A pathological panorama-oriented distributed storage and collaborative management method and system

Abstract

This invention discloses a distributed storage and collaborative management method for panoramic pathological images, comprising: acquiring the panoramic pathological image file to be uploaded, generating basic slide information and initiating a creation request; the server encodes the request as a log with an operation type identifier and replicates it to each replica node via the Raft protocol; the state machine writes the basic slide information according to the operation type; the client performs image segmentation processing and uploads the images one by one or in batches; the server encodes the segmented upload request as a corresponding log type, replicates and submits it via Raft, writes it to the segmented data table, synchronously updates the set of uploaded segment numbers and sorts and removes duplicates, and completes the upload when the number of uploaded segments reaches the total number of segments. This invention encodes slide creation and segmented upload as logs and submits them across multiple nodes, while maintaining the set of uploaded segment numbers. It supports breakpoint resumption while achieving strong distributed consistency, avoiding full retransmission and improving the reliability and recovery efficiency of uploading large pathological files.

Description

Technical Field

[0001] This invention relates to the field of medical image data storage and management technology, and in particular to a distributed storage and collaborative management method and system for pathological panoramic images. Background Technology

[0002] Whole Slide Images (WSI) are ultra-high-resolution images created by digitizing pathological slides using specialized scanning equipment. Single file sizes can reach several gigabytes or even hundreds of gigabytes, and they typically contain multi-layered pyramid structures to support random region reading at different resolutions. In clinical diagnosis and AI-assisted analysis scenarios, in addition to the original panoramic image itself, the system also needs to manage basic slide information, field-of-view annotations, small image classification results, DNA object parameters, slide-level statistical information, slide positive / negative determination results, and multiple rounds of analysis results generated under different combinations of detection and classification models. Therefore, a pathological panoramic image system is essentially a complex data management system involving original images, derived images, structured results, statistical information, and multi-role access.

[0003] Currently, existing technologies typically employ the following methods for storing and managing panoramic pathology images:

[0004] 1) Upload the panoramic image as a single large file directly to a standalone or single-node storage system, or save the original image through object storage, and then have the backend independently manage the annotations, sub-images, statistical results, and analysis results. With this method, if the large file upload is interrupted, the recovery cost is high and the retransmission efficiency is low.

[0005] 2) In a multi-node deployment environment, the correspondence between the “node where the glass slide data is located” and the “access entry node” is maintained through external gateways, scheduling services or manual configuration. The system routing is complex, and when the business entry node is inconsistent with the actual node where the data is located, an additional routing mechanism is required, which increases the complexity of the architecture.

[0006] 3) For AI detection, classification and DNA analysis results, they are usually stored in separate data tables or separate files. They lack the ability to manage the results generated by the same slide under different combinations of detection and classification models. This can easily lead to result overlap or confusion, which is not conducive to traceability and manual verification.

[0007] 4) For queries of cell minimaps and DNA objects, they are usually stored as large objects at the field level. If it is necessary to find a single cell or DNA target by object ID, it is often necessary to repeatedly perform full deserialization and traversal, resulting in poor query performance.

[0008] 5) For browsing needs such as DeepZoom and view maps, pre-cut maps, pre-generated tiles, or single-machine caching are usually relied upon. Pre-generating all thumbnails, tiles, and sub-maps will bring a significant additional storage burden.

[0009] In summary, existing technologies struggle to simultaneously achieve efficient uploading and consistent storage of large pathology panoramic image files, transparent access in a distributed environment, isolated management of multi-model analysis results, rapid querying of fine-grained objects, and unified organization of derived data. Therefore, there is an urgent need for a distributed technology solution specifically designed for the characteristics of pathology panoramic image services to improve upload efficiency, access transparency, multi-model collaborative management capabilities, and the efficiency of querying object-level results. Summary of the Invention

[0010] This invention proposes a distributed storage and collaborative management method and system for pathological panoramic images, which solves the problems of low upload and storage efficiency, complex cross-node access routing, and poor fine-grained object query performance in existing case panoramic image management systems.

[0011] The technical solution of this invention is implemented as follows:

[0012] The first aspect of this invention provides a distributed storage and collaborative management method for pathological panoramic images, including a consistent slice uploading step, specifically including the following sub-steps:

[0013] The client obtains the pathological panoramic image file to be uploaded, generates basic slide information, and sends a slide creation request to the server.

