Health data processing method and system based on distributed privacy protection learning

Abstract

The invention provides a health data processing method based on distributed privacy protection learning. The health data processing method comprises the following steps: constructing a federal learning framework to generate a machine learning model; performing an iteration process until the model converges or reaches a preset hyper-parameter; the iterative process comprises: independently training the machine learning model by local data; encrypting local data by a preset encryption algorithm, and calculating a gradient to obtain encrypted gradient information; sharing the gradient information based on multi-party computation; combining the gradient information by a preset aggregation algorithm; and updating the parameters of the machine learning model according to the aggregated gradient, and distributing the updated parameters. According to the invention, on the premise of fully protecting user data privacy, efficient and accurate distributed machine learning model training can be realized, so that the overall performance and security of a health data processing system are improved. The invention further provides a health data processing system based on distributed privacy protection learning and the electronic equipment, which have corresponding advantages.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and specifically relates to a health data processing method and system based on distributed privacy protection learning and corresponding electronic equipment. Background Art

[0002] In the field of health data analysis, the application of disease prediction models has become an important means to improve medical quality and efficiency. With the advancement of big data technology, medical institutions are able to collect a large amount of diverse health data, including electronic health records (EHR), genomic information, wearable device data, medical images, and patients' lifestyles and environmental factors. Through these data, related methods including machine learning, deep learning, and statistical analysis are widely used in the construction of disease risk assessment and prediction models. These methods can identify potential disease risks and development trends through in-depth mining of large amounts of health data. For example, the application of machine learning algorithms can analyze patients' medical history and physiological data to predict the risk of cardiovascular disease; and the application of deep learning in medical image analysis can help doctors accurately identify cancer cells at an early stage. Related technologies not only improve the accuracy of disease prediction, but also promote the realization of personalized medicine, enabling medical staff to develop more targeted prevention and treatment plans for patients, thereby significantly improving the health management effect of patients.

[0003] Due to the limitations of existing technologies, the following major problems are faced: 1. Traditional disease prediction methods rely on limited data within a single medical institution, and the relevant data often have insufficient sample size, lack of data diversity and poor representativeness. 2. Data between different medical institutions are difficult to effectively share and integrate due to restrictions such as privacy regulations and data ownership. 3. Existing data sharing methods cannot effectively guarantee data security and patient privacy, and the risk of data leakage is high. 4. When dealing with complex disease characteristics and multi-factor interactions, existing disease prediction models lack model complexity, poor adaptability and limited predictive ability. 5. Faced with massive amounts of medical data and complex model training tasks, existing computing resources and algorithms are difficult to cope with, resulting in excessively long training times for prediction models.

[0004] It can be seen that the existing technology lacks a high-security health data processing solution that can break the limitations of data silos while ensuring data security and patient privacy. At least there are specific problems to be solved, such as data encryption and sharing, model training, construction and optimization of logistic regression models, and model evaluation and verification. Therefore, it is urgent to develop a health data processing method, system and corresponding electronic equipment based on distributed privacy-preserving learning to improve the accuracy and efficiency of disease prediction, so as to promote the in-depth application of data processing technology. Summary of the invention

[0005] The present invention is to solve all or part of the problems of the above-mentioned prior art. On the one hand, the present invention proposes a health data processing method based on distributed privacy-preserving learning, so that the prediction model can learn more comprehensive and representative disease characteristics and laws, thereby providing more accurate and reliable disease prediction results. On the other hand, the present invention provides a health data processing system based on distributed privacy-preserving learning, which realizes data sharing and collaborative learning across medical institutions while ensuring data privacy and security, thereby improving the accuracy and efficiency of disease prediction.

