A cloud platform-based distributed medical image data storage method

Abstract

The present application relates to medical image data processing and cloud storage technical field, more specifically, it relates to a kind of distributed medical image data storage method based on cloud platform, for solving the problem that prior art cannot accurately distinguish focus area and normal tissue, under high compression ratio, it is prone to cause diagnostic key information loss, cause structure fracture, fuzzy or artifact, lack of differentiated perception ability to structure and texture damage, cannot realize targeted repair;The present application realizes efficient compression and high-quality reconstruction under the premise of guaranteeing diagnostic key information by the mechanism of semantic-guided differentiated compression and structure-texture separation repair, based on the semantic label of focus and anatomical region Dynamic allocation of compression parameters, low compression fidelity to focus area, high compression reduction to non-key area, according to gradient and texture feature Adaptive calculation of repair weight and corresponding regulation, accurately repair edge fracture, fuzzy and texture loss.

Description

Technical Field

[0001] This invention relates to the field of medical image data processing and cloud storage technology, and more specifically, to a distributed medical image data storage method based on a cloud platform. Background Technology

[0002] With the rapid growth of medical imaging data, traditional local storage systems have exposed problems such as poor scalability, difficulty in cross-institutional interoperability, high costs, and serious data redundancy. Even when migrating to cloud platforms, bottlenecks such as inconsistent data standards, the risk of losing critical details during compression, lack of intelligent repair capabilities, insufficient security and compliance guarantees, and rigid resource scheduling mechanisms still exist, making it difficult to achieve an effective balance between storage efficiency, image quality, and clinical usability. Most existing solutions focus on a single aspect and have not yet formed an integrated intelligent system covering acquisition, processing, storage, security, and scheduling.

[0003] The patent application with publication number CN120067353A discloses a distributed medical image data storage method based on a cloud platform. The method includes: original medical images from various medical sites are encrypted during transmission, undergo dual verification, data calibration, and metadata extraction to generate standardized medical image sequences, which are then bound to structured metadata. These sequences are divided into independent logical segments by site to form standardized sub-site data streams. The compressed segment set is input into an initial segmentation model to calculate the loss gradient, and directional and arbitrary gradient alignment is performed using historical playback data. Based on gradient alignment constraints of the joint loss function and site identifiers, synthetic medical images are generated through a diffusion model to form a structured playback dataset. The gradient direction cosine similarity between the site-specific synthetic data generated by the diffusion model and the real data is improved, as is the anatomical structure overlap rate. Combined with dynamic cue vector optimization using reinforcement learning, the efficiency of historical data playback is improved.

[0004] However, while the aforementioned reference patents employ dynamic interpolation and DICOM normalization to improve boundary sharpness and data consistency, and combine deformation sparse convolution to preserve key textures and save storage, the hot-cold layering strategy based on activity entropy effectively reduces storage costs and shortens access latency, and the introduction of a gradient alignment mechanism reduces heterogeneous bias, improving segmentation accuracy and generalization ability, and the fusion of synthetic data and cue vectors enhances gradient consistency, anatomical overlap rate, and playback efficiency, they cannot accurately distinguish between lesion areas and normal tissues. Under high compression rates, they are prone to losing key diagnostic information, causing structural breaks, blurring, or artifacts. They lack the ability to differentiate structural and texture damage, making targeted repair impossible and significantly reducing the integrity, usability, and clinical applicability of compressed images. At the same time, they cannot integrate access frequency, sensitivity, and storage cost to build an intelligent layered storage mechanism, resulting in high response latency for highly sensitive or high-frequency access data, while low-frequency data occupies high-cost storage resources for a long time, keeping the overall storage cost high. They also cannot adaptively schedule read and write requests based on the real-time load status of the storage cluster, causing uneven node load and restricting the utilization of storage resources and the scalability of system performance.

[0005] To address the aforementioned problems, this invention proposes a distributed medical image data storage method based on a cloud platform. Summary of the Invention

[0006] The purpose of this invention is to provide a distributed medical image data storage method based on a cloud platform. This method addresses the shortcomings of existing technologies, which cannot accurately distinguish between lesion areas and normal tissues, easily leading to the loss of critical diagnostic information at high compression rates, causing structural breaks, blurring, or artifacts. Furthermore, it lacks the ability to differentiate structural and texture damage, hindering targeted repair and significantly reducing the integrity, usability, and clinical applicability of compressed images. Additionally, it fails to integrate access frequency, sensitivity, and storage cost to construct an intelligent hierarchical storage mechanism, resulting in high response latency for highly sensitive or frequently accessed data, while low-frequency data occupies high-cost storage resources for extended periods, leading to persistently high overall storage costs. Moreover, it cannot adaptively schedule read / write requests based on the real-time load status of the storage cluster, causing uneven node load and restricting storage resource utilization and system performance scalability.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] A distributed medical image data storage method based on a cloud platform includes the following steps:

[0009] Step 1: Receive DICOM format medical image data from the medical image scanning and reconstruction process, including image pixel data and DICOM standard-compliant metadata. Perform automatic artifact recognition and removal processing on the image pixel data and perform standardization processing on the metadata to generate standardized DICOM image data.

[0010] Step 2: Apply semantic region-based differential compression to the standardized DICOM image data, allocate compression intensity according to the lesion area and anatomical structure, and automatically repair structural breaks, edge blurring and texture loss using a dual repair mechanism of structure and texture separation.

[0011] Step 3: Encrypt and store the compressed and repaired medical image data, and perform intelligent hierarchical management based on access frequency and sensitivity, automatically migrate to hot storage, warm storage or cold storage media, and adaptively schedule storage resources based on the real-time load of storage nodes, and balance the read and write request load among storage nodes.

[0012] Step 4: When a user accesses the hierarchically stored medical image data, perform multi-factor authentication, monitor and record user operation behavior, detect abnormal activities and trigger response mechanisms, and generate audit logs.

[0013] In a preferred embodiment of the present invention, step two, which involves differential compression based on semantic regions and allocation of compression intensity according to the lesion region and anatomical structure, includes:

[0014] The standardized DICOM medical image data is used as input. The standardized DICOM image data is accompanied by a set of semantic regions. Each semantic region has a unique semantic label. For each semantic region, the semantic region and its semantic label are read. If the semantic label indicates that the region belongs to the pathological category, the semantic region is the lesion region. If the semantic label indicates that the region belongs to the normal tissue category, the semantic region is the anatomical structure region.

[0015] The Shannon entropy of the wavelet coefficient distribution of the semantic region is calculated as the information redundancy, and the average gradient magnitude of all pixels in the semantic region is calculated as the structural complexity. The compression ratio of the semantic region is calculated based on these two features.

[0016] If the semantic region is a lesion region, calculate the area ratio of the lesion region, construct a gray-level co-occurrence matrix within the local window of the lesion region, calculate the contrast, and calculate the average contrast of all local windows of the lesion region as the texture complexity. Calculate the compression intensity of the lesion region based on the area ratio, structural complexity, and texture complexity.

[0017] If the semantic region is an anatomical structure region, the same computational structure as the lesion region is used to separate the compression strategies of the lesion region and the anatomical structure region. The lesion region outputs the compression intensity, and the anatomical structure region outputs the compression ratio. All semantic regions output the corresponding compression control parameters according to this rule, and perform partitioned compression on the standardized DICOM image data accordingly to generate compression results.

[0018] In a preferred embodiment of the present invention, the automatic repair process in step two, which employs a dual repair mechanism of structure and texture separation to address structural breakage, edge blurring, and texture loss, includes:

[0019] For each semantic region, read the gradient magnitude of all pixels within the semantic region and the gray-level co-occurrence matrix contrast of all local windows, calculate the average gradient magnitude of all pixels within the semantic region and the average gray-level co-occurrence matrix contrast of all local windows, and determine the repair weight of the semantic region based on the two calculation results.

[0020] The structural repair intensity is set to equal the repair weight, and the texture repair intensity is set to equal one minus the repair weight. The structural repair intensity is used to repair the gradient-related structural information of the semantic region, and the texture repair intensity is used to repair the contrast-related texture information of the gray-level co-occurrence matrix of the semantic region, generating the repaired semantic region content. All semantic regions are processed in this way, and all repaired semantic region contents are merged to generate the final repair result.

[0021] In a preferred embodiment of the present invention, step three, which involves encrypting and storing the compressed and repaired medical image data and performing intelligent hierarchical management based on access frequency and sensitivity, includes:

[0022] The process involves taking compressed and repaired medical image data as input, and for each compressed and repaired medical image data, obtaining the number of accesses, sensitivity score, and storage cost, and calculating the hierarchical evaluation value of the medical image data based on the number of accesses, sensitivity score, and storage cost.

[0023] Obtain a first boundary value and a second boundary value. If the first boundary value is greater than the second boundary value, compare the hierarchical evaluation value of the medical image data with the first boundary value. If the hierarchical evaluation value is greater than the first boundary value, allocate the medical image data to a hot storage medium. If the hierarchical evaluation value is not greater than the first boundary value, compare the hierarchical evaluation value of the medical image data with the second boundary value. If the hierarchical evaluation value is greater than or equal to the second boundary value, allocate the medical image data to a warm storage medium. If the hierarchical evaluation value is less than the second boundary value, allocate the medical image data to a cold storage medium.

[0024] Before writing to the allocated storage medium, the medical image data is encrypted using the AES algorithm, thus completing the encryption and storage of the medical image data.

[0025] In a preferred embodiment of the present invention, the process of automatically migrating the corresponding data to a hot storage, warm storage, or cold storage medium in step three includes:

[0026] For each medical image data, obtain the number of accesses to the medical image data in the most recent three statistical periods, check whether the number of accesses to the medical image data in the most recent three statistical periods has been continuously increasing or decreasing. If the number of accesses to the medical image data in the most recent three statistical periods has been continuously increasing or decreasing, obtain the predicted value of the number of accesses to the medical image data in the future and the equivalent cost and benefit of a single access, obtain the unit time cost increment after migrating the medical image data from the current storage level to the target storage level, and calculate the migration decision value of the medical image data based on the predicted value of the number of accesses in the future, the equivalent cost and benefit of a single access, and the unit time cost increment.

[0027] Compare the migration decision value of the medical image data with zero. If the migration decision value of the medical image data is greater than zero, obtain the current storage level, level evaluation value, first boundary value, and second boundary value of the medical image data. Based on the level evaluation value, first boundary value, and second boundary value of the medical image data, compare the current storage level of the medical image data with the target storage level. If the current storage level is not equal to the target storage level, migrate the medical image data from the current storage level to the target storage level. If the migration decision value of the medical image data is not greater than zero, do not perform the migration operation.

[0028] In a preferred embodiment of the present invention, step three, which involves adaptively scheduling storage resources based on the real-time load of storage nodes and balancing read / write request loads among storage nodes, includes:

[0029] For each new read / write request, initialize the total overall pressure value to zero. For each storage node, obtain the average resource usage of the storage node, obtain the average request response time of the storage node, obtain the baseline response time, calculate the normalized value of the average request response time of the storage node, calculate the overall pressure value of the storage node, and sum the overall pressure values ​​of the storage nodes to the total overall pressure value.

[0030] For each storage node, obtain the overall pressure value of the storage node, calculate the allocation weight of the storage node. If the allocation weight is less than zero, set it to zero and accumulate the allocation weights of the storage nodes to the total allocation weight. If the total allocation weight is greater than zero, for each storage node, obtain the allocation weight of the storage node, calculate the probability of the storage node being selected, select a storage node according to the probability of each storage node being selected, and allocate new read and write requests to the selected storage node. If the total allocation weight is equal to zero, distribute new read and write requests evenly and randomly to any storage node.

[0031] In a preferred embodiment of the present invention, step four, which involves performing multi-factor authentication and monitoring and recording user behavior when a user accesses hierarchically stored medical image data, includes:

[0032] Get the medical image data access request, get the current session credential status. If the current session credential is invalid, get the first verification result, the second verification result, and the third verification result. If the first verification result, the second verification result, or the third verification result fails, get the number of consecutive verification failures and increment the number of consecutive verification failures by one. If the number of consecutive verification failures is less than three, return a verification failure message. If the number of consecutive verification failures is greater than or equal to three, get the account lock duration, put the account in a locked state, and end the processing of this access request. If the first verification result, the second verification result, and the third verification result are all successful, reset the number of consecutive verification failures to zero and issue a session credential.

[0033] Obtain user identity, operation type, operation time, operation source Internet Protocol address, unique identifier of the device used for the operation, and unique identifier of the medical image data being operated, and generate an operation record.

[0034] In a preferred embodiment of the present invention, the process of performing abnormal activity identification and response requirements and generating audit logs in step four includes:

[0035] Get user actions and the number of abnormal matches. If the number of abnormal matches is zero, set the operation result to normal. If the number of abnormal matches is one, set the operation result to level one abnormality. If the number of abnormal matches is greater than or equal to two, set the operation result to level two abnormality.

[0036] Obtain the operation time, user identity, operation type, operation parameter summary, operation result and audit-specific voucher, calculate the integrity verification string, generate audit log, and store the audit log in a storage location that cannot be modified or deleted.

[0037] Compared with the prior art, the advantages of this invention are:

[0038] (1) In this invention, through semantically guided differential compression and structure-texture separation repair mechanism, efficient compression and high-quality reconstruction are achieved on the premise of ensuring key diagnostic information. Based on the semantic labels of lesions and anatomical regions, compression parameters are dynamically allocated by combining wavelet entropy, gradient complexity, area ratio and texture contrast. The lesion region is compressed with low fidelity, and the non-critical region is compressed with high reduction. At the same time, the repair weight is adaptively calculated according to gradient and texture features, and the structure and texture repair intensity are adjusted respectively. The edge breakage, blur and texture loss are accurately repaired, taking into account storage efficiency and image diagnostic value, and significantly improving the integrity, usability and clinical applicability of the compressed medical image.

[0039] (2) In this invention, an intelligent hierarchical storage mechanism is constructed by integrating access frequency, sensitivity and storage cost. Data is automatically allocated to hot storage, warm storage or cold storage according to the hierarchical evaluation value. AES encryption is used to ensure security and compliance. Data location is dynamically adjusted by combining access trend prediction and migration benefits to reduce storage costs. At the same time, the comprehensive load is calculated based on node resource occupation and response time. Read and write requests are scheduled by reverse weighting to achieve cluster load balancing, improve system throughput and response efficiency, and significantly optimize resource utilization and scalability while ensuring high availability of highly sensitive data. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the steps of the distributed medical image data storage method in this invention.

[0041] Figure 2 This is a schematic diagram of the logic flow in Embodiment 1 of the present invention;

[0042] Figure 3 This is a schematic diagram of the logic flow in Embodiment 2 of the present invention. Detailed Implementation

[0043] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] Example 1: As Figure 1 and Figure 2 As shown, the present invention proposes a distributed medical image data storage method based on a cloud platform, comprising the following steps:

[0045] Step 1: Receive DICOM format medical image data from the medical image scanning and reconstruction process. This data includes image pixel data and DICOM-compliant metadata. The image pixel data is a raw sampling intensity numerical matrix representing the content of the medical image, used for automatic artifact recognition and removal. The metadata includes pixel physical spacing, image layer thickness, image layer spacing, number of image rows, number of image columns, display window width, display window level, and pixel storage bit depth. Perform automatic artifact recognition and removal on the image pixel data and standardize the metadata. Standardization includes missing value detection and completion, numerical format unification, numerical calibration according to DICOM standard unit specifications, window width and window level logical verification, and compatibility verification of pixel storage bit depth with the number of image rows and columns. Generate standardized DICOM image data.

[0046] By simultaneously performing automatic artifact recognition and removal of image pixel data and standardized processing of associated metadata on the received DICOM format medical image data, the integrity, consistency and standard compliance of metadata are ensured while improving image quality. This not only effectively eliminates interference factors that affect diagnostic accuracy, but also significantly enhances the interoperability and subsequent processing efficiency of multi-source medical image data, laying a high-quality and highly reliable data foundation for advanced applications such as AI-assisted analysis, accurate diagnosis and efficient storage.

[0047] Step 2: Apply semantic region-based differential compression to the standardized DICOM image data, allocate compression intensity according to the lesion area and anatomical structure, and use a dual repair mechanism of structure and texture separation to automatically repair structural breaks, blurred edges and lost textures.

[0048] Step two involves differential compression based on semantic regions and allocating compression intensity according to the lesion region and anatomical structure, including:

[0049] Standardized DICOM medical image data is used as input. The standardized DICOM image data has completed window width and window level mapping, grayscale normalization, and non-anatomical regions have been removed. The standardized DICOM image data is accompanied by a set of semantic regions. Each semantic region is generated by a medical image semantic segmentation model and covers a continuous area in the image. Each semantic region has a unique semantic label. The semantic label indicates whether the region belongs to the pathological category or the normal tissue category. For each semantic region, the semantic region and its semantic label are read. If the semantic label indicates that the region belongs to the pathological category, the semantic region is a lesion region. If the semantic label indicates that the region belongs to the normal tissue category, the semantic region is an anatomical structure region.

[0050] The Shannon entropy of the wavelet coefficient distribution of the semantic region is calculated as the information redundancy, and the average gradient magnitude of all pixels within the semantic region is calculated as the structural complexity. Based on these two features, the compression ratio of the semantic region is calculated according to the following relationship:

[0051] ,in This represents the compression ratio of the i-th semantic region. The base compression ratio set for the system. For the region Shannon entropy of wavelet coefficients For the region The average gradient magnitude of all pixels within the range. The parameter is a positive real number.

[0052] If the semantic region is a lesion region, calculate the area ratio of the lesion region, which is the ratio of the number of pixels in the lesion region to the total number of effective pixels in the normalized DICOM image data. Construct a gray-level co-occurrence matrix within the local window of the lesion region, calculate the contrast, and average the contrast of all local windows of the lesion region as the texture complexity. Based on the area ratio, structural complexity, and texture complexity, calculate the compression strength of the lesion region according to the following relationship:

[0053] ,in Indicates the compressive strength of the lesion area. This represents the percentage of the lesion area. This represents the average gradient magnitude of all pixels within the lesion area. This represents the average contrast of the gray-level co-occurrence matrix of the lesion region. The parameter is a positive real number.

[0054] If the semantic region is an anatomical structure region, the same computational structure as the lesion region is used, but independent weight parameters are used to separate the compression strategies of the lesion region and the anatomical structure region. The lesion region outputs the compression intensity, and the anatomical structure region outputs the compression ratio. The compression intensity or compression ratio is used to perform compression processing on the semantic region. All semantic regions output the corresponding compression control parameters according to this rule, and perform partitioned compression on the standardized DICOM image data accordingly to generate compression results.

[0055] Step two involves an automatic repair process using a dual repair mechanism that separates structure and texture to address structural breaks, blurred edges, and lost textures. This process includes:

[0056] For each semantic region, the gradient magnitude of all pixels within the semantic region and the gray-level co-occurrence matrix contrast of all local windows are read. The average gradient magnitude of all pixels within the semantic region and the average gray-level co-occurrence matrix contrast of all local windows are calculated. Based on these two calculations, the restoration weight of the semantic region is determined according to the following relationship:

[0057] ,in This represents the repair weight for the i-th semantic region. For the region The average gradient magnitude of all pixels within the range. For the region The average contrast of the gray-level co-occurrence matrix. The parameter is a positive real number.

[0058] Set the structural repair intensity to equal the repair weight, and set the texture repair intensity to equal one minus the repair weight. Use the structural repair intensity to repair the gradient-related structural information of the semantic region, and use the texture repair intensity to repair the gray-level co-occurrence matrix contrast-related texture information of the semantic region. Generate the repaired semantic region content. All semantic regions are processed in this way. Merge all repaired semantic region contents to generate the final repair result.

[0059] By employing a semantically guided differential compression and structure-texture separation repair mechanism, this method achieves efficient compression and high-quality reconstruction while preserving key diagnostic information. Based on semantic labels of lesions and anatomical regions, and combined with wavelet entropy, gradient complexity, area ratio, and texture contrast, compression parameters are dynamically allocated to achieve low-compression fidelity in lesion regions and high-compression reduction in non-critical regions. Simultaneously, repair weights are adaptively calculated based on gradient and texture features to adjust the intensity of structure and texture repair, accurately repairing edge breaks, blurring, and texture loss caused by compression. This method balances storage efficiency and image diagnostic value, significantly improving the integrity, usability, and clinical applicability of compressed medical images.

[0060] Step 3: Encrypt and store the compressed and repaired medical image data, and perform intelligent hierarchical management based on access frequency and sensitivity, automatically migrate to hot storage, warm storage or cold storage media, and adaptively schedule storage resources based on the real-time load of storage nodes, and balance the read and write request load among storage nodes.

[0061] Step three involves encrypting and storing the compressed and repaired medical image data, and then implementing intelligent hierarchical management based on access frequency and sensitivity. This includes:

[0062] Using compressed and repaired medical image data as input, for each compressed and repaired medical image data set: obtain the number of accesses, sensitivity score, and storage cost. Based on the number of accesses, sensitivity score, and storage cost, calculate the hierarchical evaluation value of the medical image data according to the following relationship:

[0063] ,in For data items The hierarchical evaluation value, For data items The number of visits in the most recent statistical period. For data items The sensitivity score, ranging from zero to one, is determined by clinical diagnostic value and whether it contains identifiable patient information. For data items The unit time cost of storage on a specified medium is determined by the cost of medium purchase and maintenance, energy consumption, and read / write performance. These are non-negative parameters, corresponding to the weights of access frequency, sensitivity, and storage cost, respectively.

[0064] Obtain a first boundary value and a second boundary value. If the first boundary value is greater than the second boundary value, compare the hierarchical evaluation value of the medical image data with the first boundary value. If the hierarchical evaluation value is greater than the first boundary value, allocate the medical image data to a hot storage medium. If the hierarchical evaluation value is not greater than the first boundary value, compare the hierarchical evaluation value of the medical image data with the second boundary value. If the hierarchical evaluation value is greater than or equal to the second boundary value, allocate the medical image data to a warm storage medium. If the hierarchical evaluation value is less than the second boundary value, allocate the medical image data to a cold storage medium.

[0065] Before writing to the allocated storage medium, the medical image data is encrypted using the AES algorithm. The encryption key is determined by the security policy. This completes the encryption and storage of the medical image data. All compressed and repaired medical image data is processed in this way.

[0066] Step three, which involves automatically migrating the corresponding data to a hot, warm, or cold storage medium, includes:

[0067] For each medical image data set, the number of accesses to the medical image data in the most recent three statistical periods is obtained. It is then checked whether the number of accesses to the medical image data in the most recent three statistical periods has been continuously increasing or decreasing. If the number of accesses to the medical image data in the most recent three statistical periods has been continuously increasing or decreasing, the predicted number of accesses to the medical image data in the future and the equivalent cost-benefit per access are obtained. The incremental cost per unit time after migrating the medical image data from the current storage level to the target storage level is obtained. A positive incremental cost per unit time indicates an increase in cost, and a negative incremental cost per unit time indicates a decrease in cost. Based on the predicted number of accesses in the future, the equivalent cost-benefit per access, and the incremental cost per unit time, the migration decision value of the medical image data is calculated according to the following relationship:

[0068] ,in For data items The migration decision value, For data items The predicted number of visits at a future time t, this value is generated by a time series prediction model based on data items. Calculated from actual access records over several past statistical periods, For data items The incremental cost per unit time after migrating from the current storage tier to the target storage tier The equivalent cost-benefit of a single access at the storage tier;

[0069] Compare the migration decision value of the medical image data with zero. If the migration decision value of the medical image data is greater than zero, obtain the current storage level, level evaluation value, first boundary value, and second boundary value of the medical image data. Based on the level evaluation value, first boundary value, and second boundary value of the medical image data, determine the target storage level of the medical image data. Compare the current storage level of the medical image data with the target storage level. If the current storage level is not equal to the target storage level, migrate the medical image data from the current storage level to the target storage level. If the migration decision value of the medical image data is not greater than zero, do not perform the migration operation. All medical image data are processed in this way.

[0070] Step three, which involves adaptively scheduling storage resources based on the real-time load of storage nodes and balancing read and write request loads among storage nodes, includes:

[0071] For each new read / write request, the total initial stress value is set to zero. For each storage node, the average resource usage is obtained, calculated based on processor utilization, memory usage, disk read / write speed, and network transmission speed. The average request response time is obtained, which is the arithmetic mean of all request response times in the past statistical period. The baseline response time is obtained. The normalized average request response time is calculated, which is equal to the average request response time divided by the baseline response time. The total stress value of the storage node is calculated, which is equal to the sum of the average resource usage and the normalized average request response time. The total stress values ​​of the storage nodes are accumulated to the total total stress value, and the total weight allocation is initialized to zero.

[0072] For each storage node, obtain the overall pressure value of the storage node, calculate the allocation weight of the storage node. The allocation weight is equal to the sum of the overall pressure values ​​minus the overall pressure value of the storage node. If the allocation weight is less than zero, set it to zero and accumulate the allocation weights of the storage nodes to the total allocation weight. If the total allocation weight is greater than zero, for each storage node, obtain the allocation weight of the storage node, calculate the probability of the storage node being selected. The probability is equal to the allocation weight of the storage node divided by the total allocation weight. Based on the probability of each storage node being selected, select one storage node and allocate new read and write requests to the selected storage node. If the total allocation weight is equal to zero, randomly and evenly allocate new read and write requests to any storage node.

[0073] By integrating an intelligent tiered strategy that considers access frequency, sensitivity, and storage cost, combined with a dynamic migration mechanism and adaptive scheduling based on real-time load, secure, efficient, and elastic storage of medical imaging data is achieved. On the one hand, data is automatically allocated to hot, warm, or cold storage based on tiered evaluation values, while ensuring privacy and compliance under encryption (AES). On the other hand, by predicting access trends and migration benefits, data distribution is dynamically optimized to reduce overall storage costs. Simultaneously, based on the comprehensive pressure calculated from node resource usage and response time, read and write requests are allocated in reverse weighting to effectively balance cluster load and improve system throughput and response stability. Overall, while ensuring high availability of highly sensitive lesion data, storage resource utilization and system scalability are significantly optimized.

[0074] Example 2: The technical solution of this embodiment of the invention differs from that of Example 1 in that:

[0075] like Figure 1 and Figure 3 As shown, step four: When a user accesses medical image data stored in a hierarchical manner, multi-factor authentication is performed, user operation behavior is monitored and recorded, abnormal activity is detected and a response mechanism is triggered, and an audit log is generated at the same time.

[0076] Step four, which involves performing multi-factor authentication and monitoring and recording user behavior when a user accesses hierarchically stored medical image data, includes:

[0077] The system retrieves a request to access medical image data and obtains the current session credential status. If the current session credential is invalid, it retrieves the first, second, and third verification results. The first verification result indicates whether the user's identity credential meets the complexity and length requirements. The second verification result indicates whether the temporary code is valid and the time deviation does not exceed the limit. The third verification result indicates whether the biometric features match the registration template successfully. If the first, second, or third verification result fails, the system retrieves the number of consecutive verification failures and increments it by one. If the number of consecutive verification failures is less than three, a verification failure message is returned. If the number of consecutive verification failures is greater than or equal to three, the system retrieves the account lock duration, sets the account to a locked state for the duration of the account lock, and terminates the current access request processing. If the first, second, and third verification results are all successful, the number of consecutive verification failures is reset to zero, and a session credential is issued.

[0078] Obtain the user's identity identifier, operation type, operation time, operation source Internet Protocol address, unique identifier of the device used for the operation, and unique identifier of the medical image data being operated on; generate an operation record; calculate the verification value of the operation record; store the operation record and the verification value together; if a previous operation record exists in the current session, obtain the operation time of the previous operation record; if the current operation time is not greater than the previous operation time, reject the current operation record.

[0079] Step four, which involves executing the abnormal activity identification and response requirements and generating audit logs, includes:

[0080] Get user actions, get the number of abnormal matches. The number of abnormal matches is the number of conditions that successfully match the user actions after comparing them with the preset abnormal conditions. If the number of abnormal matches is zero, set the operation result to normal. If the number of abnormal matches is one, interrupt the current operation and require all three verifications to be completed again. The three verifications include: the first verification is the user's identity certificate, the second verification is the temporary code, and the third verification is the biometric feature. Send a notification and set the operation result to abnormal level one. If the number of abnormal matches is greater than or equal to two, terminate the current session, set the account lock status to manual unlocking, start the response process, and set the operation result to abnormal level two.

[0081] Obtain the operation time, user identity, operation type, operation parameter summary, operation result, and audit-specific voucher. The operation parameter summary is the summary value of the operation parameter after anonymization. Calculate the integrity verification string, which is generated based on the operation time, user identity, operation type, operation parameter summary, operation result, and audit-specific voucher using a one-way verification algorithm. Generate an audit log, which contains the operation time, user identity, operation type, operation parameter summary, operation result, and integrity verification string. Store the audit log in an unmodifiable and non-deletable storage location.

[0082] By employing multi-factor authentication, operational behavior monitoring, and tiered responses to abnormal activities, a secure closed loop is constructed: strong authentication is achieved by combining passwords, dynamic codes, and biometrics, with automatic account locking upon failure or exceeding limits; all operation records are logged using six elements and tamper-proof verification values ​​are generated; responses such as re-authentication, session termination, or manual locking are dynamically triggered based on the number of abnormal matches; audit logs are anonymized, verified one-way, and stored in unmodifiable storage to ensure traceability and non-repudiation, significantly improving the access security and compliance auditing capabilities of medical imaging data while ensuring access efficiency.

[0083] The above are merely preferred embodiments 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 its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A cloud platform-based distributed medical image data storage method, characterized in that, Includes the following steps: Step 1: Receive DICOM format medical image data from the medical image scanning and reconstruction process, including image pixel data and DICOM standard-compliant metadata. Perform automatic artifact recognition and removal processing on the image pixel data and perform standardization processing on the metadata to generate standardized DICOM image data. Step 2: Apply semantic region-based differential compression to the standardized DICOM image data, allocate compression intensity according to the lesion area and anatomical structure, and automatically repair structural breaks, edge blurring and texture loss using a dual repair mechanism of structure and texture separation. Step 3: Encrypt and store the compressed and repaired medical image data, and perform intelligent hierarchical management based on access frequency and sensitivity, automatically migrate to hot storage, warm storage or cold storage media, and adaptively schedule storage resources based on the real-time load of storage nodes, and balance the read and write request load among storage nodes. Step 4: When a user accesses the hierarchically stored medical image data, perform multi-factor authentication, monitor and record user operation behavior, detect abnormal activities and trigger response mechanisms, and generate audit logs.

2. The distributed medical image data storage method based on a cloud platform according to claim 1, characterized in that, The process of differential compression based on semantic regions and allocating compression intensity according to lesion regions and anatomical structures in step two includes: The standardized DICOM medical image data is used as input. The standardized DICOM image data is accompanied by a set of semantic regions. Each semantic region has a unique semantic label. For each semantic region, the semantic region and its semantic label are read. If the semantic label indicates that the region belongs to the pathological category, the semantic region is the lesion region. If the semantic label indicates that the region belongs to the normal tissue category, the semantic region is the anatomical structure region. The Shannon entropy of the wavelet coefficient distribution of the semantic region is calculated as the information redundancy, and the average gradient magnitude of all pixels in the semantic region is calculated as the structural complexity. The compression ratio of the semantic region is calculated based on these two features. If the semantic region is a lesion region, calculate the area ratio of the lesion region, construct a gray-level co-occurrence matrix within the local window of the lesion region, calculate the contrast, and calculate the average contrast of all local windows of the lesion region as the texture complexity. Calculate the compression intensity of the lesion region based on the area ratio, structural complexity, and texture complexity. If the semantic region is an anatomical structure region, the same computational structure as the lesion region is used to separate the compression strategies of the lesion region and the anatomical structure region. The lesion region outputs the compression intensity, and the anatomical structure region outputs the compression ratio. All semantic regions output the corresponding compression control parameters according to this rule, and perform partitioned compression on the standardized DICOM image data accordingly to generate compression results.

3. The distributed medical image data storage method based on a cloud platform according to claim 2, characterized in that, The automatic repair process in step two, which employs a dual repair mechanism of structure and texture separation to address structural fractures, blurred edges, and texture loss, includes: For each semantic region, read the gradient magnitude of all pixels within the semantic region and the gray-level co-occurrence matrix contrast of all local windows, calculate the average gradient magnitude of all pixels within the semantic region and the average gray-level co-occurrence matrix contrast of all local windows, and determine the repair weight of the semantic region based on the two calculation results. The structural repair intensity is set to equal the repair weight, and the texture repair intensity is set to equal one minus the repair weight. The structural repair intensity is used to repair the gradient-related structural information of the semantic region, and the texture repair intensity is used to repair the contrast-related texture information of the gray-level co-occurrence matrix of the semantic region, generating the repaired semantic region content. All semantic regions are processed in this way, and all repaired semantic region contents are merged to generate the final repair result.

