Digital pathology slide automatic scanning and fusion system based on multi-modal imaging

Abstract

The application discloses a digital pathological section automatic scanning fusion system based on multi-modal imaging and relates to the technical field of pathological section scanning.The application comprises a hardware identity authentication module, a key management module, a data encryption module, an access control module, a temporary permission token module, a cloud synchronization module and an audit log module, the hardware identity authentication module and the key management module perform identity authentication on a user, the key management module and the data encryption module perform slide scanning and encrypt WSI files, the access control module and the temporary permission token module authorize the user to access the encrypted WSI files, and the issuance and use of the token are performed, the application adopts independent encryption and binds different ciphertext policy attributes and encryption access strategies, achieves the effect that the same file is seen as different contents by different roles and is automatically kept encrypted, and through the design of a weighted temporary access token, the effect that temporary authorization automatically expires and misuse behavior is blocked in real time is achieved.

Description

Technical Field

[0001] This invention relates to the field of pathological slide scanning technology, specifically to an automatic scanning and fusion system for digital pathological slides based on multimodal imaging. Background Technology

[0002] Digital pathology scanning systems use high-resolution cameras to automatically scan biological tissue slides across the entire field of view, converting physical slides into digital whole-slide images. These images are stored in a computer system for pathologists to view, annotate, and diagnose. A whole-slide image typically contains image pyramid data at multiple magnifications, basic patient information, specimen collection and processing information, and diagnostic annotations. In clinical workflows, the digital slides generated by the scans need to be accessed by users in different roles. For example, pathologists interpret the slides for diagnosis, junior physicians use them for study and reference, clinicians review and consult on them, quality control personnel conduct quality audits, and researchers extract anonymized data for statistical analysis. Each role has different access needs for different information layers within the same whole-slide image. For instance, researchers only need the image data and should not have access to patient identification information, while junior physicians need to observe tissue morphology, but the final diagnostic conclusion should be confirmed by a senior physician.

[0003] The existing data security mechanisms of digital pathology scanning systems have the following defects: 1. Identity authentication relies on single-factor authentication using usernames and passwords. Password leakage or weak passwords can lead to unauthorized access to the system. Full-slice images are stored as a single file, either encrypted or unencrypted, failing to provide differentiated protection for information layers with different sensitivity levels. Once the decryption key is leaked, all data is exposed. Access control is based on a preset fixed role list, lacking dynamic judgment of the context of access, such as time, location, device, and case relevance, which does not comply with the minimum necessity principle. 2. Cross-role temporary authorization relies on manual activation and deactivation of account permissions, which poses the risk of long-term permission retention due to forgetting to revoke authorization. Audit logs are stored as plain text files, lacking anti-tampering mechanisms. Log records can be modified or deleted unilaterally by administrators, resulting in insufficient anti-tampering security. Data backup strategies are simplistic, lacking hierarchical synchronization mechanisms and cloud-based encrypted disaster recovery solutions. Local storage failures pose a risk of permanent data loss and excessively long recovery times. Therefore, designing an automatic scanning and fusion system for digital pathology slices based on multimodal imaging is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an automated scanning and fusion system for digital pathological slides based on multimodal imaging, which solves the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an automatic scanning and fusion system for digital pathological slides based on multimodal imaging, comprising a hardware authentication module, a key management module, a data encryption module, an access control module, a temporary access token module, a cloud synchronization module, and an audit log module. The output of the hardware authentication module is connected to the input of the key management module, the output of the key management module is connected to the input of the data encryption module, the outputs of both the key management module and the data encryption module are connected to the input of the cloud synchronization module, the port of the data encryption module establishes bidirectional communication with the port of the access control module, the output of the temporary access token module is connected to the input of the access control module, and the outputs of the hardware authentication module, key management module, data encryption module, access control module, temporary access token module, and cloud synchronization module are all connected to the input of the audit log module. The system hardware consists of a scanner host, a security workstation, a cloud security service cluster, and a management terminal. The scanner host consists of an autoloading unit, a door control panel, a camera unit, an LED light source box, a fingerprint sensor, an NFC card reader, an infrared thermal imaging sensor, and a FIDO2 slot. The security workstation consists of a motherboard with an integrated TPM security chip, a processor, memory, a hard drive, a USB controller, an Ethernet card, and a monitor. The cloud security service cluster consists of a server with an integrated HSM module and an object storage array. The management terminal consists of a PC host and an NFC card reader. The hardware authentication module and key management module perform user authentication; the key management module and data encryption module perform slide scanning and encrypt WSI files; the access control module and temporary permission token module authorize users to access encrypted WSI files and perform token issuance and use; the cloud synchronization module and key management module synchronize local encrypted data to the cloud security service cluster; and the audit log module performs continuous recording of audit logs, block construction, and chain integrity verification.

[0006] Furthermore, the hardware authentication module and key management module perform user authentication, including the hardware authentication module being connected to the fingerprint sensor, NFC card reader, infrared thermal imaging sensor, FIDO2 security key slot and door control panel respectively; The NFC reader reads the employee card UID and transmits it to the hardware authentication module via USB interface. The hardware authentication module queries the local user database and matches the user record corresponding to the employee card UID. If the match is successful, the hardware authentication module sends a collection command to the fingerprint sensor via the USB controller. The fingerprint sensor is activated by illuminating its indicator light. The user places their finger on the surface of the fingerprint sensor, and the sensor collects the fingerprint image and extracts the fingerprint feature vector. This image is then transmitted to the hardware authentication module via USB interface. The hardware authentication module compares the fingerprint feature vector with the fingerprint template stored in the user record and calculates a similarity score based on the minutiae matching algorithm. The fingerprint feature vector extracted by the fingerprint sensor is essentially a set of several minutiae. Each minutiae includes three elements: x-coordinate, y-coordinate, and orientation angle θ. The error of each of the three elements between the fingerprint feature vector and each minutiae of the fingerprint template is calculated. Subtracting the average of the errors of all three elements of the minutiae from one yields the similarity score. If the similarity score is lower than the threshold, authentication failure is returned and the failure event is recorded in the audit log chain. If the current operation involves highly sensitive operations such as modifying system parameters or batch exporting, the hardware authentication module sends a challenge value to the FIDO2 security key slot via the USB interface. The user inserts the FIDO2 physical key and completes the press confirmation. The FIDO2 key signs the challenge value and returns the signature value. The hardware authentication module uses the public key corresponding to the key to verify the signature. After all verifications are successful, the hardware authentication module transmits the UUID and authentication method encoding to the key management module. Loading the user key: The key management module receives the UUID and sends a KEY_LOAD request to the TPM. The request parameters include the UUID binding identifier. The TPM uses its internal MK to decrypt the ciphertext of the user key KEK_User to obtain the plaintext. The plaintext KEK_User is loaded into the TPM's internal key slot, and the key handle is returned to the key management module. The key management module caches the key handle in memory for subsequent calls.

