Self-adaptive privacy security calculation method and system based on medical data feature perception

Abstract

The invention discloses an adaptive privacy security calculation method and system based on medical data feature perception, and relates to the technical field of data privacy protection. The method comprises the steps that a plurality of participating nodes extract structure and semantic features of local medical data, and data feature description information is generated; operating environment parameters and privacy compliance indexes are collected to form node environment state information; the coordination node generates privacy protection strategy configuration based on the information, and determines disturbance intensity, an encryption mode, a data processing permission and a cooperative computing mode; performing differential privacy processing by the participating nodes, generating an intermediate calculation result and submitting the intermediate calculation result to the coordinating node; and the coordination node executes joint aggregation calculation, outputs a target result, and performs privacy risk assessment and strategy dynamic adjustment. The medical data privacy protection method effectively achieves differentiation, controllability and adaptivity of medical data privacy protection, is suitable for various calculation tasks such as federated learning and statistical modeling, and has high practicability and safety.

Description

Technical Field

[0001] The present invention relates to the technical field of data privacy protection, and in particular to an adaptive privacy security calculation method and system based on medical data feature perception. Background Art

[0002] With the rapid development of medical informatization, the sharing and collaborative computing of medical data have become critical requirements in clinical research, public health management, disease prediction, and other fields. Medical data not only contains a large amount of personal identity information but also includes sensitive data related to health conditions. Therefore, how to protect data privacy and prevent the leakage of sensitive information during data sharing and computing has become a pressing technical challenge.

[0003] Existing medical data privacy protection methods are mostly based on traditional encryption techniques or differential privacy algorithms, which protect data privacy by scrambling or encrypting it. However, these methods often suffer from the following issues: The uniformity of privacy protection strategies. Traditional differential privacy algorithms typically set a uniform privacy budget for all data providers, without considering the sensitivity of the data, the compliance requirements of data providers, and the differences in task requirements. This uniform privacy budget setting often results in imprecise privacy protection strength and may not meet the specific requirements of different data and tasks. The balance between privacy protection and computational efficiency is also problematic: In some privacy protection methods, excessive privacy protection strength can lead to excessive noise addition, affecting the accuracy of computational results. On the other hand, too low a protection strength can increase the risk of privacy leakage. Existing technologies lack effective ways to balance privacy protection and computational efficiency. Lack of dynamic feedback mechanisms: Many existing methods lack dynamic evaluation of the effectiveness of privacy protection strategies, nor do they have mechanisms for adjusting and optimizing them. In real-world applications, privacy protection requirements constantly change with changes in computational tasks, differences in data sources, and changes in the computing environment. Therefore, the lack of a dynamic adjustment mechanism makes privacy protection ineffective in adapting to different computing scenarios. Privacy protection and compliance issues: Different data providers have different privacy protection levels and compliance requirements. Existing technologies do not take into account the compliance differences among data providers, resulting in the inability of privacy protection measures to flexibly adapt to the compliance capabilities of each node. Especially in the medical field, data protection compliance is particularly important, and the impact of compliance scores on the strength of privacy protection cannot be ignored. Collaborative computing issues for privacy protection schemes: Most existing privacy protection schemes are suitable for single computing tasks or static computing modes. It is difficult to provide flexible privacy protection schemes for scenarios with cross-task collaboration and different computing modes. Especially in the distributed computing of medical data, the differences in computing power and compliance capabilities of data providers require privacy protection strategies to dynamically adapt to different computing modes.

[0004] Therefore, existing technologies urgently need an innovative privacy protection method that can dynamically adjust privacy protection strategies based on the characteristics of medical data, the environmental parameters of participating nodes, and privacy compliance, and optimize protection strength in real time through a feedback mechanism, while ensuring a balance between computing efficiency and privacy security. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive privacy-safe computing method and system based on medical data feature perception. By introducing mechanisms such as data feature analysis, node environment assessment, and dynamic calculation of a privacy budget, this method addresses existing issues such as the lack of adaptability of privacy protection strategies, the inadequate balance between privacy protection and computational efficiency, and the lack of dynamic feedback regulation. By comprehensively considering data sensitivity, node compliance, and computational task requirements, this invention provides a flexible and adaptive solution for medical data privacy protection, ensuring effective protection of data privacy across diverse computational tasks and optimizing the system's computational performance.

