Chronic disease patient archive data analysis method and device, storage medium and equipment

Abstract

The invention relates to the technical field of computers, and discloses a chronic disease patient archive data analysis method and device, a storage medium and equipment, and the method comprises the steps: dividing training archive data into discrete variables and continuous variables, carrying out the processing of each class value in the discrete variables through one-hot coding, and carrying out the processing of each class value in the continuous variables through one-hot coding; a user-defined normalization method is applied; the discrete variables and the continuous variables after standardization processing are combined into condition vectors, two discrete variable combinations with equal probabilities are randomly selected, all combination probabilities are calculated, and sampling is completed; the generative adversarial network framework is used for training the generative model and the discrimination model for multiple times, the discrimination model is used for discriminating the generative model to obtain a trained generative model, synthetic data are generated, and a neural network used for patient feature analysis is trained to predict chronic diseases. According to the invention, the problem that complete analysis and privacy protection of archive data of chronic disease patients cannot be achieved at the same time in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, storage medium and equipment for analyzing archive data of patients with chronic diseases. Background Art

[0002] Over the past few decades, artificial intelligence (AI) has been revolutionizing the healthcare industry through innovative computational methods that can support or even replace human roles in certain specific tasks. So far, AI has been applied to almost every aspect of daily life, such as voice assistants, car safety devices, and smart home sensors. Due to the huge amount of data and the increasing application of digital processing in clinical practice, multiple studies have increasingly explored the application of artificial intelligence in medical research.

[0003] The application of AI in chronic diseases requires the analysis of a large number of patients' archival data at various stages of illness, which are usually provided by various hospitals. Due to the huge amount of data, many patients are involved, and these data will contain the privacy of patients. Patients' privacy must not be leaked, nor can it be known by the institutions conducting the analysis. However, for many chronic diseases, archival data at various stages of illness are essential reference data for the diagnosis of the disease. For example, for patients with certain brain lesions, early data is particularly important for the research and application of brain disease prevention, early diagnosis and early intervention. If it cannot be correctly identified or misidentified, medical accidents will occur. Therefore, when analyzing chronic diseases based on AI, it is necessary to remove patient privacy in archival data. However, these privacy may be related to the patient's symptoms. Removing privacy will result in incomplete analysis data. You cannot have both. Summary of the invention

[0004] In view of the above technical problems, the present invention provides a method, device, storage medium and equipment for analyzing the archive data of chronic disease patients to solve the problem in the prior art that complete analysis of the archive data of chronic disease patients and privacy protection cannot be achieved at the same time.

[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0006] According to one aspect of the present invention, a method for analyzing archive data of patients with chronic diseases is proposed, and the method comprises:

[0007] The training file data of chronic disease patients are divided into discrete variables and continuous variables. For each category value in the discrete variable, one-hot encoding is used for processing. For the continuous variable, a custom normalization method is applied. The normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each continuous variable according to the number of modes to complete preprocessing;

[0008] The encoded discrete variables and the standardized continuous variables are combined into a conditional vector. In the data sampling stage, two combinations of discrete variables with equal probability are randomly selected for each row of data, and all combination probabilities are calculated. After the probability is amplified, a pair is randomly selected from all combinations as the final sample data to complete the sampling.

[0009] Based on Gaussian noise, conditional vectors and sample data, a generative model and a discriminative model are trained multiple times using a generative adversarial network framework, and the discriminative model is used to discriminate the generative model to obtain the trained generative model;

[0010] After performing preprocessing corresponding to the generation model on the archive data to be protected, random Gaussian noise is introduced into the corresponding conditional vector during sampling, and the conditional vector and Gaussian noise are input into the trained generation model to obtain synthetic data that is consistent with the feature distribution of the original archive data but not associated with the actual data;

[0011] Based on the synthetic data, a neural network for patient feature analysis is trained to predict chronic diseases.

[0012] Furthermore, the normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each of the continuous variables according to the number of modes, specifically including:

[0013] The continuous variable is a continuous data column. For each of the continuous data columns, a variational Gaussian mixture model is used to infer the number of patterns, and these data are fitted into a Gaussian mixture distribution.

[0014] For each of the continuous data columns, based on the fitted Gaussian mixture distribution, calculate the probability distribution of the data column belonging to each mode;

[0015] Based on the calculated probability distribution, samples are extracted from each pattern and the continuous data series are normalized using the extracted patterns.