[0007] Furthermore, the key management module and data encryption module perform slide scanning and WSI file encryption, including generating a session DEK. At the start of each scanning session, i.e. after the user clicks the scan button in the scanning software, the scanning software sends a DEK_GEN request to the key management module. The key management module sends a Tpm2_GetRandom instruction to the TPM via the LPC bus to request a random number. The TPM's internal true random number generator generates a random number, which serves as the data encryption key DEK_session for this session. The key management module returns the encrypted key DEK_session in plaintext to the data encryption module and simultaneously sends a Tpm2_Encrypt instruction to the TPM to encrypt DEK_session using KEK_User to obtain DEK_encrypted, which is then stored in memory. The image acquisition process involves the scanning software controlling the automatic loading unit to grab the slide holder from the slide compartment and move it to the stage. The camera unit acquires image data according to the focal length magnification and exposure parameters set in the scanning scheme. Bright field scanning uses an LED light source box to provide transmitted illumination, while fluorescence scanning uses an LED excitation light source of the corresponding wavelength and a filter kit. The camera transmits the image data stream to the workstation memory buffer via the USB protocol. After image acquisition, the WSI layer splitter reads the complete WSI image data from the memory buffer. The splitter extracts the 10x magnified image data into an overview layer according to the image pyramid hierarchy, merges the 20x and 40x magnified image data into a high-resolution layer, and extracts patient information, specimen number, staining method, and other data from the scan protocol metadata to construct the patient metadata layer. If there is already annotation information during the image reading process, it is extracted into an annotation and diagnosis layer. If it is a new scan, this layer is initially empty. Perform layered encryption and policy binding; WSI Secure File Encapsulation: The WSI file encapsulator serializes the payloads of the four encryption layers, DEK_encrypted, the magic number in the file header, the version number, the layer index table, and the integrity check hash value into a single file in TLV format and stores it in the D drive directory of the workstation hard drive. The file path and file name are recorded in the local database. An audit log is generated. The event capturer captures the scan completion event, records the operator's UUID, operation type, target WSI identifier, timestamp, and file hash value, and pushes it to the block builder.

[0008] Furthermore, the access control module and the temporary permission token module authorize users to access encrypted WSI files, and perform token issuance and use, including initiating an access request. The user clicks to load an image in the image viewing software, selects the target WSI file, the image viewing software reads the WSI security file header, obtains the encrypted payload list and the CP-ABE ciphertext policy digest of each layer, and the image viewing software sends ACCESS_REQUEST to the access control module. The request parameters include the user UUID, the target WSI identifier, and the request operation type. After receiving ACCESS_REQUEST, the attribute collector of the access control module begins to construct a complete attribute set for the current access. It queries the static role corresponding to the user's UUID from the role library and collects time attribute, location attribute, device attribute, case relationship attribute, and diagnosis status attribute respectively. The time attribute is obtained by querying the system clock to determine whether the current period is working or non-working. The location attribute is obtained by checking whether the IP address of the request source matches the hospital's intranet segment whitelist. The device attribute is obtained by checking whether the MAC address of the requesting device is in the list of registered devices. The case relationship attribute is obtained by querying the electronic medical record system interface to confirm whether the user is the attending physician or authorized consulting physician of the target case. The diagnosis status attribute is obtained by querying whether the diagnosis report corresponding to the WSI has been issued. The attribute collector then transmits the complete attribute set to the policy decision point. Execution strategy evaluation and hierarchical decryption; When issuing a WTAT token, if the authorizer needs to grant temporary access to the authorized person, the authorizer initiates a CREATE_TOKEN operation in the image viewing software. Upon receiving the request, the token generator calls the weight calculation engine. The weight calculation engine reads the authorizer's role level L_auth and the authorized person's role level L_grantee, reads the validity period T, and calculates the weight W_total using the following formula:

[0009] The formula result is rounded down, the token generator constructs a WTAT data structure, signs the token content using the authorizer's private key, stores the token in the local token database, and sends a token notification to the authorized person. The weight consumption is calculated in real time when the token is used.