[0006] On one hand, the present invention provides a health data processing method based on distributed privacy-preserving learning, comprising: constructing a federated learning framework to generate a machine learning model; performing an iterative process until the model converges or reaches preset hyperparameters; the iterative process comprises: independently training the machine learning model by local data; encrypting local data by a preset encryption algorithm, calculating gradients, and obtaining encrypted gradient information; sharing the gradient information based on multi-party computing; merging the gradient information by a preset aggregation algorithm; updating the parameters of the machine learning model by the aggregated gradients, and distributing the updated parameters; wherein the encryption algorithm and the aggregation algorithm are based on multi-party computing, including homomorphic encryption and secure summation; the machine learning model comprises a logistic regression model.

[0007] It integrates federated learning, homomorphic encryption and multi-party computing (MPC) technologies. Through homomorphic encryption and secure summation algorithms, it realizes gradient calculation, secure sharing and aggregate update of encrypted data without leaking the data of each participant, so as to realize distributed training of machine learning models across multiple participants under the premise of protecting data privacy. It can effectively train and update models through encrypted gradient information without directly sharing sensitive data. It not only strengthens data privacy protection, but also makes full use of the computing resources of various medical institutions in the process of model training through optimization algorithms and distributed computing strategies, reduces the training time cost, and improves the collaboration and efficiency of model training. While protecting the privacy of personal health data, it can realize cross-institutional data collaboration and machine learning model training, so that sensitive data such as health information can play its value in machine learning models while ensuring compliance and ethical standards. It provides a positive and feasible solution for data privacy protection and data analysis in the medical and health field.

[0008] In some embodiments, before the iterative process, the following are performed: collecting health data and storing it locally to obtain raw data; standardizing and normalizing the raw data to obtain the local data; the iterative process includes: evaluating and verifying the global model obtained after training; optimizing the global model based on the evaluation and verification results to obtain a final model; after the iterative process, deploying the final model, performing predictions, and generating disease prediction results.

[0009] By collecting, standardizing and normalizing raw health data to improve data quality, model training is made more effective; through model evaluation and verification, continuous optimization of model performance and accuracy of the final model are ensured; and after the iterative process, the model is deployed to actual applications, which improves the practical application value and prediction accuracy of the model.

[0010] In a specific embodiment, the model evaluation method includes: using an independent test data set to evaluate the global model; calculating performance indicators; comparing the calculation results with preset performance indicators; wherein the performance indicators include: accuracy, precision, and recall rate indicators.

[0011] By introducing a model evaluation method that includes performance indicators such as accuracy, precision, and recall, it is possible to objectively quantify the performance of the global model on the basis of distributed training that ensures data privacy protection, providing a more accurate and comprehensive basis for model optimization and decision support.

[0012] In a specific embodiment, the evaluation and verification methods include: cross-validation, confusion matrix, ROC curve; the method for optimizing the global model based on the evaluation and verification results includes: using a regularization method to prevent overfitting; adjusting the hyperparameters; the hyperparameters include learning rate and number of iterations.

[0013] In a specific embodiment, the health data processing method further includes: establishing a feedback mechanism, after generating a disease prediction result, optimizing the model based on the difference between the prediction result and the actual result.

[0014] By introducing evaluation and verification methods such as cross-validation, confusion matrix, ROC curve, and combining regularization to prevent overfitting and optimization methods such as hyperparameter adjustment (such as learning rate, number of iterations), the robustness of the technical solution and model performance are further enhanced. While protecting data privacy, it can more accurately evaluate the model effect and effectively guide the optimization process of the global model.

[0015] In some embodiments, the process of generating a machine learning model includes: each participant in the federated learning determines the initial parameters of the logistic regression model; defines the model structure; the initial parameters include weights and bias terms; wherein defining the model structure includes: determining the number of features, the intercept term, and the value range of the weight.

[0016] By clarifying the initial parameter settings and model structure definitions of each participant in federated learning when generating a machine learning model, including the weights, bias terms, and number of features, intercept terms, and weight value range of the logistic regression model, the model building process is further refined, improving the accuracy and efficiency of model initialization, and laying a solid foundation for the subsequent distributed training and performance optimization of the model.