[0006] The present invention achieves the above-mentioned purpose through the following technical solutions: In one aspect, the present invention provides an adaptive privacy-safe computing method based on medical data feature perception, comprising the following steps: S1. Multiple participating nodes extract features from local medical data, identify the data type, structure information, and sensitivity level of the medical data, and generate data feature description information; S2. Multiple participating nodes collect operating environment parameters and privacy compliance indicators, including computing power parameters, network status parameters, and security protection capability parameters. The privacy compliance indicators quantitatively score data protection measures based on preset rules. The nodes then integrate the operating environment parameters and privacy compliance indicators to generate node environmental status information. S3. The coordinating node receives the data feature description information and the node environment status information, and generates a privacy protection policy configuration based on the received information. The privacy protection policy configuration includes: a disturbance intensity parameter, an encryption method identifier, a node data processing authority, and a collaborative computing mode identifier; S4. After receiving the privacy protection policy configuration, multiple participating nodes perform privacy processing on the local medical data according to the disturbance intensity parameter and the encryption method identifier, generate intermediate calculation results, and send the intermediate calculation results to the coordination node; S5. The coordination node performs a joint aggregation calculation on the received intermediate calculation results according to the collaborative computing mode identifier to generate a target calculation result; S6. The coordinating node performs a privacy risk assessment on the target calculation result, collects the performance indicators and security status data of the participating nodes during the collaborative computing process, and updates the privacy protection policy configuration based on the privacy risk assessment results, performance indicators and security status data for a new round of privacy protection policy configuration in the next round of collaborative computing process.

[0007] A further improvement of the present invention is that the multiple participating nodes respectively extract features from local medical data, identify the data type, structure information and sensitivity level of the medical data, and generate data feature description information, specifically including: Based on the original format structure of medical data, identify field boundaries and hierarchical relationships, and extract structural information; Analyze the field name and sample content, call the field semantic annotation process, and generate field semantic attribute information based on the preset semantic label dictionary; Inputting the structural information and semantic attribute information into the sensitivity level classification process, and classifying the field sensitivity into high sensitivity, medium sensitivity and low sensitivity according to predefined rules; The structural information, semantic attributes and sensitivity level of each field are organized into a data field label set, and together with the unique identifiers of the participating nodes, the data feature description information is constructed.

[0008] A further improvement of the present invention is to integrate the operating environment parameters and the privacy compliance indicators to generate node environment status information, specifically including: Collect local system operating parameters, including CPU utilization, available memory capacity, average network transmission rate, and current network stability indicators, to characterize the node's computing and communication capabilities; Evaluate local security configuration, including whether to enable data encryption mechanisms, access control policies, identity authentication mechanisms, and local log auditing functions; Call the privacy compliance scoring process, score the security protection configuration according to the preset compliance scoring rule table, and generate a normalized compliance score value; The computing capacity parameters, network status parameters, security configuration information and the compliance score are integrated to construct structured node environment status information.

[0009] A further improvement of the present invention is to generate a privacy protection policy configuration based on the received information, specifically including: Analyze the field sensitivity level distribution information of each participating node and calculate the corresponding overall data sensitivity coefficient; Read computing resource parameters, network status parameters, and compliance score values ​​from the node environment status information; Based on the preset perturbation parameter generation rules, combined with the overall sensitivity coefficient of the data and the compliance score value, a personalized privacy budget parameter is generated for each node; Determining the corresponding differential privacy perturbation strength based on the privacy budget parameter; Selecting an applicable encryption method identifier based on the network status parameters and the compliance score; Based on the field sensitivity level and compliance score, configure the data processing permission parameters for each node to limit the scope of data that can be processed and the access level; According to the computing resource parameters and the target computing task type, the node collaborative computing mode identification is assigned, and the identification types include federated modeling, secure aggregation and anonymous statistics; The privacy budget parameters, perturbation strength, encryption method identifier, data processing permission parameters generated above are integrated with the collaborative computing mode identifier to construct the corresponding privacy protection policy configuration object and associate it with the unique identifier of the participating node.

[0010] A further improvement of the present invention is that the disturbance intensity parameter is calculated by the personalized privacy budget parameter of each participating node. , the calculation formula is: ; in: For the The node's personalized privacy budget parameters; A baseline privacy budget set for the system; Scoring node compliance; Scoring full marks for compliance; is the sensitivity index of node data; The disturbance intensity parameter is given by Derived and Gaussian differential privacy mechanism is used to calculate the standard deviation of noise injection The expression is: ; in: is the gradient clipping threshold of the node; The differential privacy failure probability parameter set for the system; The disturbance intensity parameter It is used to determine the perturbation intensity injected by participating nodes during the local collaborative computing process and is issued as part of the perturbation intensity parameter in the privacy protection policy configuration.

[0011] A further improvement of the present invention is that the disturbance intensity parameter Applies to one of the following privacy processing procedures: When the collaborative computing mode identifier indicates federated learning modeling, the participating nodes perform model training based on local medical data, clip the local model gradient using a preset gradient clipping threshold, and add Gaussian noise based on the disturbance intensity parameter to generate a disturbed model update parameter as an intermediate calculation result; When the collaborative computing mode identifier indicates statistical aggregation, the participating nodes add Gaussian noise that meets the disturbance intensity parameter to the local statistical indicators, generate a disturbed statistical value, and upload it as an intermediate calculation result; The intermediate calculation result is sent to the coordination node after the disturbance processing, and the intermediate calculation result does not contain original sensitive data or unprotected field content.

[0012] A further improvement of the present invention is that the coordination node performs the following joint aggregation processing on the received intermediate calculation results according to the collaborative computing mode identifier: When the collaborative computing mode is identified as federated modeling, the coordinating node uses a weighted secure aggregation algorithm to aggregate the perturbed model update parameters uploaded by each participating node, and generates updated global model parameters in combination with the current global model state as the target computing result; When the collaborative computing mode is identified as statistical aggregation, the coordinating node performs weighted summation, standardization, or confidence interval merging processing on the post-perturbation statistical indicators uploaded by each participating node, and generates an aggregated statistical analysis result as the target computing result; The joint aggregation process does not involve reverse decryption of intermediate calculation results or sensitive data reconstruction operations.

[0013] A further improvement of the present invention is that, after generating the target calculation result, the coordination node performs the following privacy risk assessment and feedback process: Perform residual analysis of sensitive fields and re-identification risk estimation on the target calculation results, and use a risk inference model under differential privacy to evaluate the potential leakage probability of the output results; Collect performance indicators during collaborative computing, including model convergence speed, computing latency, node upload success rate, and resource utilization; Collect node security status information, including upload behavior logs, abnormal fluctuation patterns, and policy consistency; Based on the evaluation results, the perturbation intensity parameters, encryption method identifiers, and data processing permissions in the privacy protection policy configuration are updated; The updated privacy protection policy configuration is used in the next round of collaborative computing, forming a closed-loop adaptive adjustment mechanism.

[0014] In another aspect, the present invention provides an adaptive privacy-preserving computing system based on medical data feature perception, which utilizes the aforementioned adaptive privacy-preserving computing method based on medical data feature perception. The system comprises: The data feature extraction module is used to extract field structure information and semantic attribute information from local medical data, generate data field sensitivity levels based on predefined rules, and construct structured data feature description information; The environment and compliance assessment module is used to collect the computing resource parameters, network status parameters and security protection configuration of the local node, calculate the normalized compliance score value based on the compliance scoring rule table, and construct the node environment status information; The privacy policy generation module is deployed on the coordination node and is used to receive data feature description information and node environment status information uploaded by each participating node. Based on the perturbation parameter generation rules and task requirements, it generates a personalized privacy protection policy configuration, including perturbation intensity parameters, encryption method identifiers, data processing permissions, and collaborative computing mode identifiers. The privacy protection execution module is deployed on each participating node and is used to perform privacy processing operations on local medical data according to the received privacy protection policy configuration, including differential privacy perturbation, encryption mechanism application and permission restriction, and generate intermediate calculation results and upload them to the coordination node; The federated aggregation computing module is deployed on the coordination node and is used to perform federated modeling or statistical aggregation on multiple intermediate computing results according to the collaborative computing mode to generate the target computing results; The policy feedback and control module is used to perform privacy risk assessment and performance indicator analysis on the target calculation results after each round of collaborative computing, update the privacy protection policy configuration, and realize closed-loop adaptive adjustment of the policy.

