A network security protection system based on trusted computing

Abstract

This invention discloses a network security protection system based on trusted computing, comprising a computing subsystem and a protection subsystem running in parallel. The protection subsystem uses a trusted platform control module as the root of trust, which controls the startup and operation of the computing subsystem. The system also includes a dynamic trust assessment engine, a medical data lifecycle protection module, and an AI model trust measurement module. This invention adopts a dual-architecture approach, with computing and protection running independently of the computing subsystem. Using the trusted platform control module as the active root of trust, it achieves active measurement and control throughout the computing process, effectively defending against unknown threats and changing the traditional passive defense model. Through the dynamic trust assessment engine, multi-dimensional factors such as device status, data sensitivity, and user behavior are quantified into dynamic trust scores, enabling dynamic adjustment of security policies and improving the accuracy and adaptability of protection.

Description

Technical Field

[0001] This invention relates to the field of medical information security technology, and more specifically to a network security protection system based on trusted computing. Background Technology

[0002] With the deepening development of medical informatization, core business systems such as Hospital Information Systems (HIS), Picture Archiving and Communication Systems (PACS), and Electronic Medical Records (EMR) carry massive amounts of highly sensitive patient diagnosis and treatment data. At the same time, new business models such as telemedicine and AI-assisted medical diagnosis place higher demands on cross-institutional data flow and collaborative computing. However, current cybersecurity protection in the medical field generally suffers from the following problems:

[0003] Limitations of passive defense mode: The traditional "blocking and killing" mode, which relies on firewalls, intrusion detection, and virus scanning, is a passive defense and is difficult to deal with attacks that exploit unknown system vulnerabilities or logical flaws, such as the WannaCry ransomware virus, which caused serious damage to the medical system.

[0004] Incomplete Trust Chain Construction: Existing trusted computing solutions based on Trusted Platform Modules (TPM) (Trusted Computing 2.0) treat the TPM as a passively attached external device, requiring invocation by the host software to function. Once the host is controlled by an attacker, the TPM's security functions will fail, making it impossible to effectively construct a complete trust chain from hardware startup to application execution.

[0005] Insufficient dynamic adaptability: Device status, data sensitivity, and user behavior patterns in the healthcare environment are constantly changing. Existing security strategies are mostly based on static rules or single identity authentication, which cannot dynamically adjust trust levels and protection strength according to real-time context, making it difficult to meet the dual requirements of security and convenience in smart healthcare.

[0006] Data and model security risks are prominent: Medical data faces the risk of leakage and tampering throughout its entire lifecycle, from collection and transmission to storage, processing, and sharing. Meanwhile, emerging threats such as poisoning training data for medical AI models, implanting backdoors in models, and adversarial attacks during inference processes are becoming increasingly serious, and there is a lack of effective reliability metrics and protective measures.

[0007] Therefore, there is an urgent need for a network security protection system that is proactively immune, dynamically adaptable, and reliable throughout the entire process to ensure the safe and reliable operation of medical information systems. Summary of the Invention

[0008] The purpose of this invention is to provide a network security protection system based on trusted computing to solve the problems existing in the prior art, and to achieve full-process trusted measurement and dynamic protection of medical information systems from the hardware layer to the application layer.

[0009] To achieve the above objectives, the present invention provides the following solution: The present invention provides a network security protection system based on trusted computing, comprising:

[0010] The computing subsystem and the protection subsystem operate in parallel; the protection subsystem uses the trusted platform control module as the root of trust, the trusted platform control module starts before the CPU of the computing subsystem, and controls the startup and operation of the computing subsystem;

[0011] The dynamic trust assessment engine is used to dynamically calculate entity trust scores based on real-time device status parameters, data sensitivity tags, and user behavior baselines, and adjust access control policies accordingly.

[0012] The medical data lifecycle protection module is used to dynamically select encryption algorithms and key lengths based on data sensitivity tags, and to perform trusted auditing of data operations throughout the entire process.

[0013] The AI ​​model credibility measurement module is used to measure the credibility of model data, algorithms, and environment.