[0010] Furthermore, the cloud synchronization module and key management module synchronize local encrypted data to the cloud security service cluster, including synchronization task scheduling. The synchronization scheduler checks the synchronization queue in a polling manner. When metadata and transaction log changes occur, they are immediately placed into the P0 queue. Diagnostic report issuance events trigger the enqueueing of the P1 queue. The scheduler traverses the modification timestamps of local WSI files and enqueues the modified files of the day into the P2 queue. User logout events and disk usage monitoring trigger the enqueueing of the P3 queue. The scheduler dequeues tasks in priority order from P0 to P3. Low-priority queues are not processed when high-priority queues are not empty. Bandwidth control: Before sending data, the scheduler checks the system clock. If the current time is in the first time period, the sending rate is set to 10Mbps. If the current time is in the second time period, no rate limit is set. If the current time is in the third time period, the sending rate is set to 50Mbps. Zero-knowledge upload: The zero-knowledge upload engine obtains the KEK_Sync ciphertext from the key management module. This ciphertext is protected by MK encryption. The zero-knowledge upload engine splits the KEK_Sync ciphertext into component A and component B. Component A is stored inside the workstation TPM, and component B is transmitted to the cloud HSM through a TLS1.3 encrypted channel. The cloud uses HSM to securely store component B. The zero-knowledge upload engine directly transmits the encrypted WSI file to the cloud object storage through the TLS1.3 channel. The cloud object storage only holds the AES-256-GCM encrypted ciphertext data and the KEK_Sync ciphertext component B. It cannot restore the KEK_Sync ciphertext and cannot decrypt the WSI data. Cross-region redundancy and version management: The cloud storage service writes a complete data copy. The write operation adopts a synchronous confirmation mechanism. The upload is only returned after at least two availability zones confirm that the write is successful. The version management service checks the version history of the written object. If the difference between the current time and the initial creation time of the object is less than or equal to seven days, the full version is retained. If the difference is greater than seven days and less than or equal to three years, the most recent weekly snapshot is retained and intermediate versions are deleted. In disaster recovery, when a storage failure or data corruption occurs on a local workstation, the administrator initiates a disaster recovery process through the management terminal. The recovery program downloads a specified version of the encrypted WSI file from the cloud object storage. The recovery program requests KEK_Sync component B from the cloud HSM and component A from the local TPM. The two components are merged and restored in the memory of the secure workstation to restore KEK_Sync, and then the DEK is decrypted. Finally, the WSI data is decrypted and written to local storage. After the recovery is completed, the audit log chain records the recovery event, including the recovery time, the list of recovered WSIs, and the data version.

[0011] Furthermore, the audit log module performs continuous recording of audit logs, block construction, and chain integrity verification, including event capture and preprocessing. The event capturer is embedded in the interface layer of each security module in the form of hook functions. Each time a security operation occurs, the corresponding module calls the event capturer interface, passes the event type enumeration value and the event payload, the event capturer generates a globally unique event ID, takes the system millisecond-level timestamp, calls SHA256 to calculate the event hash value of the event payload, and writes the event record into the memory event buffer. Execute the construction block; Chain integrity verification: The chain integrity verifier performs a full chain verification when the system starts up. It traverses each block backward from the end of the local audit_chain.dat file, calculates the SHA256 of each block, and compares it with the H_prev stored in the next block. If all match, the verification passes. If a mismatch occurs in the kth block, the block from k+1 to the end is marked as corrupt, triggering the recovery of that block from the cloud replica. The verifier performs incremental verification after generating ten new blocks, only verifying the consistency of the hash chain of the most recent twenty blocks. The verifier periodically performs cross-verification between local and cloud replicas, randomly selecting ten blocks and comparing the consistency between the local hash value and the cloud hash value.

[0012] Furthermore, the layered encryption and policy binding includes performing the following operations sequentially on each logical layer: the AES-256-GCM encryption engine uses DEK_session to encrypt the plaintext of the logical layer to obtain the first ciphertext AES; the encryption engine generates a random initialization vector IV; calls the AES-256-GCM encryption function; the input parameters are DEK_session, IV and the plaintext of the logical layer; the output is ciphertext and authentication tag; and the IV is prepended to the ciphertext header. The CP-ABE policy binder reads the access policy expression corresponding to this layer, uses DEK_session as the encryption object, calls the CP-ABE encryption function, takes the policy expression and DEK_session plaintext as input parameters, and outputs the second ciphertext CP-ABE. The first ciphertext AES, IV, authentication tag, and the second ciphertext CP-ABE are encapsulated as the cryptographic payload of this layer.

[0013] Furthermore, the policy evaluation and hierarchical decryption include traversing the WSI encryption layer by policy decision point, performing CP-ABE decryption attempts for each layer, inputting the user attribute set and the CP-ABE ciphertext of that layer into the CP-ABE decryption engine, and the decryption engine using the user attribute private key to attempt to decrypt the CP-ABE ciphertext. If the attribute set satisfies the policy conditions, decryption is successful and the DEK_session of that layer is obtained; otherwise, the layer is marked as inaccessible. For accessible layers, the policy decision point uses the obtained DEK_session to call the decryption function, takes DEK_session, IV and ciphertext as input, outputs plaintext layer data, verifies the authentication tag, and passes the decrypted layer data to the image viewing software for rendering and display; The above execution results show that different roles see different content in the same WSI file. Junior physicians only successfully decrypted the overview layer, while attending physicians successfully decrypted all four layers. When researchers conduct blinded studies, the patient metadata layer strategy includes the purpose_diagnosis condition, but the researcher attributes do not include the diagnosis attribute, so the patient metadata layer remains encrypted.

[0014] Furthermore, the weight consumption when the token is used includes the following: When an authorized user accesses the target WSI using a token, the consumption monitor calculates the weight consumption in real time. When an authorized user browses the overview layer, the consumption monitor deducts 1 point from the remaining weight W_remaining every minute. When an authorized user switches to a high-resolution layer, the consumption monitor deducts 3 points every minute. When an authorized user performs an export screenshot operation, the consumption monitor deducts 15 points per instance. The consumption monitor checks W_remaining after each deduction. If W_remaining is less than or equal to zero, it notifies the token generator to set the token status to EXPIRED, and the access control module rejects subsequent access requests. Anomaly detection: The anomaly detector continuously monitors token usage patterns. If it detects that the weight consumption exceeds half of W_total within five minutes, it sends an alert to the authorizing party. If it detects that the same token is used simultaneously from two different IP addresses, it immediately sets the token status to LOCKED, rejects all subsequent requests, and generates a SECURITY_ALERT event and pushes it to the audit log chain.

[0015] Furthermore, the construction of the block includes a block builder monitoring a memory event buffer. When the number of accumulated events in the buffer reaches one hundred, or when the time since the previous block was generated exceeds sixty seconds and the buffer is not empty, block construction is triggered. During block construction, the block builder extracts all event records from the buffer, sorts them in ascending order by timestamp, calculates the SHA256 hash value h_i for each event record, and constructs a Merkle tree. If N is even, the adjacent leaf nodes h_2j and h_2j+1 are concatenated and their SHA256 hashes are calculated to obtain the parent node. If N is odd, the last leaf node h_N-1 is copied and its SHA256 hash is calculated. This process is repeated until only one hash value remains to obtain the Merkle tree root, Merkle_root. The block constructor reads the SHA256 hash value H_prev of the previous block, combines it with the current timestamp and Merkle_root to form the block header. After serializing the block header and block body, the SHA256 hash value H_current of the current block is calculated. The block is written to the end of the local audit_chain.dat file and simultaneously transmitted to the cloud audit log replica library through the P0-level synchronization channel.

[0016] The present invention has the following beneficial effects: 1. By splitting each full-slice image into four independent logical layers—an overview layer, a high-resolution layer, a patient metadata layer, and a labeling and diagnosis layer—each layer employs independent encryption and is bound to different ciphertext policy attribute-based encryption access policies. The attribute private keys held by different roles can only decrypt the layers that meet the policy conditions. Compared with existing single-file overall encryption technologies, this invention achieves the effect of different roles seeing different content in the same file and the patient information layer automatically remaining encrypted when researchers conduct blinded studies.

[0017] 2. By designing a weighted temporary access token, the authorizer dynamically calculates the total weight limit based on the formula W_total = 100 multiplied by L_auth divided by L_grantee multiplied by T divided by 2 when issuing the token. The system's built-in operation sensitivity weight table automatically deducts the corresponding weight for each operation. The token automatically expires when the weight reaches zero or the time expires. The anomaly detection module triggers an alert or locks the device when more than half of the token is used within five minutes or when the same token is used by multiple IPs. Compared with the existing manual authorization and revocation management methods, this invention achieves the effect of automatic expiration of temporary authorization and real-time blocking of abuse.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a block diagram of the digital pathological slide automatic scanning and fusion system based on multimodal imaging according to the present invention. Detailed Implementation

[0021] 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.

[0022] Please see Figure 1This invention provides a technical solution: an automatic scanning and fusion system for digital pathological slides based on multimodal imaging, comprising a hardware authentication module, a key management module, a data encryption module, an access control module, a temporary permission token module, a cloud synchronization module, and an audit log module. The output of the hardware authentication module is connected to the input of the key management module, the output of the key management module is connected to the input of the data encryption module, the outputs of both the key management module and the data encryption module are connected to the input of the cloud synchronization module, the port of the data encryption module establishes bidirectional communication with the port of the access control module, the output of the temporary permission token module is connected to the input of the access control module, and the outputs of the hardware authentication module, key management module, data encryption module, access control module, temporary permission token module, and cloud synchronization module are all connected to the input of the audit log module. The hardware authentication module and key management module perform user authentication; the key management module and data encryption module perform slide scanning and encrypt WSI files; the access control module and temporary permission token module authorize users to access encrypted WSI files and perform token issuance and use; the cloud synchronization module and key management module synchronize local encrypted data to the cloud security service cluster; and the audit log module performs continuous recording of audit logs, block construction, and chain integrity verification.

[0023] Among them, the hardware authentication module and the key management module perform user authentication, including the hardware authentication module being connected to the fingerprint sensor, NFC card reader, infrared thermal imaging sensor, FIDO2 security key slot and door control panel respectively; The NFC reader reads the work card UID and transmits it to the hardware authentication module. The hardware authentication module queries the local user database and matches the user record. If the match is successful, the hardware authentication module sends a collection command to the fingerprint sensor. The user places their finger on the surface of the fingerprint sensor. The sensor collects the fingerprint image and extracts the fingerprint feature vector, which is then transmitted to the hardware authentication module. The hardware authentication module compares the fingerprint feature vector with the fingerprint template stored in the user record and calculates the similarity score based on the minutiae matching algorithm. If the similarity is lower than the threshold of 95%, authentication failure is returned and the failure event is recorded to the audit log chain. By collecting the operator's fingerprint feature vector with a fingerprint sensor, reading the unique identifier of the work card with an NFC card reader, and verifying the physical token signature with a FIDO2 security key, the three factors work together to complete hardware-level identity authentication. The infrared thermal imaging sensor detects the operator's presence every five seconds and automatically locks the interface and clears the plaintext key cache in memory after the person leaves for more than 30 seconds. Compared with the existing single-factor authentication technology of username and password, this invention achieves the effect of preventing unauthorized access caused by credential leakage. If the current operation involves highly sensitive operations such as modifying system parameters or batch exporting, the hardware authentication module sends a challenge value to the FIDO2 security key slot. The user inserts the FIDO2 physical key and completes the press confirmation. The FIDO2 key signs the challenge value and returns the signature value. The hardware authentication module uses the public key corresponding to the key to verify the signature. After all verifications are successful, the hardware authentication module transmits the UUID and authentication method encoding to the key management module. The key management module receives the UUID and sends a KEY_LOAD request to the TPM. The request parameters include the UUID binding identifier. The TPM uses its internal MK to decrypt the ciphertext of the user key KEK_User to obtain the plaintext, loads the plaintext KEK_User into the TPM's internal key slot, and returns the key handle to the key management module. The key management module caches the key handle in memory. By permanently storing the master key MK in the TPM security chip and making it non-exportable, protecting the role key and user key with encryption by MK, and generating the data encryption key independently for each scanning session, and by combining this with Shamir threshold secret sharing to split the root authorization key into five shares held by multiple parties and requiring at least three parties to be present for collaborative authorization when exporting in bulk or resetting the system, this invention achieves the effect of limiting the impact of key leakage to a single session and requiring multiple parties to cooperate in the abuse of the root key, compared to existing single-key encryption technologies.