[0016] Furthermore, the generative model and the discriminant model adopt a fully connected architecture with two hidden layers. The hidden layer of the generative model is processed by batch normalization and ReLU activation function, and its output layer combines tanh and softmax activation functions. The hidden layer of the discriminant model uses Leaky ReLU activation function. During the training process, the loss of the generative model is optimized using binary cross entropy loss combined with Sigmoid function, and the objective function of the discriminant model uses Wasserstein distance with gradient penalty. In each iteration, Gaussian noise and its covariance are used to calculate the gradient of Wasserstein distance to complete the training of the generative model.

[0017] Furthermore, the binary cross entropy loss function combined with the Sigmoid function is expressed as:

[0018]

[0019] Among them, x i is the input value of the training data, σ(yx) is the value of the Sigmoid function, is the sample data, and n is the total number of sample data.

[0020] Furthermore, the Wasserstein distance function with gradient penalty is expressed as:

[0021]

[0022] Among them, λ is the weight of the gradient penalty term, which is used to control the intensity of the gradient penalty;

[0023] P x is the distribution of the real data P r and generate data distribution P g The distribution obtained by uniform sampling between ∈ is a random number, and ∈~U[0,1];

[0024] x i is the training data;

[0025] The gradient of the discriminant model is expressed as:

[0026]

[0027] Where n is the total number of samples, x i is the i-th sample, is the gradient calculation of the discriminant model parameter ω, (D ω ,x i ) is the loss value corresponding to a single sample.

[0028] Furthermore, when predicting chronic diseases, it includes:

[0029] Collect synthetic data containing relevant characteristics of chronic diseases, standardize, encode and process missing values ​​of the synthetic data, and analyze the patient characteristics based on the constructed neural network model for patient characteristic analysis.

[0030] According to a second aspect of the present disclosure, a chronic disease patient file data analysis device is provided, comprising: a preprocessing module, used to divide the training file data of chronic disease patients into discrete variables and continuous variables, for each category value in the discrete variable, one-hot encoding is used for processing, for the continuous variable, a custom normalization method is applied, the normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each of the continuous variables according to the number of modes to complete the preprocessing; a sampling module, used to complete the discrete variables that have been encoded and the continuous variables that have been standardized are combined into a conditional vector, and in the data sampling stage, the processing of each row of data will randomly select two discrete variable combinations with equal probability, and calculate all combination probabilities to amplify the probabilities After that, a pair is randomly selected from all the combinations as the final sample data to complete the sampling; a training module is used to train the generative model and the discriminative model multiple times based on Gaussian noise, conditional vector and sample data using a generative adversarial network framework, and discriminate the generative model using the discriminative model to obtain the trained generative model; a synthesis module is used to perform preprocessing corresponding to the generative model on the archive data that needs privacy protection, introduce random Gaussian noise into the conditional vector corresponding to the sample at the time of sampling, input the conditional vector and Gaussian noise into the trained generative model, and obtain synthetic data that is consistent with the feature distribution of the original archive data but has no actual data association; a prediction module is used to train a neural network for patient feature analysis based on the synthetic data to predict chronic diseases.

[0031] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, storing a computer program, wherein when the computer program is executed by a processor, the method for analyzing the chronic disease patient archive data as described above is implemented.

[0032] According to a fourth aspect of the present disclosure, a chronic disease patient file data analysis device is provided, comprising: a processor; and a memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor executes the above-mentioned chronic disease patient file data analysis method.

[0033] The technical solution disclosed in this disclosure has the following beneficial effects:

[0034] The present invention performs modal classification on archival data and constructs a model using conditional vectors as additional inputs, thereby effectively presenting sparse categories in unbalanced archival data while maximally maintaining the dependencies between variables, so that the privacy of archival data can be effectively used for data analysis after being hidden. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of a method for analyzing chronic disease patient archive data in an embodiment of this specification;

[0036] Figure 2 This is a structural block diagram of a chronic disease patient file data analysis device in an embodiment of this specification;

[0037] Figure 3 It is a terminal device for implementing a method for analyzing the archive data of patients with chronic diseases in the embodiments of this specification;

[0038] Figure 4 It is a computer-readable storage medium storing a method for analyzing chronic disease patient archive data in an embodiment of this specification. DETAILED DESCRIPTION

