Federal learning-based medical data privacy protection analysis method

Abstract

The invention relates to the technical field of data processing, in particular to a medical data privacy protection analysis method based on federal learning. Comprising the following steps: taking a medical institution as a medical node, and obtaining and preprocessing a local data set to obtain a preprocessed local data set; training a local model based on the preprocessed local data set, performing gradient updating, and introducing an adaptive dynamic noise encryption mechanism to obtain an encrypted local gradient; and uploading the encrypted local gradient to a central server, carrying out decryption processing to obtain a decrypted gradient, carrying out weighted aggregation operation, updating the global model to obtain parameters of the global model after gradient updating, and introducing a nonlinear adjustment mechanism to adjust the parameters of the global model. The technical problems that cross-mechanism data collaboration and sharing cannot be effectively and safely carried out on the premise of protecting patient privacy, and the training efficiency and precision of a global model are low while medical data privacy protection is ensured are solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a medical data privacy protection analysis method based on federated learning. Background Art

[0002] With the increasing digitization and informatization of medical data, healthcare institutions are increasingly relying on big data analytics to improve patient outcomes, optimize clinical decision-making, and advance medical research. However, medical data contains a large amount of sensitive personal health information, and traditional approaches to processing medical data often rely on centralized data storage and analysis. While this approach can fully utilize the powerful computing power of central servers, it also carries the risk of privacy breaches.

[0003] Federated learning, a distributed machine learning method, enables collaborative training of global models among data holders (such as different medical institutions) without sharing original data. Each medical institution trains a model based on local data and uploads the updated model parameters to a central server for aggregation. This approach effectively avoids the privacy risks associated with centralized data storage. However, while federated learning addresses data privacy concerns, it still faces challenges in terms of model training efficiency, aggregation quality, and privacy protection due to differences in data volume, data distribution, and network environments across nodes.

[0004] At the same time, the above-mentioned existing technologies still have technical problems such as the inability to effectively and securely conduct cross-institutional data collaboration and sharing while protecting patient privacy, and the low training efficiency and accuracy of the global model while ensuring the privacy protection of medical data. Summary of the Invention

[0005] The present invention provides a medical data privacy protection analysis method based on federated learning to solve the technical problems of being unable to effectively and securely carry out cross-institutional data collaboration and sharing while protecting patient privacy, and the low training efficiency and accuracy of the global model while ensuring the privacy protection of medical data.

[0006] The present invention provides a medical data privacy protection analysis method based on federated learning, which specifically includes the following technical solutions: A medical data privacy protection analysis method based on federated learning includes the following steps: S1. Consider medical institutions as medical nodes and deploy local federated learning clients in each of them. Local datasets are acquired and preprocessed to obtain preprocessed local datasets. Local models are trained based on the preprocessed local datasets and gradients are updated. An adaptive dynamic noise encryption mechanism is introduced to obtain encrypted local gradients. S2. Upload the encrypted local gradient to the central server for decryption to obtain the decrypted gradient, perform weighted aggregation operations, update the global model, obtain the parameters of the global model after the gradient update, and introduce a nonlinear adjustment mechanism to adjust the global model parameters.

[0007] Preferably, the S1 specifically includes: At each medical node, a local model is trained based on its preprocessed local data set. The training goal of each medical node is to minimize the local loss function and perform gradient updates on the local loss function to obtain the gradient of the loss function calculated by the medical node under the current local model parameters with respect to the local model parameters.

[0008] Preferably, the S1 specifically includes: After local model training, each medical node will encrypt the calculated gradient and introduce an adaptive dynamic noise encryption mechanism to dynamically adjust the noise intensity based on the data volume and data distribution characteristics of each medical node and the current state of the global model.

