Intelligent medical information sharing method based on federal learning framework

Abstract

The invention discloses an intelligent medical information sharing method based on a federated learning framework, and relates to the technical field of medical information sharing. According to the invention, the original sensitive data of the medical institution is always kept locally, so that the leakage risk in the centralized storage and transmission process of the data is effectively avoided; a multi-level and end-to-end security protection system is constructed by combining a data encryption unit, a security authentication unit and an anonymization technology in a data preprocessing stage, and the privacy of a patient is ensured to be protected to the greatest extent in the whole model training and information sharing process. Data islands dispersed in different medical institutions can cooperate on the premise that original data does not need to be directly shared by utilizing a federated learning technology. A built federal learning system and a shared platform interface module support safe deployment of a global model which is trained completely and has excellent performance to each participating medical institution or external system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical information sharing, and in particular to a smart medical information sharing method based on a federated learning framework. BACKGROUND

[0002] Medical information system is an edge science integrating medicine, information, management, computer and other disciplines, which has been widely used in developed countries and has created good social and economic benefits. Medical information system is a necessary technical support and infrastructure for modern hospital operation. The purpose of implementing medical information system is to strengthen hospital management, improve work efficiency, improve medical quality and facilitate medical information management with more modern, scientific and standardized means, so as to establish a new image of modern hospital, which is also the inevitable direction of future hospital development.

[0003] The federated learning framework is a distributed machine learning paradigm that aims to protect data privacy while realizing multi-institutional collaborative training of global models.

[0004] In the prior art, the smart medical information sharing method usually relies on centralized data storage architecture, in which medical institutions upload patient electronic health records to a central server or cloud platform for unified processing and analysis. These methods include sharing systems based on traditional databases, API interface integration or blockchain technology applications, aiming to promote data interconnection and model training. However, such methods face significant risks of data privacy leakage, as sensitive medical information is vulnerable to unauthorized access or attacks during transmission and storage. At the same time, the problem of data silos is widespread, with inconsistent data formats and security policies between different institutions, resulting in low sharing efficiency, insufficient model generalization ability, and difficulty in meeting real-time needs of multi-institutional collaboration. Therefore, we propose a smart medical information sharing method based on a federated learning framework. SUMMARY

[0005] The purpose of the present application is to solve the problems of significant data privacy leakage risk, low sharing efficiency, insufficient model generalization ability, and difficulty in meeting real-time needs of multi-institutional collaboration mentioned in the background technology. The present application provides a smart medical information sharing method based on a federated learning framework.

[0006] In order to achieve the above purpose, the present application specifically adopts the following technical solution: A smart medical information sharing method based on a federated learning framework, comprising the following steps: S1, constructing a global machine learning model: the central server establishes a global machine learning model based on a pre-set neural network architecture, and after completing the model establishment, the initial parameters are generated randomly or loaded from a pre-trained model; S2. Deploy the Federated Learning Client Module: Collect raw data from multiple medical institutions and add it to the Federated Learning Network for communication with the server, local data management, and model training; S3. Model Parameter Download: At the beginning of each training cycle, the central server sends the current global model parameters to all participating institutions, and the participating institutions download the model parameters through an encrypted channel to avoid the transmission of raw data. S4. Local Model Training: After downloading the model parameters, each participating institution uses local medical data (such as electronic health records and image data) to train the model and performs data preprocessing. During the training process, stochastic gradient descent (SGD) or adaptive optimization algorithms are used to update the local model parameters. S5. Model Update Upload: After training is completed, each participating institution uploads the updated model parameters to the central server through the encrypted channel. The upload process also adopts end-to-end encryption and achieves efficient interaction based on a lightweight communication protocol to reduce network overhead. S6. Global Model Aggregation and Update: After receiving model updates from all participating institutions, the central server integrates the updated parameters, optimizes the global machine learning model, and records the operational audit information during the aggregation process to meet compliance requirements. S7. Iterative Training and Termination: The central server determines whether the global model has converged or reached the preset training round limit; if the condition is not met, it returns to S3 for the next round of training; if the condition is met, it outputs the final global model. S8. Global Model Upload: Deploy the trained global model to participating institutions or shared platforms for real-time medical information analysis and prediction services. It also supports a model update mechanism to adapt to new data or the addition of new institutions.

[0007] Furthermore, the global machine learning model includes disease prediction, diagnostic assistance, or health risk assessment.

[0008] Furthermore, the federated learning client module includes a data encryption unit and a security authentication unit. The data encryption unit performs end-to-end encryption of local medical data based on the Advanced Encryption Standard (AES) or RSA algorithm to ensure the confidentiality and integrity of the data during transmission and storage, preventing unauthorized access or leakage. The security authentication unit adopts a two-factor authentication mechanism based on Public Key Infrastructure (PKI) to verify the digital identity of each medical institution and manage its access permissions. The security authentication unit also includes user role assignment and operation auditing to ensure the security and compliance of the federated learning network.

[0009] Furthermore, the federated learning client module also integrates a data integration unit for data integration.

[0010] Furthermore, the encrypted channel is implemented based on the Advanced Encryption Standard (AES) or RSA algorithm, and combined with the Public Key Infrastructure (PKI) mechanism of the security authentication unit, it ensures the confidentiality, integrity and tamper resistance of data during the parameter download process.

[0011] Furthermore, the data preprocessing includes the following steps: S41. Data Cleaning: Remove missing values, outliers, and duplicate records from local medical data; automatically detect and correct data errors using preset rules or algorithms. S42. Data anonymization: Apply encryption or anonymization technologies to process sensitive information, such as patient names, ID numbers, and contact information, to ensure that personal privacy protection complies with regulatory requirements. S43. Data standardization: Perform normalization or standardization operations to unify the data formats and scales from different sources, including numerical range adjustment and category coding; S44. Feature Extraction: Extract key features, such as disease indicators or risk factors, from the preprocessed data to support feature input optimization for local model training.

[0012] Furthermore, the integrated update parameters use a federated averaging algorithm or a weighted aggregation method, wherein the federated averaging algorithm is based on the following function formula: global model parameters By weighted averaging the local model parameters of all participating institutions The weights are allocated based on the amount of data each institution has, specifically as follows: In the formula , representing the number of data samples for the i-th institution. This represents the sum of the total data samples from all institutions.

[0013] Furthermore, the weighted aggregation method adjusts the aggregation ratio based on preset weight coefficients (such as institutional data quality or participation frequency) to ensure fair and efficient model updates.

[0014] Furthermore, the central server performs performance evaluation on the updated global model, and calculates accuracy, recall or F1 score using the reserved validation dataset. The reserved validation dataset is derived from anonymized historical data or independent validation datasets from multiple medical institutions to ensure that the data is independent of the training set and does not overlap.

[0015] A smart healthcare information sharing system based on a federated learning framework includes: The central server is configured to build and manage global machine learning models, perform initial generation, distribution, aggregation and update of model parameters, and performance evaluation, while supporting operation audit information logging to meet compliance requirements. Multiple federated learning client modules are deployed in participating medical institutions for communicating with the central server, managing local data, training models and uploading updates. Each client module integrates a data encryption unit, a security authentication unit and a data integration unit. The encrypted communication channel module is implemented based on the Advanced Encryption Standard (AES) or RSA algorithm, combined with the Public Key Infrastructure (PKI) mechanism of the Security Authentication Unit, to ensure end-to-end encryption, data integrity and tamper resistance of model parameters during the download and upload process; The shared platform interface module is used to deploy the trained global model to participating institutions or external systems, and supports real-time medical information analysis, prediction services and model update mechanisms. The data storage module is configured to securely store local medical data, model parameters, and audit logs to ensure data confidentiality and privacy protection. The performance evaluation module, integrated into the central server, uses a reserved validation dataset to calculate accuracy, recall, or F1 score metrics to optimize global model performance.

[0016] The beneficial effects of this invention are as follows: 1. Through the core mechanism of the federated learning framework, this invention achieves local data retention and only interactive encryption of model parameters. The original sensitive data of each medical institution remains locally, effectively avoiding the risk of leakage during centralized data storage and transmission. Combined with data encryption units, security authentication units, and anonymization technology in the data preprocessing stage, a multi-layered, end-to-end security protection system is constructed to ensure maximum protection of patient privacy throughout the entire model training and information sharing process.

[0017] 2. This invention utilizes federated learning technology to enable "data silos" scattered across different medical institutions to collaborate without directly sharing raw data. Through data integration units that standardize, clean, and extract features from heterogeneous data, and through the efficient integration of local model updates from multiple parties using federated averaging algorithms or weighted aggregation methods, it can gather broader and richer medical knowledge. This not only significantly improves the accuracy and recall of the global model in key tasks such as disease prediction, diagnostic assistance, and health risk assessment, but also enhances the model's generalization ability and robustness, enabling it to adapt to medical scenarios in different institutions and among different populations, overcoming the performance bottleneck caused by limited data from a single institution.

[0018] 3. The federated learning system and shared platform interface module constructed in this invention support the secure deployment of trained, high-performance global models to participating medical institutions or external systems. This enables real-time cross-institutional medical information analysis, predictive services, and collaborative decision-making, such as assisting doctors in diagnosing rare diseases, providing personalized treatment recommendations, and predicting large-scale epidemic trends. Furthermore, it deeply integrates mechanisms for operational audit information recording, user role allocation, and dynamic permission adjustment, ensuring that all operations are traceable and monitorable. It can flexibly respond to dynamic changes in participating institutions (joining or leaving), continuously meeting compliance requirements and providing a solid and reliable technical foundation for building an efficient, secure, and interoperable smart healthcare ecosystem. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the data preprocessing method in this invention; Figure 3 This is a schematic diagram of the modules of the intelligent medical information sharing system in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0021] Please see Figure 1 - Figure 2 This invention provides a smart healthcare information sharing method based on a federated learning framework, comprising the following steps: S1. Constructing a global machine learning model: The central server builds a global machine learning model based on a preset neural network architecture. After the model is built, the initial parameters are generated randomly or loaded from a pre-trained model. S2. Deploy the Federated Learning Client Module: Collect raw data from multiple medical institutions and add it to the Federated Learning Network for communication with the server, local data management, and model training; S3. Model Parameter Download: At the beginning of each training cycle, the central server sends the current global model parameters to all participating institutions, and the participating institutions download the model parameters through an encrypted channel to avoid the transmission of raw data. S4. Local Model Training: After downloading the model parameters, each participating institution uses local medical data (such as electronic health records and image data) to train the model and performs data preprocessing. During the training process, stochastic gradient descent (SGD) or adaptive optimization algorithms are used to update the local model parameters. S5. Model Update Upload: After training is completed, each participating institution will upload the updated model parameters to the central server through an encrypted channel. The upload process also adopts end-to-end encryption and is based on a lightweight communication protocol to achieve efficient interaction and reduce network overhead. S6. Global Model Aggregation and Update: After receiving model updates from all participating institutions, the central server integrates the updated parameters, optimizes the global machine learning model, and records the operational audit information during the aggregation process to meet compliance requirements. S7. Iterative Training and Termination: The central server determines whether the global model has converged or reached the preset training round limit; if the condition is not met, it returns to S3 for the next round of training; if the condition is met, it outputs the final global model. S8. Global Model Upload: Deploy the trained global model to participating institutions or shared platforms for real-time medical information analysis and prediction services. It also supports a model update mechanism to adapt to new data or the addition of new institutions.

[0022] In this embodiment, preferably, the global machine learning model includes disease prediction, diagnostic assistance, or health risk assessment. By incorporating disease prediction, diagnostic assistance, or health risk assessment into the global machine learning model, multiple medical data resources can be effectively integrated, improving the model's accuracy and generalization ability in disease prediction, diagnostic assistance, and health risk assessment. At the same time, the distributed nature of federated learning ensures patient data privacy and security, preventing the leakage of sensitive information. In this embodiment, preferably, the federated learning client module includes a data encryption unit and a security authentication unit. The data encryption unit performs end-to-end encryption of local medical data based on the Advanced Encryption Standard (AES) or RSA algorithm, ensuring the confidentiality and integrity of the data during transmission and storage, and preventing unauthorized access or leakage. The security authentication unit adopts a two-factor authentication mechanism based on Public Key Infrastructure (PKI) to verify the digital identity of each medical institution and manage its access permissions. The security authentication unit also includes user role allocation and operation auditing to ensure the security and compliance of the federated learning network. That is, with the cooperation of the data encryption unit and the security authentication unit, the leakage, tampering, or unauthorized access of sensitive medical data during the federated learning process can be effectively prevented. At the same time, through a strict user role allocation mechanism (such as distinguishing between roles such as medical institution administrators, data operators, and auditors) and real-time operation audit log recording, all access behaviors are ensured to be traceable and monitorable, thereby meeting the data protection regulations (such as HIPAA or GDPR) requirements of the medical industry. In addition, the security authentication unit also supports dynamic permission adjustments to address access control changes when institutions join or leave, further enhancing the overall security and compliance of the federated learning network and providing a reliable foundation for the sharing of smart healthcare information.