[0014] The optimized formula for calculating entity trust scores by the dynamic trust assessment engine is as follows:

[0015] T=α*S_state+β*S_data+γ*S_behavior;

[0016] Where T is the dynamic trust score; S_state is the device's real-time status parameter score, and its calculation formula is:

[0017] S_state=Σ(w_i*(v_i / V_max_i));

[0018] Where v_i is the real-time value of the i-th state parameter, V_max_i is the preset threshold of the parameter, w_i is the weight of the parameter, and Σw_i=1;

[0019] S_data represents the data sensitivity score;

[0020] S_behavior is a user behavior deviation score, and its calculation formula is as follows:

[0021] S_behavior=1-(|B_current-B_baseline| / B_max);

[0022] Wherein, B_current is the current behavior feature vector, B_baseline is the historical behavior baseline vector, and B_max is the maximum allowable deviation threshold;

[0023] α, β, and γ are weight coefficients, and α+β+γ=1.

[0024] The optimization formula for the medical data lifecycle protection module to dynamically adjust encryption strength based on data sensitivity tags is as follows:

[0025] E_strength=Base_strength+k*(S_data-S_threshold);

[0026] Where E_strength is the actual encryption strength, Base_strength is the basic encryption strength, S_data is the data sensitivity score, S_threshold is the sensitivity threshold for triggering enhanced encryption, and k is the strength adjustment coefficient.

[0027] The optimization formula for the AI ​​model credibility measurement module to measure the credibility of the model is as follows:

[0028] M_trust=λ1*D_integrity+λ2*A_robustness+λ3*E_security;

[0029] Where M_trust is the model trustworthiness; D_integrity is the training data integrity metric; A_robustness is the model algorithm robustness metric; E_security is the runtime environment security metric; and λ1, λ2, and λ3 are weight coefficients.

[0030] The present invention discloses the following technical effects: The present invention adopts a dual-system architecture of parallel computing and protection. The protection subsystem runs independently of the computing subsystem. With the trusted platform control module as the active root of trust, it realizes active measurement and control of the entire computing process, effectively defends against unknown threats, and changes the traditional passive defense mode. Through the dynamic trust assessment engine, the device status, data sensitivity, user behavior and other multi-dimensional factors are quantified into dynamic trust scores, realizing the dynamic adjustment of security policies and improving the accuracy and adaptability of protection. 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 architecture of a network security protection system based on trusted computing according to the present invention;

[0033] Figure 2A flowchart of the dynamic trust assessment engine;

[0034] Figure 3 A flowchart illustrating the encryption strategy decision-making process for the medical data lifecycle protection module. Detailed Implementation

[0035] 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.

[0036] This invention discloses a network security protection system based on trusted computing, including a computing subsystem and a protection subsystem, which operate in parallel to form a dual-system architecture;

[0037] The protection subsystem uses the trusted platform control module as the root of trust. The trusted platform control module starts before the CPU of the computing subsystem and actively measures and controls the startup process and operation process of the computing subsystem.

[0038] The dynamic trust assessment engine is used to dynamically calculate entity trust scores based on real-time device status parameters, data sensitivity tags, and user behavior baselines, and adjust access control policies accordingly.

[0039] The medical data lifecycle protection module is used to dynamically select encryption algorithms and key lengths based on data sensitivity tags, and to perform trusted auditing of data operations throughout the entire process.

[0040] The AI ​​model credibility measurement module is used to perform credibility measurement on model data, algorithms, and environment before, during, and after training a medical AI model to ensure model credibility.

[0041] In one embodiment of the present invention, the optimized formula for the dynamic trust assessment engine to calculate the entity trust score is as follows:

[0042] T=α*S_state+β*S_data+γ*S_behavior;

[0043] Where T is the dynamic trust score; S_state is the device's real-time status parameter score, and its calculation formula is:

[0044] S_state=Σ(w_i*(v_i / V_max_i));

[0045] Where v_i is the real-time value of the i-th state parameter (such as CPU utilization, memory usage, network traffic), V_max_i is the preset threshold of the parameter, w_i is the weight of the parameter, and Σw_i=1.

[0046] S_data is a data sensitivity score, which assigns different scores based on data classification and grading labels (such as normal, sensitive, and extremely sensitive).

[0047] S_behavior is a user behavior deviation score, and its calculation formula is as follows:

[0048] S_behavior=1-(|B_current-B_baseline| / B_max);

[0049] Wherein, B_current is the current behavior feature vector, B_baseline is the historical behavior baseline vector, and B_max is the maximum allowable deviation threshold;

[0050] α, β, and γ are weight coefficients, and α+β+γ=1.