[0009] Preferably, the S1 specifically includes: In the implementation process of the adaptive dynamic noise encryption mechanism, an adjustment factor and an adaptive adjustment coefficient are introduced to adjust the noise standard deviation and control the relationship between the noise standard deviation and the global sample data volume, respectively. The ratio of the sample data volume in the preprocessed local data set of each medical node to the total number of samples in the preprocessed local data sets of all medical nodes is combined with the sum of the variances of the local model loss functions of all medical nodes to obtain the noise standard deviation used to encrypt the gradient on the medical node.

[0010] Preferably, the S1 specifically includes: In the implementation process of the adaptive dynamic noise encryption mechanism, by introducing adaptive dynamic noise, the noise size of each medical node is automatically adjusted according to its data volume and data distribution characteristics, and the gradient of each medical node after noisy is obtained.

[0011] Preferably, the S1 specifically includes: The noisy gradient is symmetrically encrypted to obtain the encrypted gradient, and then the encrypted local gradient is finally obtained.

[0012] Preferably, the S2 specifically includes: On the central server, the encrypted local gradients of each medical node are decrypted using a private key to obtain the decrypted gradients. The decrypted gradients are then weighted aggregated on the central server to update the global model and obtain the parameters of the global model after the gradient update.

[0013] Preferably, the S2 specifically includes: Based on the parameters of the global model after gradient update, a nonlinear adjustment mechanism is introduced. The nonlinear adjustment mechanism dynamically adjusts the update step size according to the current error of the global model, optimizes the global model, and obtains the global model parameters after nonlinear adjustment.

[0014] Preferably, the S2 specifically includes: After completing the update and nonlinear adjustment of the global model parameters, each medical node will continue training according to the new global model parameters until the preset stop conditions are met; and blockchain records and transparent verification are introduced to ensure that the global model update process is transparent and cannot be tampered with.

[0015] The beneficial effects of the technical solution of the present invention are: 1. By introducing an adaptive dynamic noise encryption mechanism, the noise intensity is dynamically adjusted according to the data volume and data distribution characteristics of each medical node and the current state of the global model. This approach makes privacy protection more flexible and efficient, further enhancing the ability to protect data privacy.

[0016] 2. By introducing a nonlinear adjustment mechanism, the global model update step size is dynamically adjusted according to the global model error. This mechanism controls the step size when the error is large, allowing the global model to adjust more smoothly during the convergence process, avoiding over-updates or under-updates. This dynamic adjustment mechanism effectively balances the model's training speed and convergence effect, avoiding the overfitting or underfitting problems that may occur in traditional fixed-step update methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the medical data privacy protection analysis method based on federated learning described in the present invention. DETAILED DESCRIPTION

[0018] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0020] The following describes in detail a specific scheme of a medical data privacy protection analysis method based on federated learning provided by the present invention with reference to the accompanying drawings.

[0021] Refer to the attached Figure 1 , which shows a flow chart of a medical data privacy protection analysis method based on federated learning provided by one embodiment of the present invention, the method comprising the following steps: S1. Consider medical institutions as medical nodes and deploy local federated learning clients in each of them. Local datasets are acquired and preprocessed to obtain preprocessed local datasets. Local models are trained based on the preprocessed local datasets and gradients are updated. An adaptive dynamic noise encryption mechanism is introduced to obtain encrypted local gradients. Medical institutions are used as medical nodes, and local clients for federated learning are deployed separately. At each medical node, medical data of patients are obtained through local clients, including basic information (such as age, gender, and medical history), clinical data (such as diagnostic reports and test results), and other information related to health status, as local datasets. The local datasets are preprocessed, such as handling missing values, data standardization, feature engineering, and normalization, to obtain the preprocessed local datasets, which are expressed as ,in, Indicates the The pre-processed local dataset of each medical node, express Middle The input features of samples, yes Middle The true label or target output of each sample is usually relevant disease diagnosis, treatment effectiveness, or expected health outcomes, It is Preprocessed local dataset of medical nodes The number of samples in The number of patient data participating in the training on each medical node; the preprocessing process is a technical means well known to those skilled in the art and will not be described in detail here.

