Pathological whole slide image storage method and data supply method for ai processing

Abstract

The scheme provides a pathological whole section image storage method and data supply method for AI processing, comprising: acquiring at least one multi-layer pyramid structure pathological whole section image, and storing each pyramid level image of the pathological whole section image in a storage engine by dividing the image into multiple image blocks; and constructing a first-level positioning index, a second-level prediction index and a third-level semantic index for each image block in the storage engine. The scheme adopts a three-level index linkage data supply logic, filters redundant data layer by layer, extracts only the minimum effective data required by the AI task, and improves the accuracy and efficiency of data supply.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and in particular to a method for storing and supplying whole pathological slide images for AI processing. Background Technology

[0002] With the rapid popularization of digital pathology technology, whole-slide images (WSI), with their ultra-high resolution of billions of pixels, have become a core data carrier in fields such as medical auxiliary diagnosis, cancer detection and grading, and disease prognosis prediction, and are widely used in clinical research and artificial intelligence (AI) pathology image analysis. WSI files are typically in the gigabyte range and generally use a multi-level pyramid structure for storage to adapt to the need for fast browsing at different scaling levels. However, the in-depth application of AI technology in the field of pathology has led to new demands for WSI data, including fine-grained, high-frequency, semantic, and customized data. Specific image patches or regions of interest (ROIs) need to be accurately extracted from WSI as the core data source for AI model training and inference.

[0003] In existing technologies, conventional ROI region prediction techniques are mainly designed for traditional scenarios such as doctors manually reviewing images and browsing images. They are difficult to adapt to the fine-grained data requirements of AI tasks and have significant technical limitations: First, the prediction granularity is limited to the macro-region level, lacking the ability to make fine predictions at the image patch level, and cannot meet the AI's need to extract small-sized image patches; Second, relying solely on image visual features to complete region judgment without combining pathological semantic labels for constraint and filtering, it is easy to include invalid regions such as blank areas and areas without tissue in the prediction range, generating a large amount of redundant data; Third, the prediction results can only provide regional coordinate references and are completely disconnected from the underlying storage system, lacking supporting preloading and preprocessing linkage mechanisms, and cannot directly support AI data supply; Fourth, they lack the ability to perceive AI task types. The effective image patches for different AI tasks such as tumor detection and pathological grading differ significantly for the same WSI, and conventional ROI prediction cannot achieve task-driven differentiated region prediction.

[0004] In summary, under the development trend of AI technology deeply empowering digital pathology, the integration defects of traditional ROI prediction technology and pathological image storage solutions have made the contradiction between the efficient utilization of massive WSI data and the rapid training and inference of AI models increasingly prominent. The industry urgently needs a technology that can deeply perceive the needs of AI tasks, integrate pathological semantics and task trajectories to achieve accurate image block-level prediction, and realize efficient support of pathological whole slice image data for AI processing workflow. Summary of the Invention

[0005] This application provides a method for storing and supplying pathological whole slide images for AI processing. The solution adopts a three-level index-linked data supply logic, filtering redundant data layer by layer and extracting only the minimum effective data required by the AI ​​task, thereby improving the accuracy and efficiency of data supply.

[0006] In a first aspect, embodiments of this application provide a method for storing whole pathological slide images for AI processing, the method comprising:

[0007] Acquire pathological whole-slice images with at least one multi-layer pyramid structure, and divide each pyramid level of the pathological whole-slice image into multiple image blocks and store them in the storage engine; For each image block in the storage engine, a first-level localization index, a second-level prediction index, and a third-level semantic index are constructed. The first-level localization index is based on the location information of the image block in the storage engine, the second-level prediction index is based on the access probability of the image block for each type of AI prediction task, and the third-level semantic index is based on the semantic tags of the pathological tissue content of the image block itself.

[0008] Secondly, embodiments of this application provide a method for supplying pathological whole-slice image data for AI processing, including: The target AI prediction task and corresponding preprocessing process obtained from the client are sent to the server, which is the storage engine built in Implementation Example 1; The server retrieves image blocks with valid semantics based on the three-level semantic index of each image block to form a valid image block set; based on the two-level prediction index of each image block in the valid image block set, it retrieves image blocks with a probability of being accessed greater than a set threshold for the target AI prediction task to form a high-probability image block set; based on the one-level localization index of each image block in the high-probability image block set, it retrieves the original data of the corresponding image block in the storage engine to form a predicted image block set. The server performs preprocessing on each image block in the set of predicted image blocks based on the preprocessing process of the target AI prediction task, obtains the preprocessing result, and returns the preprocessing result to the client.