[0017] In some embodiments, the process of obtaining encrypted gradient information includes: each participant preprocesses the local data and extracts features; the participants respectively encrypt the local data and intermediate calculation results using homomorphic encryption; the participants perform calculations in an encrypted state to obtain the encrypted linear combination results of the logistic regression model; and calculate the gradient from the linear combination results.

[0018] By preprocessing and extracting features from local data, encrypting local data and intermediate calculation results with homomorphic encryption, and performing calculations in an encrypted state to obtain the encrypted linear combination results of the logistic regression model, and then calculating the gradient from this, the sophistication and security of data processing are further enhanced, ensuring the privacy protection of gradient information during transmission and calculation, while improving the efficiency and accuracy of model training.

[0019] In some embodiments, the process of updating the parameters of the machine learning model includes: aggregating the linear combination results calculated by the participants by the aggregation algorithm respectively; applying the encrypted sigmoid function to the aggregated linear combination results to obtain an encrypted probability value; calculating the gradient of the loss function by the participants in an encrypted state based on the actual label value and the probability value respectively to obtain the aggregated gradient; and updating the weights and the bias terms by the gradient descent algorithm in an encrypted state.

[0020] By optimizing the process of updating machine learning model parameters, that is, applying the sigmoid function to the aggregated linear combination results in an encrypted state to calculate the probability value, and calculating the gradient of the loss function in an encrypted state based on this, and then updating the weights and bias terms by the gradient descent algorithm after aggregation, the security and privacy protection capabilities of model parameter updates are further enhanced, while maintaining the continuity and accuracy of model training, and improving the practicality and performance of the overall technical solution in distributed privacy-preserving learning scenarios.

[0021] On the other hand, the present invention provides a health data processing system based on distributed privacy-preserving learning, which is used to execute the health data processing method of the present invention, including: a communication and interactive participant module and an aggregation server module; the participant module is configured to independently train the machine learning model by local data; a preset encryption algorithm is used to encrypt local data, calculate gradients, and obtain encrypted gradient information; the aggregation server module is configured to: share the gradient information based on multi-party computing; merge the gradient information by a preset aggregation algorithm; update the parameters of the machine learning model by the aggregated gradients, and distribute the updated parameters.

[0022] Through modular design, the participant module and the aggregation server module are closely integrated to form an efficient and secure distributed privacy-preserving learning framework. The participant module can make full use of local data for model training and ensure data privacy through a preset encryption algorithm; while the aggregation server module can safely share and merge gradient information and update model parameters based on multi-party computing technology.

[0023] In some embodiments, there are multiple participant modules; different participant modules train the machine learning model in an asynchronous manner.

[0024] By allowing different parties to train machine learning models asynchronously, the flexibility and scalability of the technical solution in a distributed environment are further improved, so that model training can more efficiently adapt to the differences in computing resources and response times of different parties, thereby accelerating the model training process while maintaining data privacy protection and optimizing model performance. It breaks through the data privacy protection bottleneck of the traditional centralized learning model, allowing multiple parties to jointly train a high-performance machine learning model without sharing the original data; the system has higher training efficiency and accuracy through the optimized model training process, allowing the model to adapt to new data faster and predict health risks more accurately. It can not only protect the privacy rights of users and enhance users' trust in health data processing systems; it can also promote data sharing and cooperation in the fields of medicine and health.

[0025] On the other hand, the present invention provides an electronic device comprising a memory, a processor, and one or more computer programs, wherein the computer program is stored in the memory and is configured to be executed by several of the processors to execute the health data processing method based on distributed privacy-preserving learning provided on the one hand of the present invention.

[0026] Compared with the prior art, the main beneficial effects of the present invention are:

[0027] 1. The present invention provides a health data processing method based on distributed privacy-preserving learning. By constructing a federated learning framework and utilizing multi-party computing technologies such as homomorphic encryption and secure summation, it can effectively perform distributed training of machine learning models while protecting local data privacy, thereby comprehensively improving the efficiency and security of health data processing.

[0028] 2. The present invention provides a health data processing system based on distributed privacy-preserving learning, which realizes privacy-preserving learning in a distributed environment, thereby ensuring data privacy security and effectively promoting the improvement of model training efficiency and accuracy.