[0022] Furthermore, a local model is trained at each medical node based on its pre-processed local data set to extract disease prediction or other health-related information. The local model is a model trained by each medical node (i.e., medical institution) based on its pre-processed local data set. Its model architecture is determined according to the specific needs of each medical institution, such as classification model, regression model, deep learning model, which will not be elaborated here; the goal of training each medical node is to minimize the local loss function , and perform gradient updates on the local loss function. Based on the standard gradient descent algorithm and back propagation algorithm, the gradient update calculation formula is:

[0023] in, It is Medical nodes in the current local model parameters The loss function calculated below For local model parameters The gradient of is used to guide how the local model updates its parameters in each round of training to reduce the value of the loss function and thus improve the accuracy of the local model; It is a local model In the Input features of samples Local model parameters The gradient of the current local model represents the Input features of samples The sensitivity of is calculated by the existing back propagation algorithm; is the loss function For the first The true labels of the samples and the model predicted labels The gradient of , which reflects the contribution of the error between the true label and the model's predicted label to the loss function, such as calculated based on standard loss functions such as mean squared error (MSE) or cross-entropy loss; The local model is Input features of samples The prediction output, that is, the model prediction label, is obtained through the local model The output obtained depends on the specific model type; Represents the local model for the Input features of samples The predicted output.

[0024] Furthermore, after local model training, each medical node encrypts the calculated gradient to prevent data leakage during subsequent transmission. Unlike traditional differential privacy technologies, this method introduces an adaptive dynamic noise encryption mechanism. Specifically, the noise intensity is dynamically adjusted based on the data volume and data distribution characteristics of each medical node, as well as the current state of the global model. The standard deviation of the noise is The calculation formula is as follows:

[0025] in, It is in The standard deviation of the noise used to encrypt the gradient on each medical node; It is an adjustment factor used to adjust the size of the noise standard deviation. It is selected according to the privacy protection requirements in actual applications. The reference value range is ,The specific value depends on the privacy protection requirements and computing resource limitations; It is the total number of samples in the preprocessed local datasets of all medical nodes. It represents the sum of the sample data of all medical nodes in the entire federated learning process and plays a normalization role, making the relationship between the noise standard deviation of each medical node and the global sample data volume more reasonable. It is an adaptive adjustment coefficient used to control the relationship between the noise standard deviation and the global sample data volume. It is determined according to expert experience and the reference value range is ; Indicates the total number of medical nodes; It is The variance of the local model loss function corresponding to each medical node represents the distribution of the training error of the local model on the preprocessed local dataset; Describes the ratio of the sample data volume in the preprocessed local dataset of each medical node to the total number of samples in the preprocessed local datasets of all medical nodes, reflecting the data distribution characteristics of each medical node, that is, the ratio of the sample data volume in the preprocessed local dataset of each medical node to the global sample data volume; The sum of the variances of the local model loss functions of all medical nodes is calculated, which reflects the current state of the global model. By reflecting the changes in the local model loss function of each medical node, it helps to dynamically adjust the noise intensity of each medical node, thereby improving the efficiency of privacy protection.

[0026] By introducing adaptive dynamic noise, the noise level of each medical node can be automatically adjusted based on the amount of data and data distribution characteristics. For example, medical nodes with large amounts of data will have lower noise levels to avoid affecting the efficiency of local model training due to excessive noise, while medical nodes with smaller amounts of data will have higher noise levels to enhance privacy protection.

[0027] The gradient after adding noise to each medical node Calculated by the following formula:

[0028] in, The mean is 0 and the variance is Gaussian noise, is the identity matrix.

[0029] In this way, the medical node not only completes the gradient calculation, but also adds sufficient noise to the gradient to obtain the noisy gradient.

[0030] Furthermore, the noisy gradient is further protected by existing symmetric encryption (such as AES) to obtain the encrypted gradient, thus avoiding any information leakage during the local model data transmission process. Finally, the encrypted local gradient is obtained. .