4. The distributed medical image data storage method based on a cloud platform according to claim 1, characterized in that, Step three involves encrypting and storing the compressed and repaired medical image data, and then implementing intelligent hierarchical management based on access frequency and sensitivity. This includes: The process involves taking compressed and repaired medical image data as input, and for each compressed and repaired medical image data, obtaining the number of accesses, sensitivity score, and storage cost, and calculating the hierarchical evaluation value of the medical image data based on the number of accesses, sensitivity score, and storage cost. Obtain a first boundary value and a second boundary value. If the first boundary value is greater than the second boundary value, compare the hierarchical evaluation value of the medical image data with the first boundary value. If the hierarchical evaluation value is greater than the first boundary value, allocate the medical image data to a hot storage medium. If the hierarchical evaluation value is not greater than the first boundary value, compare the hierarchical evaluation value of the medical image data with the second boundary value. If the hierarchical evaluation value is greater than or equal to the second boundary value, allocate the medical image data to a warm storage medium. If the hierarchical evaluation value is less than the second boundary value, allocate the medical image data to a cold storage medium. Before writing to the allocated storage medium, the medical image data is encrypted using the AES algorithm, thus completing the encryption and storage of the medical image data.

5. A distributed medical image data storage method based on a cloud platform according to claim 4, characterized in that, The process of automatically migrating the corresponding data to a hot storage, warm storage, or cold storage medium in step three includes: For each medical image data, obtain the number of accesses to the medical image data in the most recent three statistical periods, check whether the number of accesses to the medical image data in the most recent three statistical periods has been continuously increasing or decreasing. If the number of accesses to the medical image data in the most recent three statistical periods has been continuously increasing or decreasing, obtain the predicted value of the number of accesses to the medical image data in the future and the equivalent cost and benefit of a single access, obtain the unit time cost increment after migrating the medical image data from the current storage level to the target storage level, and calculate the migration decision value of the medical image data based on the predicted value of the number of accesses in the future, the equivalent cost and benefit of a single access, and the unit time cost increment. Compare the migration decision value of the medical image data with zero. If the migration decision value of the medical image data is greater than zero, obtain the current storage level, level evaluation value, first boundary value, and second boundary value of the medical image data. Based on the level evaluation value, first boundary value, and second boundary value of the medical image data, compare the current storage level of the medical image data with the target storage level. If the current storage level is not equal to the target storage level, migrate the medical image data from the current storage level to the target storage level. If the migration decision value of the medical image data is not greater than zero, do not perform the migration operation.

6. A distributed medical image data storage method based on a cloud platform according to claim 5, characterized in that, Step three, which involves adaptively scheduling storage resources based on the real-time load of storage nodes and balancing read and write request loads among storage nodes, includes: For each new read / write request, initialize the total overall pressure value to zero. For each storage node, obtain the average resource usage of the storage node, obtain the average request response time of the storage node, obtain the baseline response time, calculate the normalized value of the average request response time of the storage node, calculate the overall pressure value of the storage node, and sum the overall pressure values ​​of the storage nodes to the total overall pressure value. For each storage node, obtain the overall pressure value of the storage node, calculate the allocation weight of the storage node. If the allocation weight is less than zero, set it to zero and accumulate the allocation weights of the storage nodes to the total allocation weight. If the total allocation weight is greater than zero, for each storage node, obtain the allocation weight of the storage node, calculate the probability of the storage node being selected, select a storage node according to the probability of each storage node being selected, and allocate new read and write requests to the selected storage node. If the total allocation weight is equal to zero, distribute new read and write requests evenly and randomly to any storage node.

7. The distributed medical image data storage method based on a cloud platform according to claim 1, characterized in that, The process of performing multi-factor authentication and monitoring and recording user behavior when a user accesses hierarchically stored medical image data in step four includes: Get the medical image data access request, get the current session credential status. If the current session credential is invalid, get the first verification result, the second verification result, and the third verification result. If the first verification result, the second verification result, or the third verification result fails, get the number of consecutive verification failures and increment the number of consecutive verification failures by one. If the number of consecutive verification failures is less than three, return a verification failure message. If the number of consecutive verification failures is greater than or equal to three, get the account lock duration, put the account in a locked state, and end the processing of this access request. If the first verification result, the second verification result, and the third verification result are all successful, reset the number of consecutive verification failures to zero and issue a session credential. Obtain user identity, operation type, operation time, operation source Internet Protocol address, unique identifier of the device used for the operation, and unique identifier of the medical image data being operated, and generate an operation record.

8. A distributed medical image data storage method based on a cloud platform according to claim 7, characterized in that, The process of performing abnormal activity identification and response requirements and generating audit logs in step four includes: Get user actions and the number of abnormal matches. If the number of abnormal matches is zero, set the operation result to normal. If the number of abnormal matches is one, set the operation result to level one abnormality. If the number of abnormal matches is greater than or equal to two, set the operation result to level two abnormality. Obtain the operation time, user identity, operation type, operation parameter summary, operation result and audit-specific voucher, calculate the integrity verification string, generate audit log, and store the audit log in a storage location that cannot be modified or deleted.

Images