A medical image storage space optimization method and system

Abstract

The application relates to a medical image storage space optimization method and system, and belongs to the technical field of medical image data storage optimization. The method comprises the following steps: acquiring initial medical image data, and performing index mapping on medical image structured information to obtain medical image label parameters; performing adaptive compression parameter regulation on a storage layered unit, and performing layer-by-layer forward coding based on a cross-layer storage mapping algorithm to obtain a differentiated storage compression strategy; independently controlling a compression channel on each storage node, independently saving differentiated incremental data, and controlling the compression parameters of the storage node; automatically migrating data between different storage levels according to the access frequency and the time threshold, automatically adjusting the storage weight of a high-value area, superimposing labels on a manually marked area, and regularly performing integrity checking and redundancy cleaning.

Description

Technical Field

[0001] This invention belongs to the field of medical image data storage optimization technology, specifically relating to a method and system for optimizing medical image storage space. Background Technology

[0002] With the rapid development of medical imaging equipment and the continuous improvement of hospital information technology, the application of medical imaging equipment in clinical diagnosis is becoming increasingly widespread. Medical institutions generate a large amount of high-resolution medical image data every day, especially in large hospitals and regional medical platforms, where image data shows a continuous growth trend.

[0003] Because medical images typically have characteristics such as large data volume, long storage period, and significant differences in access frequency, traditional medical image storage methods are gradually facing problems such as insufficient storage space, decreased data access efficiency, and increased system operating costs.

[0004] In existing technologies, medical images are typically stored through centralized management and by using uniform compression or fixed hierarchical storage methods.

[0005] However, traditional compression methods mostly use uniform compression parameters, which cannot differentiate images according to the importance of different tissue regions. This can easily lead to the loss of detailed information in key diagnostic areas, thus affecting the accuracy of subsequent diagnosis and analysis.

[0006] For frequently accessed hot image data, the existing system lacks dynamic heat identification and adaptive storage scheduling mechanisms, which means that high-frequency access data may still be stored on low-speed media, affecting doctors' access efficiency.

[0007] Furthermore, as historical image data accumulates over a long period, a large amount of duplicate and redundant data can easily appear between different storage layers.

[0008] Existing technologies typically rely solely on static redundant backups for data management, lacking dynamic detection and incremental update mechanisms for inter-layer data redundancy, resulting in wasted storage resources. Furthermore, traditional storage systems often employ fixed-cycle migration strategies during data migration, failing to intelligently adjust based on image access frequency, clinical importance, and manually annotated information, thus failing to meet the demands of modern medical scenarios for efficient, low-cost, and highly reliable storage management.

[0009] On the other hand, during the long-term archiving of medical images, doctors or administrators often manually mark key lesion areas, critical tissue areas, and high-value diagnostic areas. However, existing systems lack targeted storage weight management and compression protection mechanisms for manually marked areas, which may result in significant compression loss during subsequent compression or migration, reducing the fidelity of critical image information.

[0010] Therefore, how to construct a medical image storage space optimization method that can realize structured analysis, differentiated compression, adaptive hierarchical storage, dynamic detection of inter-layer redundancy, and closed-loop feedback optimization has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0011] To address the aforementioned problems in the existing technology, this invention provides a method for optimizing medical image storage space. The objective of this invention can be achieved through the following technical solutions: S1: Obtain initial medical image data, perform metadata parsing on the medical images, extract structured information from the medical images, and index and map the structured information of the medical images to obtain medical image tag parameters; S2: Convert the medical image label parameters into storage layer weight vectors, perform adaptive compression parameter adjustment on the storage layer unit, and perform layer-by-layer forward encoding based on the cross-layer storage mapping algorithm to obtain a differentiated storage compression strategy. S3: Based on the differentiated storage compression strategy, the compression channel is independently controlled on each storage node, the differential incremental data is independently saved, the redundancy value of each layer of data for the lower layer of storage is calculated in parallel, and the inter-layer redundancy is detected in real time by combining the reverse mapping matrix, and the compression parameters of the storage node are controlled. S4: Establish a closed-loop feedback optimization mechanism to automatically migrate data between different storage tiers based on access frequency and time thresholds, automatically adjust the storage weight of high-value areas, overlay labels on manually marked areas, and periodically perform integrity checks and redundancy cleanup.

[0012] Specifically, the process of extracting structured information from medical images includes the following steps: The initial medical image data is parsed and preprocessed to extract the structural features of different medical images at different spatial resolutions. The initial medical imaging data includes: computed tomography (CT) images, magnetic resonance imaging (MRI) images, and positron emission tomography (PET) images. By combining the feature semantic segmentation window, pixel-level and region-level features in medical images are aligned, and the prior image tissue relationships and anatomical structures are mapped to a high-dimensional feature space to construct structured information of medical images. Image index mapping and storage weight calculation are also performed. The structured medical image information includes: image examination type information, image modality information, and patient anonymity attribute information.