[0029] 3. The electronic device provided by the present invention can automatically implement the health data processing method based on distributed privacy protection learning of the present invention, has corresponding advantages, and is conducive to further promotion in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Schematic diagram of the iterative process of the health data processing method according to an embodiment of the present invention.

[0031] Figure 2 Schematic diagram of a health data processing system according to an embodiment of the present invention.

[0032] Figure 3 The figure is a flow chart of a disease prediction method based on federated learning MPC-logistic regression according to an embodiment of the present invention.

[0033] Figure 4 A schematic diagram of a local scene according to an embodiment of the present invention.

[0034] Figure 5 Schematic diagram of a secure interaction scenario according to an embodiment of the present invention.

[0035] Figure 6 Schematic diagram of a disease prediction and result feedback scenario according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical solutions in the specific embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0037] Through the health data processing method based on distributed privacy protection learning, the health data processing system and the electronic device based on distributed privacy protection learning provided in this embodiment, the dual improvement of data privacy protection and model training efficiency and accuracy in a distributed environment is achieved, which has a wide range of application scenarios and important social value, and can promote data sharing and cooperation in the fields of medical care, health, etc., while protecting user privacy and data security. Specifically: between different medical institutions, due to data privacy and regulatory restrictions, it is often difficult to achieve direct data sharing. However, using the example method and system, each medical institution can jointly train a machine learning model for disease prediction, diagnosis or treatment optimization without sharing the original patient data. For example, multiple hospitals can jointly train a prediction model for a rare disease, and by aggregating their respective data gradient information, improve the prediction accuracy of the model while protecting patient privacy. Health data research institutions usually need large-scale data sets to train and optimize models to discover new health risk factors or disease patterns. However, obtaining these data often faces challenges in privacy protection and data security. By applying the example method and system, research institutions can cooperate with other data providers (such as insurance companies, health management agencies, etc.) to jointly train models without directly accessing the original data. With the popularity of smart wearable devices and mobile health applications, the collection and analysis of personal health data has become increasingly important. However, this data often contains sensitive information, such as heart rate, sleep quality, etc. Using the example method and system, individuals can upload their health data to a secure distributed platform and train a more accurate health risk assessment model with other users while protecting personal privacy. This is not limited to specific situations.

[0038] The health data processing method based on distributed privacy protection learning provided in this embodiment can be implemented by the electronic device of this embodiment. The example electronic device includes a memory, a processor, and one or more computer programs, wherein the computer program is stored in the memory and is configured to be executed by one or more processors. The specific health data processing method based on distributed privacy protection learning provided in this embodiment.

[0039] The present invention is illustrated by way of the following embodiments to illustrate that the present invention relates to a health data processing method based on distributed privacy protection learning, a health data processing system based on distributed privacy protection learning, and an electronic device, wherein the terms involved are described as follows by way of example and not by way of limitation.

[0040] Federated Learning: Federated Learning is a distributed machine learning method that aims to jointly train models among multiple parties while protecting data privacy.

[0041] MPC: MPC (Multi-Party Computation) is a cryptographic technique that allows multiple parties to jointly compute the output of a function while keeping their inputs private.

[0042] Homomorphic encryption: Homomorphic encryption is an encryption technology that allows calculations to be performed on encrypted data, and the results of the calculations are the same as those calculated directly on the original data after decryption. It provides a strong privacy protection mechanism and is particularly suitable for scenarios that require processing on confidential data.

[0043] Logistic Regression: Logistic Regression is a widely used statistical method for dealing with binary classification problems. It performs classification by predicting the probability of an event occurring.