[0031] S2. Upload the encrypted local gradient to the central server for decryption to obtain the decrypted gradient, perform weighted aggregation operations, update the global model, obtain the parameters of the global model after the gradient update, and introduce a nonlinear adjustment mechanism to adjust the global model parameters.

[0032] The encrypted local gradients of each medical node are uploaded to the central server, which is a server that plays a coordination and aggregation role in the federated learning framework. Its main responsibility is to manage and coordinate the operation of the entire federated learning architecture, receive local model updates from each medical node, and perform global aggregation to update the global model. Specifically, on the central server, the encrypted local gradients of each medical node are decrypted using a private key to obtain the decrypted gradients. , and perform weighted aggregation on the decrypted gradients in the central server, and further, update the global model through the existing gradient descent algorithm. The purpose of weighted aggregation is to calculate the number of samples in the pre-processed local dataset of each medical node. To determine its influence in the global model update, medical nodes with large data volumes should have a greater impact on the global model. After the central server performs weighted aggregation on the decrypted gradients of each medical node, it uses the existing gradient descent algorithm to obtain the update formula for the global model:

[0033] in, are the parameters of the current global model, i.e., the model parameters after the gradient update obtained by weighted aggregation; are the global model parameters of the previous round, that is, the parameters used by the global model before this round of update; Is the learning rate, which is used to control the step size of global model parameter update and determines the amplitude of each gradient descent. Too large a value will lead to excessive updates, while too small a value may lead to slow convergence. It is determined based on expert experience and the specific global model type. The reference value range is ; It is The weight of each medical node in the global model update process is calculated based on the ratio of the data volume of any medical node to the total data volume of all medical nodes. ; Furthermore, to avoid over- or under-updates, a nonlinear adjustment mechanism is introduced that dynamically adjusts the update step size based on the current error of the global model. If the global model error is large, a larger step size is allowed; otherwise, a smaller step size is allowed, thereby smoothly optimizing the global model.

[0034] The specific nonlinear adjustment formula is:

[0035] in, are the global model parameters after nonlinear adjustment; It is the adjustment coefficient, which is used to control the amplitude of nonlinear adjustment and determines the influence of error size on step size adjustment. It can smooth the update process of the global model and avoid over-adjustment of the global model. The reference value range is determined according to the expert experience method. ; It is the hyperbolic tangent function, a common activation function with a smooth S-shaped curve. Its function is to suppress extreme values ​​when the global model is updated to ensure that the adjustment amplitude is not too large; is the gradient of the global model, It is the loss function of the global model, which is obtained by weighted combination of the local model loss functions of all medical nodes; is the L2 norm; It is the L2 norm of the global model gradient, which indicates the size of the current gradient of the global model and measures the size of the training error of the global model. When the gradient norm of the global model is large, it means that the current local model error is large. Is the error threshold, which determines when to perform nonlinear adjustment. When the L2 norm of the global model gradient exceeds the error threshold, the nonlinear adjustment will increase the step size, thereby speeding up the update of the global model. Otherwise, the step size will be relatively small to avoid over-adjustment. It is determined according to the expert experience method. The reference material range is ; The ratio between the L2 norm of the global model gradient and the preset error threshold measures the size of the current training error of the global model relative to the error threshold; is a weighted term used to adjust the amplitude of nonlinear updates; By using hyperbolic tangent mapping, the influence range is limited to In this way, the step size is updated smoothly to avoid excessive gradient updates. This can prevent the update step size from being too high when the error is too large, which can lead to unstable training. Represents the final adjustment factor, which controls the dynamic adjustment of the global model update step size.

[0036] After completing the update and nonlinear adjustment of the global model parameters, each medical node will continue to train according to the new global model parameters until the stopping conditions preset according to the expert experience method are met.

[0037] Furthermore, blockchain recording and transparent verification are introduced to ensure that the global model update process is transparent and tamper-proof. After each aggregation, the central server records the global model's hash value, the IDs of participating medical nodes, version information, and update timestamps on the blockchain, ensuring the immutability and traceability of the update process.