[0024] The key management module and data encryption module perform slide scanning and WSI file encryption, including generating the session DEK. At the start of each scanning session, i.e. after the user clicks the scan button in the MD9 scanning software, the scanning software sends a DEK_GEN request to the key management module. The key management module sends a Tpm2_GetRandom command to the TPM via the LPC bus to request a 256-bit random number. The TPM's internal true random number generator generates a 256-bit random number, which is used as the data encryption key DEK_session for this session. The key management module returns the encrypted key DEK_session in plaintext to the data encryption module and simultaneously sends a Tpm2_Encrypt command to the TPM to encrypt DEK_session using KEK_User to obtain DEK_encrypted, which is stored in memory. The MD9 scanning software controls the autoloading unit to grab the slide holder from the slide compartment and move it to the stage. The camera unit acquires image data, and the camera transmits the image data stream to the workstation memory buffer. After image acquisition is completed, the WSI layer splitter reads the complete WSI image data from the memory buffer. The splitter extracts the 10x magnified image data into an overview layer according to the image pyramid hierarchy, merges the 20x and 40x magnified image data into a high-resolution layer, and extracts patient information, specimen number, staining method and other data from the scan protocol metadata to construct the patient metadata layer. If there is already annotation information during the image reading process, it is extracted into an annotation and diagnosis layer. If it is a new scan, this layer is initially empty. Perform layered encryption and policy binding; The WSI file wrapper serializes the payloads of the four encryption layers, DEK_encrypted, the magic number in the file header, the version number, the layer index table, and the integrity check hash value into a single file in TLV format and stores it in the KR directory on the D drive of the workstation. The file path and file name are recorded in the local database. The event capturer captures the scan completion event, records the operator UUID, operation type SCAN_COMPLETE, target WSI identifier, timestamp, and file hash value, and pushes it to the block builder.

[0025] The access control module and the temporary permission token module authorize users to access encrypted WSI files and execute the issuance and use of tokens. This includes the user clicking to load an image in the MD9 image viewing software, selecting the target WSI file, the image viewing software reading the WSI security file header, obtaining the four-layer encrypted payload list and the CP-ABE ciphertext policy digest for each layer, and the image viewing software sending ACCESS_REQUEST to the access control module. The request parameters include the user's UUID, the target WSI identifier, and the request operation type, such as browsing or exporting. After receiving ACCESS_REQUEST, the attribute collector of the access control module begins to construct a complete set of attributes for the current access. It queries the role library for the static role corresponding to the user's UUID, such as chief_physician, and collects time attributes, location attributes, device attributes, case relationship attributes, and diagnosis status attributes respectively. The complete set of attributes is then passed to the policy decision point. By overlaying five runtime dynamic attributes—time, location, device, case relationship, and diagnosis status—on top of six static roles, access decisions depend on both role identity and context. Compared to existing access control technologies that rely solely on a fixed list of roles, this invention achieves the effect of granting access in accordance with the minimum necessary principle and automatically denying access in abnormal environments. Execute CP-ABE strategy evaluation and hierarchical decryption; When an authorizer needs to grant temporary access to an authorized person, such as a chief physician temporarily authorizing a junior physician to view a high-resolution layer of a WSI (Web Image Source System), the authorizer initiates a CREATE_TOKEN operation in the MD9 image viewing software. After receiving the request, the token generator calls the weight calculation engine. The weight calculation engine reads the authorizer's role level L_auth and the authorized person's role level L_grantee, reads the validity period T, and calculates the weight W_total. The calculation formula is as follows:

[0026] The following is a calculation example: The chief physician L_auth equals 5 issues a token with a validity period of T equals 2 hours to the junior physician L_grantee equals 3, with weight W_total=166. The token generator constructs a WTAT data structure, signs the token content using the authorizer's attribute private key, stores the token in the local token database, and sends a token notification to the authorized person. The weight consumption is calculated in real time when the token is used.

[0027] Among them, the cloud synchronization module and the key management module synchronize local encrypted data to the cloud security service cluster. This includes the synchronization scheduler checking the synchronization queue in a polling manner, putting metadata and transaction log changes into the P0 queue immediately, the diagnostic report issuance event triggering the P1 queue to be enqueued, at 2:00 AM every day, the scheduler traverses the local WSI file modification timestamps and enqueues the modified files of the day into the P2 queue, user logout events and disk usage monitoring triggering the P3 queue to be enqueued, and the scheduler dequeues tasks in priority order from P0 to P3. Low priority queues are not processed when high priority queues are not empty. Before sending data, the scheduler checks the system clock. If the current time is in the first time period (8:00 to 20:00), the sending rate is set to 10Mbps. If the current time is in the second time period (0:00 to 6:00), no rate limit is set. If the current time is in the third time period (6:00 to 8:00 or 20:00 to 0:00), the sending rate is set to 50Mbps. Through a four-level priority scheduling strategy of real-time synchronization of metadata and transaction logs, immediate synchronization after diagnostic report issuance, daily early morning synchronization of full-slice images, and event-driven synchronization, the scheduler limits the rate during working hours to ensure the bandwidth of scanning services. The zero-knowledge upload engine obtains the KEK_Sync ciphertext from the key management module. This ciphertext is protected by MK encryption. The zero-knowledge upload engine splits the KEK_Sync ciphertext into component A and component B. Component A is stored inside the workstation TPM, and component B is transmitted to the cloud HSM through a TLS1.3 encrypted channel. The cloud uses HSM to securely store component B. The zero-knowledge upload engine directly transmits the encrypted WSI file to the cloud object storage through the TLS1.3 channel. The cloud object storage only holds the AES-256-GCM encrypted ciphertext data and the KEK_Sync ciphertext component B. It cannot restore the KEK_Sync ciphertext and cannot decrypt the WSI data. The cloud storage service writes complete data copies to three availability zones. The write operation adopts a synchronous confirmation mechanism. The upload is only returned after at least two availability zones confirm that the write is successful. The version management service checks the version history of the written object. If the difference between the current time and the initial creation time of the object is less than or equal to seven days, the full version is retained. If the difference is greater than seven days and less than or equal to three years, the most recent weekly snapshot is retained and intermediate versions are deleted. The cloud-based zero-knowledge architecture, which stores keys separately in the host TPM and cloud hardware security modules, prevents any single party from decrypting the stored data. The redundant storage across three availability zones, combined with the version retention strategy of full version within seven days to weekly snapshots within three years, achieves the effect of data recovery point target not exceeding five minutes and recovery time target not exceeding twelve hours, compared to existing single full backup or local backup-only technologies. When a local workstation experiences storage failure or data corruption, the administrator initiates a disaster recovery process through the management terminal. The recovery program downloads a specified version of the encrypted WSI file from the cloud object storage. The recovery program requests KEK_Sync component B from the cloud HSM and component A from the local TPM. The two components are merged and restored in the memory of the secure workstation to restore KEK_Sync, thereby decrypting DEK. Finally, the WSI data is decrypted and written to local storage. After the recovery is completed, the audit log chain records the recovery event, including the recovery time, the list of recovered WSIs, and the data version.

[0028] The audit log module performs continuous recording of audit logs, block construction, and chain integrity verification. This includes an event catcher embedded in the interface layer of each security module in the form of hook functions. Each time a security operation occurs, the corresponding module calls the event catcher interface, passes the event type enumeration value and the event payload, the event catcher generates a globally unique event ID, takes the system millisecond-level timestamp, calls SHA256 to calculate the event hash value of the event payload, and writes the event record into the memory event buffer. Execute the construction block; The chain integrity verifier performs a full chain check when the system starts up. It traverses each block backward from the end of the local audit_chain.dat file, calculates the SHA256 for each block, and compares it with the H_prev stored in the next block. If all match, the check passes. If a mismatch occurs in the kth block, the block from k+1 to the end is marked as corrupt, triggering the recovery of that block from the cloud replica. The verifier performs incremental verification after generating ten new blocks, only verifying the consistency of the hash chain of the most recent twenty blocks. The verifier periodically performs cross-verification between local and cloud replicas, randomly selecting ten blocks and comparing the consistency between the local hash value and the cloud hash value.

[0029] Among them, layered encryption and policy binding include performing the following operations sequentially on each logical layer: the AES-256-GCM encryption engine uses DEK_session to encrypt the plaintext of the logical layer to obtain the first ciphertext AES; the encryption engine generates a 96-bit random initialization vector IV; calls the AES-256-GCM encryption function with DEK_session, IV and plaintext of the logical layer as input parameters; and outputs ciphertext and a 128-bit authentication tag. The IV is prepended to the ciphertext header. The CP-ABE policy binder reads the access policy expression corresponding to this layer, uses DEK_session as the encrypted object, calls the CP-ABE encryption function, takes the policy expression and DEK_session plaintext as input parameters, and outputs the second ciphertext CP-ABE. This CP-ABE ciphertext means that only users whose attribute set satisfies the policy expression can use their attribute private key to decrypt and obtain DEK_session. The first ciphertext AES, IV, authentication tag, and the second ciphertext CP-ABE are encapsulated as the cryptographic payload of this layer.

[0030] The strategy evaluation and hierarchical decryption process includes traversing the four encryption layers of WSI at the strategy decision point, performing a CP-ABE decryption attempt for each layer, inputting the user attribute set and the CP-ABE ciphertext of that layer into the CP-ABE decryption engine, and the decryption engine using the user attribute private key to attempt to decrypt the CP-ABE ciphertext. If the attribute set meets the policy conditions, decryption is successful and the DEK_session of that layer is obtained; otherwise, the layer is marked as inaccessible. For accessible layers, the policy decision point uses the obtained DEK_session to call the AES-256-GCM decryption function, takes DEK_session, IV and ciphertext as input, outputs plaintext layer data, verifies the authentication tag, and passes the decrypted layer data to the MD9 image viewing software for rendering and display.

[0031] The weight consumption when using tokens includes the following: When an authorized user accesses the target WSI using a token, the consumption monitor calculates the weight consumption in real time. When an authorized user browses the overview layer, the consumption monitor deducts 1 point from the remaining weight W_remaining every minute. When an authorized user switches to a high-resolution layer, the consumption monitor deducts 3 points every minute. When an authorized user performs an export screenshot operation, the consumption monitor deducts 15 points per instance. The consumption monitor checks W_remaining after each deduction. If W_remaining is less than or equal to zero, it notifies the token generator to set the token status to EXPIRED, and the access control module rejects subsequent access requests. The anomaly detector continuously monitors token usage patterns. If it detects that the weight consumption exceeds half of W_total within five minutes, it sends an alert to the authorizing person. If it detects that the same token is used simultaneously from two different IP addresses, it immediately sets the token status to LOCKED, rejects all subsequent requests, and generates a SECURITY_ALERT event to push to the audit log chain.

[0032] The block construction process includes the block builder monitoring the memory event buffer. When the number of accumulated events in the buffer reaches one hundred, or when the time since the previous block was generated exceeds sixty seconds and the buffer is not empty, block construction is triggered. During block construction, the block builder extracts all event records from the buffer, sorts them in ascending order by timestamp, calculates the SHA256 hash value h_i for each event record, and constructs a Merkle tree. If N is even, the adjacent leaf nodes h_2j and h_2j+1 are concatenated and their SHA256 hashes are calculated to obtain the parent node. If N is odd, the last leaf node h_N-1 is copied and its SHA256 hash is calculated. This process is repeated until only one hash value remains to obtain the Merkle tree root, Merkle_root. The block constructor reads the SHA256 hash value H_prev of the previous block, combines it with the current timestamp and Merkle_root to form the block header. After serializing the block header and block body, the SHA256 hash value H_current of the current block is calculated. The block is written to the end of the local audit_chain.dat file and simultaneously transmitted to the cloud audit log replica library through the P0 level synchronization channel. By calculating the SHA256 hash value for each audit event, a chain structure is formed every sixty seconds or one hundred events, which includes the hash value of the previous block and the Merkle tree root. Modifying any historical record will destroy the hash consistency of all subsequent blocks. Local and cloud dual copies are cross-checked periodically. Compared with the existing technology of storing audit logs in text files, this invention achieves the effect of audit records being tamper-proof and having judicial-grade evidence integrity.