[0014] The server encodes the creation request into a log with an operation type identifier and replicates it to each replica node via the Raft protocol;

[0015] After the log is committed, the state machine calls the data layer to write the new slide basic information according to the operation type, and deletes the old historical data under the same barcode;

[0016] The client segments the original image into segments based on the segment size and sends multiple segments and their corresponding upload requests to the server one by one or in batches.

[0017] The server encodes the slice upload request into the corresponding log type, copies and submits it via the Raft protocol, and writes it to the slice data table; it synchronously updates the set of uploaded slice sequence numbers and sorts and removes duplicates; when the number of uploaded slices reaches the total number of slices, it updates the slide upload completion status.

[0018] Specifically, the basic information of the slide includes at least several of the following: barcode, format, file summary, field of view rows and columns, file size, slide size, and total number of slides.

[0019] Preferably, the method further includes a virtual whole-image reconstruction and random access step, specifically including the following sub-steps:

[0020] The server maps the original images that have been stored in the database into logically complete panoramic image files through a virtual file system.

[0021] When the image reading module initiates a read request based on file offset, the virtual file system calculates the range of fragments that need to be hit based on the target offset and length;

[0022] The server uses a batch read method to obtain multiple data segments, and concatenates the multiple data segments according to the file offset order to generate a continuous byte stream;

[0023] The image wrapper performs random access, pyramid-level reading, and region cropping of a continuous byte stream as a regular panoramic image file.

[0024] Preferably, the method further includes a cross-node transparent access step, specifically comprising the following sub-steps:

[0025] Users initiate pathological image data requests through a unified URL, and the current node parses the request to extract the slide barcode;

[0026] The current node determines whether the target slide data exists locally. If it does, the current node processes it and returns the result.

[0027] If it does not exist, obtain a list of similar HTTP service nodes from the service registry, forward the original HTTP request to each candidate node one by one and attach a forwarded flag to prevent circular forwarding, until a node returns a valid response, and return the response to the requester as is.

[0028] Preferably, the method further includes a multi-model result collaborative management step, specifically including the following sub-steps:

[0029] The server generates model keys based on the combination of the detection model and the classification model;

[0030] The cell minimap results are stored in a structured manner using a combination of "slide barcode + model key + field of view coordinates + category identifier", while updating slide statistics and model relationship information.

[0031] When a query request is received, the analysis results under the corresponding model combination are accurately located based on the model key;

[0032] When a manual review or reclassification request is received, the objects are grouped according to the old and new classifications. After locating the target item based on the object ID, the migration and rewriting are performed, and the cache of the corresponding model and field of view is cleared.

[0033] Preferably, the management method further includes a rapid object localization step at the cell or DNA level, specifically comprising the following sub-steps:

[0034] When caching cell results, the server establishes a mapping relationship between object ID and image information, and when caching DNA parameters, it establishes a mapping relationship between object ID and image region, contour point set, and scaling factor.

[0035] When an image request is received by querying by object ID, the cache mapping is queried first. If a match is found, the image cropping area and contour information are obtained directly, and the corresponding image is generated. If no match is found, the corresponding field of view object data is read from the database, the cache is filled, and the localization is re-executed.

[0036] Preferably, the management method further includes a step for batch extraction of simplified DNA results from entire glass slides, specifically including the following sub-steps:

[0037] Based on the basic information of the slide, obtain the coordinate range of the entire field of view, generate the DNA data query key for each field of view, and use the underlying database batch read interface to obtain the DNA data of the entire field of view at once.

[0038] Deserialization is performed in parallel using a thread pool. For each object, simplified fields including object ID, category ID, DNA index, area, and field coordinates are extracted. All results are aggregated, sorted according to preset indicators, and the field-level DNA data is written to a cache.

[0039] Preferably, the management method further includes a dynamic DeepZoom mapping and tile generation step, specifically including the following sub-steps:

[0040] When a DeepZoom request is received, parse the target level and tile index in the request;

[0041] Establish a mapping relationship from the target layer to the original layer based on the pyramid layer information of the original panoramic image;

[0042] Select the original layer that is closest to the target layer as the source layer and calculate the corresponding clipping region;

[0043] The target image is generated by cropping the image from the source layer, scaling and encoding it.