[0038] Through blockchain technology, all global model update operations will be recorded and verified, increasing the transparency and trust of the global model training process.

[0039] In summary, a medical data privacy protection analysis method based on federated learning has been completed.

[0040] The order in which the embodiments of the invention are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0041] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0042] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A medical data privacy protection analysis method based on federated learning, characterized by: The following steps are involved: S1. Consider medical institutions as medical nodes and deploy local federated learning clients in each of them. Local datasets are acquired and preprocessed to obtain preprocessed local datasets. Local models are trained based on the preprocessed local datasets and gradients are updated. An adaptive dynamic noise encryption mechanism is introduced to obtain encrypted local gradients. S2. Upload the encrypted local gradient to the central server for decryption to obtain the decrypted gradient, perform weighted aggregation operations, update the global model, obtain the parameters of the global model after the gradient update, and introduce a nonlinear adjustment mechanism to adjust the global model parameters.

2. A medical data privacy protection analysis method based on federated learning according to claim 1, characterized in that: Said S1 specifically includes: At each medical node, a local model is trained based on its preprocessed local data set. The training goal of each medical node is to minimize the local loss function and perform gradient updates on the local loss function to obtain the gradient of the loss function calculated by the medical node under the current local model parameters with respect to the local model parameters.

3. A medical data privacy protection analysis method based on federated learning according to claim 2, characterized in that: Said S1 specifically includes: After local model training, each medical node will encrypt the calculated gradient and introduce an adaptive dynamic noise encryption mechanism to dynamically adjust the noise intensity based on the data volume and data distribution characteristics of each medical node and the current state of the global model.

4. A medical data privacy protection analysis method based on federated learning according to claim 3, characterized in that: Said S1 specifically includes: In the implementation process of the adaptive dynamic noise encryption mechanism, an adjustment factor and an adaptive adjustment coefficient are introduced to adjust the noise standard deviation and control the relationship between the noise standard deviation and the global sample data volume, respectively. The ratio of the sample data volume in the preprocessed local data set of each medical node to the total number of samples in the preprocessed local data sets of all medical nodes is combined with the sum of the variances of the local model loss functions of all medical nodes to obtain the noise standard deviation used to encrypt the gradient on the medical node.

5. A medical data privacy protection analysis method based on federated learning according to claim 4, characterized in that: Said S1 specifically includes: In the implementation process of the adaptive dynamic noise encryption mechanism, by introducing adaptive dynamic noise, the noise size of each medical node is automatically adjusted according to its data volume and data distribution characteristics, and the gradient of each medical node after noisy is obtained.

6. A medical data privacy protection analysis method based on federated learning according to claim 5, characterized in that: Said S1 specifically includes: The noisy gradient is symmetrically encrypted to obtain the encrypted gradient, and then the encrypted local gradient is finally obtained.

7. The medical data privacy protection analysis method based on federated learning according to claim 1 is characterized in that: Said S2 specifically includes: On the central server, the encrypted local gradients of each medical node are decrypted using a private key to obtain the decrypted gradients. The decrypted gradients are then weighted aggregated on the central server to update the global model and obtain the parameters of the global model after the gradient update.

8. A medical data privacy protection analysis method based on federated learning according to claim 7, characterized in that: Said S2 specifically includes: Based on the parameters of the global model after gradient update, a nonlinear adjustment mechanism is introduced. The nonlinear adjustment mechanism dynamically adjusts the update step size according to the current error of the global model, optimizes the global model, and obtains the global model parameters after nonlinear adjustment.

9. A medical data privacy protection analysis method based on federated learning according to claim 7, characterized in that: Said S2 specifically includes: After completing the update and nonlinear adjustment of the global model parameters, each medical node will continue training according to the new global model parameters until the preset stop conditions are met; and blockchain records and transparent verification are introduced to ensure that the global model update process is transparent and cannot be tampered with.

Images