[0013] Specifically, the index mapping method is as follows: Establish a multidimensional index table for medical images, mapping the examination type, image modality, time information, and regional features in the structured information of medical images to the corresponding index dimensions, and generating a unique index identifier for the images; The unique image index identifier includes at least: associated image modality type, examination time, patient anonymity identifier, lesion region number, and storage level number; A dynamic heat matrix is ​​constructed based on the index access frequency to statistically analyze the access activity of different medical image data in real time, and a corresponding fast index channel is established for high-frequency access images. Combining the unique image index identifier, the index result is bound to the storage node number, and according to the fast index channel, low-latency storage paths and high-speed cache resources are preferentially allocated to hot images to form medical image tag parameters; The medical image labeling parameters include: image modality label, examination type label, and tissue importance label.

[0014] Specifically, the execution process of the adaptive compression parameter adjustment is as follows: Based on the storage layer weight vector, the compression parameters corresponding to each storage layer are dynamically weighted and adjusted to obtain a set of data compression parameters adapted to different image regions. The data compression parameter set includes: compression ratio parameter, quantization parameter, and encoding precision parameter; The adaptive regulation uses the tissue density change gradient as the dominant factor, assigning a lower compression loss weight to regions with drastic density changes and a higher compression ratio weight to regions with uniform density. The tissue density change gradient includes: the grayscale change rate of adjacent pixels, the region edge intensity change rate, and the tissue texture change rate; Based on the data compression parameter set, regionally differentiated compression processing is performed on medical image data, retaining high-precision image information in key tissue areas and increasing the compression rate in non-critical background areas to reduce overall storage usage.

[0015] Specifically, the layer-by-layer forward encoding process includes the following steps: Medical image data is divided into a basic layer, an enhancement layer, and a redundancy check layer according to storage levels. The base layer data is subjected to main compression encoding to generate the first layer of compressed data; the enhancement layer data is subjected to differential features between the base layer and the enhancement layer data, and differential data packets are generated using incremental encoding; the redundancy check layer is subjected to check encoding to generate a check vector for integrity recovery. The system performs layer-by-layer forward encoding according to a preset hierarchical order to form multi-level compressed storage data. Based on a cross-layer storage mapping algorithm, it optimizes the inter-layer association mapping and storage path allocation of the multi-level compressed storage data. The system dynamically configures compression parameters and storage locations according to the redundancy of data at different levels to obtain a differentiated storage compression strategy.

[0016] Specifically, the process of independently controlling the compression channel on each storage node is as follows: On each storage node, an independent data compression channel and an incremental cache channel are set up to perform original image compression and differential data update, respectively. The data contribution ratio of each compression channel is independently controlled according to the storage layer weight vector. The independently controlled compression channels include: a core diagnostic data-dominated channel and a background data-dominated channel; The core diagnostic data dominant channel controls low-loss compression parameters; the background data dominant channel controls high compression ratio parameters.

[0017] Specifically, the method for controlling the compression parameters of the storage node is as follows: The inter-layer redundancy parameters obtained from the analysis are weighted and mapped to node weights that characterize compression priority, redundant objects, and storage constraints. Based on the node weights, the quantization parameters and compression ratio are dynamically adjusted. When the redundancy value exceeds the preset threshold, an incremental update is triggered. The redundancy data distribution is decomposed into feature nodes, and the redundancy information of the current layer data is propagated back to the lower layer storage node. The inter-layer compression compensation vector is calculated based on the redundancy value, the compression parameters of the storage node are controlled, and the compensation node and its weight are compiled into a weighted directed graph structure to construct the reverse mapping matrix.

[0018] Specifically, the process of migrating data between different storage tiers includes the following steps: Data migration priority scores are calculated based on the frequency of access to medical images, the time of most recent access, and the clinical importance score. When the priority score is lower than the hot storage threshold, the data will be automatically migrated to the warm storage layer or the cold storage layer. During the migration process, keep the metadata index updated in sync and increase the storage weight of high-value diagnostic regions.

[0019] Specifically, the closed-loop feedback optimization mechanism includes: hot storage feedback mode and warm storage feedback mode; The hot storage feedback mode is used to monitor the number of accesses to medical images in real time, dynamically increase the storage weight of the corresponding medical image data according to the real-time changes in popularity, and automatically allocate high-speed cache resources and low-latency storage channels. The warm storage feedback mode is used to perform a downgrade migration of medical image data whose access frequency has decreased and exceeded a preset time threshold, migrating the corresponding data from the hot storage layer to the warm storage layer, and simultaneously adjusting the compression parameters and redundancy retention ratio. When a sudden high-frequency access occurs, the temperature stratum data is migrated back to the hot storage layer, and manual intervention and overlay annotation are introduced, giving higher weight and priority to manually marked high-value areas.

[0020] Specifically, the method for calculating the lower-level storage redundancy value is as follows: Perform feature similarity matching between the current layer data and the lower layer stored data, calculate the overlap of repeated encoded regions, quantify inter-layer data redundancy, and generate corresponding inter-layer redundancy coefficients. By combining data access frequency and compression loss parameters, the inter-layer redundancy coefficient is dynamically corrected, and the redundancy coefficient is backpropagated to the corresponding storage node to obtain the lower layer storage redundancy value.

[0021] Specifically, the process of overlaying annotations includes the following steps: Receive manual marking instructions for the target medical imaging region and obtain the location coordinates, tissue type, and diagnostic interest information of the corresponding region; Based on location coordinates, the marked regions are mapped to the corresponding image layers and storage nodes, and an independent high-weight annotation index is established for the marked regions; Access popularity tracking is performed based on the high-weighted labeled index, compression loss is reduced and redundancy retention ratio is increased for labeled areas, and the corresponding storage weight is dynamically updated based on access behavior.

[0022] Specifically, a medical image storage space optimization system includes: Medical image parsing and mapping module: acquires initial medical image data, performs metadata parsing on the medical images, extracts structured information from the medical images, and indexes and maps the structured information to obtain medical image tag parameters; Layered storage compression coding module: converts the medical image tag parameters into storage layer weight vectors, performs adaptive compression parameter adjustment on the storage layer unit, and performs layer-by-layer forward encoding based on the cross-layer storage mapping algorithm to obtain a differentiated storage compression strategy; Storage layer redundancy detection module: Based on the differentiated storage compression strategy, the compression channel is independently controlled on each storage node, the differential incremental data is independently saved, the redundancy value of each layer of data for the lower layer of storage is calculated in parallel, and the inter-layer redundancy is detected in real time by combining the reverse mapping matrix, and the compression parameters of the storage node are controlled. Closed-loop hierarchical migration management module: Establishes a closed-loop feedback optimization mechanism, automatically migrates data between different storage tiers based on access frequency and time thresholds, automatically adjusts the storage weight of high-value areas, overlays markings on manually marked areas, and periodically performs integrity checks and redundancy cleanup.

[0023] The beneficial effects of this invention are as follows: This invention provides a method for optimizing medical image storage space. By performing structured parsing, index mapping, and hierarchical compression management of medical image data, it achieves efficient storage and intelligent scheduling of medical image data. Compared with traditional uniform compression and fixed storage methods, this invention can dynamically adjust compression parameters according to the importance of different image regions, effectively reducing overall storage space usage and improving the storage efficiency of medical image data while ensuring the image quality of key diagnostic areas.

[0024] By establishing a storage layer weight vector and a cross-layer storage mapping mechanism, layer-by-layer forward encoding is performed on the data of the base layer, enhancement layer, and redundancy check layer. Combined with a differential incremental data storage method, the storage volume of repeatedly encoded data is reduced, inter-layer data redundancy is decreased, and storage resource utilization is improved. Furthermore, by real-time detection of inter-layer redundancy and dynamic adjustment of compression parameters using a reverse mapping matrix, the adaptive optimization capability of the storage system can be further enhanced.

[0025] A closed-loop feedback optimization mechanism was constructed, which can dynamically adjust the data storage hierarchy based on the frequency of medical image access, time threshold, and clinical value, enabling rapid access to hot image data and low-cost archiving of low-frequency data, thereby improving the overall access efficiency of the system.

[0026] For high-value areas manually marked by doctors or administrators, this invention can also increase the storage weight of the corresponding areas and reduce compression losses, thereby ensuring the integrity and reliability of critical diagnostic information. At the same time, this invention also has integrity verification and redundancy cleanup functions, which can enhance data security and stability during long-term storage of medical images. Attached Figure Description

[0027] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0028] Figure 1 This is a schematic diagram of the framework of a medical image storage space optimization system according to the present invention.

[0029] Figure 2 This is a schematic diagram illustrating the execution of layer-by-layer forward encoding in a medical image storage space optimization system of the present invention. Detailed Implementation

[0030] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0031] Please see Figure 1 This invention also provides a medical image storage space optimization system, specifically including: Medical image parsing and mapping module: acquires initial medical image data, performs metadata parsing on the medical images, extracts structured information from the medical images, and indexes and maps the structured information to obtain medical image tag parameters; Layered storage compression coding module: converts the medical image tag parameters into storage layer weight vectors, performs adaptive compression parameter adjustment on the storage layer unit, and performs layer-by-layer forward encoding based on the cross-layer storage mapping algorithm to obtain a differentiated storage compression strategy; Storage layer redundancy detection module: Based on the differentiated storage compression strategy, the compression channel is independently controlled on each storage node, the differential incremental data is independently saved, the redundancy value of each layer of data for the lower layer of storage is calculated in parallel, and the inter-layer redundancy is detected in real time by combining the reverse mapping matrix, and the compression parameters of the storage node are controlled. Closed-loop hierarchical migration management module: Establishes a closed-loop feedback optimization mechanism, automatically migrates data between different storage tiers based on access frequency and time thresholds, automatically adjusts the storage weight of high-value areas, overlays markings on manually marked areas, and periodically performs integrity checks and redundancy cleanup.

[0032] The specific implementation process is as follows: Medical image parsing and mapping module: acquires initial medical image data, performs metadata parsing on the medical images, extracts structured information from the medical images, and indexes and maps the structured information to obtain medical image tag parameters.

[0033] The initial medical imaging data includes at least one of CT images, MRI images, PET images, ultrasound images, and digital pathological slides; the medical imaging data format includes at least DICOM format, NIfTI format, or RAW medical imaging format.

[0034] The format parsing and modal preprocessing process is as follows: The acquired initial medical image data is identified in terms of data format, and the image header information, metadata information and pixel data in the corresponding image file are parsed. The image head information includes at least the image modality type, layer thickness parameter, spatial resolution parameter, window width and window level parameter, acquisition time parameter, and device source information; Perform corresponding modality preprocessing operations on medical image data according to different image modalities; For CT images, grayscale normalization, metal artifact suppression, and inter-slice resampling are performed. For MRI images, bias field correction, noise filtering, and multi-sequence registration are performed. For PET images, standard acquisition value normalization and low count noise correction are performed. For ultrasound images, speckle noise suppression and edge enhancement processing are performed. For digital pathology slide images, color normalization, multi-magnification pyramid stratification, and background area removal are performed.

[0035] In practice, the system first receives raw medical image data uploaded from the hospital's PACS system through the image access module, and then performs format parsing and modal preprocessing on the medical images. The modal preprocessing includes grayscale normalization, noise filtering, spatial resampling, and slice thickness unification to eliminate imaging differences between different devices.

[0036] The process of extracting structured information from medical images includes the following steps: A pre-trained medical image segmentation model was used to identify tissue regions in images at different spatial resolutions, and to extract organ boundaries, lesion regions, tissue density changes and texture distribution features. The pre-trained medical image segmentation model can be a deep learning model based on U-Net structure, 3D U-Net structure, nnU-Net structure or Swin Transformer structure. The input of the model is standardized medical image data, and the output is the pixel-level segmentation result of the corresponding tissue region. The system employs a semantic coding model to perform high-dimensional semantic vector mapping of image tissue relationships and anatomical structures. Specifically: The segmented tissue region labels, organ categories, and lesion attributes are input into a pre-trained semantic coding network (such as Medical-BERT, BioBERT, or CLIP-Medical multimodal coding model) to obtain the corresponding structural semantic feature vectors. Vectorization encoding is performed on the regional texture features, grayscale statistical features, and spatial location information in the image to form regional content feature vectors. Semantic spatial alignment and fusion processing is then performed on the structural semantic feature vectors and the regional content feature vectors.

[0037] The semantic space alignment and fusion operations can also be implemented using a lightweight multilayer perceptron network (MLP): The structural semantic feature vector and the regional content feature vector are concatenated and then input into the fusion network. After nonlinear mapping, the fused structured feature vector is output to enhance the semantic consistency between different modalities of medical images.

[0038] Based on the above fusion results, the system constructs a set of structured medical image information and performs image index mapping and storage weight calculation.

[0039] The index mapping process is as follows: Establish a multidimensional index table for medical images, mapping examination type, image modality, examination time, lesion area, tissue category and patient anonymity identifier to the corresponding index dimensions, and generating a unique index identifier for the image. A dynamic heat matrix is ​​constructed by statistically analyzing the access frequency of different medical image data based on historical access logs. Establish a fast indexing channel for frequently accessed images and prioritize binding hot images to low-latency storage nodes; The final result is a medical image tag parameter that includes image level, access frequency, organizational importance, and storage priority.

[0040] Layered storage compression encoding module: The medical image label parameters are converted into storage layer weight vectors, adaptive compression parameter adjustment is performed on the storage layer unit, and a differentiated storage compression strategy is obtained by forward encoding layer by layer based on the cross-layer storage mapping algorithm.

[0041] The storage tiering unit is used to: hierarchically manage the storage of medical image data according to access frequency, clinical importance, compression requirements and storage timeliness, and perform corresponding data compression, cache scheduling and storage path allocation according to the data characteristics of different tiers.

[0042] The cross-layer storage mapping algorithm is used to: establish inter-layer association mapping relationships for medical image data, and dynamically allocate corresponding storage paths and storage nodes according to the access frequency, compression requirements, redundancy level, and storage priority of the corresponding medical image data.

[0043] In the specific implementation process, the system first establishes a storage layer weight vector based on the medical image label parameters.

[0044] The storage layer weight vector includes at least: Image access popularity weight: used to measure the frequency of medical images being accessed and used; Tissue importance weighting: used to distinguish the diagnostic value of different anatomical tissues; Lesion sensitivity weight: Used to assess the sensitivity of a region to lesion detection and diagnosis; Time decay weight: used to describe the decrease in the importance of an image over time; Compression tolerance weight: Used to measure the acceptable level of distortion in different regions.

[0045] The execution process of the adaptive compression parameter adjustment is as follows: Based on the storage layer weight vector, the compression parameters corresponding to each storage layer are dynamically weighted and adjusted to form a set of data compression parameters adapted to different image regions. Among them, areas with high diagnostic value, such as tumor margins, vascular structures, and fracture areas, are assigned lower compression loss weights. For background tissue, uniform soft tissue regions, and blank regions, assign higher compression ratio weights.

[0046] In practice, the system adopts a region-aware differentiated compression method: Low-loss compression algorithms (such as JPEG2000 lossless mode or HEVC Intra low quantization mode) are used for critical tissue regions. High compression ratio encoding is used for non-critical background areas to reduce overall storage usage.

[0047] The layer-by-layer forward encoding process includes the following steps: Medical image data is divided into a base layer, an enhancement layer, and a redundancy check layer; in: The base layer is used to store the original master image information; The enhancement layer is used to store high-frequency detail regions and incremental difference information; The redundancy check layer is used to generate data recovery check vectors.

[0048] In practice: Perform primary compression encoding on the base layer data to generate the first layer of compressed data; The difference features between the enhancement layer data and the base layer data are extracted, and incremental encoding is used to generate difference data packets; Perform check encoding processing on the redundant check layer to generate a check vector for integrity recovery.

[0049] The system performs layer-by-layer forward encoding according to a preset hierarchical order, and performs inter-layer association mapping and path optimization allocation on multi-level compressed storage data based on a cross-layer storage mapping algorithm. The system dynamically configures compression parameters and storage locations according to the redundancy of data at different levels to obtain a differentiated storage compression strategy.

[0050] Storage inter-layer redundancy detection module: Based on the differentiated storage compression strategy, the compression channel is independently controlled on each storage node, the differential incremental data is independently saved, the redundancy value of each layer of data for the lower layer of storage is calculated in parallel, and the inter-layer redundancy is detected in real time by combining the reverse mapping matrix, and the compression parameters of the storage node are controlled.

[0051] In this embodiment, as Figure 2 As shown, the layer-by-layer forward encoding process includes the following steps: Medical image data is divided into a basic layer, an enhancement layer, and a redundancy check layer according to storage levels. The base layer data is subjected to main compression encoding to generate the first layer of compressed data; the enhancement layer data is subjected to differential features between the base layer and the enhancement layer data, and differential data packets are generated using incremental encoding; the redundancy check layer is subjected to check encoding to generate a check vector for integrity recovery. The system performs layer-by-layer forward encoding according to a preset hierarchical order to form multi-level compressed storage data. Based on a cross-layer storage mapping algorithm, it optimizes the inter-layer association mapping and storage path allocation of the multi-level compressed storage data. The system dynamically configures compression parameters and storage locations according to the redundancy of data at different levels to obtain a differentiated storage compression strategy.

[0052] In practice, an independent data compression channel and an incremental caching channel are set up on each storage node.

[0053] in: The data compression channel is used to perform compression on the raw image; Incremental cache channels are used to perform differential data updates.

[0054] The system dynamically allocates bandwidth resources and compression priorities for each compression channel based on the storage layer weight vector.

[0055] The execution process of the data compression channel is as follows: Perform main compression encoding on basic medical imaging data; The compression accuracy is dynamically adjusted based on the importance of the image region. Generate compressed storage data at the corresponding level.

[0056] The execution process of the incremental cache channel is as follows: Real-time caching of incremental difference data generated during the updating of medical images; Perform local incremental coding on newly added or modified areas; The differential incremental data is then associated and mapped with the original compressed data to reduce the amount of data that is stored repeatedly.

[0057] The method for calculating the redundancy value of the lower-level storage is as follows: Perform feature similarity matching between the current layer data and the lower layer stored data; Statistical analysis of the overlap of repeated coding regions; Quantify the data redundancy between layers and generate the corresponding inter-layer redundancy coefficient.

[0058] In practice: The cosine similarity or structural similarity index (SSIM) is used to calculate the degree of data similarity between layers; The redundancy coefficient is dynamically corrected by combining access frequency and compression loss parameters, and the corrected redundancy coefficient is backpropagated to the corresponding storage node to obtain the lower-level storage redundancy value.

[0059] When the redundancy value exceeds the preset threshold: Trigger the incremental update mechanism; Perform feature node decomposition on redundant data distribution; The redundant information of the current layer data is propagated back to the lower layer storage node; The inter-layer compression compensation vector is calculated based on the redundancy value, and the compensation nodes and their weights are compiled into a weighted directed graph structure to construct a reverse mapping matrix to control the compression parameters of the storage nodes in real time.

[0060] Closed-loop hierarchical migration management module: Establish a closed-loop feedback optimization mechanism to automatically migrate data between different storage tiers based on access frequency and time thresholds, automatically adjust the storage weight of high-value areas, overlay labels on manually marked areas, and periodically perform integrity checks and redundancy cleanup.

[0061] Data migration priority scores are calculated based on the frequency of access to medical images, the time of most recent access, and clinical importance scores.

[0062] When the priority score is below the hot storage threshold: Automatically migrate data to a warm or cold storage layer; During the migration process, the metadata index is updated synchronously, and the storage weight of high-value diagnostic regions is increased.

[0063] The closed-loop feedback optimization mechanism includes hot storage feedback mode and warm storage feedback mode.

[0064] in: The hot storage feedback mode is used to monitor the number of accesses to medical images in real time and dynamically increase the storage weight of the corresponding medical image data. It also automatically allocates high-speed cache resources and low-latency storage channels.

[0065] Warm storage feedback mode is used for: Perform a downgrade migration for medical images whose access frequency has decreased and exceeded a preset time threshold; The corresponding data is migrated from the hot storage layer to the warm storage layer; Simultaneously adjust compression parameters and redundancy retention ratio.

[0066] When a sudden high-frequency access occurs: The temperature sphere data is automatically migrated back to the thermal storage layer, and a manual intervention overlay annotation mechanism is introduced to assign higher weight and priority to manually marked high-value areas.

[0067] The overlay annotation process includes the following steps: Receive manual marking instructions for the target medical imaging region; Obtain the location coordinates, tissue type, and diagnostic information for the corresponding region; The marked areas are mapped to the corresponding image layers and storage nodes based on their location coordinates; Create independent high-weighted annotation indexes for the marked regions, and perform access popularity tracking based on the high-weighted annotation indexes; Reduce compression loss and increase redundancy retention for marked areas, and dynamically update the corresponding storage weights based on access behavior.

[0068] In this embodiment, the calculation process for the data migration priority score is as follows: Obtain the historical access frequency, most recent access time, clinical importance score, storage duration, and corresponding storage level weight of the corresponding medical image data; The above parameters are standardized to eliminate the dimensional differences between different parameters; Among them, access frequency is used to characterize the historical access activity of medical image data, recent access time is used to characterize the time interval between medical image data and the current time, clinical importance score is used to characterize the value level of the corresponding medical image data in clinical diagnosis, storage duration is used to characterize the archiving time of medical image data, and storage level weight is used to characterize the importance of the current storage level of the corresponding medical image data. Further, based on the preset weight ratio, the standardized parameters are comprehensively weighted and calculated to obtain the corresponding data migration priority score; Among them, medical imaging data with high access frequency, short recent access time, and high clinical importance score have higher data migration priority scores. Medical image data that has been stored for a long time and has not been accessed for a long period of time has a lower data migration priority score; The storage level of medical imaging data is dynamically adjusted based on the data migration priority score. When the data migration priority score is higher than the preset hot storage threshold, the corresponding medical image data will be retained in the hot storage layer, and high-speed cache resources and low-latency storage channels will be allocated. When the data migration priority score is lower than the preset warm storage threshold, the corresponding medical image data will be automatically migrated to the warm storage layer or the cold storage layer, and the corresponding metadata index, storage path mapping relationship and compression parameter configuration will be updated synchronously.

[0069] Example 1: Application of layered compression storage of brain CT images; This embodiment specifically illustrates the application process of the method of this application in the storage of brain CT images.

[0070] The hospital's PACS system generates a large amount of brain CT scan data daily, with each scan containing approximately 300 to 800 slice images. The system first receives the raw DICOM format brain CT images and performs preprocessing operations, including grayscale normalization, skull region separation, and brain tissue segmentation.

[0071] The system uses a pre-trained 3D U-Net model to automatically segment the brain parenchyma, ventricle, and suspected hemorrhage areas, and extracts the corresponding tissue semantic feature vectors using a Medical-BERT model.

[0072] Semantic fusion is then performed on the lesion area and the background area to form a set of structured image information.

[0073] During the storage phase: A low-loss compression strategy was used for key tissues such as the hemorrhage area and brainstem area. A high compression ratio encoding method is used for the background area.

[0074] The base layer data is stored on hot storage nodes, the enhancement layer differential data is stored on warm storage nodes, and the redundancy check layer is stored on cold backup nodes.

[0075] In subsequent clinical use: If a cerebral hemorrhage case is frequently accessed by doctors, the system will automatically increase the storage popularity weight of that case and migrate it to a high-speed cache node. If the device is not accessed within a preset time, it will automatically be downgraded and migrated to a warm storage layer.