[0044] A health data processing method based on distributed privacy-preserving learning illustrated in an embodiment of the present invention includes: constructing a federated learning framework to generate a machine learning model; and performing an iterative process until the model converges or reaches a preset hyperparameter. Figure 1 As shown, the iterative process includes: independently training the machine learning model with local data; encrypting the local data with a preset encryption algorithm, calculating the gradient, and obtaining encrypted gradient information; sharing the gradient information based on multi-party computing; merging the gradient information with a preset aggregation algorithm; updating the parameters of the machine learning model with the aggregated gradient, and distributing the updated parameters; wherein the encryption algorithm and the aggregation algorithm are based on multi-party computing, including homomorphic encryption and secure summation; the machine learning model includes a logistic regression model.

[0045] In some embodiments, before the iterative process, the following are performed: collecting health data and storing it locally to obtain raw data; standardizing and normalizing the raw data to obtain local data; the iterative process includes: evaluating and verifying the global model obtained after training; optimizing the global model based on the evaluation and verification results to obtain the final model; after the iterative process, deploying the final model, performing predictions, and generating disease prediction results.

[0046] In a specific embodiment, the model evaluation method includes: using an independent test data set to evaluate the global model; calculating performance indicators; comparing the calculation results with preset performance indicators; wherein the performance indicators include: accuracy, precision, and recall indicators.

[0047] In a specific embodiment, the evaluation and verification methods include: cross-validation, confusion matrix, ROC curve; the method of optimizing the global model based on the evaluation and verification results includes: using regularization methods to prevent overfitting; adjusting hyperparameters; hyperparameters include learning rate and number of iterations.

[0048] In a specific embodiment, the health data processing method further includes: establishing a feedback mechanism, after generating a disease prediction result, optimizing the model based on the difference between the prediction result and the actual result.

[0049] In some embodiments, the process of generating a machine learning model includes: each participant in the federated learning determines the initial parameters of the logistic regression model; defines the model structure; the initial parameters include weights and bias terms; wherein defining the model structure includes: determining the number of features, the intercept term, and the value range of the weight.

[0050] In some embodiments, the process of obtaining encrypted gradient information includes: each participant preprocesses local data and extracts features; the participants encrypt local data and intermediate calculation results using homomorphic encryption; the participants perform calculations in an encrypted state to obtain encrypted linear combination results of the logistic regression model; and the gradient is calculated from the linear combination results.

[0051] In some embodiments, the process of updating the parameters of the machine learning model includes: aggregating the linear combination results calculated by the participants by an aggregation algorithm; applying an encrypted sigmoid function to the aggregated linear combination results to obtain an encrypted probability value; calculating the gradient of the loss function by the actual label value and probability value participants in an encrypted state to obtain the aggregated gradient; and updating the weights and bias terms by a gradient descent algorithm in an encrypted state.

[0052] refer to Figure 2 As shown, an embodiment of the present invention also provides a health data processing system based on distributed privacy-preserving learning, which is used to execute the health data processing method of the above embodiment, including: multiple participant modules for communication interaction (participant module A, participant module B, participant module C...participant module N) and an aggregation server module. Each participant module is configured to independently train a machine learning model with local data; encrypt local data with a preset encryption algorithm, calculate gradients, and obtain encrypted gradient information; the aggregation server module is configured to: share gradient information based on multi-party calculations; merge gradient information with a preset aggregation algorithm; update the parameters of the machine learning model with the aggregated gradients, and distribute the updated parameters. The example aggregation server module is composed of a central coordination server, but is not limited to it.

[0053] In some embodiments, there are multiple participant modules; different participant modules train the machine learning model in an asynchronous manner. Training can also be done in a synchronous manner, which is not limited here.

[0054] The following reference Figures 2 to 6 , the specific technical solutions of the health data processing method and system based on distributed privacy protection learning in disease prediction in the embodiment are introduced. Figure 3As shown in the figure, the main process of the disease prediction method based on federated learning MPC-logistic regression includes:

[0055] 1. Data collection: Each participant (such as hospitals and research institutions) collects patient data and ensures that the data is stored locally and is not shared.

[0056] 2. Model initialization: Determine the initial parameters and structure of the logistic regression model and each participant configures their own computing environment.

[0057] 3. Local training: Each participant trains the logistic regression model on local data and calculates local gradients or parameter updates.

