Intelligent medical big data privacy desensitization methods, systems, electronic devices, and storage media
By constructing a field-level directed acyclic graph of data lineage and distributed edge execution, the dynamic adaptability and high concurrency performance issues of compliant data anonymization in medical big data in existing technologies are solved, realizing automated management and efficient anonymization of sensitive data.
Patent Information
- Application Number
- CN202611232525.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies suffer from problems such as rigid architecture, poor dynamic adaptability, limited concurrent performance, high operation and maintenance costs, and hard-coded and solidified desensitization logic in the process of cross-domain sharing and compliant desensitization of medical big data. They cannot meet the high-concurrency, dynamic, and compliant desensitization requirements of multi-source heterogeneous medical big data.
A field-level directed acyclic graph of data lineage is constructed using AST hierarchical syntax parsing and DAG incremental coloring propagation algorithm. Atomic desensitization operators are dynamically assembled through three-dimensional nonlinear weighted calculation and distributed edge local execution is performed to achieve automatic perception and fine-grained control of sensitive data.
It enables automated perception and real-time updates of sensitive data, eliminates manual operation and maintenance costs and blind spots in data anonymization protection, increases the throughput limit of data flow, reduces operation and maintenance and iteration costs, and balances data security and business availability.
Smart Images

Figure CN123069418A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the technical field of medical big data processing, specifically relating to an intelligent medical big data privacy data desensitization method, system, electronic device, and storage medium. Background Technology
[0002] In scenarios involving cross-domain sharing, scientific research analysis, and business operations of medical big data, compliant desensitization of multi-source, heterogeneous patient privacy data and sensitive medical data is a rigid compliance requirement for the circulation of data elements. With the deepening of medical informatization and the rapid development of health and medical big data applications, medical institutions generate and circulate massive amounts of sensitive data containing patients' personal privacy information in daily diagnosis and treatment, scientific research collaboration, and cross-institutional data sharing scenarios. How to ensure the normal circulation and use of data while ensuring that privacy data complies with the compliance requirements of laws, regulations, and industry standards such as the *Data Security Law*, the *Personal Information Protection Law*, and the *Standardization Maturity Assessment Scheme for Information Interconnection and Interoperability of Medical Institutions* has become a core issue that urgently needs to be addressed in the field of medical big data governance. Existing data desensitization technologies include four technical routes.
[0003] The first category is static view permission masking technology based on RBAC roles. This technology relies on the application-layer permission system to build a two-dimensional mapping relationship between "user roles and data fields," and outputs masked data to different roles through fixed masking rules. The second category is fixed batch masking technology based on ETL tasks. This technology, during the data aggregation and ETL cleaning stages, uses manually written masking SQL scripts and custom UDF functions to perform unified fixed masking on all data and then persistently stores it, belonging to a "single masking, permanent effect" static processing mode. The third category is regular expression batch masking technology based on a unified gateway. This technology uses proxy gateways such as Nginx and APISIX as the processing core, and uses preset regular expressions to perform string replacement masking on plaintext request and response messages flowing through the gateway. All data streams must be centrally processed through the central gateway. The fourth category is static tag cascading masking technology based on metadata. This technology relies on the data asset catalog to configure fixed sensitive tags for source sensitive data tables and fields, and synchronizes the tags to downstream derived data through static replication.
[0004] However, these four technical approaches generally suffer from common shortcomings such as rigid architecture, poor dynamic adaptability, limited concurrent performance, and high operation and maintenance costs. They cannot support the full-link, dynamic, and high-concurrency compliance masking requirements of multi-source heterogeneous medical big data, which manifests in four core defects. Defect one is that the two-dimensional static control model cannot adapt to the dynamic scenario of the entire link. Traditional masking technology only relies on the two-dimensional dimension of "subject user - object data" to build control strategies, without incorporating the key risk variable of data physical flow scenarios (such as internal development and debugging, internal verification and preview, external API sharing, cross-institutional collaboration, etc.). The leakage risk of the same sensitive data varies significantly in different links, but existing technologies can only adopt uniform masking rules, resulting in the coexistence of problems such as excessive masking damaging business availability and insufficient masking causing compliance risks. At the same time, the resulting siloed single-point protection architecture has low strategy reuse efficiency. The second defect is that manual static labeling creates gaps in desensitization during lineage iteration. In existing technologies, sensitive data identification, tag binding, and strategy association all rely on manual maintenance. During data processing, after SQL operations such as JOIN, UNION, field alias replacement, and table splitting and reorganization, newly generated derived fields cannot automatically inherit the source sensitive attributes, and the original static sensitive tags cannot be automatically synchronized to the derived data, creating large-scale desensitization protection blind spots. Manual relabeling is inefficient, has high overall maintenance costs, and poses a significant risk of privacy leakage. The third defect is that centralized gateway architectures have performance bottlenecks and stability risks. Gateway-based desensitization solutions use a centralized processing architecture. All plaintext medical data streams must be completely transmitted to the central gateway for desensitization before flowing back to the business system. In scenarios with high concurrency flow of massive amounts of medical big data, this generates high network I / O overhead, easily causing link blockage and interface response timeouts. Centralized computation of large-scale data can easily lead to server memory overflow, significantly reducing the data flow throughput limit and failing to support large-scale medical big data element circulation business. The fourth drawback is that the de-identification logic is hard-coded and fixed, resulting in high costs for cross-platform adaptation and iteration. Traditional de-identification rules are fixed in the form of hard-coded underlying code, and the de-identification operators, execution order, and de-identification strength cannot be dynamically adjusted. For heterogeneous business components such as data collection, offline development, online preview, and cross-domain sharing, separate development of adapted de-identification logic is required, resulting in a large amount of repetitive development work. Strategy iteration and rule updates require modification of the underlying code and service restart, resulting in high operation and maintenance costs and insufficient business flexibility. Summary of the Invention
[0005] To address one or more of the aforementioned problems, this application provides a first aspect of a method for desensitizing privacy data in intelligent medical big data, a second aspect of an intelligent system for desensitizing privacy data in intelligent medical big data, a third aspect of an electronic device, and a fourth aspect of a storage medium for efficiently desensitizing privacy data.
[0006] The technical solution of this application is as follows.
[0007] In a first aspect, this application provides a method for desensitizing privacy data in intelligent medical big data, comprising acquiring multi-source heterogeneous medical big data processing scripts and receiving user data requests; further comprising: Perform AST hierarchical syntax parsing on medical big data processing scripts to construct corresponding field-level data lineage directed acyclic graphs. Using the original sensitive field as the topological root node of the directed acyclic graph, the DAG incremental coloring propagation algorithm is used to propagate the sensitive attributes and sensitivity levels to the downstream derived fields along the directed path of the directed acyclic graph; Based on the user's data request, reverse tracing of the directed acyclic graph from the target field is performed to obtain the sensitivity level of the corresponding object asset. The sensitivity level of the object assets is integrated with the security level of the main user extracted from the user's data request and the risk coefficient of the data physical flow process. A three-dimensional nonlinear weighted calculation is then performed to obtain the desensitization severity index. Atom desensitization operators are dynamically assembled based on the desensitization severity index to form a desensitization execution tree; The de-identification execution tree is sent down to the edge nodes of the directed acyclic graph, where the edge nodes perform zero-copy de-identification operations and output compliant de-identified data.
[0008] As one implementation method of the first aspect, AST layered syntax parsing is performed on the medical big data processing script, including: Perform one or more of the following on medical big data processing scripts: lexical analysis, syntax tree construction, semantic parsing, and automatic conflict correction. Extract the relationships between field mappings, table associations, and field derivation transformations, and output the field-level lineage topology edge structure.
[0009] As one implementation of the first aspect, sensitive attributes and sensitivity levels are propagated downstream along directed paths in a directed acyclic graph to derived fields, including: Perform a depth-first traversal of the directed acyclic graph so that downstream derived fields inherit the sensitivity level parameters of the original sensitive fields; When the topology of a directed acyclic graph changes, incremental topology change detection is used to perform local recoloring only on the subgraphs affected by the change.
[0010] As one implementation of the first aspect, reverse ancestry tracing is performed by traversing the directed acyclic graph backwards from the target field, including: Starting from the target field, search upstream parent nodes along the directed edges of the directed acyclic graph until the root node of the topology. Read the sensitivity levels of pre-marked object assets in the topology root node and return them to the retrieval starting point.
[0011] As one implementation of the first aspect, a three-dimensional nonlinear weighted calculation is performed to obtain the desensitization severity index, including: The DAG reverse lineage tracing algorithm is used to obtain the sensitivity level of the object asset. The security level of the main user is extracted from the user's data request using a link probe. Link probes are used to extract risk factors of the physical flow of data from user data requests; The desensitization severity index is obtained by substituting the sensitivity level of the object asset, the security level of the subject user, and the risk coefficient of the data physical flow process into the three-dimensional nonlinear weighted calculation formula. The formula for three-dimensional nonlinear weighted calculation is: D index = max(0, ω1·G asset - ω2·C user + ω3·S type ); Among them, D index G is the desensitization severity index. asset For the sensitivity level of the object assets, C user For the primary user security level, S type ω1, ω2, and ω3 are the risk coefficients for the physical flow of data, and ω1 + ω2 + ω3 = 1.
[0012] As one implementation of the first aspect, the atomic desensitization operators are dynamically assembled according to the desensitization severity index to form a desensitization execution tree, including: Basic atomic desensitization operators are matched from the atomic desensitization operator pool based on the desensitization severity index; Dependency topology sorting based on operator execution constraints; Construct a tree-like execution structure with hierarchical dependencies and execution priorities; The tree-shaped execution structure is serialized and compressed to obtain the de-identified execution tree.
[0013] As one implementation of the first aspect, the edge node performs zero-copy de-identification operations and outputs compliant de-identified data, including: It takes over the local data stream's original memory pointer of multi-source heterogeneous medical big data, directly performs atomic de-identification operations in the local original memory address, and outputs a compliant de-identified data stream.
[0014] Secondly, an intelligent medical big data privacy data desensitization system includes an acquisition module for obtaining multi-source heterogeneous medical big data processing scripts and receiving user data requests; it also includes: The lineage dynamic transmission module is used to perform AST layered syntax parsing on medical big data processing scripts and construct corresponding field-level data lineage directed acyclic graphs. Using the original sensitive fields as the topological root nodes of the directed acyclic graph, the DAG incremental coloring transmission algorithm is used to transmit sensitive attributes and sensitivity levels to downstream derived fields along the directed path of the directed acyclic graph. The three-dimensional matrix decision module is used to trace the lineage of objects by traversing the directed acyclic graph in reverse from the target field based on the response to the user's data request, and obtain the sensitivity level of the object assets corresponding to the target field; it integrates the sensitivity level of the object assets with the security level of the main user and the risk coefficient of the data physical flow process extracted from the user's data request, and performs three-dimensional nonlinear weighted calculation to obtain the desensitization severity index. The distributed execution control module is used to dynamically assemble atomic desensitization operators to form a desensitization execution tree based on the desensitization severity index; the desensitization execution tree is then distributed to the edge nodes of the directed acyclic graph, where the edge nodes perform zero-copy desensitization operations and output compliant desensitized data.
[0015] Thirdly, an electronic device includes a memory, a processor, and a computer program stored on the memory and running on the processor, wherein the processor executes the computer program to implement the intelligent medical big data privacy data desensitization method of the first aspect.
[0016] Fourthly, a storage medium is a computer-readable storage medium storing computer instructions for enabling a computer to implement the intelligent medical big data privacy data desensitization method of the first aspect.
[0017] Compared with the prior art, the advantages of this application are as follows: It can automatically detect sensitive privacy data, eliminating manual operation and maintenance costs and blind spots in de-identification protection; it can achieve refined de-identification management in multiple scenarios; and it avoids the high-concurrency performance bottleneck of centralized gateways through distributed edge local execution. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating one of the intelligent medical big data privacy data desensitization methods of this application.
[0019] Figure 2 This is a schematic diagram of the structural framework of one of the intelligent medical big data privacy data desensitization systems in this application.
[0020] Figure 3 This is a schematic diagram of the frame of one of the electronic devices in this application. Detailed Implementation
[0021] Referring to the illustrations, the principles of this application are illustrated by way of example implementation in a suitable operating environment. The following description is based on the illustrative specific embodiments of this application and should not be construed as limiting other specific embodiments not detailed herein, such as corresponding adjustments to the order of steps of the method of this application based on the principles of the technical concept.
[0022] It should be noted that this application only standardizes the names of examination items in the medical institution's examination data, and does not involve sensitive medical big data such as patient personal identification information, medical record content, and diagnostic conclusions. It complies with the relevant requirements of laws, regulations and ethical norms regarding de-identification and minimal processing of personal information.
[0023] In existing technologies, two-dimensional static management and control models cannot adapt to dynamic scenarios across the entire chain, manual static labeling causes desensitization gaps during lineage iteration, centralized gateway architectures have performance bottlenecks and stability risks, and hard-coded desensitization logic has high costs for cross-platform adaptation and iteration.
[0024] like Figure 1 As shown in the figure, this application discloses a method for desensitizing privacy data in intelligent medical big data, which includes the following steps.
[0025] S1. Obtain multi-source heterogeneous medical big data processing scripts and receive user data requests.
[0026] In this embodiment, the medical big data processing script comes from multi-source heterogeneous medical big data raw data tables, which can be obtained by accessing multi-source heterogeneous medical data information systems such as HIS / EMR / LIS / PACS / RIS.
[0027] Optionally, in this embodiment, a distributed, lightweight link probe is used to deploy non-intrusive front-end probes and transaction listeners on nodes such as data acquisition terminals, offline data development platforms, real-time computing engines, data preview terminals, external API sharing gateways, and cross-domain data synchronization nodes. This enables the acquisition of medical big data processing scripts and data requests. The distributed, lightweight link probe is based on a bypass listening mechanism, which does not consume the main business link's computing power or modify the original data processing flow. It can capture real-time transaction behaviors such as data read / write requests, SQL processing tasks, field changes, data export, interface calls, and cross-node data synchronization. It has the capabilities of request message parsing, scenario parameter extraction, user identity verification, and data flow feature capture. It can standardize and encapsulate five-tuple request parameters containing user identity, permission level, data asset identifier, flow scenario, and data flow characteristics, providing complete and real-time raw data support for further data processing at the upper layer and enabling automatic perception of data flow.
[0028] S2. Perform AST layered syntax parsing on the medical big data processing scripts and construct the corresponding field-level data lineage directed acyclic graph.
[0029] In this embodiment, the AST hierarchical parsing of the medical big data processing script includes one or more of the following: lexical analysis, syntax tree construction, semantic parsing, and automatic conflict correction. From this, the relationships of field mapping, table association, and field derivation transformation in the original data tables of multi-source heterogeneous medical big data are extracted, and the field-level lineage topology edge structure is output.
[0030] Optionally, in this embodiment, the script's lexical, syntactic, and semantic structures can be parsed layer by layer using AST layered syntax parsing technology. This allows for the extraction of processing behaviors such as data table creation, field addition, deletion, renaming, table JOIN, field aggregation, data filtering, alias mapping, and table splitting and merging. This can overcome the limitations of traditional coarse-grained table-level lineage and thus construct a field-level traceable DAG topology.
[0031] In this embodiment, optionally, lineage verification, conflict detection, and automatic completion of missing links can be performed to correct deviations in AST hierarchical syntax parsing, and then standardized structured field-level DAG lineage topology data can be output to provide a standardized data foundation for sensitive field pair coloring, reverse tracing, and risk assessment.
[0032] S3. Using the original sensitive field as the topological root node of the directed acyclic graph, the DAG incremental coloring propagation algorithm is used to propagate the sensitive attributes and sensitivity levels to the downstream derived fields along the directed path of the directed acyclic graph.
[0033] In this embodiment, sensitive attributes and sensitivity levels are transmitted to downstream derived fields along the directed path of the directed acyclic graph. This includes performing a depth-first traversal of the directed acyclic graph so that the downstream derived fields inherit the sensitivity level parameters of the original sensitive fields. When the topology of the directed acyclic graph changes, the topology change is detected through incremental topology change detection, and local recoloring is performed only on the subgraphs involved in the change.
[0034] Optionally, in this embodiment, the original sensitive field is used as the root node of the topology. Based on the preset five-level sensitivity level standard, the entire domain is traversed along the directed path of the directed acyclic graph. The sensitive attributes and sensitivity levels of the original sensitive field are automatically transmitted to the downstream derived fields, aggregate fields, alias fields, etc., thereby updating the metadata in real time. After processing and iteration, the downstream data can be updated with sensitive attributes and sensitivity levels synchronously without manual intervention, thereby eliminating the desensitization gap and privacy leakage blind spots at the underlying level.
[0035] S4. Based on the user's data request, reverse the tracing of the directed acyclic graph from the target field to obtain the sensitivity level of the corresponding object asset.
[0036] In this embodiment, reverse tracing of the directed acyclic graph from the target field is performed. This includes starting from the target field, searching upstream parent nodes along the directed edges of the directed acyclic graph until the topological root node, and then reading the pre-marked object asset sensitivity level in the topological root node and sending it back to the search starting point.
[0037] S5. The sensitivity level of the object asset is integrated with the security level of the main user extracted from the user's data request and the risk coefficient of the data physical flow link, and a three-dimensional nonlinear weighted calculation is performed to obtain the desensitization severity index.
[0038] In this embodiment, a three-dimensional nonlinear weighted calculation is performed to obtain the desensitization severity index. This includes calling the DAG reverse lineage tracing algorithm to obtain the sensitivity level of the object asset, using a link probe to extract the security level of the main user and the risk coefficient of the data physical flow link from the user's data request, and substituting the object asset sensitivity level, the security level of the main user, and the risk coefficient of the data physical flow link into the three-dimensional nonlinear weighted calculation formula to obtain the desensitization severity index.
[0039] In this embodiment, the three-dimensional nonlinear weighted calculation formula is: Dindex = max(0, ω1·Gasset - ω2·Cuser + ω3·Stype); Where Dindex is the desensitization severity index, Gasset is the sensitivity level of the object asset, Cuser is the security level of the subject user, Stype is the risk coefficient of the data physical flow process, and ω1, ω2, and ω3 are weighting coefficients, and ω1+ω2+ω3=1.
[0040] In this embodiment, the sensitivity level of the object asset is specifically represented by the sensitivity classification value of the original sensitive fields obtained through reverse kinship tracing; the security level of the subject user is specifically represented by the quantified classification value of trust based on user identity authentication and permissions; the risk coefficient of the data physical flow process is specifically represented by the standardized risk quantification value of the physical scenario based on real-time data stream; after obtaining the de-identification severity index, differentiated de-identification strategies can be matched according to the index range to control the de-identification depth, coverage, and execution intensity. High-risk scenarios can automatically increase the de-identification intensity, while low-risk scenarios can appropriately relax the de-identification restrictions, thereby maximizing the preservation of business data characteristics while meeting medical privacy regulations and balancing data privacy security and business availability.
[0041] S6. Dynamically assemble atomic desensitization operators to form a desensitization execution tree based on the desensitization severity index.
[0042] In this embodiment, as one implementation of the first aspect, dynamically assembling atomic desensitization operators to form a desensitization execution tree based on the desensitization severity index includes: S61. Match basic atomic desensitization operators from the atomic desensitization operator pool according to the desensitization severity index; S62. Dependency topology sorting based on operator execution constraints; S63. Construct a tree-like execution structure with hierarchical dependencies and execution priorities; S64. Serialize and compress the tree-shaped execution structure to obtain the de-identified execution tree.
[0043] Optionally, in this embodiment, dynamic assembly employs an adaptive desensitization decision component. This component incorporates a three-dimensional parameter rule base, a weighting model, and a nonlinear weighted calculation engine. It receives user five-tuple request parameters reported in real-time by the link-aware system. First, it performs reverse lineage tracing, traversing the topology from the corresponding request fields downstream to locate the original sensitive fields and extract the sensitivity level of the object assets. Then, it combines the pre-configured weights of the main user's security level and the risk coefficients of the data physical flow process, substituting them into the three-dimensional nonlinear weighted calculation formula to calculate the real-time desensitization severity index. Next, based on the desensitization severity index, it divides the desensitization intensity range into multiple levels and automatically matches the corresponding desensitization rules, operator combinations, and execution priorities.
[0044] S7. The de-identification execution tree is sent down to the edge nodes of the directed acyclic graph, and the edge nodes perform zero-copy de-identification operations and output compliant de-identified data.
[0045] In this embodiment, the edge node performs zero-copy de-identification operations and outputs compliant de-identified data. This includes taking over the original memory pointer of the local data stream of multi-source heterogeneous medical big data, directly performing atomic de-identification operations in the local original memory address, and outputting compliant de-identified data stream.
[0046] In this embodiment, steps S6 and S7 are optionally executed based on the dynamic assembly engine of the desensitization execution tree, the atomic desensitization operator resource pool, the distributed function message bus, and the edge instruction scheduling module. The atomic desensitization operator resource pool covers multiple desensitization methods such as field masking, partial desensitization, full hashing, mean perturbation, numerical offset, field masking, and reversible decryption. The desensitization methods are combined and dynamically arranged according to the desensitization severity index. The dynamic assembly engine of the desensitization execution tree and the distributed function message bus support real-time memory construction, serialization compression and packaging, and high-speed asynchronous bus distribution of the desensitization execution tree. The edge instruction scheduling module deployed on each business node can receive lightweight instructions and complete the desensitization operation in the local memory of the business through a zero-copy memory operation algorithm, realizing unified management and control of desensitization strategy, distributed release of computing power, and local data processing.
[0047] In this embodiment, optionally, the atomic desensitization operator resource pool decomposes all desensitization capabilities into independent, reusable, and composable minimum operational units, supporting operator order adjustment and custom parameter configuration. After obtaining the desensitization severity index, the desensitization execution tree dynamic assembly algorithm is executed to complete operator matching, hierarchical sorting, and dependency relationship construction, generating a tree-shaped execution structure with parent-child hierarchy, execution priority, and pre- and post-dependency constraints. Different desensitization severity indices are matched with different execution tree depths and operator stacking layers; low-intensity desensitization corresponds to shallow simple operator combinations, while high-intensity desensitization corresponds to multi-layer composite operator logic. After the execution tree is constructed, it is serialized and compressed, and then asynchronously issued as lightweight instructions through a distributed function message bus.
[0048] In this embodiment, optionally, after receiving the serialization execution tree instruction issued by the distributed function message bus, the edge node performs deserialization parsing to restore the complete desensitization operation logic, and takes over the local data read / write interface of the business to intercept the plaintext data before it flows out of the business memory.
[0049] Compared with the prior art, the advantages of this application are as follows: It can automatically detect sensitive privacy data, eliminating manual operation and maintenance costs and blind spots in desensitization protection. This application abandons the traditional manual sorting, manual labeling configuration, and static synchronous protection mode. Instead, it relies on the AST layered syntax parsing algorithm and DAG incremental coloring propagation algorithm to realize the automated perception of the entire process of data processing, derivation, and flow, and the automatic hierarchical transmission of sensitive attributes. For the high-frequency iteration, data table reorganization, and frequent field changes of medical business big data, it can automatically capture data processing change behavior after connecting to a multi-source heterogeneous medical big data information system, and update sensitive assets and lineage topology in real time. There is no need for security operation and maintenance personnel to manually maintain asset lists and configure desensitization strategies. It completely solves the pain points of desensitization failure of derived fields, lack of protection for newly added fields, lag in manual maintenance, and high operation and maintenance costs, and realizes real-time automated management and control of sensitive assets. It enables refined de-identification management across multiple scenarios. This application is not limited to traditional two-dimensional static de-identification control, but adds a three-dimensional dynamic decision-making method based on the dimension of data physical flow scenarios. It unifies risk control for three core factors: the sensitivity level of the object asset, the security level of the subject user, and the risk coefficient of the data physical flow process. It can dynamically adjust the combination of atomic de-identification operators and the de-identification execution intensity for the same sensitive privacy data based on the differences in the security level of the operator and the data flow risk. This enables differentiated management to ensure business availability in development scenarios and ensure data security in external sharing scenarios. It solves the contradiction between excessive de-identification that damages business and insufficient de-identification that causes compliance risks in traditional fixed de-identification rules, achieving a balance between security compliance and business availability, and adapting to the needs of complex links, multiple scenarios, and high flow of medical big data. By employing a distributed edge-local execution approach, this application avoids the high-concurrency performance bottleneck of centralized gateways. Instead of the traditional centralized gateway approach of processing all data in transit, the central node is only responsible for complex strategy calculations, execution tree assembly, and instruction scheduling, without bearing the transmission and cleaning operations of massive amounts of plaintext data. This avoids performance defects such as network I / O blocking, excessive bandwidth consumption, server memory overflow, and interface response timeouts caused by centralized processing of large-scale data. Combined with edge zero-copy to eliminate cross-memory data transfer overhead, it improves the speed and throughput limit of de-identification processing, and can stably support the high-concurrency, high-volume, and low-latency flow requirements of medical big data. By employing a dynamic orchestration mechanism for atomic desensitization operators, adaptation costs are reduced. This application overcomes the shortcomings of traditional desensitization logic, which relies on hard-coded, fixed, and unadjustable underlying code. It breaks down all desensitization capabilities into standardized and reusable atomic desensitization operators. Relying on a dynamic assembly algorithm based on the desensitization execution tree, it can achieve code-free, visual orchestration of desensitization logic. For differentiated business scenarios, sensitivity levels, and compliance requirements, operators can be freely combined, execution order adjusted, and desensitization strength customized. Online policy iteration can be completed without modifying the underlying code or stopping and restarting the service. This enables unified effect of a single policy configuration across all heterogeneous business components, eliminating the need to develop separate adaptation logic for multiple types of businesses such as data collection, development, preview, and sharing, thus reducing development, operation, and iteration costs.