[0023] In this embodiment, preferably, the federated learning client module also integrates a data integration unit for data integration. This unit effectively standardizes and cleans heterogeneous data from different medical institutions, including removing redundant information, unifying data formats (such as HL7 or FHIR standards), and ensuring data consistency and usability, thereby improving the efficiency and accuracy of federated learning model training. Furthermore, the data integration unit implements anonymization mechanisms during data processing, such as k-anonymization or differential privacy technology, to further protect sensitive patient information, prevent privacy leaks, and ensure that the integrated data complies with medical industry regulations (such as HIPAA or GDPR). Combined with the aforementioned data encryption and security authentication units, a complete data processing chain is formed, optimizing the overall performance and reliability of smart healthcare information sharing while ensuring data security.

[0024] In this embodiment, preferably, the encrypted channel is implemented based on the Advanced Encryption Standard (AES) or RSA algorithm, combined with the Public Key Infrastructure (PKI) mechanism of the security authentication unit to ensure the confidentiality, integrity and tamper resistance of data during the parameter download process.

[0025] In this embodiment, preferably, the data preprocessing includes the following steps: S41. Data Cleaning: Remove missing values, outliers, and duplicate records from local medical data, and automatically detect and correct data errors through preset rules or algorithms; S42. Data Desensitization: Apply encryption or anonymization technology to process sensitive information, such as patient names, ID numbers, and contact information, to ensure that personal privacy protection complies with regulatory requirements; S43. Data standardization: Perform normalization or standardization operations to unify the data formats and scales from different sources, including numerical range adjustment and category coding; S44. Feature Extraction: Extract key features, such as disease indicators or risk factors, from the preprocessed data to support feature input optimization for local model training.

[0026] Data preprocessing effectively improves the efficiency and accuracy of local model training, reduces data noise and anomalies, and optimizes feature input quality, thereby enhancing the generalization ability and prediction accuracy of federated learning models in smart healthcare information sharing. Simultaneously, this process ensures data consistency and availability, supports the implementation of privacy protection mechanisms, further reduces compliance risks, and provides a reliable data foundation for subsequent model aggregation.

[0027] In this embodiment, preferably, the integrated update parameters use a federated averaging algorithm or a weighted aggregation method, wherein the federated averaging algorithm is based on the following function formula: global model parameters By weighted averaging the local model parameters of all participating institutions The weights are allocated based on the amount of data each institution has, specifically as follows: In the formula , representing the number of data samples for the i-th institution. This represents the sum of all data samples from all institutions. The federated averaging algorithm efficiently aggregates local model updates to obtain a globally optimized model while protecting the data privacy of each institution. Simultaneously, by assigning weights based on data volume, it ensures that institutions with abundant data contribute more to the global model, thereby improving the model's accuracy and generalization ability in tasks such as medical diagnosis and disease prediction, and supporting collaborative decision-making and information sharing in smart healthcare systems.

[0028] In this embodiment, preferably, the weighted aggregation method adjusts the aggregation ratio based on preset weight coefficients (such as institutional data quality or participation frequency) to ensure fair and efficient model updates. The weighted aggregation method can dynamically optimize the aggregation process according to preset weight coefficients (such as institutional data quality or participation frequency), thereby significantly improving the fairness and efficiency of the model while protecting data privacy. At the same time, by flexibly adjusting the weights, it can adapt to the heterogeneous needs of different medical scenarios. For example, in rare disease diagnosis or multi-center collaboration, it ensures that institutions with high data quality or frequent participation gain reasonable influence, thereby enhancing the stability and generalization ability of the model in medical prediction tasks and supporting efficient information sharing and collaborative decision-making in smart healthcare systems.

[0029] In this embodiment, preferably, the central server performs performance evaluation on the updated global model, using a reserved validation dataset to calculate accuracy, recall, or F1 score. The reserved validation dataset originates from anonymized historical data or independent validation datasets from multiple medical institutions, ensuring that the data is independent and non-overlapping with the training set. This performance evaluation allows for real-time monitoring of model performance changes, identification of potential biases or overfitting issues, and timely optimization of model parameters or adjustment of the federated learning strategy. Based on the evaluation results, the central server can dynamically adjust weight coefficients or select high-contribution institutions to participate in subsequent iterations, enhancing the model's robustness and generalization ability in tasks such as medical diagnosis and disease prediction. Ultimately, this supports the smart healthcare system in achieving more accurate collaborative decision-making and efficient information sharing, improving the overall quality of medical services.