[0039] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0040] In addition, the accompanying drawings are only schematic illustrations of the present disclosure. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0041] The present invention provides a method for analyzing the archive data of patients with chronic diseases. Figure 1FIG. 1 is a flow chart of a method for analyzing chronic disease patient profile data of a product provided in an embodiment of the present invention. The method can be applied to electronic devices such as personal computers, servers, and tablet computers. The method can be performed by a device, which can be implemented by software and / or hardware. The method can specifically include the following steps S101 to S105:

[0042] In step S101, the training file data of chronic disease patients are divided into discrete variables and continuous variables. For each category value in the discrete variable, one-hot encoding is used for processing. For the continuous variable, a custom normalization method is applied. The normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each of the continuous variables according to the number of modes to complete preprocessing.

[0043] Discrete variables refer to variables whose values ​​are discrete, such as gender or the prevalence of a certain disease. In real-world data sets, continuous variables often present multimodal distribution characteristics, such as the height (or body length) of men and women. Traditional normalization methods, such as the minimum-maximum transformation, usually compress these continuous variables into the range of [-1,1], but this may lead to information loss or introduce artifacts into the data. In order to solve this problem, the present invention adopts a custom normalization method, that is, a variational Gaussian mixture model is used to estimate the number of modes of continuous variables, and each continuous variable is standardized according to the estimated result.

[0044] Among them, the continuous variable is a continuous data column. For each of the continuous data columns, a variational Gaussian mixture model is used to infer the number of its modes, and these data are fitted into a Gaussian mixture distribution; for each of the continuous data columns, the probability distribution of each mode is calculated based on the fitted Gaussian mixture distribution; according to the calculated probability distribution, samples are extracted from each mode, and the continuous data column is normalized using the extracted mode.

[0045] Specifically, the model of Gaussian mixture distribution can be expressed as:

[0046]

[0047] The Gaussian mixture distribution model is used to represent the continuous data column c i The model can be composed of multiple modes, each with a weight μ k and a standard deviation φ k , where N(c i,j ; η k ,φ k ) is a k and standard deviation φ kThen, for the continuous data column c i Each value c in i,j , calculate the probability from each mode. The probability is calculated as:

[0048] ρ k =μ k N(c i,j ; η k ,φ k );

[0049] where ρ k It is c i,j The probability of coming from the kth pattern.

[0050] The rows of continuous variables can be converted into a concatenation of continuous and discrete columns. The expression for the concatenation is:

[0051]

[0052] where d i,j It is a one-hot encoding representation of discrete variables.

[0053] In step S102, the discrete variables that have been encoded and the continuous variables that have been standardized are combined into a conditional vector. In the data sampling stage, two combinations of discrete variables with equal probabilities are randomly selected for each row of data, and all combination probabilities are calculated. After the probabilities are amplified, a pair is randomly selected from all the combinations as the final sample data to complete the sampling.

[0054] In the prior art, the processing of archival data often processes each variable separately, which leads to ignoring the interdependence between variables in the data. In this embodiment, each value of the discrete variable is encoded and used as an additional conditional vector for training the generative model and the discriminant model. During the training process, the input training data is resampled according to the conditional vector to ensure that those less frequent category values ​​can be observed during the training process. After the categorical variable conversion, each category value is converted into an encoding vector. For each row of sampled data, two categorical variables with the same probability are randomly selected, such as "gender" and "diabetes". Next, the probability mass of all possible combinations of categorical values ​​in these two variables is calculated. Based on these calculated probabilities, a pair is randomly selected from all possible combinations, for example, the combination of "male and no diabetes" is selected and represented as a specific encoding form, such as [01,100,2]. Since patient file data is often unbalanced, some rare categorical values ​​appear less frequently. In order to increase the chance of capturing the dependencies between these rare categorical values, the logarithm of the probability can be calculated, which can amplify these probability values. This logarithmic form of probability can be expressed as log(P(x)).

[0055] In step S103, based on Gaussian noise, conditional vectors and sample data, the generative model and the discriminative model are trained multiple times using a generative adversarial network framework, and the discriminative model is used to discriminate the generative model to obtain the trained generative model.