[0050] like Figure 2 As shown in the figure, this application discloses an intelligent medical big data privacy data desensitization system, including: The acquisition module is used to acquire multi-source heterogeneous medical big data processing scripts and receive user data requests; The lineage dynamic transmission module is used to perform AST layered syntax parsing on medical big data processing scripts and construct corresponding field-level data lineage directed acyclic graphs. Using the original sensitive fields as the topological root nodes of the directed acyclic graph, the DAG incremental coloring transmission algorithm is used to transmit sensitive attributes and sensitivity levels to downstream derived fields along the directed path of the directed acyclic graph. The three-dimensional matrix decision module is used to trace the lineage of objects by traversing the directed acyclic graph in reverse from the target field based on the response to the user's data request, and obtain the sensitivity level of the object assets corresponding to the target field; it integrates the sensitivity level of the object assets with the security level of the main user and the risk coefficient of the data physical flow process extracted from the user's data request, and performs three-dimensional nonlinear weighted calculation to obtain the desensitization severity index. The distributed execution control module is used to dynamically assemble atomic desensitization operators to form a desensitization execution tree based on the desensitization severity index; the desensitization execution tree is then distributed to the edge nodes of the directed acyclic graph, where the edge nodes perform zero-copy desensitization operations and output compliant desensitized data.
[0051] The system described in the above embodiments is used to execute the corresponding intelligent medical big data privacy data desensitization method in the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0052] like Figure 3 As shown, based on the same application concept, and corresponding to any of the above embodiments, this application also discloses an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements the above-mentioned intelligent medical big data privacy data desensitization method.
[0053] Specifically, the device includes: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected within the device via the bus 1050.
[0054] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), GPU (Graphics Processing Unit), or one or more integrated circuits, to implement relevant programs and achieve the technical solutions provided in the embodiments of this specification.
[0055] The memory 1020 can be implemented in the form of ROM (Read-Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1020 and called by the processor 1010. The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, projectors, speakers, vibrators, indicator lights, etc.
[0056] The communication interface 1040 is used to connect the communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB (Universal Serial Bus), network cable, etc.) or wireless means (such as mobile network, WIFI (Wireless Fidelity), Bluetooth, etc.).
[0057] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.
[0058] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.
[0059] The electronic devices described above are used to implement the corresponding intelligent medical big data privacy data desensitization method in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0060] Based on the same application concept, and corresponding to any of the above embodiments, this application also discloses a computer-readable storage medium that stores computer instructions for enabling a computer to implement the above-described intelligent medical big data privacy data desensitization method.
[0061] The computer-readable storage medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium, which can be used to store information accessible by a computing device. The computer instructions stored in the storage medium of the above embodiments are used to enable the computer to implement the intelligent medical big data privacy data desensitization method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0062] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments or combinations thereof, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.
Claims
1. A method for desensitizing privacy data in intelligent medical big data, comprising acquiring multi-source heterogeneous medical big data processing scripts and receiving user data requests; characterized in that, The method further includes: The medical big data processing script is subjected to AST layered syntax parsing to construct the corresponding field-level data lineage directed acyclic graph; Using the original sensitive field as the topological root node of the directed acyclic graph, the DAG incremental coloring propagation algorithm is used to propagate the sensitive attributes and sensitivity levels to the downstream derived fields along the directed path of the directed acyclic graph; Based on the response to the user's data request, reverse tracing of the directed acyclic graph from the target field is performed to obtain the sensitivity level of the object asset corresponding to the target field; The sensitivity level of the object assets is combined with the security level of the main user extracted from the user's data request and the risk coefficient of the data physical flow process to obtain the desensitization severity index through three-dimensional nonlinear weighted calculation. Based on the aforementioned desensitization severity index, atomic desensitization operators are dynamically assembled to form a desensitization execution tree; The de-identification execution tree is sent down to the edge nodes of the directed acyclic graph, where the edge nodes perform zero-copy de-identification operations and output compliant de-identified data.