[0076] Tests have shown that, compared to traditional uniform compression schemes, the method proposed in this application can reduce overall storage usage while maintaining stable image quality in key lesion areas, thus meeting clinical diagnostic requirements.

[0077] Example 2: Incremental storage application of pathological slides; This embodiment illustrates the application of the method of this application in the storage of digital pathological slides.

[0078] Digital pathology slides typically have ultra-high resolution, with single image data sizes reaching several gigabytes or more. The system first performs multi-magnification region segmentation on the pathology slides and identifies cancerous areas, normal tissue areas, and blank background areas.

[0079] After the pathologist manually marked the cancerous areas: The system automatically creates a high-weighted label index and reduces the compression loss parameter in that region while increasing the redundancy retention ratio.

[0080] When only a local area is modified during subsequent pathological analysis: The system performs local incremental encoding only on the modified area and saves the difference data through the incremental cache channel to avoid repeatedly saving the complete slice data.

[0081] After multiple iterations and updates: The inter-layer redundancy detection mechanism automatically identifies duplicate areas and performs compression compensation and redundancy cleanup on redundant data.

[0082] While ensuring the clarity of the pathological diagnosis area, the storage space required for long-term archiving of pathological images has been significantly reduced.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for optimizing medical image storage space, characterized in that, include S1: Obtain initial medical image data, perform metadata parsing on the medical images, extract structured information from the medical images, and index and map the structured information of the medical images to obtain medical image tag parameters; S2: Convert the medical image label parameters into storage layer weight vectors, perform adaptive compression parameter adjustment on the storage layer unit, and perform layer-by-layer forward encoding based on the cross-layer storage mapping algorithm to obtain a differentiated storage compression strategy. S3: Based on the differentiated storage compression strategy, the compression channel is independently controlled on each storage node, the differential incremental data is independently saved, the redundancy value of each layer of data for the lower layer of storage is calculated in parallel, and the inter-layer redundancy is detected in real time by combining the reverse mapping matrix, and the compression parameters of the storage node are controlled. S4: Establish a closed-loop feedback optimization mechanism to automatically migrate data between different storage tiers based on access frequency and time thresholds, automatically adjust the storage weight of high-value areas, overlay labels on manually marked areas, and periodically perform integrity checks and redundancy cleanup.

2. The method according to claim 1, characterized in that, The process of extracting structured information from medical images includes the following steps: The initial medical image data is parsed and preprocessed to extract the structural features of different medical images at different spatial resolutions. By combining the feature semantic segmentation window, pixel-level and region-level features in medical images are aligned, and the prior image tissue relationships and anatomical structures are mapped to a high-dimensional feature space to construct a set of structured information of medical images. Image index mapping and storage weight calculation are then performed.

3. The method according to claim 1, characterized in that, The index mapping method is as follows: Establish a multidimensional index table for medical images, mapping the examination type, image modality, time information, and regional features in the structured information of medical images to the corresponding index dimensions, and generating a unique index identifier for the images; A dynamic heat matrix is ​​constructed based on the index access frequency to statistically analyze the access activity of different medical image data in real time, and a corresponding fast index channel is established for high-frequency access images. By combining the unique image index identifier, the index result is bound to the storage node number, and according to the fast index channel, low-latency storage paths and high-speed cache resources are preferentially allocated to hot images to form medical image tag parameters.

4. The method according to claim 1, characterized in that, The execution process of the adaptive compression parameter adjustment is as follows: Based on the storage layer weight vector, the compression parameters corresponding to each storage layer are dynamically weighted and adjusted to obtain a set of data compression parameters adapted to different image regions. The adaptive regulation uses the tissue density change gradient as the dominant factor, assigning a lower compression loss weight to regions with drastic density changes and a higher compression ratio weight to regions with uniform density. Based on the data compression parameter set, regionally differentiated compression processing is performed on medical image data, retaining high-precision image information in key tissue areas and increasing the compression rate in non-critical background areas to reduce overall storage usage.

5. The method according to claim 1, characterized in that, The layer-by-layer forward encoding process includes the following steps: Medical image data is divided into a basic layer, an enhancement layer, and a redundancy check layer according to storage levels. Perform primary compression encoding on the base layer data and generate the first layer of compressed data; The difference features between the enhancement layer data and the base layer are extracted, and incremental encoding is used to generate difference data packets; the redundancy check layer is subjected to check encoding processing to generate check vectors for integrity recovery. The system performs layer-by-layer forward encoding according to a preset hierarchical order to form multi-level compressed storage data. Based on a cross-layer storage mapping algorithm, it optimizes the inter-layer association mapping and storage path allocation of the multi-level compressed storage data. The system dynamically configures compression parameters and storage locations according to the redundancy of data at different levels to obtain a differentiated storage compression strategy.