[0015] The beneficial effects of the present invention are as follows: the present invention dynamically generates privacy protection strategies through data feature extraction and node environment assessment, thereby achieving differentiated and personalized privacy protection configuration. Compared with the existing technology that adopts a unified static parameter approach, the present invention combines the structural information of medical data, the sensitivity level and the compliance score of participating nodes to generate a policy configuration containing multi-dimensional parameters such as disturbance intensity, encryption method, processing authority and collaborative mode, which significantly improves the pertinence and flexibility of privacy protection. The present invention also introduces a closed-loop risk assessment and policy update mechanism, which can automatically optimize the protection strategy according to the privacy risk and performance indicators after each round of collaborative calculation, and achieve continuous adaptation and dynamic adjustment. This mechanism effectively avoids the degradation of computing performance caused by over-protection, or the risk of privacy leakage caused by insufficient protection, and improves the overall security and efficiency of the system. By combining data sensitivity with node compliance to allocate privacy budgets, the present invention improves the accuracy of model training and statistical analysis while protecting the privacy of medical data. It is applicable to a variety of collaborative computing modes and has good scalability and engineering application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them: Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a system modular diagram of the present invention. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of 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 described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0018] like Figure 1 FIG. 1 is an embodiment of the present invention, which provides an adaptive privacy-safe computing method based on medical data feature perception, including the following steps: Step S1. Multiple participating nodes extract features from local medical data, identify the data type, structure information, and sensitivity level of the medical data, and generate data feature description information; In a preferred embodiment, the plurality of participating nodes respectively extract features from local medical data, identify the data type, structure information and sensitivity level of the medical data, and generate data feature description information, specifically including: Based on the original format structure of medical data, identify field boundaries and hierarchical relationships, and extract structural information; Analyze the field name and sample content, call the field semantic annotation process, and generate field semantic attribute information based on the preset semantic label dictionary; Inputting the structural information and semantic attribute information into the sensitivity level classification process, and classifying the field sensitivity into high sensitivity, medium sensitivity and low sensitivity according to predefined rules; The structural information, semantic attributes and sensitivity level of each field are organized into a data field label set, and together with the unique identifiers of the participating nodes, the data feature description information is constructed.

[0019] This embodiment combines structural boundary identification, semantic attribute annotation, and sensitivity level grading to transform medical data fields into a set of quantifiable data feature labels, which serve as the core input for subsequent policy configuration. Compared to existing approaches that extract sensitive fields based solely on rule templates or keyword filtering, this implementation improves the accuracy of identifying highly sensitive information and the granularity of label generation through structured modeling and field semantic awareness, laying a precise feature foundation for adaptive policy generation.

[0020] Step S2. Multiple participating nodes collect operating environment parameters and privacy compliance indicators. The operating environment parameters include computing power parameters, network status parameters, and security protection capability parameters. The privacy compliance indicators quantitatively score data protection measures based on preset rules. The operating environment parameters and privacy compliance indicators are integrated to generate node environmental status information. In a preferred embodiment, the operating environment parameters and the privacy compliance indicators are integrated to generate node environment status information, specifically including: Collect local system operating parameters, including CPU utilization, available memory capacity, average network transmission rate, and current network stability indicators, to characterize the node's computing and communication capabilities; Evaluate local security configuration, including whether to enable data encryption mechanisms, access control policies, identity authentication mechanisms, and local log auditing functions; Call the privacy compliance scoring process, score the security protection configuration according to the preset compliance scoring rule table, and generate a normalized compliance score value; The computing capacity parameters, network status parameters, security configuration information and the compliance score are integrated to construct structured node environment status information.

[0021] In this embodiment, the node environmental status information generated through the above-mentioned approach not only covers system resource parameters and network status indicators, but also introduces a structured compliance scoring mechanism, comprehensively characterizing the operational capabilities and privacy compliance of each participating node. Compared to the existing approach of static node capability classification based solely on device performance or network bandwidth, this invention incorporates data security configuration and compliance factors into the node status modeling process. This enables the coordinating node to comprehensively consider the dual dimensions of "technical capability + regulatory risk" in subsequent privacy policy generation, improving the accuracy of policy adaptation and risk management capabilities.

[0022] Step S3. The coordinating node receives the data feature description information and the node environment status information, and generates a privacy protection policy configuration based on the received information. The privacy protection policy configuration includes: a disturbance intensity parameter, an encryption method identifier, a node data processing permission, and a collaborative computing mode identifier; In a preferred embodiment, generating a privacy protection policy configuration based on the received information specifically includes: Analyze the field sensitivity level distribution information of each participating node and calculate the corresponding overall data sensitivity coefficient; Read computing resource parameters, network status parameters, and compliance score values ​​from the node environment status information; Based on the preset perturbation parameter generation rules, combined with the overall sensitivity coefficient of the data and the compliance score value, a personalized privacy budget parameter is generated for each node; Determining the corresponding differential privacy perturbation strength based on the privacy budget parameter; Selecting an applicable encryption method identifier based on the network status parameters and the compliance score; Based on the field sensitivity level and compliance score, configure the data processing permission parameters for each node to limit the scope of data that can be processed and the access level; According to the computing resource parameters and the target computing task type, the node collaborative computing mode identification is assigned, and the identification types include federated modeling, secure aggregation and anonymous statistics; The privacy budget parameters, perturbation strength, encryption method identifier, data processing permission parameters generated above are integrated with the collaborative computing mode identifier to construct the corresponding privacy protection policy configuration object and associate it with the unique identifier of the participating node.