[0056] Specifically, the generative model and the discriminant model adopt a fully connected architecture with two hidden layers. The hidden layer of the generative model is processed by batch normalization and ReLU activation function, and its output layer combines tanh and softmax activation functions. The hidden layer of the discriminant model uses Leaky ReLU activation function. During the training process, the loss of the generative model is optimized using binary cross entropy loss combined with Sigmoid function. The objective function of the discriminant model adopts Wasserstein distance with gradient penalty. In each iteration, Gaussian noise and its covariance are used to calculate the gradient of Wasserstein distance to complete the training of the generative model.

[0057] In the generative model, batch normalization and Relu activation function are applied to each hidden layer of the generative model to improve efficiency and stability to solve the problems of gradient disappearance and data sparsity. In order to generate a mixture of categorical and continuous features, tanh and softmax activation functions are applied to the output layer. In the discriminative model, leaky Relu function is applied to the hidden layer, which can handle negative input values ​​and performs better than Relu and dropout of each layer in the discriminative model.

[0058] In addition, this embodiment uses a discriminator to perform classification learning on the input classification vector and continuous vector to distinguish between real data and synthetic data output by the generation model. By optimizing the classification loss, the generation model is trained to produce more realistic synthetic data. In addition to learning from the loss feedback of the discriminator, the generation model also introduces an additional penalty mechanism to ensure that the generated data can reflect the variables and values ​​sampled in the conditional vector, thereby more accurately simulating the conditional vector of the real data. Since the conditional vector contains multiple variables, we introduced a method to capture the dependencies between these variables. To this end, we use a binary cross entropy loss function combined with a Sigmoid layer to optimize the performance of the generation model, where the binary cross entropy loss function combined with the Sigmoid function is expressed as:

[0059]

[0060] Among them, x i is the input value of the training data, σ(y i ) is the value of Sigmoid function, is the sample data, and n is the total number of sample data.

[0061] When training a generative adversarial network (GAN), two major problems may be encountered, which may cause the training process to fail. First, if the performance of the discriminator is too perfect and can correctly distinguish between real data and synthetic data, the gradient information provided will become insignificant because the gradient almost disappears and cannot provide useful information for further training of the generator. Second, the generator may fall into generating a small number of outputs that are misclassified as real data by the discriminator with high confidence, causing the generator to repeatedly generate the same synthetic data points. This phenomenon is called "model collapse". In order to overcome these problems, this embodiment proposes to use the Wasserstein distance with gradient penalty as the objective function in the discriminant model. The Wasserstein distance is a method to measure the difference between two probability distributions. It quantifies the minimum cost required to convert a random sample from a distribution to a target distribution, ensuring that the discriminant model can always provide meaningful gradients for the generative model, which helps to stabilize the training process and prevent the gradient from disappearing. Specifically, the Wasserstein distance function with gradient penalty is expressed as:

[0062]

[0063] Among them, λ is the weight of the gradient penalty term, which is used to control the intensity of the gradient penalty;

[0064] P x is the distribution of the real data P r and generate data distribution P g The distribution obtained by uniform sampling between ∈ is a random number, and ∈~U[0,1];

[0065] x i is the training data;

[0066] The gradient of the discriminator is expressed as:

[0067]

[0068] Where n is the total number of samples, x i is the i-th sample, is the gradient calculation of the discriminator parameter ω, (D ω ,x i ) is the loss value corresponding to a single sample.

[0069] In step S104, after the archival data to be privacy protected is preprocessed according to the generation model, random Gaussian noise is introduced into the corresponding conditional vector during sampling, and the conditional vector and Gaussian noise are input into the trained generation model to obtain synthetic data that is consistent with the feature distribution of the original archival data but has no association with actual data.

[0070] Specifically, in the process of synthetic data generation, differential privacy replacement can be used to ensure that replacing or deleting the privacy information of a single patient has the least impact on the synthetic data, while protecting personal privacy and generating a statistically significant synthetic data set. The effectiveness of privacy protection can be evaluated by calculation. Specifically, if there are two data sets D and D' that differ only in a single data point, then for any output subset S, it is expressed as:

[0071] P(M p (D)∈S)≤e ∈ ·P(M p (D′)∈S)+δ;

[0072] Among them, M p (D represents the original training data, M p (D′ represents the generated synthetic data, p represents the randomness introduced by Gaussian noise, and e ∈ is the privacy budget parameter.

[0073] When synthetic data is generated, the privacy budget e ∈ It can be understood as a privacy loss measure for a small change in synthetic data when a data instance is added or deleted from the real training data. The smaller the values ​​of ∈ and δ, the smaller the difference in the output probability of the model when processing datasets D and D'', thus providing stronger privacy protection.

[0074] In step S105, a neural network for patient feature analysis is trained based on the synthesized data to predict chronic diseases.

[0075] Specifically, synthetic data containing relevant characteristics of chronic diseases are collected, the synthetic data are standardized, encoded and missing value processed, and the patient characteristics are analyzed based on the constructed neural network model for patient characteristic analysis.

[0076] Based on the same idea, Figure 2As shown, a chronic disease patient file data analysis device is provided, the device includes: a preprocessing module 201, used to divide the training file data of chronic disease patients into discrete variables and continuous variables, for each category value in the discrete variable, one-hot encoding is used for processing, for the continuous variable, a custom normalization method is applied, the normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each of the continuous variables according to the number of modes to complete the preprocessing; a sampling module 202, used to complete the discrete variables that have been encoded and the continuous variables that have been standardized are combined into a conditional vector, in the data sampling stage, the processing of each row of data will randomly select two combinations of discrete variables with equal probability, and calculate all the combination probabilities, after amplifying the probabilities, from all A pair is randomly selected from the combination as the final sample data to complete the sampling; a training module 203 is used to train the generative model and the discriminant model multiple times based on Gaussian noise, conditional vector and sample data using a generative adversarial network framework, and discriminate the generative model using the discriminant model to obtain the trained generative model; a synthesis module 204 is used to perform preprocessing corresponding to the generative model on the archive data requiring privacy protection, introduce random Gaussian noise into the conditional vector corresponding to the sample at the time of sampling, input the conditional vector and Gaussian noise into the trained generative model, and obtain synthetic data that is consistent with the feature distribution of the original archive data but is not associated with actual data; a prediction module 205 is used to train a neural network for patient feature analysis based on the synthetic data to predict chronic diseases.

[0077] The above-mentioned device constructs a model by performing modal classification on the archival data and taking the conditional vector as an additional input, thereby effectively presenting the sparse categories in the unbalanced archival data, while maintaining the dependencies between variables to the greatest extent, so that the privacy of the archival data can be effectively used for data analysis after being hidden.

[0078] The specific details of each module / unit in the above-mentioned device have been described in detail in the implementation method of the method part. The undisclosed details can be found in the implementation method of the method part, so they will not be repeated here.

[0079] Based on the same idea, the present specification also provides a chronic disease patient file data analysis device, such as Figure 3 shown.

[0080] The chronic disease patient file data analysis device may be the terminal device or server provided in the above embodiment.

[0081] The chronic disease patient archive data analysis device may have relatively large differences due to different configurations or performances, and may include one or more processors 301, memory 302 and a bus, and the memory 302 may store one or more storage applications or data. Among them, the memory 302 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) and / or a cache memory unit, that is, for example, a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on an electronic device, and may further include a read-only storage unit. The application stored in the memory 302 may include one or more program modules (not shown in the figure), such program modules include but are not limited to: an operating system, one or more application programs, other program modules and program data, each of these examples or a certain combination may include the implementation of a network environment. Further, the processor 301 may be configured to communicate with the memory 302 to execute a series of computer executable instructions in the memory 302 on the chronic disease patient archive data analysis device. The chronic disease patient profile data analysis device may also include one or more power supplies 303, one or more wired or wireless network interfaces 304, one or more I / O interfaces (input and output interfaces) 305, one or more external devices 306 (such as a keyboard), and may also communicate with one or more devices that enable a user to interact with the device, and / or communicate with any device that enables the device to communicate with one or more other computing devices (such as a router, a network switch, etc.). Such communication may be performed through the I / O interface 305. In addition, the device may also communicate with one or more networks (such as a local area network (LAN)) through the wired or wireless interface 304.

[0082] The processor 301 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 301 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect the various components of the entire electronic device, and executes or executes the programs or modules stored in the memory 11 (such as the product chronic disease patient file data analysis program, etc.), and calls the data stored in the memory 11 to perform various functions of the processing device and process data.

[0083] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 302 and at least one processor 301, etc.

[0084] The power supply can be logically connected to the at least one processor 301 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power supply can also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 can also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0085] Figure 3 Only the chronic disease patient archive data analysis device with components is shown, and those skilled in the art can understand that Figure 3 The structure shown does not constitute a limitation on the chronic disease patient archive data analysis device, and may include fewer or more components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0086] Specifically in this embodiment, the chronic disease patient file data analysis device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the chronic disease patient file data analysis device, and the one or more programs are configured to be executed by one or more processors, including the following computer executable instructions:

[0087] The training file data of chronic disease patients are divided into discrete variables and continuous variables. For each category value in the discrete variable, one-hot encoding is used for processing. For the continuous variable, a custom normalization method is applied. The normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each continuous variable according to the number of modes to complete preprocessing;

[0088] The encoded discrete variables and the standardized continuous variables are combined into a conditional vector. In the data sampling stage, two combinations of discrete variables with equal probability are randomly selected for each row of data, and all combination probabilities are calculated. After the probability is amplified, a pair is randomly selected from all combinations as the final sample data to complete the sampling.

[0089] Based on Gaussian noise, conditional vectors and sample data, a generative model and a discriminative model are trained multiple times using a generative adversarial network framework, and the discriminative model is used to discriminate the generative model to obtain the trained generative model;

[0090] After performing preprocessing corresponding to the generation model on the archive data to be protected, random Gaussian noise is introduced into the corresponding conditional vector during sampling, and the conditional vector and Gaussian noise are input into the trained generation model to obtain synthetic data that is consistent with the feature distribution of the original archive data but not associated with the actual data;

[0091] Based on the synthetic data, a neural network for patient feature analysis is trained to predict chronic diseases.

[0092] Based on the same idea, the exemplary embodiment of the present invention also provides a computer-readable storage medium, on which a program product capable of implementing the above method of this specification is stored. In some possible implementations, various aspects of the present disclosure can also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.

[0093] refer to Figure 4 As shown, a program 500 for implementing the above method according to an exemplary embodiment of the present disclosure is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0094] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0095] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0096] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, CSS, HTML, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0097] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the exemplary implementation of the present disclosure.

[0098] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiment of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0099] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the exemplary embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0100] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0101] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for analyzing archive data of patients with chronic diseases, characterized in that: The analytical method comprises: The training file data of chronic disease patients are divided into discrete variables and continuous variables. For each category value in the discrete variable, one-hot encoding is used for processing. For the continuous variable, a custom normalization method is applied. The normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each continuous variable according to the number of modes to complete preprocessing; The encoded discrete variables and the standardized continuous variables are combined into a conditional vector. In the data sampling stage, two combinations of discrete variables with equal probability are randomly selected for each row of data, and all combination probabilities are calculated. After the probability is amplified, a pair is randomly selected from all combinations as the final sample data to complete the sampling. Based on Gaussian noise, conditional vectors and sample data, a generative model and a discriminative model are trained multiple times using a generative adversarial network framework, and the discriminative model is used to discriminate the generative model to obtain the trained generative model; After performing preprocessing corresponding to the generation model on the archive data to be protected, random Gaussian noise is introduced into the corresponding conditional vector during sampling, and the conditional vector and Gaussian noise are input into the trained generation model to obtain synthetic data that is consistent with the feature distribution of the original archive data but not associated with the actual data; Based on the synthetic data, a neural network for patient feature analysis is trained to predict chronic diseases.

2. The method for analyzing chronic disease patient archive data according to claim 1, characterized in that: The normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each of the continuous variables according to the number of modes, specifically including: The continuous variable is a continuous data column. For each of the continuous data columns, a variational Gaussian mixture model is used to infer the number of patterns, and these data are fitted into a Gaussian mixture distribution. For each of the continuous data columns, based on the fitted Gaussian mixture distribution, calculate the probability distribution of the data column belonging to each mode; Based on the calculated probability distribution, samples are extracted from each pattern and the continuous data series are normalized using the extracted patterns.

3. The method for analyzing chronic disease patient archive data according to claim 1, characterized in that: The generative model and the discriminant model adopt a fully connected architecture with two hidden layers. The hidden layer of the generative model is processed by batch normalization and ReLU activation function, and its output layer combines tanh and softmax activation functions. The hidden layer of the discriminant model uses Leaky ReLU activation function. During the training process, the loss of the generative model is optimized using binary cross entropy loss combined with Sigmoid function. The objective function of the discriminant model adopts Wasserstein distance with gradient penalty. In each iteration, Gaussian noise and its covariance are used to calculate the gradient of Wasserstein distance to complete the training of the generative model.

4. The method for analyzing chronic disease patient archive data according to claim 3, characterized in that: The binary cross entropy loss function combined with the Sigmoid function is expressed as: Among them, x i is the input value of the training data, σ(y i ) is the value of Sigmoid function, is the sample data, and n is the total number of sample data.

5. The method for analyzing chronic disease patient archive data according to claim 3, characterized in that: The Wasserstein distance function with gradient penalty is expressed as: Among them, λ is the weight of the gradient penalty term, which is used to control the intensity of the gradient penalty; P x is the distribution of the real data P r and generate data distribution P g The distribution obtained by uniform sampling between ∈ is a random number, and ∈~U [ 0,1 ] ; x i is the training data; The gradient of the discriminant model is expressed as: Where n is the total number of samples, x i is the i-th sample, is the gradient calculation of the discriminant model parameter ω, (D ω ,x i ) is the loss value corresponding to a single sample.

6. The method for analyzing chronic disease patient archive data according to claim 1, characterized in that: When predicting chronic diseases, include: Collect synthetic data containing relevant characteristics of chronic diseases, standardize, encode and process missing values ​​of the synthetic data, and analyze the patient characteristics based on the constructed neural network model for patient characteristic analysis.

7. A chronic disease patient file data analysis device, characterized in that: include: A preprocessing module, used to divide the training file data of chronic disease patients into discrete variables and continuous variables, and use one-hot encoding to process each category value in the discrete variables, and use a custom normalization method for the continuous variables. The normalization method infers the number of modes of the continuous variables through a variational Gaussian mixture model, and performs a standardization operation on each of the continuous variables according to the number of modes to complete the preprocessing; The sampling module is used to complete the combination of the encoded discrete variables and the standardized continuous variables into a conditional vector. In the data sampling stage, two discrete variable combinations with equal probability are randomly selected for each row of data, and all combination probabilities are calculated. After the probability is amplified, a pair is randomly selected from all combinations as the final sample data to complete the sampling; A training module, used to train the generative model and the discriminative model multiple times based on Gaussian noise, conditional vectors and sample data using a generative adversarial network framework, and discriminate the generative model using the discriminative model to obtain the trained generative model; A synthesis module, for performing preprocessing corresponding to the generation model on the archive data requiring privacy protection, introducing random Gaussian noise into the corresponding conditional vector during sampling, and inputting the conditional vector and Gaussian noise into the trained generation model to obtain synthetic data that is consistent with the feature distribution of the original archive data but not associated with the actual data; The prediction module is used to train a neural network for patient feature analysis based on the synthetic data to predict chronic diseases.

8. A chronic disease patient file data analysis device, characterized in that: The device comprises: a processor; and a memory arranged to store computer executable instructions which, when executed, cause the processor to: The training file data of chronic disease patients are divided into discrete variables and continuous variables. For each category value in the discrete variable, one-hot encoding is used for processing. For the continuous variable, a custom normalization method is applied. The normalization method infers the number of modes of the continuous variable through a variational Gaussian mixture model, and performs a standardization operation on each continuous variable according to the number of modes to complete preprocessing; The encoded discrete variables and the standardized continuous variables are combined into a conditional vector. In the data sampling stage, two combinations of discrete variables with equal probability are randomly selected for each row of data, and all combination probabilities are calculated. After the probability is amplified, a pair is randomly selected from all combinations as the final sample data to complete the sampling. Based on Gaussian noise, conditional vectors and sample data, a generative model and a discriminative model are trained multiple times using a generative adversarial network framework, and the discriminative model is used to discriminate the generative model to obtain the trained generative model; After performing preprocessing corresponding to the generation model on the archive data to be protected, random Gaussian noise is introduced into the corresponding conditional vector during sampling, and the conditional vector and Gaussian noise are input into the trained generation model to obtain synthetic data that is consistent with the feature distribution of the original archive data but not associated with the actual data; Based on the synthetic data, a neural network for patient feature analysis is trained to predict chronic diseases.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the chronic disease patient file data analysis method as described in any one of claims 1 to 6.

Images