[0044] A second aspect of the present invention provides a distributed storage and collaborative management system for pathological panoramic images, comprising:

[0045] The client is used to perform client operations in the method;

[0046] A distributed server cluster, comprising at least a Raft service layer, a state machine layer, a data layer, a virtual file system layer, an image wrapper layer, an HTTP access layer, and a cache layer, is used to execute the server operations in the method described above.

[0047] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention introduces a Raft log replication mechanism based on operation type identifier, which encodes slide creation and fragment upload as logs and commits them among multiple replica nodes. At the same time, it maintains the set of uploaded fragment sequence numbers and sorts and deduplicates them. It supports breakpoint resume after upload interruption on the basis of achieving strong distributed consistency. This mechanism avoids the drawback of full retransmission after traditional single-node upload failure, reduces the impact of network fluctuations on the upload process, and improves the upload reliability and recovery efficiency of large pathological panoramic files in weak network environment. (2) This invention maps multiple fragments stored in the database in a scattered manner into a logically complete panoramic image file through a virtual file system, and responds to read requests based on file offsets, batch hits the fragment range, and splices them in order to generate a continuous byte stream, so that the upper-level image wrapper can perform pyramid-level reading and region cropping without being aware of the underlying storage structure; it retains the advantages of writing and copying brought by fragmented storage, and is compatible with the random access interface of existing image processing libraries, realizing transparent access to pathological panoramic images without pre-generating the full image, reducing storage and development adaptation costs; (3) The present invention can access pathological image data through a unified URL entry point. The current node automatically determines the data ownership and forwards the request to the actual storage node. There is no need for external gateways or manual maintenance of routing information. This avoids the tedious operation of frequently adjusting access configuration due to changes in data distribution in traditional multi-node deployments and reduces the complexity of system operation and maintenance. (4) This invention generates a unique model key by combining a detection model and a classification model, and stores the cell minimap results in a structured manner of “slide barcode + model key + field coordinates + category identifier”. At the same time, it maintains the slide statistical information and model relationship information, so that the analysis results of the same slide under different model combinations can coexist without interference. It supports accurate query by model key and allows the results to be migrated and rewritten according to the object ID during manual review. Compared with the traditional single table or single file overwrite storage, this invention significantly enhances the traceability, isolation and maintainability of multi-round AI analysis results, and meets the business needs of model comparison and result review in pathological AI-assisted diagnosis. (5) By pre-establishing the mapping relationship between object ID and image information, image region, contour point set and scaling factor in the cache, the target image is generated directly by prioritizing the cache hit during query. Only when the cache is missing will the underlying field object data be read and the cache be filled. This avoids the overhead of repeated full deserialization and traversal when querying by ID under the traditional field-level large object storage method, and significantly improves the image acquisition speed of a single cell or DNA object. (6) When receiving a DeepZoom request, the present invention dynamically establishes a mapping relationship between the target layer and the closest original layer based on the target layer and tile index in the request and the pyramid layer information of the original panoramic image. It then crops, scales and encodes the required tiles in real time, avoiding the large amount of additional storage overhead caused by the traditional pre-cutting method. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A panoramic view in an embodiment of the present invention Figure 1 Flowchart for uploading homogeneous slices.

[0051] Figure 2 This is a flowchart illustrating the cross-node transparent HTTP access process in an embodiment of the present invention.

[0052] Figure 3 This is a flowchart illustrating the collaborative management of multi-model results in an embodiment of the present invention.

[0053] Figure 4 This is a flowchart illustrating the rapid location of cell-level or DNA-level objects in an embodiment of the present invention.

[0054] Figure 5 This is a flowchart of dynamic DeepZoom mapping and tile generation in an embodiment of the present invention.

[0055] Figure 6 This is an architecture diagram of the distributed storage and collaborative management system for pathological panoramic images according to the present invention. Detailed Implementation

[0056] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0057] This invention addresses the technical challenges of high upload interruption costs for large pathology panoramic images, complex access routing in distributed environments, difficulty in managing multiple model results concurrently, poor performance of fine-grained object queries, and high pre-storage overhead for derived images. It proposes a distributed storage and collaborative management method for pathology panoramic images. This method centers on consistent slice uploads, combining virtual whole-image reconstruction and random access, transparent cross-node access, and collaborative management of multiple model results. Furthermore, it incorporates optimized strategies such as rapid object localization, batch DNA retrieval, and dynamic DeepZoom generation, forming a complete distributed storage and management solution.

[0058] Reference Figure 1 The first aspect of this invention provides a distributed storage and collaborative management method for pathological panoramic images, comprising:

[0059] The consistent slice upload step includes the following sub-steps:

[0060] S101, the client obtains the pathological panoramic image file to be uploaded, generates basic slide information, and sends a slide creation request to the server; the basic slide information includes at least several of the following: barcode, format, file summary, field of view rows and columns, file size, slide size, and total number of slides;

[0061] S102, the server encodes the creation request into a log with an operation type identifier and replicates it to each replica node via the Raft protocol;

[0062] S103, after the log is committed, the state machine calls the data layer to write the new slide basic information according to the operation type and deletes the historical old data under the same barcode;

[0063] S104, the client divides the original image into segments according to the segment size, and sends multiple segments and corresponding segment upload requests to the server one by one or in batches;

[0064] S105, the server encodes the slice upload request into the corresponding log type, and writes it into the slice data table after copying and submitting it via the Raft protocol; it synchronously updates the set of uploaded slice sequence numbers and sorts and removes duplicates; when the number of uploaded slices reaches the total number of slices, it updates the slide upload completion status.

[0065] Through the above methods, a segmented, recoverable, and consistent distributed storage for ultra-large pathological panoramic images is achieved. This invention utilizes the Raft consensus protocol to ensure strong consistency between slide metadata and segmented data across multiple nodes. At the same time, by managing the uploaded segment sequence number set, it supports breakpoint resumption, significantly reducing the cost of recovering from upload interruptions caused by network fluctuations, and improving the reliability and efficiency of uploading large pathological panoramic image files.

[0066] Preferably, the method further includes:

[0067] The virtual whole-map reconstruction and random access step specifically includes the following sub-steps:

[0068] S201, the server maps the original image fragments that have been stored in the database into logically complete panoramic image files through the virtual file system;

[0069] S202, When the image reading module initiates a read request based on file offset, the virtual file system calculates the range of segments that need to be hit based on the target offset and length;

[0070] S203, the server uses a batch reading method to obtain multiple data segments, and concatenates the multiple data segments according to the file offset order to generate a continuous byte stream;

[0071] S204, the image wrapper performs random access, pyramid-level reading, and region cropping of the continuous byte stream as a regular panoramic image file.

[0072] Through the above method, without changing the calling method of the upper-layer image library, the fragmented panoramic images in the distributed KV storage can have access capabilities similar to local full-image files. This invention retains the write and copy advantages brought by fragmented storage, and is compatible with the random access interface of existing image processing libraries. It achieves transparent random access to the underlying fragmented storage without pre-generating the full image, reducing storage and development adaptation costs.

[0073] Preferably, the method further includes:

[0074] Transparent cross-node access steps, such as Figure 2 As shown, this step specifically includes the following sub-steps:

[0075] S301, users initiate pathological image data requests (including field of view, thumbnail, DeepZoom tile, cell image or DNA image data requests) through a unified URL, and the current node parses the request to extract the slide barcode;

[0076] S302, the current node determines whether the target slide data exists locally. If it exists, the current node processes it directly and returns the result.

[0077] S303 If it does not exist, obtain a list of similar HTTP service nodes from the service registry, forward the original HTTP request to each candidate node one by one and attach a forwarded flag to prevent circular forwarding, until a node returns a valid response, and return the response to the requester as is.

[0078] The above method enables users to access pathological image data through a unified entry point without being aware of the location of the underlying nodes; it eliminates the need for external gateways or manual maintenance of routing information, avoiding the cumbersome operation of frequently adjusting access configurations due to changes in data distribution in traditional multi-node deployments, and reducing the complexity of system operation and maintenance.

[0079] Preferably, the method further includes:

[0080] Multi-model result collaborative management steps, such as Figure 3 As shown, this step specifically includes the following sub-steps:

[0081] S401, The server generates a model key based on the combination of the detection model and the classification model;

[0082] S402 uses a combination of "slide barcode + model key + field of view coordinates + category identifier" to store cell minimap results in a structured manner, while updating slide statistical information and model relationship information.

[0083] S403, When a query request is received, the analysis results under the corresponding model combination are accurately located based on the model key;

[0084] S404: When a manual review or reclassification request is received, the objects are grouped according to the old and new classifications. After locating the target item based on the object ID, the object is migrated and rewritten, and the cache in the corresponding model and field of view is cleared.