[0030] The above methods and steps can effectively address the challenges mentioned in the background, such as the risk of medical data privacy leakage, the problem of data silos between institutions, and the difficulty of integrating heterogeneous data.

[0031] Specifically, within a federated learning framework, this invention ensures that original medical data remains within the participating institutions through local data training and encrypted parameter transmission mechanisms, preventing the leakage of sensitive information. Simultaneously, it integrates multi-party knowledge using global model aggregation techniques (such as federated averaging algorithms or weighted aggregation methods), significantly improving the accuracy and generalization ability of disease prediction, diagnostic assistance, and health risk assessment, overcoming model performance bottlenecks caused by data fragmentation. Furthermore, the data integration unit, in conjunction with standardized processes (such as HL7 or FHIR format unification) and anonymization processing (such as k-anonymization or differential privacy), efficiently processes heterogeneous data sources, reduces noise interference, optimizes feature input quality, and supports real-time medical information analysis services. The dynamic access control of the security authentication unit (such as PKI-based two-factor authentication and role assignment) further enhances network compliance, adapts to dynamic changes within institutions, and ensures the entire system complies with regulations such as HIPAA or GDPR. Ultimately, this invention achieves efficient and secure sharing of smart healthcare information, promotes cross-institutional collaborative decision-making, and improves the overall quality and accessibility of healthcare services.

[0032] Please see Figure 3 The present invention also provides a smart medical information sharing system based on a federated learning framework, comprising: The central server is configured to build and manage global machine learning models, perform initial generation, distribution, aggregation and update of model parameters, and performance evaluation, while supporting operation audit information logging to meet compliance requirements. Multiple federated learning client modules are deployed in participating medical institutions for communicating with the central server, managing local data, training models and uploading updates. Each client module integrates a data encryption unit, a security authentication unit and a data integration unit. The encrypted communication channel module is implemented based on the Advanced Encryption Standard (AES) or RSA algorithm, combined with the Public Key Infrastructure (PKI) mechanism of the Security Authentication Unit, to ensure end-to-end encryption, data integrity and tamper resistance of model parameters during the download and upload process; The shared platform interface module is used to deploy the trained global model to participating institutions or external systems, and supports real-time medical information analysis, prediction services and model update mechanisms. The data storage module is configured to securely store local medical data, model parameters, and audit logs to ensure data confidentiality and privacy protection. The performance evaluation module, integrated into the central server, uses a reserved validation dataset to calculate accuracy, recall, or F1 score metrics to optimize global model performance.

[0033] The synergistic cooperation of the above modules enables efficient medical data sharing and privacy protection. Specifically, this includes: collaboratively training a global machine learning model across multiple medical institutions without directly exchanging sensitive raw data, thus significantly reducing the risk of privacy leaks; supporting real-time medical information analysis, such as disease prediction, diagnostic assistance, and personalized treatment recommendations, improving the efficiency of clinical decision-making; ensuring the confidentiality, integrity, and tamper resistance of data transmission and storage through end-to-end encryption and security authentication mechanisms, complying with medical data compliance standards such as HIPAA or GDPR; optimizing model performance by periodically adjusting parameters using a performance evaluation module to improve prediction accuracy and robustness, ultimately promoting the overall quality and collaborative level of smart healthcare services.

[0034] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for sharing smart healthcare information based on a federated learning framework, characterized in that, Includes the following steps: S1. Constructing a global machine learning model: The central server builds a global machine learning model based on a preset neural network architecture. After the model is built, the initial parameters are generated randomly or loaded from a pre-trained model. S2. Deploy the Federated Learning Client Module: Collect raw data from multiple medical institutions and add it to the Federated Learning Network for communication with the server, local data management, and model training; S3. Model Parameter Download: At the beginning of each training cycle, the central server sends the current global model parameters to all participating institutions, and the participating institutions download the model parameters through an encrypted channel to avoid the transmission of raw data. S4. Local Model Training: After downloading the model parameters, each participating institution uses local medical data (such as electronic health records and image data) to train the model and performs data preprocessing. During the training process, stochastic gradient descent (SGD) or adaptive optimization algorithms are used to update the local model parameters. S5. Model Update Upload: After training is completed, each participating institution uploads the updated model parameters to the central server through the encrypted channel. The upload process also adopts end-to-end encryption and achieves efficient interaction based on a lightweight communication protocol to reduce network overhead. S6. Global Model Aggregation and Update: After receiving model updates from all participating institutions, the central server integrates the updated parameters, optimizes the global machine learning model, and records the operational audit information during the aggregation process to meet compliance requirements. S7. Iterative Training and Termination: The central server determines whether the global model has converged or reached the preset training round limit. If the conditions are not met, return to S3 for the next round of training; if the conditions are met, output the final global model. S8. Global Model Upload: Deploy the trained global model to participating institutions or shared platforms for real-time medical information analysis and prediction services. It also supports a model update mechanism to adapt to new data or the addition of new institutions.

