Compression with Probabilistic Circuits
US20220319635A1 (en) Generating minority-class examples for training data
US20240185043A1 (en) Generating Synthetic Heterogenous Time-Series Data
CN115063500B (en) Magnetic nanoparticle imaging reconstruction method based on generation countermeasure network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
TA01 Transfer of patent application right

Effective date of registration: 20220921

Address after: Room 2601 (Unit 07), Qianhai Free Trade Building, No. 3048, Xinghai Avenue, Nanshan Street, Qianhai Shenzhen-Hong Kong Cooperation Zone, Shenzhen, Guangdong 518000

Applicant after: Shenzhen Ping An Smart Healthcare Technology Co.,Ltd.

Address before: 1-34 / F, Qianhai free trade building, 3048 Xinghai Avenue, Mawan, Qianhai Shenzhen Hong Kong cooperation zone, Shenzhen, Guangdong 518000

Applicant before: Ping An International Smart City Technology Co.,Ltd.

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant