CN118152898A - Electrocardiogram classification method and system based on deep learning and data privacy protection - Google Patents
Electrocardiogram classification method and system based on deep learning and data privacy protection Download PDFInfo
- Publication number
- CN118152898A CN118152898A CN202410579496.6A CN202410579496A CN118152898A CN 118152898 A CN118152898 A CN 118152898A CN 202410579496 A CN202410579496 A CN 202410579496A CN 118152898 A CN118152898 A CN 118152898A
- Authority
- CN
- China
- Prior art keywords
- client
- data
- parameters
- server
- model
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013135 deep learning Methods 0.000 title claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 61
- 238000012549 training Methods 0.000 claims abstract description 49
- 238000012360 testing method Methods 0.000 claims abstract description 34
- 230000002776 aggregation Effects 0.000 claims abstract description 24
- 238000004220 aggregation Methods 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000012795 verification Methods 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000001351 cycling effect Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims 1
- 230000006870 function Effects 0.000 claims 1
- 238000010606 normalization Methods 0.000 claims 1
- 230000005540 biological transmission Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Landscapes
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention relates to the technical field of electrocardiographic classification, in particular to an electrocardiographic classification method and system for protecting data privacy based on deep learning, wherein the method comprises the following steps: acquiring 12-lead electrocardiogram data and a label; constructing a convolutional neural network model, and processing the output characteristics by using a plurality of full-connection layers; the server side generates a pair of secret keys and distributes the public key and the private key to each client side; each client encrypts the training data set by using the public key; the server side and the client side end deploy convolutional neural network models; the client parameters are subjected to data aggregation through the federal learning server, and the decrypted local parameters are imported into convolutional neural network models corresponding to the clients; and (3) performing iterative training, namely storing parameters and models of the server as Model, and classifying the test set. The invention realizes information sharing among different source data, and improves the efficiency and accuracy of electrocardio classification efficiently on the basis of reducing data transmission.
Description
Technical Field
The invention relates to the technical field of electrocardiographic classification, in particular to an electrocardiographic classification method and system for protecting data privacy based on deep learning.
Background
When the traditional model is trained, all data need to be concentrated on one side, and when model parameters are learned, information used during training is easy to leak, so federal learning based on privacy protection becomes a research hotspot. Federal learning is a solution idea between parties holding data in multiple parties and effectively protecting local data privacy of each party. Based on a plurality of modes of client training and server aggregation, the FedAvg model proposed by Google utilizes intermediate information such as model parameters and the like to replace original data for transmission, so that the data privacy is protected to a certain extent. However, the intermediate information is usually extracted from information contained in the original data, and when the information is leaked, the privacy of the original data is equivalent to that of the original data, especially in the field of electrocardiosignal classification, when part of medical information of the subject is leaked, other key information of the subject can be deduced from the leaked part. Therefore, it has become a trend to protect data privacy in conjunction with encryption technology. Different from the existing method, the invention adopts the grid encryption and relaxation differential privacy algorithm as the data encryption technology, combines federal learning to protect the privacy disclosure problem of the electrocardio classification model during training, and realizes electrocardio high-performance classification by using a small amount of data of multiple parties.
Disclosure of Invention
Aiming at the defects of the prior art, the invention develops an electrocardio classification method and system for protecting data privacy based on deep learning.
The technical scheme for solving the technical problems is as follows:
in one aspect, the invention provides an electrocardiographic classification method for protecting data privacy based on deep learning, which comprises the following steps:
1) Acquiring lead electrocardiogram data and a label, and dividing the data set into a training set, a verification set and a test set according to the proportion;
2) Constructing a convolutional neural network model, and processing the characteristics output in the step 1) by using a plurality of full connection layers;
3) The server side generates a pair of keys, including a public key and a private key, the server side distributes the public key and the private key to each client side, each client side encrypts a training data set by using the public key, then a random value of addition or multiplication is set, and the value P needing encryption is subjected to random addition or random multiplication to obtain encrypted data;
4) The method comprises the steps that convolutional neural network models are arranged at a server side and a client side, encrypted data are uploaded to the server side and serve as initial parameters of the server, the initial parameters are distributed to all clients by the server, all clients decrypt the parameters by using private keys to obtain decrypted parameters, and then the decrypted parameters are imported into all client side models to obtain models initialized by all clients;
5) The client parameters are subjected to data aggregation through the federal learning server, and the decrypted local parameters are imported into convolutional neural network models corresponding to the clients;
6) Iterative training, namely saving parameters and models of a server as Model, and classifying test sets;
7) And inputting the test set on each client into a Model for Model evaluation and classification, and finally evaluating the classification precision jointly by the test results of the N clients.
In a specific embodiment, step 1) is specifically as follows:
1-1) carrying out standardization processing on the data, cutting each signal into segments according to 10s, wherein each segment comprises 3600 sample points, and one record is divided into 30 x 60/10=180 samples;
1-2) performing IID processing on the standardized data, and setting N clients, wherein each client holds 8 records, and total 8×180=1440 samples;
1-3) converting the label of the signal into One-hot form, and obtaining a data set after data preprocessing And a tag y in which c=5 categories,/>Respectively representing samples held by 6 clients;
1-4) training, verifying and dividing the test set according to the proportion of 8:1:1 for each client sample.
In a specific embodiment, step 2) is specifically as follows:
2-1) defining a convolutional neural network model comprising 4 stacked CNN layers, each layer following one batchNorm, relu layers, the first two CNN layers following one Maxpool layer, to achieve a reduction in data dimension;
2-2) respectively inputting the data processed in the step 1) into two full-connection layers with the unit number of 10 and the unit number of 5 to obtain the output of a final model, setting the core size of 4 CNN layers to be 3, setting the number of filters to be 32,32,64,64 respectively, and setting the core size of the maximum pooling layer to be 2.
In a specific embodiment, step 3) is specifically as follows:
3-1) generating a pair of keys including a public key pub and a private key prv by the server terminal based on the grid structure and the parameter characteristics;
3-2) the server distributes the public key and the private key to each client;
3-3) each client encrypts the training data by using the public key, and the obtained encrypted data is:
，
Wherein, Respectively represent the training data sets encrypted by each client after public key processing, namelyI represents the i-th customer service end, pub ()' represents the public key,/>Representing an i-th client encrypted training data set;
3-4) setting a random value of addition or multiplication, and carrying out random addition or random multiplication on the value P to be encrypted to obtain encrypted data P'.
In a specific embodiment, step 4) is specifically as follows:
4-1) the convolutional neural network model Net is respectively deployed at the server side and the client side, and the data after encryption of each client side after public key processing is carried out Inputting the parameters into the Net model corresponding to the client, and initializing the parameters to obtain the parameters/>;
4-2) Parameters of the reactionRandom addition or random multiplication is carried out to obtain encrypted data/>Data is processedUploading to a server side as an initial parameter/>, of the server；
4-3) The server will parametersDistributing the parameters to each client, and decrypting the parameters by each client by using a private key to obtain decrypted parameters/>Then, the parameters/> are imported into each client model NetAnd obtaining the model initialized by each client.
In a specific embodiment, step 5) is specifically as follows:
5-1) Gauss-based relaxed differential privacy algorithm if for any two neighboring data sets The method meets the following conditions:
，
Wherein, ,/>Is a constant used to represent differential privacy to/>Probability of being broken, pr [ ] represents probability, D and/>Is two adjacent data sets, only 1 data in the two data sets is different, and other data are identical; /(I)Representing a random algorithm,/>R represents real number domain, S represents arbitrary data,/>Representing a privacy budget;
5-2) for the convolutional neural network model corresponding to the ith client, using the client dataset Training the model to obtain the parameter/>；
5-3) Using the idea of differential privacy to enable the client to add disturbance to local parameters obtained by local training, and uploading a model after disturbance to a server, so that the aggregation process of each round meets the differential privacy, namely:
，
Wherein, Refers to the parameters obtained by the client i after differential privacy protection, and the parameters are compared with each otherIn terms of having a disturbing nature,/>Representing privacy budget,/>Representing privacy budget as/>To/>A differential privacy algorithm that breaks privacy;
5-4) server aggregation:
The server corresponds to N clients Aggregation is carried out to obtain updated parameters/>, at the server sideThe method comprises the following steps:
；
5-5) will Distributing to each client, executing the steps 5-4) and 5-3), and importing the decrypted local parameters into the convolutional neural network model corresponding to each client.
In a specific embodiment, step 6) is specifically as follows:
sequentially cycling the operations of step 5), learning characteristics of samples and preventing overfitting by a client random scheduling strategy, randomly selecting from N clients Aggregation of parameters involved in each round,/>The aggregate number representing the parameters involved in each round is calculated as follows:
，
I.e. starting from the second round, the strategy randomly selects N/2 parameters participating in each round from N of all clients at a time to obtain aggregated parameters ，/>The aggregate number representing the parameters involved in each round is calculated as follows:
，
And then performing iteration, checking the loss of the verification set tested at each client in sequence, stopping training the client after the loss of the verification set of a certain client is no longer reduced, ending training the Model after all clients stop training, and storing parameters and the Model of the server as models for classifying the test set.
In a specific embodiment, after a Model is obtained, a test set on each client is input to the Model for Model evaluation and classification, and finally, classification accuracy is evaluated by test results of N clients together, namely:
，
where N refers to the number of clients, Refers to the classification accuracy of the ith client on its corresponding test set.
In a specific embodiment, the convolutional neural network structure includes 4 stacked CNN layers, each layer is followed by one batchNorm, relu layers, the first two CNN layers are followed by one Maxpool layer, the kernel size of the CNN layers is 3, the number of filters is 32,32,64,64, and the kernel size of the maximum pooling layer is 2.
On the other hand, the invention also provides an electrocardio classification method based on data privacy protection of deep learning, which comprises a server side and a client side, and further comprises the following steps:
The data acquisition processing module is used for acquiring lead electrocardiogram data and labels and dividing a data set into a training set, a verification set and a test set according to the proportion;
The convolutional neural network model comprises 4 stacked CNN layers, each layer follows one batchNorm, relu layers, and the first two CNN layers follow one Maxpool layer;
The convolution neural network model is deployed on the server side and the client side, and the server side generates a pair of keys comprising a public key and a private key; each client encrypts the training data set by using the public key, sets a random value of addition or multiplication, and performs random addition or random multiplication on the value P needing encryption to obtain encrypted data;
The aggregation module is used for realizing data aggregation of the client parameters and importing the decrypted local parameters into the convolutional neural network model corresponding to each client;
And the calculation module is used for storing the parameters and the Model of the server as Model by using a preset iterative algorithm and classifying the test set.
The effects provided in the summary of the invention are merely effects of embodiments, not all effects of the invention, and the above technical solution has the following advantages or beneficial effects:
The invention provides an electrocardio classification method and system based on data privacy protection of deep learning based on federal learning thought of multi-client training and server aggregation, wherein a plurality of models consistent with convolutional neural networks are respectively deployed at a plurality of clients and a server, model weight intermediate information corresponding to each client is obtained after model training is carried out by utilizing local data of each client, an improved grid encryption algorithm is designed, and improved grid encryption and combined encryption of loose differential privacy are utilized, so that the accurate classification of electrocardio signals is realized under the federal learning framework, the privacy data of the weight intermediate information is efficiently protected, the addition of a plurality of clients is supported, the method has good expandability, and the problem that a data model cannot be trained due to island of medical data information is effectively solved.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
FIG. 1 is a schematic diagram of the federal learning process according to the present invention.
FIG. 2 is a diagram of a deep learning model architecture of the present invention.
Detailed Description
In order to clearly illustrate the technical features of the present solution, the present invention will be described in detail below with reference to the following detailed description and the accompanying drawings.
Example 1
An electrocardio classification method based on data privacy protection of deep learning comprises the following steps:
1) Acquiring lead electrocardiogram data and a label, and dividing the data set into a training set, a verification set and a test set according to the proportion;
2) Constructing a convolutional neural network model, and processing the characteristics output in the step 1) by using a plurality of full connection layers;
3) The server side generates a pair of keys, including a public key and a private key, the server side distributes the public key and the private key to each client side, each client side encrypts a training data set by using the public key, then a random value of addition or multiplication is set, and the value P needing encryption is subjected to random addition or random multiplication to obtain encrypted data;
4) The method comprises the steps that convolutional neural network models are arranged at a server side and a client side, encrypted data are uploaded to the server side and serve as initial parameters of the server, the initial parameters are distributed to all clients by the server, all clients decrypt the parameters by using private keys to obtain decrypted parameters, and then the decrypted parameters are imported into all client side models to obtain models initialized by all clients;
5) The client parameters are subjected to data aggregation through the federal learning server, and the decrypted local parameters are imported into convolutional neural network models corresponding to the clients;
6) Iterative training, namely saving parameters and models of a server as Model, and classifying test sets;
7) And inputting the test set on each client into a Model for Model evaluation and classification, and finally evaluating the classification precision jointly by the test results of the N clients.
In a specific embodiment, step 1) is specifically as follows:
1-1) carrying out standardization processing on the data, cutting each signal into segments according to 10s, wherein each segment comprises 3600 sample points, and one record is divided into 30 x 60/10=180 samples;
1-2) performing IID processing on the standardized data, and setting N clients, wherein each client holds 8 records, and total 8×180=1440 samples;
1-3) converting the label of the signal into One-hot form, and obtaining a data set after data preprocessing And a tag y in which c=5 categories,/>Respectively representing samples held by 6 clients;
1-4) training, verifying and dividing the test set according to the proportion of 8:1:1 for each client sample.
In a specific embodiment, step 2) is specifically as follows:
2-1) defining a convolutional neural network model comprising 4 stacked CNN layers, each layer following one batchNorm, relu layers, the first two CNN layers following one Maxpool layer, to achieve a reduction in data dimension;
2-2) respectively inputting the data processed in the step 1) into two full-connection layers with the unit number of 10 and the unit number of 5 to obtain the output of a final model, setting the core size of 4 CNN layers to be 3, setting the number of filters to be 32,32,64,64 respectively, and setting the core size of the maximum pooling layer to be 2.
In a specific embodiment, step 3) is specifically as follows:
3-1) generating a pair of keys including a public key pub and a private key prv by the server terminal based on the grid structure and the parameter characteristics;
3-2) the server distributes the public key and the private key to each client;
3-3) each client encrypts the training data by using the public key, and the obtained encrypted data is:
，
Wherein, Respectively represent the training data sets encrypted by each client after public key processing, namelyI represents the i-th customer service end, pub ()' represents the public key,/>Representing an i-th client encrypted training data set;
3-4) setting a random value of addition or multiplication, and carrying out random addition or random multiplication on the value P to be encrypted to obtain encrypted data P'.
In a specific embodiment, step 4) is specifically as follows:
4-1) the convolutional neural network model Net is respectively deployed at the server side and the client side, and the data after encryption of each client side after public key processing is carried out Inputting the parameters into the Net model corresponding to the client, and initializing the parameters to obtain the parameters/>;
4-2) Parameters of the reactionRandom addition or random multiplication is carried out to obtain encrypted data/>Data is processedUploading to a server side as an initial parameter/>, of the server；
4-3) The server will parametersDistributing the parameters to each client, and decrypting the parameters by each client by using a private key to obtain decrypted parameters/>Then, the parameters/> are imported into each client model NetAnd obtaining the model initialized by each client.
In a specific embodiment, step 5) is specifically as follows:
5-1) Gauss-based relaxed differential privacy algorithm if for any two neighboring data sets The method meets the following conditions:
，
Wherein, ,/>Is a constant used to represent differential privacy to/>Probability of being broken, pr [ ] represents probability, D and/>Is two adjacent data sets, only 1 data in the two data sets is different, and other data are identical; /(I)Representing a random algorithm,/>R represents real number domain, S represents arbitrary data,/>Representing a privacy budget;
5-2) for the convolutional neural network model corresponding to the ith client, using the client dataset Training the model to obtain the parameter/>；
5-3) Using the idea of differential privacy to enable the client to add disturbance to local parameters obtained by local training, and uploading a model after disturbance to a server, so that the aggregation process of each round meets the differential privacy, namely:
，
Wherein, Refers to the parameters obtained by the client i after differential privacy protection, and the parameters are compared with each otherIn terms of having a disturbing nature,/>Representing privacy budget,/>Representing privacy budget as/>To/>A differential privacy algorithm that breaks privacy;
5-4) server aggregation:
The server corresponds to N clients Aggregation is carried out to obtain updated parameters/>, at the server sideThe method comprises the following steps:
；
5-5) will Distributing to each client, executing the steps 5-4) and 5-3), and importing the decrypted local parameters into the convolutional neural network model corresponding to each client.
In a specific embodiment, step 6) is specifically as follows:
sequentially cycling the operations of step 5), learning characteristics of samples and preventing overfitting by a client random scheduling strategy, randomly selecting from N clients Aggregation of parameters involved in each round,/>The aggregate number representing the parameters involved in each round is calculated as follows:
，
I.e. starting from the second round, the strategy randomly selects N/2 parameters participating in each round from N of all clients at a time to obtain aggregated parameters ，/>The aggregate number representing the parameters involved in each round is calculated as follows:
，
And then performing iteration, checking the loss of the verification set tested at each client in sequence, stopping training the client after the loss of the verification set of a certain client is no longer reduced, ending training the Model after all clients stop training, and storing parameters and the Model of the server as models for classifying the test set.
In a specific embodiment, after a Model is obtained, a test set on each client is input to the Model for Model evaluation and classification, and finally, classification accuracy is evaluated by test results of N clients together, namely:
，
where N refers to the number of clients, Refers to the classification accuracy of the ith client on its corresponding test set.
In a specific embodiment, the convolutional neural network structure includes 4 stacked CNN layers, each layer is followed by one batchNorm, relu layers, the first two CNN layers are followed by one Maxpool layer, the kernel size of the CNN layers is 3, the number of filters is 32,32,64,64, and the kernel size of the maximum pooling layer is 2.
The invention provides a specific embodiment:
As can be seen from table 1, introducing privacy protection into the model may be at the cost of a certain degree of precision reduction, for example, the federal learning model of homomorphic encryption and differential privacy is lower in classification precision than the CNN model without privacy protection, but the data encryption technology based on the combination of grid encryption and relaxation differential privacy provided by the invention can achieve a result equivalent to or even better than the CNN model without privacy protection while achieving privacy protection. This further illustrates the advantages of the proposed method of the invention.
Table 1 classification results obtained after application of the embodiment of the present invention to MITDB datasets
。
Example 2
An electrocardiograph system based on data privacy protection of deep learning performs an electrocardiograph classification method based on data privacy protection of deep learning, and has a server side and a client side, and further includes:
The data acquisition processing module is used for acquiring lead electrocardiogram data and labels and dividing a data set into a training set, a verification set and a test set according to the proportion;
The convolutional neural network model comprises 4 stacked CNN layers, each layer follows one batchNorm, relu layers, and the first two CNN layers follow one Maxpool layer;
The convolution neural network model is deployed on the server side and the client side, and the server side generates a pair of keys comprising a public key and a private key; each client encrypts the training data set by using the public key, sets a random value of addition or multiplication, and performs random addition or random multiplication on the value P needing encryption to obtain encrypted data;
The aggregation module is used for realizing data aggregation of the client parameters and importing the decrypted local parameters into the convolutional neural network model corresponding to each client;
And the calculation module is used for storing the parameters and the Model of the server as Model by using a preset iterative algorithm and classifying the test set.
While the foregoing description of the embodiments of the present invention has been presented with reference to the drawings, it is not intended to limit the scope of the invention, but rather, it is apparent that various modifications or variations can be made by those skilled in the art without the need for inventive work on the basis of the technical solutions of the present invention.
Claims (10)
1. An electrocardio classification method based on data privacy protection of deep learning is characterized by comprising the following steps:
1) Acquiring lead electrocardiogram data and a label, and dividing the data set into a training set, a verification set and a test set according to the proportion;
2) Constructing a convolutional neural network model, and processing the characteristics output in the step 1) by using a plurality of full connection layers;
3) The server side generates a pair of keys, including a public key and a private key, the server side distributes the public key and the private key to each client side, each client side encrypts a training data set by using the public key, then a random value of addition or multiplication is set, and the value P needing encryption is subjected to random addition or random multiplication to obtain encrypted data;
4) The method comprises the steps that convolutional neural network models are arranged at a server side and a client side, encrypted data are uploaded to the server side and serve as initial parameters of the server, the initial parameters are distributed to all clients by the server, all clients decrypt the parameters by using private keys to obtain decrypted parameters, and then the decrypted parameters are imported into all client side models to obtain models initialized by all clients;
5) The client parameters are subjected to data aggregation through the federal learning server, and the decrypted local parameters are imported into convolutional neural network models corresponding to the clients;
6) Iterative training, namely saving parameters and models of a server as Model, and classifying test sets;
7) And inputting the test set on each client into a Model for Model evaluation and classification, and finally evaluating the classification precision jointly by the test results of the N clients.
2. The method for classifying electrocardiographs based on deep learning data privacy protection according to claim 1, wherein the step 1) is specifically as follows:
1-1) carrying out standardization processing on the data, cutting each signal into segments according to 10s, wherein each segment comprises 3600 sample points, and one record is divided into 30 x 60/10=180 samples;
1-2) performing IID processing on the standardized data, and setting N clients, wherein each client holds 8 records, and total 8×180=1440 samples;
1-3) converting the label of the signal into One-hot form, and obtaining a data set after data preprocessing And a tag y in which c=5 categories,/>Respectively representing samples held by 6 clients;
1-4) training, verifying and dividing the test set according to the proportion of 8:1:1 for each client sample.
3. The method for classifying electrocardiographs based on deep learning data privacy protection according to claim 2, wherein the step 2) is specifically as follows:
2-1) defining a convolutional neural network model, which comprises 4 stacked convolutional neural network CNN layers, wherein each layer follows a normalization BatchNorm layer and an activation function Relu layer, and the first two CNN layers follow a maximum pooling Maxpool layer so as to realize the reduction of data dimension;
2-2) inputting the data processed in the step 1) into two full connection layers FC with the unit number of 10 and the unit number of 5 respectively, obtaining the output of a final model, setting the core size of 4 CNN layers to be 3, setting the number of filters to be 32,32,64,64 respectively, and setting the core size of the maximum pooling layer to be 2.
4. The method for classifying electrocardiographic signals based on deep learning data privacy protection according to claim 3, wherein the step 3) is specifically as follows:
3-1) generating a pair of keys including a public key pub and a private key prv by the server terminal based on the grid structure and the parameter characteristics;
3-2) the server distributes the public key and the private key to each client;
3-3) each client encrypts the training data by using the public key, and the obtained encrypted data is:
，
Wherein, Respectively represent the training data sets encrypted by each client after public key processing, namelyI represents the i-th customer service end, pub ()' represents the public key,/>Representing an i-th client encrypted training data set;
3-4) setting a random value of addition or multiplication, and carrying out random addition or random multiplication on the value P to be encrypted to obtain encrypted data P'.
5. The method for classifying electrocardiographic signals based on deep learning data privacy protection of claim 4, wherein the step 4) is specifically as follows:
4-1) the convolutional neural network model Net is respectively deployed at the server side and the client side, and the data after encryption of each client side after public key processing is carried out Inputting the parameters into the Net model corresponding to the client, and initializing the parameters to obtain the parameters/>;
4-2) Parameters of the reactionRandom addition or random multiplication is carried out to obtain encrypted data/>Data/>Uploading to a server side as an initial parameter/>, of the server；
4-3) The server will parametersDistributing the parameters to each client, and decrypting the parameters by each client by using a private key to obtain decrypted parameters/>Then, the parameters/> are imported into each client model NetAnd obtaining the model initialized by each client.
6. The method for classifying electrocardiographic signals based on deep learning data privacy protection of claim 5, wherein the step 5) is specifically as follows:
5-1) Gauss-based relaxed differential privacy algorithm if for any two neighboring data sets The method meets the following conditions:
，
Wherein, ,/>Is a constant used to represent differential privacy to/>Probability of being broken, pr [ ] represents probability, D and/>Is two adjacent data sets, only 1 data in the two data sets is different, and other data are identical; /(I)Representing a random algorithm,/>R represents real number domain, S represents arbitrary data,/>Representing a privacy budget;
5-2) for the convolutional neural network model corresponding to the ith client, using the client dataset Training the model to obtain the parameter/>；
5-3) Using the idea of differential privacy to enable the client to add disturbance to local parameters obtained by local training, and uploading a model after disturbance to a server, so that the aggregation process of each round meets the differential privacy, namely:
，
Wherein, Refers to the parameters obtained by the client i after differential privacy protection, and the parameters are compared with each other/>In terms of having a disturbing nature,/>Representing privacy budget,/>Representing privacy budget as/>To/>A differential privacy algorithm that breaks privacy;
5-4) server aggregation:
The server corresponds to N clients Aggregation is carried out to obtain updated parameters/>, at the server sideThe method comprises the following steps:
；
5-5) will Distributing to each client, executing the steps 5-4) and 5-3), and importing the decrypted local parameters into the convolutional neural network model corresponding to each client.
7. The method for classifying electrocardiographic signals based on deep learning data privacy protection of claim 6, wherein the step 6) is specifically as follows:
sequentially cycling the operations of step 5), learning characteristics of samples and preventing overfitting by a client random scheduling strategy, randomly selecting from N clients Aggregation of parameters involved in each round,/>The aggregate number representing the parameters involved in each round is calculated as follows:
，
I.e. starting from the second round, the strategy randomly selects N/2 parameters participating in each round from N of all clients at a time to obtain aggregated parameters ，/>The aggregate number representing the parameters involved in each round is calculated as follows:
，
And then performing iteration, checking the loss of the verification set tested at each client in sequence, stopping training the client after the loss of the verification set of a certain client is no longer reduced, ending training the Model after all clients stop training, and storing parameters and the Model of the server as models for classifying the test set.
8. The method for classifying electrocardiographs based on deep learning data privacy protection according to claim 7, wherein the method is characterized in that:
After the Model is obtained, the test set on each client is input to the Model for Model evaluation and classification, and the final classification accuracy is evaluated by the test results of N clients together, namely:
，
where N refers to the number of clients, Refers to the classification accuracy of the ith client on its corresponding test set.
9. The method for electrocardiographic classification based on deep learning data privacy protection of claim 8 wherein the convolutional neural network structure comprises 4 stacked CNN layers, each layer is followed by one batchNorm and one Relu layer, the first two CNN layers are followed by one Maxpool layer, the kernel size of the CNN layers is 3, the number of filters is 32,32,64,64, and the kernel size of the largest pooling layer is 2.
10. An electrocardiographic classification system based on deep learning data privacy protection, characterized in that an electrocardiographic classification method based on deep learning data privacy protection as claimed in any one of claims 1-9 is performed, and the electrocardiographic classification system has a server side and a client side, and further comprises:
The data acquisition processing module is used for acquiring lead electrocardiogram data and labels and dividing a data set into a training set, a verification set and a test set according to the proportion;
The convolutional neural network model comprises 4 stacked CNN layers, each layer follows one batchNorm, relu layers, and the first two CNN layers follow one Maxpool layer;
The convolution neural network model is deployed on the server side and the client side, and the server side generates a pair of keys comprising a public key and a private key; each client encrypts the training data set by using the public key, sets a random value of addition or multiplication, and performs random addition or random multiplication on the value P needing encryption to obtain encrypted data;
The aggregation module is used for realizing data aggregation of the client parameters and importing the decrypted local parameters into the convolutional neural network model corresponding to each client;
And the calculation module is used for storing the parameters and the Model of the server as Model by using a preset iterative algorithm and classifying the test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410579496.6A CN118152898B (en) | 2024-05-11 | Electrocardiogram classification method and system based on deep learning and data privacy protection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410579496.6A CN118152898B (en) | 2024-05-11 | Electrocardiogram classification method and system based on deep learning and data privacy protection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118152898A true CN118152898A (en) | 2024-06-07 |
| CN118152898B CN118152898B (en) | 2024-07-26 |
Family
ID=
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190147188A1 (en) * | 2017-11-16 | 2019-05-16 | Microsoft Technology Licensing, Llc | Hardware protection for differential privacy |
| CN110572253A (en) * | 2019-09-16 | 2019-12-13 | 济南大学 | Method and system for enhancing privacy of federated learning training data |
| WO2020233260A1 (en) * | 2019-07-12 | 2020-11-26 | 之江实验室 | Homomorphic encryption-based privacy-protecting multi-institution data classification method |
| US20210019428A1 (en) * | 2019-07-19 | 2021-01-21 | Fuzhou University | Preservation system for preserving privacy of outsourced data in cloud based on deep convolutional neural network |
| CN112949741A (en) * | 2021-03-18 | 2021-06-11 | 西安电子科技大学 | Convolutional neural network image classification method based on homomorphic encryption |
| CN113850272A (en) * | 2021-09-10 | 2021-12-28 | 西安电子科技大学 | Local differential privacy-based federal learning image classification method |
| CN115169413A (en) * | 2022-07-21 | 2022-10-11 | 浙江工业大学 | Electrocardiosignal accurate classification method based on improved convolutional neural network |
| CN115549888A (en) * | 2022-09-29 | 2022-12-30 | 南京邮电大学 | Block chain and homomorphic encryption-based federated learning privacy protection method |
| CN117421762A (en) * | 2023-08-21 | 2024-01-19 | 北京理工大学 | Federal learning privacy protection method based on differential privacy and homomorphic encryption |
| CN117640253A (en) * | 2024-01-25 | 2024-03-01 | 济南大学 | Federal learning privacy protection method and system based on homomorphic encryption |
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190147188A1 (en) * | 2017-11-16 | 2019-05-16 | Microsoft Technology Licensing, Llc | Hardware protection for differential privacy |
| WO2020233260A1 (en) * | 2019-07-12 | 2020-11-26 | 之江实验室 | Homomorphic encryption-based privacy-protecting multi-institution data classification method |
| US20210019428A1 (en) * | 2019-07-19 | 2021-01-21 | Fuzhou University | Preservation system for preserving privacy of outsourced data in cloud based on deep convolutional neural network |
| CN110572253A (en) * | 2019-09-16 | 2019-12-13 | 济南大学 | Method and system for enhancing privacy of federated learning training data |
| CN112949741A (en) * | 2021-03-18 | 2021-06-11 | 西安电子科技大学 | Convolutional neural network image classification method based on homomorphic encryption |
| CN113850272A (en) * | 2021-09-10 | 2021-12-28 | 西安电子科技大学 | Local differential privacy-based federal learning image classification method |
| CN115169413A (en) * | 2022-07-21 | 2022-10-11 | 浙江工业大学 | Electrocardiosignal accurate classification method based on improved convolutional neural network |
| CN115549888A (en) * | 2022-09-29 | 2022-12-30 | 南京邮电大学 | Block chain and homomorphic encryption-based federated learning privacy protection method |
| CN117421762A (en) * | 2023-08-21 | 2024-01-19 | 北京理工大学 | Federal learning privacy protection method based on differential privacy and homomorphic encryption |
| CN117640253A (en) * | 2024-01-25 | 2024-03-01 | 济南大学 | Federal learning privacy protection method and system based on homomorphic encryption |
Non-Patent Citations (2)
| Title |
|---|
| 毛典辉;李子沁;蔡强;薛子育;: "基于DCGAN反馈的深度差分隐私保护方法", 北京工业大学学报, no. 06, 24 April 2018 (2018-04-24) * |
| 芈小龙;隋景鹏;: "面向深度学习的差分隐私保护方法", 舰船电子工程, no. 09, 20 September 2020 (2020-09-20) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110288094B (en) | Model parameter training method and device based on federal learning | |
| Dridi et al. | Cryptography of medical images based on a combination between chaotic and neural network | |
| Wu et al. | Local Shannon entropy measure with statistical tests for image randomness | |
| CN103270516B (en) | System and method for securing virtual machine computing environments | |
| CN107770405A (en) | Image encryption method and device | |
| CN111428887B (en) | Model training control method, device and system based on multiple computing nodes | |
| CN103258312B (en) | There is the digital image encryption method of fast key stream generting machanism | |
| CN113434898B (en) | Non-interactive privacy protection logistic regression federal training method and system | |
| CN113362048A (en) | Data label distribution determining method and device, computer equipment and storage medium | |
| CN115392487A (en) | Privacy protection nonlinear federal support vector machine training method and system based on homomorphic encryption | |
| CN116738444B (en) | Xia Puli value-based multi-party contribution degree evaluation method for data security sharing platform | |
| Wang et al. | Heterogeneous defect prediction based on federated transfer learning via knowledge distillation | |
| WO2020144118A1 (en) | Method for creating avatars for protecting sensitive data | |
| CN117390657A (en) | Data encryption method, device, computer equipment and storage medium | |
| CN114912631A (en) | Block chain medical data sharing-based federal learning method | |
| US20230161899A1 (en) | Data processing for release while protecting individual privacy | |
| CN111177746A (en) | Efficient visual secret sharing method with core participants | |
| CN118152898B (en) | Electrocardiogram classification method and system based on deep learning and data privacy protection | |
| CN118152898A (en) | Electrocardiogram classification method and system based on deep learning and data privacy protection | |
| CN105095774B (en) | Data ciphering method and system | |
| CN107133994B (en) | Image encryption method | |
| US11537742B2 (en) | Sampling from a remote dataset with a private criterion | |
| CN107145974B (en) | Method for implementing negative investigation and reconstructing positive data | |
| CN114022708B (en) | Encryption image classification method and device based on convolutional neural network | |
| Duan | Digital marketing solutions based on consumer data and homomorphic encryption |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |