Large model localization deployment and security isolation device

Abstract

The invention discloses a large model localization deployment and security isolation device for a hospital closed network environment, which consists of a model deployment unit, a security isolation unit, a data management unit, a security control unit and an intelligent updating unit, and realizes independent operation of a large model in a hospital. According to the system, safe input of medical data is achieved through data cleaning, desensitization and formatting processing, full-process safety of data and model files is guaranteed through authority control, access recording and encrypted storage, and credible upgrading and abnormal rollback of a model are completed through a one-way isolation and offline updating mechanism. The large model can be operated and maintained under the condition that an external network is not connected, the requirements of hospitals for data privacy and network security are met, and the method is suitable for various types of intelligent medical services.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and information technology, and more specifically to a device for localized deployment and secure isolation of large models. Background Technology

[0002] With the rapid development of artificial intelligence technology, the application scenarios of large-scale models in the healthcare industry are constantly expanding, including clinical auxiliary diagnosis, medical image analysis, medical record generation, intelligent triage, nursing consultation, and in-hospital operation analysis. Large-scale models, through inference and learning from massive amounts of data, can significantly improve the service quality and work efficiency of hospitals. However, hospital data is highly sensitive and privacy-sensitive; diagnostic data, imaging materials, and patient personal information are strictly protected by laws and regulations, and any leakage could lead to serious legal and management risks. Therefore, the application of large-scale models within hospitals must strictly adhere to information security, data security, and privacy protection requirements, and cannot rely on external cloud environments or transmit patient data to external servers.

[0003] In current technologies, hospitals mostly use traditional servers, local databases, and simple encrypted storage schemes to ensure data security. However, these solutions are insufficient to meet the high computing power requirements of running large models, and also fail to meet requirements such as continuous updates, differential maintenance, and compliance auditing. Because hospital intranets are generally closed and strictly isolated from the external network, update packages, parameter files, or corpus packages for large models cannot be directly imported from the external network. This makes it difficult for large model systems to maintain long-term effectiveness and technological advancement in practical use. Some hospitals have attempted to update models through manual import or special channel transmission, but these methods are not only inefficient but also often pose security risks, failing to meet the high data protection requirements of hospitals.

[0004] In terms of data processing, hospital business systems generate data in complex formats and of diverse types, including structured medical records, semi-structured imaging reports, and multimodal clinical records. Without effective data cleaning and privacy anonymization mechanisms, directly inputting data into large models will not only reduce inference performance but may also lead to privacy leaks. Existing solutions lack unified standards for data anonymization and formatting, making it difficult to meet the data collaboration needs of different departments and multiple systems.

[0005] In terms of access management, hospitals have various roles, including doctors, nurses, administrators, and maintenance personnel, each with different permissions. Access to certain content needs to be strictly controlled, as unauthorized access can pose serious risks. Traditional access management methods often rely on the business system's own permission system, making it difficult to achieve a unified, traceable, and auditable access control mechanism for large-scale systems, and thus failing to meet the hospital's requirements for network security and operational standards.

[0006] Regarding model updates and system maintenance, due to the sheer size of large models and their frequent updates, hospitals lacking a robust offline update mechanism are prone to experiencing critical function interruptions and disruptions to medical services if errors occur during the update or loading process. Furthermore, many existing hospital internal systems lack professional model version control and rollback mechanisms, making it impossible to guarantee a rapid restoration to the old version after an update failure.

[0007] In summary, hospitals currently face several challenges when introducing large-scale models, including difficulties in data discharge, high local computing power requirements, challenges in importing external updates, insufficient security protection, inconsistent access control, weak log traceability, and unrecoverable updates after failures. Existing technologies do not provide an integrated solution for achieving local deployment, secure isolation, standardized data processing, access control, encrypted storage, and reliable offline updates of large-scale models within a hospital's closed network environment. Therefore, it is necessary to propose a new technical solution to address these pain points. Summary of the Invention

[0008] The purpose of this invention is to provide a large-scale model localization deployment and security isolation device. Through the collaboration of the model deployment host, security isolation gateway, data management module, security control layer and intelligent update module, it can achieve high-security deployment of large models in the hospital internal environment, compliant processing of hospital data, one-way import of external update files and trusted writing of model files, and ensure the security and hospital business continuity during the large model update process.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A large-scale model localization deployment and security isolation device is applied to the internal network environment of a hospital, including a model deployment host, a security isolation gateway, a data management module, a security control layer, and an intelligent update module;

[0011] The model deployment host includes an AI computing module, a task scheduling module, and an interface adaptation module, which are used to perform large-scale model inference within the hospital and application access to the hospital system (HIS / EMR / PACS);

[0012] The security isolation gateway includes an intranet protection unit, a relay buffer unit, and an external network input unit, which are used to realize the one-way secure transmission of external update files to the hospital's intranet.