[0023] In this embodiment, the privacy protection policy configuration generated by the above method not only includes personalized disturbance strength and encryption method, but also covers multi-dimensional control parameters such as data processing permissions and calculation mode identification, forming a complete policy object that is structured, executable, and can be issued. Compared with the practice of setting policies only for a single privacy budget or a unified encryption method in the prior art, the present invention introduces a joint evaluation mechanism for data sensitivity and node compliance scores at the policy generation level, which can generate differentiated policy configurations for different participating nodes, fully reflecting the dynamic privacy management concept of "different for data, different for node". In addition, the present invention adopts a modular policy configuration structure, which lays a scalable foundation for automatic execution and policy feedback updates in subsequent calculation processes.

[0024] Step S4. After receiving the privacy protection policy configuration, the multiple participating nodes perform privacy processing on the local medical data according to the disturbance intensity parameter and the encryption method identifier, generate intermediate calculation results, and send the intermediate calculation results to the coordination node; In a preferred embodiment, the disturbance intensity parameter is calculated by the personalized privacy budget parameter of each participating node. , the calculation formula is: ; in: For the The node's personalized privacy budget parameters; A baseline privacy budget set for the system; Scoring node compliance; Scoring full marks for compliance; is the sensitivity index of node data; The disturbance intensity parameter is given by Derived and Gaussian differential privacy mechanism is used to calculate the standard deviation of noise injection The expression is: ; in: is the gradient clipping threshold of the node; The differential privacy failure probability parameter set for the system; The disturbance intensity parameter It is used to determine the perturbation intensity injected by participating nodes during the local collaborative computing process and is issued as part of the perturbation intensity parameter in the privacy protection policy configuration.

[0025] The disturbance intensity parameter Applies to one of the following privacy processing procedures: When the collaborative computing mode identifier indicates federated learning modeling, the participating nodes perform model training based on local medical data, clip the local model gradient using a preset gradient clipping threshold, and add Gaussian noise based on the disturbance intensity parameter to generate a disturbed model update parameter as an intermediate calculation result; When the collaborative computing mode identifier indicates statistical aggregation, the participating nodes add Gaussian noise that meets the disturbance intensity parameter to the local statistical indicators, generate a disturbed statistical value, and upload it as an intermediate calculation result; The intermediate calculation result is sent to the coordination node after the disturbance processing, and the intermediate calculation result does not contain original sensitive data or unprotected field content.

[0026] In this embodiment, through the above-mentioned method, the perturbation intensity parameter realizes personalized calculation under the dual constraints of participating node compliance and data sensitivity, and is linked with the node local model gradient clipping mechanism, combined with the standard derivation formula of the Gaussian mechanism in differential privacy theory, to accurately control the noise injection amplitude. Unlike the scheme of adopting a unified perturbation parameter or static privacy budget setting in the prior art, the present invention realizes the fine matching of perturbation intensity and node risk level through mathematical modeling, and improves the system's dynamic balance ability between protection strength and calculation accuracy. In addition, combined with the collaborative computing mode identification, the privacy perturbation processing flow realizes flexible adaptation of the processing path at the task level, ensuring that the privacy protection and validity control of the intermediate results can be achieved in both typical tasks of model training and statistical aggregation.

[0027] Step S5. The coordination node performs a joint aggregation calculation on the received intermediate calculation results according to the collaborative computing mode identifier to generate a target calculation result; In a preferred embodiment, the coordination node performs the following joint aggregation processing on the received intermediate calculation results according to the collaborative computing mode identifier: When the collaborative computing mode is identified as federated modeling, the coordinating node uses a weighted secure aggregation algorithm to aggregate the perturbed model update parameters uploaded by each participating node, and generates updated global model parameters in combination with the current global model state as the target computing result; When the collaborative computing mode is identified as statistical aggregation, the coordinating node performs weighted summation, standardization, or confidence interval merging processing on the post-perturbation statistical indicators uploaded by each participating node, and generates an aggregated statistical analysis result as the target computing result; The joint aggregation process does not involve reverse decryption of intermediate calculation results or sensitive data reconstruction operations.

[0028] In this embodiment, through the above-mentioned joint aggregation method, the coordination node can complete the global merging of model update parameters or statistical indicators without decrypting the original data or intermediate contribution values ​​of each participating node, ensuring the realization of cross-node collaborative computing goals while protecting the data privacy of each node. Compared with the existing methods that rely on plaintext aggregation or merging models through trusted third parties, the present invention completes global output without destroying the disturbance intensity protection boundary through the "disturbance protection + pattern guidance" processing mechanism, thereby improving privacy security and system reliability. Under the guidance of pattern recognition, the coordination node can automatically adapt the aggregation strategy according to the task type, thereby improving the versatility and adaptability of the system in multi-task collaborative computing scenarios.