2. The intelligent medical big data privacy data desensitization method according to claim 1, characterized in that, The step of performing AST layered syntax parsing on the medical big data processing script includes: The medical big data processing script is subjected to one or more of the following: lexical analysis, syntax tree construction, semantic parsing, and automatic conflict correction. Relationships such as field mapping, table association, and field derivation transformation are extracted from the data, and the field-level lineage topology edge structure is output.
3. The intelligent medical big data privacy data desensitization method according to claim 1, characterized in that, The transmission of sensitive attributes and sensitivity levels to downstream derived fields along the directed path of the directed acyclic graph includes: A depth-first traversal is performed on the directed acyclic graph so that the downstream derived fields inherit the sensitivity level parameters of the original sensitive fields; When the topology of the directed acyclic graph changes, local recoloring is performed only on the subgraphs affected by the change through incremental topology change detection.
4. The intelligent medical big data privacy data desensitization method according to claim 1, characterized in that, The reverse ancestry tracing by traversing the directed acyclic graph in reverse from the target field includes: Starting from the target field, the upstream parent node is searched in reverse along the directed edges of the directed acyclic graph until the root node of the topology is reached; Read the pre-marked object asset sensitivity level in the topology root node and return it to the retrieval starting point.
5. The intelligent medical big data privacy data desensitization method according to claim 1, characterized in that, The process of obtaining the desensitization severity index through three-dimensional nonlinear weighted calculation includes: The DAG reverse lineage tracing algorithm is used to obtain the sensitivity level of the object asset. The security level of the principal user is extracted from the user's data request using a link probe. Link probes are used to extract risk factors of the physical flow of data from user data requests; The desensitization severity index is obtained by substituting the object asset sensitivity level, the subject user security level, and the risk coefficient of the data physical flow process into the three-dimensional nonlinear weighted calculation formula. The formula for three-dimensional nonlinear weighted calculation is: D index = max(0, ω1·G asset - ω2·C user + ω3·S type ); Among them, D index G is the desensitization severity index. asset For the sensitivity level of the object assets, C user For the primary user security level, S type ω1, ω2, and ω3 are the risk coefficients for the physical flow of data, and ω1 + ω2 + ω3 = 1.
6. The intelligent medical big data privacy data desensitization method according to claim 1, characterized in that, The step of dynamically assembling atomic desensitization operators to form a desensitization execution tree based on the desensitization severity index includes: Basic atomic desensitization operators are matched from the atomic desensitization operator pool based on the desensitization severity index; Dependency topology sorting based on operator execution constraints; Construct a tree-like execution structure with hierarchical dependencies and execution priorities; The tree-shaped execution structure is serialized and compressed to obtain a de-identified execution tree.
7. The intelligent medical big data privacy data desensitization method according to claim 1, characterized in that, The edge nodes perform zero-copy de-identification operations and output compliant de-identified data, including: It takes over the local data stream's original memory pointer of multi-source heterogeneous medical big data, directly performs atomic de-identification operations in the local original memory address, and outputs a compliant de-identified data stream.
8. An intelligent medical big data privacy data anonymization system, comprising an acquisition module for acquiring multi-source heterogeneous medical big data processing scripts and receiving user data requests; characterized in that, The system also includes: The lineage dynamic transmission module is used to perform AST layered syntax parsing on the medical big data processing script and construct the corresponding field-level data lineage directed acyclic graph; taking the original sensitive field as the topological root node of the directed acyclic graph, the DAG incremental coloring transmission algorithm is used to transmit sensitive attributes and sensitivity levels to downstream derived fields along the directed path of the directed acyclic graph; The three-dimensional matrix decision module is used to perform reverse lineage tracing by traversing the directed acyclic graph in reverse from the target field based on the response to the user's data request, and obtain the object asset sensitivity level corresponding to the target field; and to perform three-dimensional nonlinear weighted calculation by integrating the object asset sensitivity level with the main user security level and the risk coefficient of the data physical flow process extracted from the user's data request to obtain the desensitization severity index. The distributed execution control module is used to dynamically assemble atomic desensitization operators to form a desensitization execution tree according to the desensitization severity index; and to distribute the desensitization execution tree to the edge nodes of the directed acyclic graph, whereby the edge nodes perform zero-copy desensitization operations and output compliant desensitized data.
9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the intelligent medical big data privacy data desensitization method according to any one of claims 1-7.
10. A storage medium, which is a computer-readable storage medium, characterized in that, The device stores computer instructions, which are used to enable a computer to implement the intelligent medical big data privacy data desensitization method according to any one of claims 1-7.