[0009] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a pathological whole slide image storage method for AI processing or a pathological whole slide image data supply method for AI processing.

[0010] Fourthly, embodiments of this application provide a readable storage medium storing a computer program, which, when executed by a processor, implements a method for storing pathological whole-slice images for AI processing or a method for supplying pathological whole-slice image data for AI processing.

[0011] The main contributions and innovations of this invention are as follows: This application embodiment trains an access probability prediction model based on historical access data, constructs a two-level prediction index with the access probability of image patches for various types of AI prediction tasks as the core, realizes task-driven differentiated region prediction, and can preload and preprocess image patches with high access probability in a targeted manner, providing a basis for probability screening of subsequent data supply and reducing invalid data processing; this application embodiment adopts a three-level index linkage data supply logic, filtering redundant data layer by layer, extracting only the minimum effective data required by the AI ​​task, improving the accuracy and efficiency of data supply; this application embodiment compiles the target AI task and preprocessing process into a structured data requirement descriptor, and transmits it to the server through the data query API, realizing the standardization and structured transmission of task requirements, improving the communication efficiency between the client and the server; this application embodiment preprocesses image patches on the server based on pluggable image processing operators, which can cover a variety of preprocessing operations such as color normalization and geometric transformation, flexibly adapting to the preprocessing requirements of different AI prediction tasks, directly generating tensor data that can be input into neural networks, without secondary processing by the client.

[0012] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a method for storing pathological whole-slice images for AI processing, according to an embodiment of this application. Figure 2 This is a schematic flowchart of a method for supplying pathological whole slide image data for AI processing according to an embodiment of this application; Figure 3 This is a schematic diagram of a structured data requirement descriptor according to an embodiment of this application; Figure 4 This is a schematic diagram of a process for preprocessing image blocks according to an embodiment of this application; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0015] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0016] Example 1 This application provides a method for storing whole-slice images of pathological tissue for AI processing. It employs a three-level index-linked data supply logic, filtering redundant data layer by layer to extract only the minimum effective data required for the AI ​​task, thereby improving the accuracy and efficiency of data supply. Specifically, refer to... Figure 1 The method includes: Acquire pathological whole-slice images with at least one multi-layer pyramid structure, and divide each pyramid level of the pathological whole-slice image into multiple image blocks and store them in the storage engine; For each image block in the storage engine, a first-level localization index, a second-level prediction index, and a third-level semantic index are constructed. The first-level localization index is based on the location information of the image block in the storage engine, the second-level prediction index is based on the access probability of the image block for each type of AI prediction task, and the third-level semantic index is based on the semantic tags of the pathological tissue content of the image block itself.

[0017] In the current embodiment, pathological whole slide images are generally stored in a multi-layer pyramid structure, with each pyramid level corresponding to a pathological whole slide image of different resolution.

[0018] Specifically, this solution divides pathological whole-slice images of different resolutions into multiple image blocks to adapt to the resolution requirements of different AI prediction tasks.

[0019] Furthermore, a target image block size is preset for each pyramid level image, and the corresponding pyramid level image is divided into multiple image blocks based on the preset target image block size.

[0020] For example, for a high-resolution pyramid-level image, if the preset target image block size is 512×512 pixels, then the pyramid image is divided into multiple image blocks of 512×512 pixels; for a low-resolution pyramid-level image, if the preset target block size is 256×256 pixels, then the pyramid image is divided into multiple image blocks of 256×256 pixels.

[0021] In other words, the higher the resolution of the pyramid-level image, the larger the pre-defined target image block size.

[0022] In the current embodiment, a unique block ID is assigned to each image block, and a mapping relationship between the block ID and the corresponding image block location information is constructed as a first-level index. The location information includes the pathological whole slice image to which it belongs, the pyramid level to which it belongs, the spatial coordinates, and the data storage address.

[0023] Specifically, by establishing a mapping relationship between block IDs and corresponding image block location information, nanosecond-level rapid location of single or multiple image blocks corresponding to arbitrary coordinates and pyramid levels in a pathological whole-slice image can be achieved.

[0024] In the current embodiment, historical access data of each image block for historical AI prediction tasks is obtained, and an access probability prediction model is pre-trained based on the historical access data. The access probability prediction model takes the block ID of the image block, the type of AI prediction task, and the semantic label of the pathological tissue content of the image block as input, and outputs the access probability of the image block for each type of AI prediction task as a secondary prediction index.