[0029] Step S6. The coordinating node performs a privacy risk assessment on the target calculation result, and collects the performance indicators and security status data of the participating nodes during the collaborative computing process. The coordinating node updates the privacy protection policy configuration based on the privacy risk assessment results, performance indicators and security status data, for use in a new round of privacy protection policy configuration for the next round of collaborative computing process.

[0030] In a preferred embodiment, after generating the target calculation result, the coordination node performs the following privacy risk assessment and feedback process: Perform residual analysis of sensitive fields and re-identification risk estimation on the target calculation results, and use a risk inference model under differential privacy to evaluate the potential leakage probability of the output results; Collect performance indicators during collaborative computing, including model convergence speed, computing latency, node upload success rate, and resource utilization; Collect node security status information, including upload behavior logs, abnormal fluctuation patterns, and policy consistency; Based on the evaluation results, the perturbation intensity parameters, encryption method identifiers, and data processing permissions in the privacy protection policy configuration are updated; The updated privacy protection policy configuration is used in the next round of collaborative computing, forming a closed-loop adaptive adjustment mechanism.

[0031] In this embodiment, through the aforementioned feedback process, the coordination node can not only dynamically assess the privacy risks of the output results, but also comprehensively analyze the performance and security situation during task execution, thereby optimizing and updating policy elements such as perturbation intensity, encryption method, and data permissions in real time, achieving continuous policy evolution and risk-driven control. Compared to the existing approach that only sets privacy protection configuration based on initial task parameters and lacks a feedback mechanism, this invention introduces a closed-loop logic of "output evaluation + status monitoring + policy adjustment", which enables the system to have continuous self-adaptation capabilities. It can actively adjust based on the historical operation status of collaborative computing and the risk inferred from sensitive information, thereby improving the long-term stability of the system and the targeted data protection.

[0032] like Figure 2 FIG. 1 is another embodiment of the present invention, which provides an adaptive privacy-safe computing system based on medical data feature perception, and applies the adaptive privacy-safe computing method based on medical data feature perception described above. The system includes: The data feature extraction module is used to extract field structure information and semantic attribute information from local medical data, generate data field sensitivity levels based on predefined rules, and construct structured data feature description information; The environment and compliance assessment module is used to collect the computing resource parameters, network status parameters and security protection configuration of the local node, calculate the normalized compliance score value based on the compliance scoring rule table, and construct the node environment status information; The privacy policy generation module is deployed on the coordination node and is used to receive data feature description information and node environment status information uploaded by each participating node. Based on the perturbation parameter generation rules and task requirements, it generates a personalized privacy protection policy configuration, including perturbation intensity parameters, encryption method identifiers, data processing permissions, and collaborative computing mode identifiers. The privacy protection execution module is deployed on each participating node and is used to perform privacy processing operations on local medical data according to the received privacy protection policy configuration, including differential privacy perturbation, encryption mechanism application and permission restriction, and generate intermediate calculation results and upload them to the coordination node; The federated aggregation computing module is deployed on the coordination node and is used to perform federated modeling or statistical aggregation on multiple intermediate computing results according to the collaborative computing mode to generate the target computing results; The policy feedback and control module is used to perform privacy risk assessment and performance indicator analysis on the target calculation results after each round of collaborative computing, update the privacy protection policy configuration, and realize closed-loop adaptive adjustment of the policy.

[0033] In summary, this invention, by constructing an adaptive privacy-safe computing architecture based on medical data feature perception, achieves a closed-loop dynamic process from data sensitivity identification, node capability assessment, privacy policy generation, local disturbance processing, joint aggregate computing, and policy feedback updates. This technical solution fully integrates multiple mechanisms such as differential privacy, distributed computing, compliance scoring, and multi-dimensional policy scheduling. While ensuring data privacy and security, it also takes into account the performance requirements of collaborative computing and engineering deployability, significantly improving the accuracy, flexibility, and intelligence of privacy protection.

[0034] This invention can not only be widely used in highly sensitive computing scenarios such as medical data sharing, clinical joint modeling, and remote health analysis, but also has good scalability and platform adaptability, and can serve the privacy collaborative computing needs between institutions of different sizes. It has significant practical application value and industrial promotion prospects.