[0051] The optimization formula for the medical data lifecycle protection module to dynamically adjust encryption strength based on data sensitivity tags is as follows:

[0052] E_strength=Base_strength+k*(S_data-S_threshold);

[0053] Where E_strength is the actual encryption strength (such as key length), Base_strength is the basic encryption strength, S_data is the data sensitivity score, S_threshold is the sensitivity threshold for triggering enhanced encryption, and k is the strength adjustment coefficient.

[0054] The AI ​​model credibility measurement module uses the following optimization formula to measure the credibility of the model:

[0055] M_trust=λ1*D_integrity+λ2*A_robustness+λ3*E_security;

[0056] Where M_trust is the model trustworthiness; D_integrity is the training data integrity metric; A_robustness is the model algorithm robustness metric; E_security is the runtime environment security metric, obtained by measuring the trustworthiness of the computing environment (such as a container or virtual machine) in which the model runs; λ1, λ2, and λ3 are weight coefficients.

[0057] In one embodiment of the present invention, a network security protection method based on trusted computing includes the following steps:

[0058] S1: When the system is powered on, the trusted platform control module in the protection subsystem starts up before the CPU of the computing subsystem and acts as the root of trust.

[0059] S2: The protection subsystem actively measures the BIOS, bootloader, operating system kernel, system services and applications of the computing subsystem through a trusted software base to build a chain of trust;

[0060] S3: During the operation of the computing subsystem, the dynamic trust assessment engine collects device status, data access behavior, and user operation logs in real time to calculate the dynamic trust score T.

[0061] S4: Based on the dynamic trust score T, dynamically adjust the access control policy to implement stricter permission restrictions or isolation for low-trust entities;

[0062] S5: When medical data is accessed or transmitted, the medical data lifecycle protection module dynamically calculates the encryption strength E_strength based on its sensitivity label and implements encryption, while recording the operation log.

[0063] S6: When the medical AI model is being trained or inferred, the AI ​​model trust measurement module measures the data, algorithm, and environment respectively, and calculates the model trustworthiness M_trust. The model is allowed to be loaded or run only when M_trust is higher than a preset threshold.

[0064] In Embodiment 1 of the present invention, as Figure 1 As shown, this method is applied to top-tier hospitals to enhance the security of their core HIS and PACS systems.

[0065] System Deployment:

[0066] Trusted computing platforms are deployed at key nodes in the hospital's data center, including servers, doctor workstations, and network switches. Each platform comprises a computing subsystem and a protection subsystem. The protection subsystem is based on a trusted motherboard with a built-in TPCM chip. The TPCM chip incorporates domestically developed cryptographic algorithms (SM2 / SM3 / SM4).

[0067] Workflow:

[0068] Trust Chain Construction: After the server powers on, TPCM starts first to measure the integrity of the server BIOS. If the BIOS hash value matches the baseline value stored in TPCM, the CPU is allowed to boot and load the BIOS. Subsequently, TPCM, through the measurement agent in the Trusted Software Base, sequentially measures the integrity of the OSLoader, operating system kernel, system services, and HIS / PACS applications. If any step fails to measure, TPCM will prevent booting or issue an alarm according to its policy, ensuring that the system is in a trusted state from the start.

[0069] Dynamic trust assessment:

[0070] Device status monitoring: On the doctor's workstation terminal, the dynamic trust assessment engine continuously monitors parameters such as CPU utilization, memory usage, disk I / O, and network connection count. For example, when the CPU utilization of a terminal continues to rise abnormally (possibly due to a mining virus infection), its S_state score decreases.

[0071] Data sensitivity awareness: When doctors access patients' electronic medical records, the system automatically identifies the sensitivity labels of the medical record data (such as "patient genetic information - extremely sensitive") and assigns a high S_data score.

[0072] Behavioral baseline comparison: The system establishes a normal behavioral baseline for each doctor account (such as routine department visits and query time patterns). If an account attempts to download a large amount of patient image data in batches outside of working hours, its S_behavior score will significantly decrease.