2. The method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The global machine learning model includes disease prediction, diagnostic assistance, or health risk assessment.

3. The method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The federated learning client module includes a data encryption unit and a security authentication unit. The data encryption unit performs end-to-end encryption of local medical data based on the Advanced Encryption Standard (AES) or RSA algorithm to ensure the confidentiality and integrity of the data during transmission and storage, and to prevent unauthorized access or leakage. The security authentication unit adopts a two-factor authentication mechanism based on Public Key Infrastructure (PKI) to verify the digital identity of each medical institution and manage its access permissions. The security authentication unit also includes user role assignment and operation auditing to ensure the security and compliance of the federated learning network.

4. The method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The federated learning client module also integrates a data integration unit for data integration.

5. The method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The encrypted channel is implemented based on the Advanced Encryption Standard (AES) or RSA algorithm, and combined with the Public Key Infrastructure (PKI) mechanism of the security authentication unit, it ensures the confidentiality, integrity and tamper resistance of data during the parameter download process.

6. The method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The data preprocessing includes the following steps: S41. Data Cleaning: Remove missing values, outliers, and duplicate records from local medical data; automatically detect and correct data errors using preset rules or algorithms. S42. Data anonymization: Apply encryption or anonymization technologies to process sensitive information, such as patient names, ID numbers, and contact information, to ensure that personal privacy protection complies with regulatory requirements. S43. Data standardization: Perform normalization or standardization operations to unify the data formats and scales from different sources, including numerical range adjustment and category coding; S44. Feature Extraction: Extract key features, such as disease indicators or risk factors, from the preprocessed data to support feature input optimization for local model training.

7. The method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The integrated update parameters use either a federated averaging algorithm or a weighted aggregation method. The federated averaging algorithm is based on the following function formula: global model parameters. By weighted averaging the local model parameters of all participating institutions The weights are allocated based on the amount of data each institution has, specifically as follows: In the formula , representing the number of data samples for the i-th institution. This represents the sum of the total data samples from all institutions.

8. A method for sharing smart medical information based on a federated learning framework according to claim 7, characterized in that, The weighted aggregation method adjusts the aggregation ratio based on preset weight coefficients (such as institutional data quality or participation frequency) to ensure fair and efficient model updates.

9. A method for sharing smart medical information based on a federated learning framework according to claim 1, characterized in that, The central server performs performance evaluation on the updated global model, and uses the reserved validation dataset to calculate accuracy, recall or F1 score. The reserved validation dataset is derived from anonymized historical data or independent validation datasets from multiple medical institutions to ensure that the data is independent of the training set and does not overlap.

10. A smart medical information sharing system based on a federated learning framework, characterized in that, include: The central server is configured to build and manage global machine learning models, perform initial generation, distribution, aggregation and update of model parameters, and performance evaluation, while supporting operation audit information logging to meet compliance requirements. Multiple federated learning client modules are deployed in participating medical institutions for communicating with the central server, managing local data, training models and uploading updates. Each client module integrates a data encryption unit, a security authentication unit and a data integration unit. The encrypted communication channel module is implemented based on the Advanced Encryption Standard (AES) or RSA algorithm, combined with the Public Key Infrastructure (PKI) mechanism of the Security Authentication Unit, to ensure end-to-end encryption, data integrity and tamper resistance of model parameters during the download and upload process; The shared platform interface module is used to deploy the trained global model to participating institutions or external systems, and supports real-time medical information analysis, prediction services and model update mechanisms. The data storage module is configured to securely store local medical data, model parameters, and audit logs to ensure data confidentiality and privacy protection. The performance evaluation module, integrated into the central server, uses a reserved validation dataset to calculate accuracy, recall, or F1 score metrics to optimize global model performance.

Images