[0035] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0036] Any process or method description in the flowchart or otherwise described herein can be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process. The scope of the preferred embodiments of the present application includes additional implementations in which the functions may be performed in a different order than shown or discussed, including in a substantially simultaneous manner or in a reverse order depending on the functions involved.

[0037] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An adaptive privacy-safe computing method based on medical data feature perception, characterized in that: The following steps are involved: S1. Multiple participating nodes extract features from local medical data, identify the data type, structure information, and sensitivity level of the medical data, and generate data feature description information; S2. Multiple participating nodes collect operating environment parameters and privacy compliance indicators, including computing power parameters, network status parameters, and security protection capability parameters. The privacy compliance indicators quantitatively score data protection measures based on preset rules. The nodes then integrate the operating environment parameters and privacy compliance indicators to generate node environmental status information. S3. The coordinating node receives the data feature description information and the node environment status information, and generates a privacy protection policy configuration based on the received information. The privacy protection policy configuration includes: a disturbance intensity parameter, an encryption method identifier, a node data processing authority, and a collaborative computing mode identifier; S4. After receiving the privacy protection policy configuration, multiple participating nodes perform privacy processing on the local medical data according to the disturbance intensity parameter and the encryption method identifier, generate intermediate calculation results, and send the intermediate calculation results to the coordination node; S5. The coordination node performs a joint aggregation calculation on the received intermediate calculation results according to the collaborative computing mode identifier to generate a target calculation result; S6. The coordinating node performs a privacy risk assessment on the target calculation result, collects the performance indicators and security status data of the participating nodes during the collaborative computing process, and updates the privacy protection policy configuration based on the privacy risk assessment results, performance indicators and security status data for a new round of privacy protection policy configuration in the next round of collaborative computing process.

2. The adaptive privacy-safe computing method based on medical data feature perception according to claim 1, characterized in that: In step S1, the plurality of participating nodes respectively extract features from the local medical data, identify the data type, structure information and sensitivity level of the medical data, and generate data feature description information, specifically including: Based on the original format structure of medical data, identify field boundaries and hierarchical relationships, and extract structural information; Analyze the field name and sample content, call the field semantic annotation process, and generate field semantic attribute information based on the preset semantic label dictionary; Inputting the structural information and semantic attribute information into the sensitivity level classification process, and classifying the field sensitivity into high sensitivity, medium sensitivity and low sensitivity according to predefined rules; The structural information, semantic attributes and sensitivity level of each field are organized into a data field label set, and together with the unique identifiers of the participating nodes, the data feature description information is constructed.

3. The adaptive privacy-safe computing method based on medical data feature perception according to claim 1, characterized in that: In step S2, the operating environment parameters and the privacy compliance indicators are integrated to generate node environment status information, which specifically includes: Collect local system operating parameters, including CPU utilization, available memory capacity, average network transmission rate, and current network stability indicators, to characterize the node's computing and communication capabilities; Evaluate local security configuration, including whether to enable data encryption mechanisms, access control policies, identity authentication mechanisms, and local log auditing functions; Call the privacy compliance scoring process, score the security protection configuration according to the preset compliance scoring rule table, and generate a normalized compliance score value; The computing capacity parameters, network status parameters, security configuration information and the compliance score are integrated to construct structured node environment status information.

4. The adaptive privacy-safe computing method based on medical data feature perception according to claim 1, characterized in that: In step S3, a privacy protection policy configuration is generated based on the received information, specifically including: Analyze the field sensitivity level distribution information of each participating node and calculate the corresponding overall data sensitivity coefficient; Read computing resource parameters, network status parameters, and compliance score values ​​from the node environment status information; Based on the preset perturbation parameter generation rules, combined with the overall sensitivity coefficient of the data and the compliance score value, a personalized privacy budget parameter is generated for each node; Determining the corresponding differential privacy perturbation strength based on the privacy budget parameter; Selecting an applicable encryption method identifier based on the network status parameters and the compliance score; Based on the field sensitivity level and compliance score, configure the data processing permission parameters for each node to limit the scope of data that can be processed and the access level; According to the computing resource parameters and the target computing task type, the node collaborative computing mode identification is assigned, and the identification types include federated modeling, secure aggregation and anonymous statistics; The privacy budget parameters, perturbation strength, encryption method identifier, data processing permission parameters generated above are integrated with the collaborative computing mode identifier to construct the corresponding privacy protection policy configuration object and associate it with the unique identifier of the participating node.

5. The adaptive privacy-safe computing method based on medical data feature perception according to claim 4, characterized in that: In step S4, the disturbance intensity parameter is calculated by the personalized privacy budget parameter of each participating node. , the calculation formula is: ； in: For the The node's personalized privacy budget parameters; A baseline privacy budget set for the system; Scoring node compliance; Scoring full marks for compliance; is the sensitivity index of node data; The disturbance intensity parameter is given by Derived and Gaussian differential privacy mechanism is used to calculate the standard deviation of noise injection The expression is: ； in: is the gradient clipping threshold of the node; The differential privacy failure probability parameter set for the system; The disturbance intensity parameter It is used to determine the perturbation intensity injected by participating nodes during the local collaborative computing process and is issued as part of the perturbation intensity parameter in the privacy protection policy configuration.

6. The adaptive privacy-safe computing method based on medical data feature perception according to claim 5, characterized in that: The disturbance intensity parameter Applies to one of the following privacy processing procedures: When the collaborative computing mode identifier indicates federated learning modeling, the participating nodes perform model training based on local medical data, clip the local model gradient using a preset gradient clipping threshold, and add Gaussian noise based on the disturbance intensity parameter to generate a disturbed model update parameter as an intermediate calculation result; When the collaborative computing mode identifier indicates statistical aggregation, the participating nodes add Gaussian noise that meets the disturbance intensity parameter to the local statistical indicators, generate a disturbed statistical value, and upload it as an intermediate calculation result; The intermediate calculation result is sent to the coordination node after the disturbance processing, and the intermediate calculation result does not contain original sensitive data or unprotected field content.

7. The adaptive privacy-safe computing method based on medical data feature perception according to claim 6, characterized in that: In step S5, the coordination node performs the following joint aggregation processing on the received intermediate calculation results according to the collaborative computing mode identifier: When the collaborative computing mode is identified as federated modeling, the coordinating node uses a weighted secure aggregation algorithm to aggregate the perturbed model update parameters uploaded by each participating node, and generates updated global model parameters in combination with the current global model state as the target computing result; When the collaborative computing mode is identified as statistical aggregation, the coordinating node performs weighted summation, standardization, or confidence interval merging processing on the post-perturbation statistical indicators uploaded by each participating node, and generates an aggregated statistical analysis result as the target computing result; The joint aggregation process does not involve reverse decryption of intermediate calculation results or sensitive data reconstruction operations.

8. The adaptive privacy-safe computing method based on medical data feature perception according to claim 1, characterized in that: In step S6, after generating the target calculation result, the coordination node performs the following privacy risk assessment and feedback process: Perform residual analysis of sensitive fields and re-identification risk estimation on the target calculation results, and use a risk inference model under differential privacy to evaluate the potential leakage probability of the output results; Collect performance indicators during collaborative computing, including model convergence speed, computing latency, node upload success rate, and resource utilization; Collect node security status information, including upload behavior logs, abnormal fluctuation patterns, and policy consistency; Based on the evaluation results, the perturbation intensity parameters, encryption method identifiers, and data processing permissions in the privacy protection policy configuration are updated; The updated privacy protection policy configuration is used in the next round of collaborative computing, forming a closed-loop adaptive adjustment mechanism.

9. An adaptive privacy-protected computing system based on medical data feature perception, applying the adaptive privacy-protected computing method based on medical data feature perception as claimed in any one of claims 1 to 8, characterized in that: The system comprises: The data feature extraction module is used to extract field structure information and semantic attribute information from local medical data, generate data field sensitivity levels based on predefined rules, and construct structured data feature description information; The environment and compliance assessment module is used to collect the computing resource parameters, network status parameters and security protection configuration of the local node, calculate the normalized compliance score value based on the compliance scoring rule table, and construct the node environment status information; The privacy policy generation module is deployed on the coordination node and is used to receive data feature description information and node environment status information uploaded by each participating node. Based on the perturbation parameter generation rules and task requirements, it generates a personalized privacy protection policy configuration, including perturbation intensity parameters, encryption method identifiers, data processing permissions, and collaborative computing mode identifiers. The privacy protection execution module is deployed on each participating node and is used to perform privacy processing operations on local medical data according to the received privacy protection policy configuration, including differential privacy perturbation, encryption mechanism application and permission restriction, and generate intermediate calculation results and upload them to the coordination node; The federated aggregation computing module is deployed on the coordination node and is used to perform federated modeling or statistical aggregation on multiple intermediate computing results according to the collaborative computing mode to generate the target computing results; The policy feedback and control module is used to perform privacy risk assessment and performance indicator analysis on the target calculation results after each round of collaborative computing, update the privacy protection policy configuration, and realize closed-loop adaptive adjustment of the policy.

Images