[0058] 4. Secure exchange: Use the MPC protocol to perform secure exchange of local updates, ensuring that only calculation results are shared, not raw data.

[0059] 5. Aggregate computing: Aggregate updates from all parties, merge these updates through MPC methods, and calculate the global model.

[0060] 6. Parameter update: global model update, update the parameters of the global model, and return the new parameters to all participants.

[0061] 7. Iterative training: Repeat the process of local training and secure aggregation until the model converges or reaches the set performance standard.

[0062] 8. Model evaluation: Evaluate the performance of the global model on the test set and calculate indicators such as accuracy, precision, and recall.

[0063] 9. Disease prediction: Use the trained logistic regression model to predict new data. Generate disease prediction results to provide support for clinical decision-making.

[0064] 10. Result feedback: Collect the predicted results and compare them with the actual results to optimize the model. In the example, adjust the model parameters based on the feedback information to improve the prediction accuracy.

[0065] Figure 4 It illustrates the scenarios of data collection, model initialization, and local training of the system. Figure 5 It illustrates the scenarios of the system performing secure data exchange, aggregate calculation, parameter update, and iterative training. Figure 6 It illustrates the scenarios of the system performing model evaluation, disease prediction, and result feedback.

[0066] The implementation process of the example aggregation calculation includes: 1. Preparation stage: All participants in federated learning determine the specific logical model: including the number of features, intercept terms, and weight value ranges.

[0067] 2. Encryption calculation stage.

[0068] 1. Each participant preprocesses the local data and extracts features X1, X2, …, Xn.

[0069] 2. Each participant uses encryption technology such as homomorphic encryption to encrypt local data and intermediate calculation results.

[0070] 3. For the linear combination part β0+β1X1+β2X2+…+βnXn in the logistic regression, each participant performs calculations in an encrypted state to obtain the encrypted linear combination result.

[0071] 3. Aggregation stage.

[0072] 1. Each participant sends the encrypted linear combination result to an aggregation server (which can be a trusted third party or a central coordination server for distributed aggregation through a specific protocol).

[0073] 2. The aggregation server aggregates the encrypted linear combination results from each participant without decryption. For example, the total linear combination result can be obtained by secure summation or other methods.

[0074] 3. The aggregation server applies the encrypted sigmoid function to the aggregated encrypted result to calculate the encrypted probability value P(Y=1|X). The logistic regression training of the example uses the sigmoid function (logistic function, which converts the input of the linear combination into a probability value: where β0 is the intercept, β1, β2...βn are the weights of the features, and X1, X2,..., Xn are the input features.

[0075] 4. Update phase.

[0076] 1. Based on the encrypted probability value and the actual label value, each participant can calculate the gradient of the loss function in an encrypted state.

[0077] 2. Use a secure gradient descent algorithm to update the weights β1, β2, …, βn and the intercept β0 in an encrypted state.

[0078] In the example, in the data encryption and sharing link, each medical institution uses specific encryption software and tools to encrypt local data. These encryption tools use high-intensity encryption algorithms, such as the AES encryption algorithm, combined with digital certificates and identity authentication mechanisms to ensure that only legitimate medical institutions can encrypt and decrypt data. The encrypted data is transmitted through a secure data transmission channel, such as a virtual private network (VPN), to ensure the security of the data during transmission. In the model training phase, a central coordination server is established to coordinate the allocation of computing tasks and model parameter synchronization among various medical institutions. The computing nodes of each medical institution use local encrypted data for local model training calculations according to the assigned tasks. For example, by calculating the gradient information of the encrypted data and uploading it to the central coordination server. The central coordination server aggregates and processes the encrypted gradient information uploaded by each medical institution without obtaining plaintext data, and then distributes the updated model parameters to each medical institution to achieve collaborative training of the model. For the construction and optimization of the logistic regression model, feature engineering technology is used to preprocess the medical data and extract key disease features. For example, extract features such as age, gender, medical history, symptoms, etc. from the patient's medical records, and perform numerical and standardized processing. During the model training process, use regularization methods such as L1 and L2 regularization to prevent the model from overfitting. At the same time, optimize the performance of the model by adjusting the model's hyperparameters, such as learning rate, number of iterations, etc. In the model evaluation and verification phase: use cross-validation, confusion matrix, ROC curve and other techniques to evaluate and verify the trained model. Collect independent test data sets from various medical institutions to evaluate the predictive performance of the model. If the model performance does not meet the requirements, return to the model training stage for adjustment and optimization.