[0013] The data management module includes a data cleaning unit, a privacy desensitization unit, and a formatted transmission unit, which are used to clean, desensitize, and standardize the data in the hospital's business system.

[0014] The security control layer includes an access control module and an encrypted storage module. The access control module includes an identity authentication unit, a permission hierarchical unit, and an access log unit. The encrypted storage module includes a file encryption unit, a signature verification unit, and a log traceability unit.

[0015] The intelligent update module includes an update receiving unit, an incremental loading unit, and a version rollback unit, which are used to perform security updates, integrity checks, and abnormal rollbacks of the large model within the institute.

[0016] Preferably, the external network input unit is used to receive model patch packages, corpus files and configuration files from external update media (portable hard drives, dedicated USB flash drives, etc.) and transmit them unidirectionally to the relay buffer unit.

[0017] Preferably, the transit buffer unit adopts a physical isolation structure and a dual-buffer queue to ensure that data flows only unidirectionally from the outside to the hospital's intranet, preventing any reverse data leakage.

[0018] Preferably, the data cleaning unit is used to perform field validation, outlier removal, and format standardization on structured and semi-structured data in the HIS / EMR / PACS system.

[0019] Preferably, the privacy desensitization unit performs desensitization on sensitive medical data such as patient name, hospital number, image sequence number, ID number, and telephone number.

[0020] Preferably, the formatted transmission unit automatically converts the desensitized medical data into a structured vector format and synchronizes it to the model deployment host according to the model input requirements.

[0021] Preferably, the identity authentication unit is bound to the hospital doctor terminal, department management terminal, and information technology maintenance terminal to realize multi-role hierarchical access management, and the access log unit records all access behaviors in a timestamp chain.

[0022] Preferably, the signature verification unit adopts a dual verification method of digital signature and hash verification to ensure that the trusted source and content of the file to be updated have not been tampered with.

[0023] Preferably, the incremental loading unit is used to compare the differences between the current model version and the update patch, and only load the changed sections to improve model update efficiency and reduce system downtime.

[0024] Preferably, the version rollback unit restores the old version file in the model and data storage area when the update is abnormal, the verification fails, or the loading is interrupted, so as to ensure the continuity of the hospital's critical business.

[0025] Compared with existing technologies, this invention achieves localized, secure, and reliable operation of large models within a hospital environment through the collaborative design of a model deployment unit, a security isolation unit, a data management unit, a security control unit, and an intelligent update unit. The system can perform large model inference without relying on the external network, enabling medical data processing within the hospital's closed network. This avoids the risk of leakage of patient privacy, medical records, and imaging data during transmission, effectively meeting the hospital's regulatory requirements for data security.

[0026] The secure isolation structure provided by this invention ensures that update packages, corpus packages, and configuration files from external media can only flow unidirectionally into the internal network during the import process, eliminating potential threats from external networks to the hospital system. This structure avoids the predicament of traditional large-scale hospital model systems being unable to be updated or requiring reliance on external networks for updates, enabling large models to maintain maintainability and scalability even in a completely isolated environment.

[0027] The data management structure of this invention can perform cleaning, anonymization, and format standardization on various types of data from hospital business systems, making the data received by the model more standardized and stable, and fully compliant with medical data privacy protection requirements. This process can ensure the quality of data input to the model inference, and improve the accuracy and reliability of model inference.

[0028] The access control and encrypted storage structure established in this invention ensures data security during storage and access from multiple dimensions, including user roles, access permissions, log traceability, and file encryption. The system automatically assigns permissions based on user type, all access behaviors are fully recorded, model files and inference data are encrypted during storage, and integrity verification is performed before reading, effectively preventing unauthorized access and file tampering.

[0029] This invention also provides hospitals with a mechanism for model upgrades and maintenance without an internet connection through an intelligent update structure. The system can perform signature verification and integrity verification on update files, load only the differential content to complete the rapid update, and automatically roll back to the old version in case of anomalies, ensuring the continuity of hospital operations and the stability of large model services.

[0030] In summary, this invention constructs a complete security closed-loop system from data input, model operation, security protection, access control, file storage to model updates, enabling hospitals to safely use large model technology under strict compliance requirements, significantly improving the hospital's intelligence level and business processing efficiency, and has significant technical advantages and application value. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a schematic diagram of the overall structure of the large-scale local deployment and security isolation device of the present invention.

[0033] Figure 2 This is a schematic diagram of the linkage structure between the data management module and the security isolation gateway.

[0034] Figure 3 This is a schematic diagram of the logical structure of the access control module and the encrypted storage module.

[0035] Figure 4 A diagram illustrating the offline update process for the smart update module.

[0036] The annotations in the attached figures are explained as follows:

[0037] 101-Model Deployment Host; 1011-AI-Computing Module; 1012-Task Scheduling Module; 1013-Interface Adaptation Module; 102-Security Isolation Gateway; 1021-Internal Network Protection Unit; 1022-Relay Buffer Unit; 1023-External Network Input Unit; 103-Data Management Module; 1031-Data Cleaning Unit; 1032-Privacy Desensitization Unit; 1033-Formatted Transmission Unit; 104-Access Control Module; 1041-Identity Authentication Unit; 1042-Permission Hierarchy Unit; 1043-Access Log Unit; 105-Encrypted Storage Module; 1051-File Encryption Unit; 1052-Signature Verification Unit; 1053-Log Traceability Unit; 106-Intelligent Update Module; 1061-Update Receiving Unit; 1062-Incremental Loading Unit; 1063-Version Rollback Unit. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1

[0040] This embodiment provides a large-scale localized deployment and security isolation device suitable for hospital closed network environments. This embodiment combines... Figures 1 to 4 This description is provided so that those skilled in the art can clearly understand the overall structure of the device, the data flow, the security control methods, and the offline update process.

[0041] like Figure 1 As shown, the system in this embodiment is installed in a dedicated computer room within the hospital. It consists of a model deployment unit, a security isolation unit, a data management unit, a security control unit, and an intelligent update unit. These modules are connected sequentially according to the data flow, forming a large model system operating in a closed loop within the hospital's intranet. The model deployment unit, located at the center of the structural diagram, is responsible for the large model's inference calculations, task scheduling, and interface adaptation. The security isolation unit, located on the external input side of the system, is used to enable unidirectional import of update files and external data. The data management unit, located between the hospital's business system and the model deployment unit, is used to preprocess business data. The security control unit operates throughout the entire process of data access and file storage. The intelligent update unit is independently configured to perform offline update operations.

[0042] like Figure 2 As shown, a clear data flow relationship exists between the data management unit and the security isolation unit. The security isolation unit receives update packages, corpus files, or configuration files from external media and transmits them unidirectionally to the hospital's intranet. After transmission, the data first enters the data management unit for processing. In this embodiment, the data management unit can clean and convert the format of large amounts of diagnostic and treatment data aggregated from systems such as HIS, EMR, and PACS, while also de-identifying sensitive fields involving patient privacy. The processed data flows to the model deployment unit, enabling large models to perform inference tasks on a secure and standardized data foundation.

[0043] like Figure 3 As shown, this embodiment manages all access behaviors and data operations during model deployment through a security control unit. The security control unit includes an access control structure and an encrypted storage structure. The access control structure implements tiered permission management for doctors, department administrators, and hospital maintenance personnel, ensuring that different roles can only access functions or data within their authorized scope, preventing unauthorized access. All access behaviors are recorded in the log system and are recorded and traced in a chain-like manner. The encrypted storage structure encrypts and verifies the integrity of model files, update packages, inference data, and log records, ensuring the security and immutability of files during reading, transmission, and updating.

[0044] like Figure 4As shown, this embodiment uses an intelligent update unit to achieve offline updates of the large model within the hospital's closed network. After the administrator imports the update file into the update portal, the system automatically performs signature verification and integrity checks to confirm the trustworthiness of the update package's source and the completeness of its content. Upon successful verification, the system executes a differential update method, loading only the parts that differ from the old version, significantly improving update efficiency and reducing the impact on hospital operations. If an anomaly occurs during the update process, verification fails, or the file is corrupted, the system automatically triggers a version rollback function, restoring the large model to its pre-update state and ensuring the continuous operation of the hospital system. After the update is complete, the new model parameters are written to an encrypted storage structure and reloaded by the model deployment unit.

[0045] In this embodiment, the inference, data processing, security management, and version maintenance of the entire large model are all completed autonomously within the hospital's intranet environment, without the need for external servers or cloud services, thus ensuring that medical data remains within the hospital and is not leaked. Figure 1 The overall architecture shown Figure 2 The data processing flow shown Figure 3 The permission and encryption control structure shown and Figure 4 The offline update mechanism shown in this embodiment builds a complete, secure, and reliable large model operation system within the hospital, solving the industry pain points of not being able to securely use large models, update models, and ensure data compliance in the hospital environment.

[0046] This embodiment has the advantages of flexible deployment, high security level, reliable updates, and stable operation. It can be promoted and applied in multiple departments and business systems of hospitals, and is suitable for the intelligent construction of the medical industry with strong supervision and high privacy requirements.

[0047] Example 2

[0048] This embodiment addresses the scenario of concurrent access from multiple departments in a large general hospital, providing a localized deployment and secure isolation system for large models operating in a multi-data source environment. This embodiment also incorporates... Figures 1 to 4 This explanation is intended to demonstrate the collaborative processing capabilities and operating mechanism of the device of the present invention when a hospital needs to process a large amount of data of different types simultaneously and when terminals of different roles access the model service at the same time.

[0049] like Figure 1As shown, this embodiment deploys a large-scale model localization system in the hospital information center, establishing connections with multiple business terminals within a single hospital area via a local area network. The model deployment unit has been expanded to simultaneously respond to multiple inference requests from the emergency department, outpatient department, inpatient department, medical imaging department, and nursing unit. The model deployment unit dynamically allocates resources through an internal task scheduling structure, enabling the model to simultaneously handle different types of tasks such as symptom question answering, image-assisted interpretation, drug interaction checks, and medical record writing optimization. In high-concurrency scenarios, the scheduling structure automatically monitors the model load and prioritizes requests from each department through an asynchronous queue, ensuring that high-priority tasks such as those from the emergency department are processed first.

[0050] like Figure 2 As shown, the data management unit in this embodiment accesses more diverse data sources, including outpatient electronic medical record systems, inpatient medical order systems, laboratory systems, and ultrasound and radiology imaging systems. Due to differences in data formats and field structures across different systems, the data management unit places higher demands on data cleaning, field mapping, and format standardization. The data cleaning process automatically matches cleaning strategies based on department type; for example, it retains medical terminology in imaging reports, breaks down medical records into paragraphs, and labels outliers in test results. Simultaneously, a desensitized structure uniformly implements privacy protection when data flows in from multiple departments, ensuring that all content input to the model undergoes rigorous desensitization processing, guaranteeing that the system meets regulatory requirements even in complex business environments. The processed data is then converted into a standardized structure that can be directly processed by larger models through a formatted structure, ensuring consistency in inference input.

[0051] like Figure 3 As shown, to accommodate the access needs of multiple roles and terminals, this embodiment enhances the security control unit. The access control structure configures permission levels according to the hospital's role system, assigning doctors, nurses, pharmacists, department heads, administrators, and hospital IT maintenance personnel to different levels. The system automatically restricts the scope of model services and data categories accessible to users based on their roles. For example, doctors can access medical record generation, diagnostic and treatment assistance suggestions, and image interpretation services; nurses can access nursing-related tasks; and administrators can only view system operation status and model status. All access behaviors are recorded through a log chain system, with access records for different roles stored separately in the log system for subsequent departmental-level tracing and auditing. Regarding file protection, the encrypted storage structure performs real-time encryption on the intermediate results of each model call, preventing unauthorized access to the inference process content on the server.

[0052] like Figure 4As shown, the intelligent update unit in this embodiment has important applications in large hospital scenarios. When the hospital needs to iteratively optimize the model, the administrator imports the update content offline into the system entry point. After receiving the update package, the system selects an appropriate update window based on the hospital's current high-concurrency access situation and automatically enters a phased update process. First, the update content undergoes integrity verification and signature verification. Then, an incremental update method is used, replacing only the changed parts in the old version. If business requests still arrive during the model update process, the system automatically redirects the requests to the old version model, and smoothly switches to the new version after the update is completed. If a verification failure or loading anomaly occurs during the update process, the system's version rollback structure will immediately restore the old version model, ensuring that the hospital's business during peak hours is not affected.

[0053] Overall, Example 2 demonstrates the invention's capabilities in handling concurrent access across multiple departments, heterogeneous processing of multiple data sources, enhanced access control, and uninterrupted updates in complex hospital environments. The system maintains stable operation under high loads from various business processes, ensuring that all data flows within the hospital are controlled and secure through data processing and security management. This example is suitable for large general hospitals, tertiary hospitals, or multi-campus centralized management scenarios, and can meet the needs of high-intensity, large-scale intelligent medical applications.

[0054] Example 3

[0055] This embodiment provides a large-scale localized deployment and secure isolation system for hospital environments with higher security levels, operating under deeply isolated network conditions. This embodiment combines... Figures 1 to 4 The description focuses on demonstrating the enhanced security and stability of the device of this invention in hospitals with highly sensitive data types, strict internal and external network isolation systems, and high security audit requirements.

[0056] like Figure 1 As shown, the overall architecture of this embodiment still consists of a model deployment unit, a security isolation unit, a data management unit, a security control unit, and an intelligent update unit. However, the system operating environment is more closed. The model deployment unit is placed within the hospital's core security zone, physically isolated from other network segments providing external services. In this embodiment, the model deployment unit performs highly sensitive tasks such as deep medical text generation, complex disease course analysis, large-scale image feature embedding, and scientific research data structuring. The computing module also employs a higher level of access restrictions. All data from the hospital's business systems must undergo strict review through a secure channel before entering the model deployment environment, ensuring that the large model cannot directly access unprocessed raw sensitive information.

[0057] like Figure 2As shown, the data management process is further strengthened in high-security scenarios. The data management unit does not merely perform cleaning, desensitization, and formatting; instead, it introduces a strict data flow whitelist mechanism, allowing only authorized data types to enter the model system. In this embodiment, the data management unit is commonly used to process highly sensitive content such as psychiatric medical records, tumor reports, infectious disease cases, and genetic disease records. During the cleaning phase, the system uses a combination of rule matching and model recognition to remove non-compliant information. In the desensitization phase, algorithms replace sensitive content with internally unified codes, ensuring that the data cannot be reconstructed to identify the patient's individual identity before entering the model. The formatting phase further recombines the data to ensure that the input received by the model does not contain any personally identifiable content.

[0058] like Figure 3 As shown, this embodiment strengthens both access control and the encrypted storage structure. The access control structure employs multi-factor authentication, implementing two-factor or multi-factor login verification for doctors, researchers, administrators, and operations personnel. Permissions are more granular; each access action is not only logged but also automatically generates a complete operation fingerprint for security auditing. The system continuously monitors access behavior, automatically triggering alarms when abnormal access patterns occur, and temporarily suspending model calls if necessary. The encrypted storage structure also employs a multi-layered key system, with model files, configuration files, and inference results encrypted using different keys, and no decryption relationship exists between keys. The signature verification mechanism in this embodiment incorporates a dual verification structure, requiring both internal signature verification and external authorization verification before model reading to ensure absolute file trustworthiness. The log tracking mechanism is also enhanced, using a chain structure to ensure each operation has an immutable chain fingerprint, allowing for complete reconstruction afterward.