[0085] The above methods can support the simultaneous existence of analysis results for the same slide under multiple model combinations, and support continuous management after manual revision; enhance the traceability, isolation and maintainability of multi-round AI analysis results, and meet the business needs of model comparison and result verification in pathological AI-assisted diagnosis.

[0086] Preferably, the method further includes:

[0087] Rapid localization steps for cellular or DNA-level objects, such as... Figure 4 As shown, this step specifically includes the following sub-steps:

[0088] S501, when caching cell results, the server establishes a mapping relationship between object ID and image information, and when caching DNA parameters, it establishes a mapping relationship between object ID and image region, contour point set and scaling factor.

[0089] S502: When an image request is received by querying by object ID, the cache mapping is queried first. If a match is found, the image cropping area and contour information are obtained directly, and the corresponding image is generated. If no match is found, the corresponding field of view object data is read from the database, the cache is filled, and the positioning is re-executed.

[0090] The above methods can significantly reduce the overhead of large object deserialization and repeated traversal during object-level queries; greatly shorten the image acquisition time for fine-grained objects; and improve the interactive experience for pathologists when reviewing single cells or DNA targets.

[0091] Preferably, the method further includes:

[0092] The step for batch extraction of simplified DNA results from a whole slide includes the following sub-steps:

[0093] S601: Based on the basic information of the slide, obtain the coordinate range of the entire field of view, generate the DNA data query key for each field of view, and use the underlying database batch read interface to obtain the DNA data of the entire field of view at once.

[0094] S602 uses a thread pool to perform deserialization in parallel. For each object, it extracts simplified fields including object ID, category ID, DNA index, area, and field coordinates. After summarizing all results, it sorts and outputs them according to preset indicators and writes the field-level DNA data into the cache.

[0095] The above method enables efficient DNA summary query and result export for whole slides; it combines multiple field-of-view queries into a single batch read, and with parallel processing, it significantly improves the efficiency of DNA summary statistics at the whole slide level, making it suitable for pathology report generation and large-scale data analysis scenarios.

[0096] Preferably, the method further includes:

[0097] Dynamic DeepZoom mapping and tile generation steps, such as Figure 5 As shown, this step specifically includes the following sub-steps:

[0098] S701: When a DeepZoom request is received, the server does not directly read the pre-generated tiles, but instead parses the target level and tile index in the request.

[0099] S702, Establish the mapping relationship from the target layer to the original layer based on the pyramid layer information of the original panoramic image;

[0100] S703, Select the original layer closest to the target layer as the source layer and calculate the corresponding clipping region;

[0101] S704 cropped the image from the source layer, scaled and encoded it to generate the target tile.

[0102] If the request is for a description file, the .dzi description content will be generated dynamically.

[0103] The above method enables on-demand browsing without pre-generating all DeepZoom tiles, generating only the required tiles in real time. This avoids the significant additional storage overhead associated with traditional pre-tiled methods, while maintaining full browsing and interactive capabilities. It is particularly suitable for pathology system deployment environments with limited storage resources.

[0104] A second aspect of this invention provides a distributed storage and collaborative management system for pathological panoramic images, such as... Figure 6 As shown, the system includes:

[0105] WsiCtl client / control terminal is preferred as an upload control terminal, download terminal or data retrieval terminal, responsible for initiating operations such as slide creation, slide upload, field of view download, annotation download and result retrieval;

[0106] The WsiServer distributed server cluster includes the following components:

[0107] The Raft service layer is used to forward, replicate, and commit write requests;

[0108] The state machine layer is used to map different business requests to corresponding state machine application operations;

[0109] The RocksDB data layer is used to persist basic slide information, original slices, annotations, statistics, thumbnails, DNA parameters, and model relationships.

[0110] Image wrapper layer for loading panoramic images, capturing field of view, thumbnails, DeepZoom tiles, and cell region images;

[0111] The HTTP access layer is used to provide unified URL access capabilities to the outside world;

[0112] The virtual file system layer is used to reconstruct the underlying fragmented data into a logical full image file that can be transparently read by image processing libraries;

[0113] A caching layer is used to accelerate access to the view map, thumbnails, image information, and DNA parameters.

[0114] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distributed storage and collaborative management method for pathological panoramic images, characterized in that, This includes the consistent slice upload step, which specifically includes the following sub-steps: The client obtains the pathological panoramic image file to be uploaded, generates basic slide information, and sends a slide creation request to the server. The server encodes the creation request into a log with an operation type identifier and replicates it to each replica node via the Raft protocol; After the log is committed, the state machine calls the data layer to write the new slide basic information according to the operation type, and deletes the old historical data under the same barcode; The client segments the original image into segments based on the segment size and sends multiple segments and their corresponding upload requests to the server one by one or in batches. The server encodes the slice upload request into the corresponding log type, copies and submits it via the Raft protocol, and writes it to the slice data table; it synchronously updates the set of uploaded slice sequence numbers and sorts and removes duplicates; when the number of uploaded slices reaches the total number of slices, it updates the slide upload completion status.

2. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The basic information of the slide includes at least several of the following: barcode, format, file summary, field of view rows and columns, file size, slide size, and total number of slides.

3. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The method also includes virtual whole-map reconstruction and random access steps, specifically including the following sub-steps: The server maps the original images that have been stored in the database into logically complete panoramic image files through a virtual file system. When the image reading module initiates a read request based on file offset, the virtual file system calculates the range of fragments that need to be hit based on the target offset and length; The server uses a batch read method to obtain multiple data segments, and concatenates the multiple data segments according to the file offset order to generate a continuous byte stream; The image wrapper performs random access, pyramid-level reading, and region cropping of a continuous byte stream as a regular panoramic image file.

4. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The method also includes a cross-node transparent access step, which specifically includes the following sub-steps: Users initiate pathological image data requests through a unified URL, and the current node parses the request to extract the slide barcode; The current node determines whether the target slide data exists locally. If it does, the current node processes it and returns the result. If it does not exist, obtain a list of similar HTTP service nodes from the service registry, forward the original HTTP request to each candidate node one by one and attach a forwarded flag to prevent circular forwarding, until a node returns a valid response, and return the response to the requester as is.

5. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The method also includes a multi-model result collaborative management step, which specifically includes the following sub-steps: The server generates model keys based on the combination of the detection model and the classification model; The cell minimap results are stored in a structured manner using a combination of "slide barcode + model key + field of view coordinates + category identifier", while updating slide statistics and model relationship information. When a query request is received, the analysis results under the corresponding model combination are accurately located based on the model key; When a manual review or reclassification request is received, the objects are grouped according to the old and new classifications. After locating the target item based on the object ID, the migration and rewriting are performed, and the cache of the corresponding model and field of view is cleared.

6. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The method also includes a rapid object localization step at the cell or DNA level, specifically comprising the following sub-steps: When caching cell results, the server establishes a mapping relationship between object ID and image information, and when caching DNA parameters, it establishes a mapping relationship between object ID and image region, contour point set, and scaling factor. When an image request is received by querying by object ID, the cache mapping is queried first. If a match is found, the image cropping area and contour information are obtained directly, and the corresponding image is generated. If no match is found, the corresponding field of view object data is read from the database, the cache is filled, and the localization is re-executed.

7. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The method also includes a step for batch extraction of simplified DNA results from whole slides, specifically comprising the following sub-steps: Based on the basic information of the slide, obtain the coordinate range of the entire field of view, generate the DNA data query key for each field of view, and use the underlying database batch read interface to obtain the DNA data of the entire field of view at once. Deserialization is performed in parallel using a thread pool. For each object, simplified fields including object ID, category ID, DNA index, area, and field coordinates are extracted. All results are aggregated, sorted according to preset indicators, and the field-level DNA data is written to a cache.

8. The distributed storage and collaborative management method for pathological panoramic images as described in claim 1, characterized in that, The method also includes a dynamic DeepZoom mapping and tile generation step, specifically including the following sub-steps: When the server receives a DeepZoom request, it parses the target level and tile index in the request; Establish a mapping relationship from the target layer to the original layer based on the pyramid layer information of the original panoramic image; Select the original layer that is closest to the target layer as the source layer and calculate the corresponding clipping region; The target image is generated by cropping the image from the source layer, scaling and encoding it.

9. A distributed storage and collaborative management system for pathological panoramic images, characterized in that, include: A client, used to perform client operations in the method of any one of claims 1 to 8; A distributed server cluster, comprising at least a Raft service layer, a state machine layer, a data layer, a virtual file system layer, an image wrapper layer, an HTTP access layer, and a cache layer, for performing server operations in the method of any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images