[0079] The example provides the technical architecture and process of the entire disease prediction method based on federated learning MPC-logistic regression, the specific technical implementation and related algorithms of data encryption sharing, and the application and optimization of the logistic regression model in disease prediction. Through the application of federated learning MPC technology in the integration of human cost data, multi-source data fusion is achieved while ensuring data security and privacy. Advanced data analysis algorithms and models are used to deeply explore the relationship between human cost and business performance. The generated precise cost optimization strategies and suggestions are operational in combination with the actual situation of the participants. Specifically: First, through the federated learning MPC technology, data encryption sharing among multiple medical institutions is realized under the premise of ensuring data security and patient privacy. Advanced encryption algorithms are used to encrypt the data of each institution, so that the data is always in an encrypted state during transmission and calculation. Only authorized legal calculations can obtain useful information, effectively avoiding the risk of data leakage. Secondly, the powerful classification ability and interpretability of the logistic regression model are combined with the federated learning MPC framework to build a more accurate disease prediction model. The logistic regression model can effectively handle multi-factor disease characteristics. Through distributed computing and collaborative training of encrypted data, it can automatically learn and capture the complex relationships between different factors, thereby improving the prediction accuracy of the model. Furthermore, during the model training process, by optimizing algorithms and distributed computing strategies, the computing resources of various medical institutions are fully utilized, greatly improving the training efficiency. The time cost of model training is reduced, and effective prediction models can be obtained more quickly, providing timely support for clinical decision-making. Finally, through joint training and data sharing among multiple medical institutions, richer and more diverse disease data are integrated, breaking the limitations of data silos. This enables the model to learn more comprehensive and representative disease characteristics and laws, thereby providing more accurate and reliable disease prediction results.

[0080] It can be seen that the disease prediction method based on federated learning MPC-logistic regression can bring many significant benefits, including at least the following aspects.

[0081] In terms of prediction accuracy, by integrating diverse data from multiple medical institutions, the model can capture more comprehensive and detailed disease characteristics and patterns, thereby significantly improving the accuracy of disease prediction. Practical applications have shown that compared with traditional methods, the prediction accuracy of this method can be increased by 15%-20%, providing a more reliable basis for early diagnosis and intervention of diseases.

[0082] In terms of data security and privacy protection, strict encryption mechanisms and distributed computing models ensure the confidentiality and integrity of patient data. This not only eliminates the concerns of medical institutions about data sharing, promotes the effective use of medical data, but also avoids legal liability and social trust crises that may be caused by data leakage.

[0083] In terms of the rational allocation of medical resources, accurate disease prediction helps to plan and allocate medical resources in advance, such as beds, medicines, and medical staff, etc. It can effectively avoid waste and shortage of resources and improve the operating efficiency and service quality of the medical system.

[0084] In addition, by implementing the example health data processing method, it is also conducive to promoting cooperation and communication between medical institutions. By jointly participating in the training and optimization of the model, different institutions can share experience and expertise and promote the coordinated development and innovation of medical technology. For individual patients, more accurate and timely disease prediction can help them take preventive and treatment measures earlier, reduce the harm and treatment costs of the disease, and improve the quality of life and recovery opportunities.

[0085] In order to facilitate the description, some common English nouns or letters and concepts used in the present invention are only used for exemplary reference rather than restrictive interpretation or specific usage, and the protection scope of the present invention should not be limited by their Chinese translation or specific letters.