[0033] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An automated scanning and fusion system for digital pathological slides based on multimodal imaging, characterized in that: The system includes a hardware authentication module, a key management module, a data encryption module, an access control module, a temporary permission token module, a cloud synchronization module, and an audit log module. The output of the hardware authentication module is connected to the input of the key management module, the output of the key management module is connected to the input of the data encryption module, the outputs of both the key management module and the data encryption module are connected to the input of the cloud synchronization module, the port of the data encryption module establishes bidirectional communication with the port of the access control module, the output of the temporary permission token module is connected to the input of the access control module, and the outputs of the hardware authentication module, key management module, data encryption module, access control module, temporary permission token module, and cloud synchronization module are all connected to the input of the audit log module. The hardware authentication module and key management module perform user authentication; the key management module and data encryption module perform slide scanning and encrypt WSI files; the access control module and temporary permission token module authorize users to access encrypted WSI files and perform token issuance and use; the cloud synchronization module and key management module synchronize local encrypted data to the cloud security service cluster; and the audit log module performs continuous recording of audit logs, block construction, and chain integrity verification.

2. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 1, characterized in that, The hardware authentication module and key management module perform user authentication, including the hardware authentication module being connected to the fingerprint sensor, NFC card reader, infrared thermal imaging sensor, FIDO2 security key slot and door control panel respectively; The NFC reader reads the work card UID and transmits it to the hardware authentication module. The hardware authentication module queries the local user database and matches the user record. If the match is successful, the hardware authentication module sends a collection command to the fingerprint sensor. The user places their finger on the surface of the fingerprint sensor. The sensor collects the fingerprint image and extracts the fingerprint feature vector, which is then transmitted to the hardware authentication module. The hardware authentication module compares the fingerprint feature vector with the fingerprint template stored in the user record and calculates the similarity score. If the similarity is lower than the threshold, authentication failure is returned and the failure event is recorded to the audit log chain. If the current operation involves highly sensitive operations such as modifying system parameters or batch exporting, the hardware authentication module sends a challenge value to the FIDO2 security key slot. The user inserts the FIDO2 physical key and completes the press confirmation. The FIDO2 key signs the challenge value and returns the signature value. The hardware authentication module uses the public key corresponding to the key to verify the signature. After all verifications are successful, the hardware authentication module transmits the UUID and authentication method encoding to the key management module. The key management module receives the UUID and sends a KEY_LOAD request to the TPM. The request parameters include the UUID binding identifier. The TPM uses its internal MK to decrypt the ciphertext of the user key KEK_User to obtain the plaintext, loads the plaintext KEK_User into the TPM's internal key slot, and returns the key handle to the key management module. The key management module caches the key handle in memory.

3. The automated scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 1, characterized in that, The key management module and data encryption module perform slide scanning and WSI file encryption. This includes the scanning software sending a DEK_GEN request to the key management module at the start of each scanning session, the key management module sending a Tpm2_GetRandom command to the TPM to request a random number, the TPM generating a random number as the data encryption key DEK_session for this session, the key management module returning the encrypted key DEK_session in plaintext to the data encryption module, and simultaneously sending a Tpm2_Encrypt command to the TPM to encrypt DEK_session using KEK_User to obtain DEK_encrypted, which is then stored in memory. The scanning software controls the automatic loading unit to grab the slide holder from the slide compartment and move it to the stage. The camera unit acquires image data and the camera transmits the image data stream to the workstation memory buffer. After image acquisition is completed, the WSI layer splitter reads the complete WSI image data from the memory buffer. The splitter extracts the 10x magnified image data into an overview layer according to the image pyramid hierarchy, merges the 20x and 40x magnified image data into a high-resolution layer, and extracts patient information, specimen number, staining method and other data from the scan protocol metadata to construct the patient metadata layer. If there is already annotation information during the image reading process, it is extracted into an annotation and diagnosis layer. If it is a new scan, this layer is initially empty. Perform layered encryption and policy binding; The WSI file wrapper serializes the payloads of the four encryption layers, DEK_encrypted, the magic number in the file header, the version number, the layer index table, and the integrity check hash value into a single file, stores it in a directory, and records the file path and file name in a local database. The event capturer captures the scan completion event, records the operator's UUID, operation type, target WSI identifier, timestamp, and file hash value, and pushes it to the block builder.

4. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 1, characterized in that, The access control module and the temporary permission token module authorize users to access encrypted WSI files, and perform token issuance and use, including when the user clicks to load an image, selects a target WSI file, the image viewing software reads the WSI security file header, obtains the encrypted payload list and the CP-ABE ciphertext policy digest of each layer, and sends an ACCESS_REQUEST to the access control module. The request parameters include the user's UUID, the target WSI identifier, and the request operation type. After receiving ACCESS_REQUEST, the access control module constructs a complete set of attributes for the current access, queries the role library for the static role corresponding to the user's UUID, and collects time attributes, location attributes, device attributes, case relationship attributes, and diagnosis status attributes respectively, and then transmits the complete set of attributes to the policy decision point. Execution strategy evaluation and hierarchical decryption; When the authorizer needs to grant temporary access to the authorized person, the authorizer initiates a creation operation in the image viewing software. After receiving the request, the token generator calls the weight calculation engine to calculate the weight. The token generator constructs a WTAT data structure, signs the token content using the authorizer's attribute private key, stores the token in the local token database, and sends a token notification to the authorized person. The weight consumption is calculated in real time when the token is used.

5. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 1, characterized in that, The cloud synchronization module and key management module synchronize local encrypted data to the cloud security service cluster. This includes the synchronization scheduler checking the synchronization queue in a polling manner, immediately placing metadata and transaction log changes into the P0 queue, triggering the P1 queue for enqueueing when a diagnostic report issuance event occurs, the scheduler traversing the local WSI file modification timestamps and enqueuing the modified files of the day into the P2 queue, and triggering the P3 queue for enqueueing when a user logout event and disk usage monitoring occur. The scheduler dequeues tasks in priority order from P0 to P3, and does not process low-priority queues when high-priority queues are not empty. Before sending data, the scheduler checks the system clock. If the current time is in the first time period, the sending rate is set to 10Mbps. If the current time is in the second time period, no rate limit is set. If the current time is in the third time period, the sending rate is set to 50Mbps. The zero-knowledge upload engine obtains the KEK_Sync ciphertext from the key management module. The zero-knowledge upload engine splits the KEK_Sync ciphertext into component A and component B. Component A is stored inside the workstation TPM, and component B is transmitted to the cloud HSM. The cloud uses HSM to securely store component B. The zero-knowledge upload engine transmits the encrypted WSI file to the cloud object storage. The cloud object storage only holds the AES-256-GCM encrypted ciphertext data and the KEK_Sync ciphertext component B. It cannot restore the KEK_Sync ciphertext and therefore cannot decrypt the WSI data. The cloud storage service writes a complete data copy. The write operation adopts a synchronous confirmation mechanism. It will only return the upload completion after at least two availability zones confirm that the write is successful. The version management service checks the version history of the written object. If the difference between the current time and the initial creation time of the object is less than or equal to seven days, the full version is retained. If the difference is greater than seven days and less than or equal to three years, the most recent weekly snapshot is retained and intermediate versions are deleted. When a local workstation experiences storage failure or data corruption, the administrator initiates a disaster recovery process through the management terminal. The recovery program downloads a specified version of the encrypted WSI file from the cloud object storage. The recovery program requests KEK_Sync component B from the cloud HSM and component A from the local TPM. The two components are merged and restored in the memory of the secure workstation to restore KEK_Sync, thereby decrypting DEK. Finally, the WSI data is decrypted and written to local storage. After the recovery is completed, the audit log chain records the recovery event, including the recovery time, the list of recovered WSIs, and the data version.

6. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 1, characterized in that, The audit log module performs continuous recording of audit logs, block construction, and chain integrity verification. This includes an event capturer embedded in the interface layer of each security module. Each time a security operation occurs, the corresponding module calls the event capturer interface, passes the event type enumeration value and the event payload, the event capturer generates a globally unique event ID, takes the system timestamp, calculates the event hash value for the event payload, and writes the event record into the memory event buffer. Execute the construction block; The chain integrity verifier performs a full chain check when the system starts up. It traverses each block from the end of the local file backward, calculates the SHA256 of each block, and compares it with the H_prev stored in the next block. If all match, the check passes. If a mismatch occurs in the kth block, the block from k+1 to the end is marked as corrupt, triggering the recovery of that block from the cloud copy. The verifier performs incremental verification after generating ten new blocks, only verifying the consistency of the hash chain of the most recent twenty blocks. The verifier periodically performs cross-verification between local and cloud replicas, randomly selecting ten blocks and comparing the consistency between the local hash value and the cloud hash value.

7. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 3, characterized in that, The layered encryption and policy binding includes performing the following operations sequentially on each logical layer: the AES-256-GCM encryption engine uses DEK_session to encrypt the plaintext of the logical layer to obtain the first ciphertext AES; the encryption engine generates a random initialization vector IV; calls the AES-256-GCM encryption function with DEK_session, IV and plaintext of the logical layer as input parameters; and outputs ciphertext and authentication tag, with IV prepended to the ciphertext header. The CP-ABE policy binder reads the access policy expression corresponding to this layer, uses DEK_session as the encryption object, calls the CP-ABE encryption function, takes the policy expression and DEK_session plaintext as input parameters, and outputs the second ciphertext CP-ABE. The first ciphertext AES, IV, authentication tag, and the second ciphertext CP-ABE are encapsulated as the cryptographic payload of this layer.

8. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 4, characterized in that, The policy evaluation and hierarchical decryption process includes traversing the four encryption layers of WSI at the policy decision point, performing a CP-ABE decryption attempt for each layer, inputting the user attribute set and the CP-ABE ciphertext of that layer into the CP-ABE decryption engine, and the decryption engine using the user attribute private key to attempt to decrypt the CP-ABE ciphertext. If the attribute set meets the policy conditions, decryption is successful and the DEK_session of that layer is obtained; otherwise, the layer is marked as inaccessible. For accessible layers, the policy decision point uses the obtained DEK_session to call the decryption function, takes DEK_session, IV and ciphertext as input, outputs plaintext layer data, verifies the authentication tag, and passes the decrypted layer data to the image viewing software for rendering and display.

9. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 4, characterized in that, The weight consumption when the token is used includes the following: When an authorized user accesses the target WSI using a token, the consumption monitor calculates the weight consumption in real time. When an authorized user browses the overview layer, the consumption monitor deducts 1 point from the remaining weight W_remaining every minute. When an authorized user switches to a high-resolution layer, the consumption monitor deducts 3 points every minute. When an authorized user performs an export screenshot operation, the consumption monitor deducts 15 points per instance. The consumption monitor checks W_remaining after each deduction. If W_remaining is less than or equal to zero, it notifies the token generator to set the token status to EXPIRED, and the access control module rejects subsequent access requests. The anomaly detector continuously monitors token usage patterns. If it detects that the weight consumption exceeds half of W_total within five minutes, it sends an alert to the authorizing person. If it detects that the same token is used simultaneously from two different IP addresses, it immediately sets the token status to LOCKED, rejects all subsequent requests, and generates a SECURITY_ALERT event to push to the audit log chain.

10. The automatic scanning and fusion system for digital pathological slides based on multimodal imaging according to claim 6, characterized in that, The construction of the block includes a block builder monitoring a memory event buffer. When the number of accumulated events in the buffer reaches one hundred, or when the time since the previous block was generated exceeds sixty seconds and the buffer is not empty, the block construction is triggered. During block construction, the block builder extracts all event records from the buffer, sorts them in ascending order by timestamp, calculates the SHA256 hash value h_i for each event record, and constructs a Merkle tree. If N is even, the adjacent leaf nodes h_2j and h_2j+1 are concatenated and their SHA256 hashes are calculated to obtain the parent node. If N is odd, the last leaf node h_N-1 is copied and its SHA256 hash is calculated. This process is repeated until only one hash value remains to obtain the Merkle tree root, Merkle_root. The block constructor reads the SHA256 hash value H_prev of the previous block, combines it with the current timestamp and Merkle_root to form the block header. After serializing the block header and block body, the SHA256 hash value H_current of the current block is calculated. The block is written to the end of the local file and simultaneously transmitted to the cloud audit log replica library through the P0-level synchronous channel.

Images