[0025] Specifically, historical access data is obtained by continuously recording the access of each image block in the whole pathological slide image by different types of AI prediction tasks. The historical access data includes the ID of the whole pathological slide image accessed by different types of AI prediction tasks, the block ID of the image block, the pyramid level to which the image block belongs, the access time, the preprocessing method, and the feedback of the model inference results.

[0026] For example, the types of AI prediction tasks mentioned in this solution include tumor detection, pathological grading, and prognostic analysis.

[0027] Specifically, collaborative filtering networks, lightweight neural networks, or statistical analysis models are used as the basic architecture for the access probability prediction model. During training, the model learns the preferences of different types of AI prediction tasks for different pyramid levels, image patch sizes, pathological tissue content of the image patch itself, access frequency of different image patches, and access patterns from historical access data.

[0028] In other words, by using a well-trained access probability prediction model, the access probability of each image patch for different types of AI prediction tasks can be accurately predicted. This enables the preloading and preprocessing of image patches with high access probabilities in a targeted manner according to the task type in the subsequent AI processing workflow, thereby improving the efficiency of data supply.

[0029] In the current embodiment, the pathological tissue region within each image block is obtained using a tissue region segmentation model, and the pathological tissue content of the pathological tissue region is identified by a tissue recognition model using semantic analysis. Based on the pathological tissue content, semantic tags are constructed as a three-level semantic index for the corresponding image block.

[0030] Specifically, the pathological tissue content of an image patch may include one or more combinations of tumor epithelium, stroma, background, and necrotic areas.

[0031] In the current embodiment, there is no fixed order in which the first-level localization index, the second-level prediction index, and the third-level semantic index are constructed. It is worth mentioning that before constructing the second-level prediction index, the third-level semantic index can be used to remove semantically invalid image blocks such as blanks, disorganized content, and dead regions, thereby reducing invalid predictions from the source.

[0032] Example 2 This application also proposes a method for supplying pathological whole-section image data for AI processing, referencing... Figure 2 ,include: The target AI prediction task and corresponding preprocessing process obtained from the client are sent to the server, which is the storage engine built in Implementation Example 1; The server retrieves image blocks with valid semantics based on the three-level semantic index of each image block to form a valid image block set; based on the two-level prediction index of each image block in the valid image block set, it retrieves image blocks with a probability of being accessed greater than a set threshold for the target AI prediction task to form a high-probability image block set; based on the one-level localization index of each image block in the high-probability image block set, it retrieves the original data of the corresponding image block in the storage engine to form a predicted image block set. The server performs preprocessing on each image block in the set of predicted image blocks based on the preprocessing process of the target AI prediction task, obtains the preprocessing result, and returns the preprocessing result to the client.

[0033] In the current embodiment, the client uses a structured language to compile the target AI prediction task and the corresponding preprocessing flow to obtain a structured data requirement descriptor. The structured data requirement descriptor is then sent to the server via a data query API. A schematic diagram of the structured data requirement descriptor is shown below. Figure 3 As shown.

[0034] Specifically, in the client, the task requirements of the target AI prediction task are compiled into a structured language, such as JSON or Protobuf. The task requirements of the target AI prediction task include the AI ​​prediction task type, the pathological whole slide image ID, the target resolution level, the image size, the preprocessing process, etc.

[0035] For example, the preprocessing process includes color normalization, random rotation, and standardization in AI prediction tasks.

[0036] Specifically, the Data Query API is a declarative interface for communication between the client and the server, used to transmit structured data requirement descriptors.

[0037] Furthermore, since the secondary prediction index in this scheme is based on the access probability of image blocks for each type of AI prediction task, if the structured data demand descriptor does not specify pathological whole slide images or image blocks, it can also automatically call image blocks with access probability greater than a set threshold for the target AI prediction task to form a high-probability image block set, thereby realizing intelligent data supply without specifying a region.

[0038] In the current embodiment, image patches in the high-probability image patch set that do not match the task requirements of the target AI prediction task are deleted based on a three-level semantic index.

[0039] For example, if the target AI prediction task requires tumor grading, then image patches with semantic labels of normal tissue in the high-probability image patch set are deleted.