6. The method according to claim 1, characterized in that, The process of independently controlling the compression channel on each storage node is as follows: On each storage node, an independent data compression channel and an incremental cache channel are set up to perform raw image compression and differential data update, respectively. The bandwidth resources and compression priority of each compression channel are dynamically allocated according to the storage layer weight vector, and the data contribution ratio of each compression channel is independently controlled. The data compression channel is executed as follows: the basic medical image data is subjected to main compression encoding processing, and the compression accuracy is dynamically adjusted according to the importance of the image area to generate compressed storage data of the corresponding level. The execution process of the incremental caching channel is as follows: real-time caching of differential incremental data generated during the medical image update process, performing local incremental encoding on newly added or modified areas, and associating and mapping the differential incremental data with the original compressed data to reduce the amount of duplicate data storage.

7. The method according to claim 1, characterized in that, The method for controlling the compression parameters of the storage node is as follows: The inter-layer redundancy parameters obtained from the analysis are weighted and mapped to node weights that characterize compression priority, redundant objects, and storage constraints. Based on the node weights, the quantization parameters and compression ratio are dynamically adjusted. When the redundancy value exceeds the preset threshold, an incremental update is triggered. The redundancy data distribution is decomposed into feature nodes, and the redundancy information of the current layer data is propagated back to the lower layer storage node. The inter-layer compression compensation vector is calculated based on the redundancy value, the compression parameters of the storage node are controlled, and the compensation node and its weight are compiled into a weighted directed graph structure to construct the reverse mapping matrix.

8. The method according to claim 1, characterized in that, The process of migrating data between different storage tiers includes the following steps: Data migration priority scores are calculated based on the frequency of access to medical images, the time of most recent access, and the clinical importance score. When the priority score is lower than the hot storage threshold, the data will be automatically migrated to the warm storage layer or the cold storage layer. During the migration process, keep the metadata index updated in sync and increase the storage weight of high-value diagnostic regions.

9. The method according to claim 1, characterized in that, The closed-loop feedback optimization mechanism includes: hot storage feedback mode and warm storage feedback mode; The hot storage feedback mode is used to monitor the number of accesses to medical images in real time, dynamically increase the storage weight of the corresponding medical image data according to the real-time changes in popularity, and automatically allocate high-speed cache resources and low-latency storage channels. The warm storage feedback mode is used to perform a downgrade migration of medical image data whose access frequency has decreased and exceeded a preset time threshold, migrating the corresponding data from the hot storage layer to the warm storage layer, and simultaneously adjusting the compression parameters and redundancy retention ratio. When a sudden high-frequency access occurs, the temperature stratum data is migrated back to the hot storage layer, and manual intervention and overlay annotation are introduced, giving higher weight and priority to manually marked high-value areas.

10. The method according to claim 1, characterized in that, The method for calculating the lower-level storage redundancy value is as follows: Perform feature similarity matching between the current layer data and the lower layer stored data, calculate the overlap of repeated encoded regions, quantify inter-layer data redundancy, and generate corresponding inter-layer redundancy coefficients. By combining data access frequency and compression loss parameters, the inter-layer redundancy coefficient is dynamically corrected, and the redundancy coefficient is backpropagated to the corresponding storage node to obtain the lower layer storage redundancy value.

11. The method according to claim 1, characterized in that, The process of overlaying annotations includes the following steps: Receive manual marking instructions for the target medical imaging region and obtain the location coordinates, tissue type, and diagnostic interest information of the corresponding region; Based on location coordinates, the marked regions are mapped to the corresponding image layers and storage nodes, and an independent high-weight annotation index is established for the marked regions; Access popularity tracking is performed based on the high-weighted labeled index, compression loss is reduced and redundancy retention ratio is increased for labeled areas, and the corresponding storage weight is dynamically updated based on access behavior.

12. A medical image storage space optimization system, used to perform the method as described in any one of claims 1-11, characterized in that, Medical image parsing and mapping module: acquires initial medical image data, performs metadata parsing on the medical images, extracts structured information from the medical images, and indexes and maps the structured information to obtain medical image tag parameters; Layered storage compression coding module: converts the medical image tag parameters into storage layer weight vectors, performs adaptive compression parameter adjustment on the storage layer unit, and performs layer-by-layer forward encoding based on the cross-layer storage mapping algorithm to obtain a differentiated storage compression strategy; Storage layer redundancy detection module: Based on the differentiated storage compression strategy, the compression channel is independently controlled on each storage node, the differential incremental data is independently saved, the redundancy value of each layer of data for the lower layer of storage is calculated in parallel, and the inter-layer redundancy is detected in real time by combining the reverse mapping matrix, and the compression parameters of the storage node are controlled. Closed-loop hierarchical migration management module: Establishes a closed-loop feedback optimization mechanism, automatically migrates data between different storage tiers based on access frequency and time thresholds, automatically adjusts the storage weight of high-value areas, overlays markings on manually marked areas, and periodically performs integrity checks and redundancy cleanup.

Images