[0086] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solutions provided in the embodiments of the present invention are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0087] In this article, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0088] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A health data processing method based on distributed privacy-preserving learning, characterized in that: include: Build a federated learning framework to generate machine learning models; The iterative process is carried out until the model converges or reaches the preset hyperparameters; The iterative process includes: independently training the machine learning model by local data; The preset encryption algorithm encrypts the local data, calculates the gradient, and obtains the encrypted gradient information; Sharing the gradient information based on multi-party computing; merging the gradient information by a preset aggregation algorithm; Updating the parameters of the machine learning model using the aggregated gradients and distributing the updated parameters; Among them, the encryption algorithm and the aggregation algorithm are based on multi-party computing, including homomorphic encryption and secure summation; the machine learning model includes a logistic regression model.

2. The health data processing method based on distributed privacy-preserving learning according to claim 1 is characterized in that: Before the iterative process, the following steps are performed: collecting health data and storing them locally to obtain raw data; Standardizing and normalizing the original data to obtain the local data; The iterative process includes: evaluating and verifying the global model obtained after training; optimizing the global model based on the evaluation and verification results to obtain a final model; after the iterative process, deploying the final model, performing prediction, and generating disease prediction results.

3. The health data processing method based on distributed privacy-preserving learning according to claim 2 is characterized in that: Methods for model evaluation include: evaluating the global model using an independent test dataset; calculating performance metrics; Compare the calculation results with the preset performance indicators; The performance indicators include accuracy, precision and recall.

4. The health data processing method based on distributed privacy-preserving learning according to claim 2 is characterized in that: The evaluation and verification methods include: cross-validation, confusion matrix, ROC curve; The method for optimizing the global model based on the evaluation and verification results includes: using a regularization method to prevent overfitting; adjusting the hyperparameters; The hyperparameters include learning rate and number of iterations.

5. The health data processing method based on distributed privacy-preserving learning according to claim 1, characterized in that: It also includes: establishing a feedback mechanism, after generating disease prediction results, optimizing the model based on the difference between the prediction results and the actual results.

6. The health data processing method based on distributed privacy-preserving learning according to any one of claims 1 to 5, characterized in that: The process of generating a machine learning model involves: Each participant in the federated learning determines the initial parameters of the logistic regression model; Define the model structure; The initial parameters include weights and bias terms; defining the model structure includes: determining the number of features, the intercept term and the value range of the weight.

7. The health data processing method based on distributed privacy-preserving learning according to claim 6, characterized in that: The process of obtaining encrypted gradient information includes: Each of the participants pre-processes the local data to extract features; The participants encrypt the local data and the intermediate calculation results respectively by homomorphic encryption; The participants perform calculations in an encrypted state to obtain the encrypted linear combination results of the logistic regression model; and calculate the gradient based on the linear combination results.

8. The health data processing method based on distributed privacy-preserving learning according to claim 7 is characterized in that: The process of updating the parameters of a machine learning model involves: Aggregating the linear combination results calculated by the participants respectively by the aggregation algorithm; Apply the encrypted sigmoid function to the aggregated linear combination results to obtain the encrypted probability value; The actual label value and the probability value are used by the participants to calculate the gradient of the loss function in an encrypted state to obtain the gradient after the gathering; The weight and the bias term are updated in an encrypted state by a gradient descent algorithm.

9. A health data processing system based on distributed privacy-preserving learning, characterized in that: Used to execute the health data processing method based on distributed privacy protection learning as described in any one of claims 1 to 8; comprising: a communication interaction participant module and an aggregation server module; The participant module is configured to independently train the machine learning model using local data; encrypt the local data using a preset encryption algorithm, calculate the gradient, and obtain encrypted gradient information; The aggregation server module is configured to: share the gradient information based on multi-party computing; merge the gradient information by a preset aggregation algorithm; update the parameters of the machine learning model by the aggregated gradient, and distribute the updated parameters.

10. The health data processing system based on distributed privacy-preserving learning according to claim 9, characterized in that: There are multiple participant modules; different participant modules train the machine learning model in an asynchronous manner.

Images