[0059] like Figure 4 As shown, the intelligent update unit in this embodiment employs a strict multi-stage verification mechanism during the model update process. When the hospital needs to update the model, the administrator submits the update file offline to the secure isolation entry point. The system first verifies the source of the update file, and then enters a multi-level integrity verification process. The update process only begins if all verifications pass. During the update process, a partitioned loading method is used, dividing the model into multiple logical modules for separate loading to avoid the overall risks that might result from a one-time update. The system automatically activates a secure shadow mode during the update, pre-running a local verification task of the new model before model switching. Only when the verification results are consistent with expectations will the new model be put into formal use. If inconsistencies, file corruption, or operational abnormalities occur during the update process, the system will immediately perform a rollback recovery and send an update failure message to the operations and maintenance center.

[0060] In summary, Example 3 demonstrates the operation of this invention in a high-security hospital environment. Through stricter data flow protection, more granular access control, a more complex encryption system, and a more rigorous update verification scheme, the large model can operate securely in a highly sensitive medical data environment. This example is suitable for hospital scenarios with extremely high requirements for data security and auditing capabilities, and can meet the security and compliance needs of medical institutions with strict regulations and high privacy sensitivity when introducing large models.

[0061] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A large-scale model localization deployment and security isolation device, applied in a hospital internal network environment, comprising a model deployment host (101), a security isolation gateway (102), a data management module (103), a security control layer (104, 105), and an intelligent update module (106), characterized in that: The model deployment host (101) includes an AI computing module (1011), a task scheduling module (1012), and an interface adaptation module (1013), which are used to perform large-scale inference within the hospital and application access to the hospital system (HIS / EMR / PACS); The security isolation gateway (102) includes an intranet protection unit (1021), a relay buffer unit (1022), and an external network input unit (1023) in sequence, which are used to realize the one-way secure transmission of external update files to the hospital intranet; The data management module (103) includes a data cleaning unit (1031), a privacy desensitization unit (1032), and a formatted transmission unit (1033), which are used to clean, desensitize, and standardize the data of the hospital business system. The security control layer includes an access control module (104) and an encrypted storage module (105). The access control module (104) includes an identity authentication unit (1041), a permission leveling unit (1042), and an access log unit (1043). The encrypted storage module (105) includes a file encryption unit (1051), a signature verification unit (1052), and a log traceability unit (1053). The intelligent update module (106) includes an update receiving unit (1061), an incremental loading unit (1062), and a version rollback unit (1063), which are used to perform security updates, integrity checks, and abnormal rollbacks of the large model within the institute.

2. The apparatus according to claim 1, characterized in that, The external network input unit (1023) is used to receive model patch packages, corpus files and configuration files from external update media (portable hard drives, dedicated USB flash drives, etc.) and transmit them unidirectionally to the relay buffer unit (1022).

3. The apparatus according to claim 1, characterized in that, The transit buffer unit (1022) adopts a physical isolation structure and a dual buffer queue to ensure that data flows only unidirectionally from the outside to the hospital's intranet, preventing any reverse data leakage.

4. The apparatus according to claim 1, characterized in that, The data cleaning unit (1031) is used to perform field validation, outlier removal, and format standardization on structured and semi-structured data in the HIS / EMR / PACS system.

5. The apparatus according to claim 1, characterized in that, The privacy desensitization unit (1032) performs desensitization on sensitive medical data such as patient name, hospital number, image sequence number, ID number, and telephone number.

6. The apparatus according to claim 1, characterized in that, The formatted transmission unit (1033) automatically converts the desensitized medical data into a structured vector format and synchronizes it to the model deployment host (101) according to the model input requirements.

7. The apparatus according to claim 1, characterized in that, The identity authentication unit (1041) is bound to the hospital doctor terminal, department management terminal and information technology maintenance terminal to realize multi-role hierarchical access management, and the access log unit (1043) records all access behaviors in a timestamp chain.

8. The apparatus according to claim 1, characterized in that, The signature verification unit (1052) adopts a dual verification method of digital signature and hash verification to ensure that the trusted source and content of the file to be updated have not been tampered with.

9. The apparatus according to claim 1, characterized in that, The incremental loading unit (1062) is used to compare the differences between the current model version and the update patch, and only load the changed sections to improve the model update efficiency and reduce system downtime.

10. The apparatus according to claim 1, characterized in that, The version rollback unit (1063) restores the old version file in the model and data storage area when the update is abnormal, the verification fails, or the loading is interrupted, thus ensuring the continuity of the hospital's critical business.

Images