[0073] Overall Scoring and Policy Execution: The engine calculates the overall trust score T based on the formula T=α*S_state+β*S_data+γ*S_behavior. For requests to access highly sensitive data, if T is below the threshold (e.g., due to device malfunction or suspicious behavior), the system will automatically trigger secondary authentication (e.g., SMS verification code + facial recognition) or restrict access permissions, allowing only the viewing of anonymized data.

[0074] Dynamic Data Encryption: When doctors access patient CT images through the PACS system, the medical data lifecycle protection module identifies the image data as "sensitive." Based on the formula E_strength = Base_strength + k * (S_data - S_threshold), the system decides to use a 256-bit key with the SM4 algorithm for encrypted transmission, instead of the basic 128-bit key. Simultaneously, all data access operations (including the accessor, time, operation type, and data identifier) ​​are recorded in a blockchain-based trusted audit log, ensuring that operations are tamper-proof and traceable.

[0075] Continuous protection: During system operation, the protection subsystem continuously monitors critical resources of the computing subsystem, such as memory, processes, and network connections, through TPCM to prevent attacks such as malicious code injection and illegal process creation, achieving "testability and controllability throughout the computing process".

[0076] Implementation results:

[0077] After deployment, the system successfully defended against multiple scans and penetration attempts targeting the hospital's intranet. In a simulated ransomware attack drill, the protection subsystem, before the virus attempted to encrypt files, promptly identified and blocked the attack process by measuring its behavioral characteristics (abnormal file operations) and code integrity, ensuring the continuous operation of the HIS / PACS system. Simultaneously, the dynamic trust assessment mechanism effectively reduced the risk of unauthorized access to sensitive data by internal personnel, significantly decreasing the incidence of data breaches.

[0078] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0079] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A network security protection system based on trusted computing, characterized by, include: The computing subsystem and the protection subsystem operate in parallel; the protection subsystem uses the trusted platform control module as the root of trust, the trusted platform control module starts before the CPU of the computing subsystem, and controls the startup and operation of the computing subsystem; The dynamic trust assessment engine is used to dynamically calculate entity trust scores based on real-time device status parameters, data sensitivity tags, and user behavior baselines, and adjust access control policies accordingly. The medical data lifecycle protection module is used to dynamically select encryption algorithms and key lengths based on data sensitivity tags, and to perform trusted auditing of data operations throughout the entire process. The AI ​​model credibility measurement module is used to measure the credibility of model data, algorithms, and environment.

2. The network security protection system based on trusted computing according to claim 1, characterized in that: The optimized formula for calculating entity trust scores by the dynamic trust assessment engine is as follows: T=α*S_state+β*S_data+γ*S_behavior; Where T is the dynamic trust score; S_state is the device's real-time status parameter score, and its calculation formula is: S_state=Σ(w_i*(v_i / V_max_i)); Where v_i is the real-time value of the i-th state parameter, V_max_i is the preset threshold of the parameter, w_i is the weight of the parameter, and Σw_i=1; S_data represents the data sensitivity score; S_behavior is a user behavior deviation score, and its calculation formula is as follows: S_behavior=1-(|B_current-B_baseline| / B_max); Wherein, B_current is the current behavior feature vector, B_baseline is the historical behavior baseline vector, and B_max is the maximum allowable deviation threshold; α, β, and γ are weight coefficients, and α+β+γ=1.

3. The network security protection system based on trusted computing according to claim 1, characterized in that: The optimization formula for the medical data lifecycle protection module to dynamically adjust encryption strength based on data sensitivity tags is as follows: E_strength=Base_strength+k*(S_data-S_threshold); Where E_strength is the actual encryption strength, Base_strength is the basic encryption strength, S_data is the data sensitivity score, S_threshold is the sensitivity threshold for triggering enhanced encryption, and k is the strength adjustment coefficient.

4. The network security protection system based on trusted computing according to claim 1, characterized in that: The optimization formula for the AI ​​model credibility measurement module to measure the credibility of the model is as follows: M_trust=λ1*D_integrity+λ2*A_robustness+λ3*E_security; Where M_trust is the model trustworthiness; D_integrity is the training data integrity metric; A_robustness is the model algorithm robustness metric; E_security is the runtime environment security metric; and λ1, λ2, and λ3 are weight coefficients.

Images