[0040] In the current embodiment, a flowchart illustrating the preprocessing of image blocks is shown below. Figure 4 As shown, each image patch in the predicted image patch set is preprocessed based on pluggable image processing operators to generate tensor data that can be directly input into the neural network. The image processing operators can be decoders, croppers, color normalizers, geometric transformers, tensor generators, etc., to cover a variety of preprocessing operations.

[0041] In the current embodiment, a set number of original image blocks with access probabilities from high to low in the preprocessing results are cached to facilitate quick access by the client.

[0042] Specifically, a hot data cache pool is built on the server side. A set number of image block raw data with access probabilities from high to low in the preprocessing results are added to the hot data cache pool. For example, the hot data cache pool is memory or SSD, and the set number is the maximum capacity of the hot data cache pool.

[0043] Furthermore, the hot data cache pool adopts a dynamic caching strategy. Specifically, the tensor data in the hot data cache pool is sorted by a combination of access probability value and access frequency. Image blocks with high probability values ​​and high access frequencies have the highest priority and are preferentially retained.

[0044] Specifically, the hot data cache pool implements hierarchical storage based on the access probability value of image blocks—high probability (≥0.8) hot data resides in memory, medium probability (0.5-0.8) warm data resides in SSD, and low probability (<0.5) cold data is stored only in the original block and archived; when the cache is replaced, low probability and low access frequency blocks are removed first, and the space after replacement is used to load image blocks with higher probability.

[0045] Specifically, within the hot data cache pool, dedicated partitions are created for different types of AI prediction tasks to prevent tensor data from crowding each other out and to improve the cache hit rate for the same type of task.

[0046] In the current embodiment, after the client receives the preprocessing result, there is no need to perform secondary data processing; the preprocessing result can be directly applied to the target AI prediction task.

[0047] In the current embodiment, this solution achieves precise image block-level localization through the linkage of three-level indexes. Combined with probabilistic predictive preprocessing and preloading, data preprocessing is offloaded from the computing node to the storage node, transmitting only the minimum effective tensor required by AI. Compared with conventional ROI prediction + traditional storage solutions, the data iteration cycle of AI tasks is shortened by more than 60%, significantly accelerating model training and inference. At the same time, it realizes the refined and semantic application of conventional ROI prediction technology in AI pathological image processing scenarios, bridging the technical gap between prediction results and the fine-grained data requirements of AI.

[0048] In some other embodiments, a cluster of storage servers is deployed in a data center, and the software system of the present invention is installed. The WSI raw files (such as in SVS format) are stored in a distributed object storage or high-performance file system. Indexes and metadata are stored in a cache database (such as Redis) or a relational database.

[0049] Workflow: 1. The WSI generated by the pathology scanner is uploaded to the system. The system automatically parses its pyramid structure and establishes a block spatial index. Simultaneously, it calls the built-in lightweight tissue segmentation model to generate tissue region masks and establish semantic indexes. 2. An AI training task for breast cancer detection is initiated. Its data loader submits a request to this system: json{wsi_id”:“BRCA-2024-001”,“region_type”:“semantic”,“semantic_label”:“invasive_carcinoma”,“level”:0, / / maximum resolution“patch_size”:[512,512],“preprocessing”:[{“op”:“color_normalization”,“target_mean”:[0.485,0.456,0.406],“target_std”:[0.229,0.224,0.225]},{“op”:“random_rotate”,“angle_range”:[-90,90]},{“op”:“to_tensor”}]}.

[0050] 3. The AI ​​service interface module parses the request and randomly selects a coordinate within the region in WSI based on the semantic index "invasive_carcinoma".

[0051] 4. The preprocessing pipeline reads the corresponding original image patches from the storage engine based on the coordinates and levels, and performs the following operations in sequence: a. Color normalization (e.g., normalizing to the target mean and variance using the Macenko method). b. Random rotation by an angle. c. Conversion to PyTorch tensor format (CxHxW, value range [0,1]).

[0052] 5. The processed tensors are written directly to the memory address specified by the training server via the RDMA network.

[0053] 6. The system records this access pattern and caches the processing result (tensor). Subsequent similar requests can retrieve the data directly from the cache, resulting in faster speeds.

[0054] In some other embodiments, the preprocessing pipeline portion, which includes fixed operators such as color normalization and format conversion, is offloaded to the smart network interface card (NIC). After data is read from the storage medium, it is processed directly on the NIC and encapsulated into network packets for transmission, further reducing host CPU overhead and memory copying frequency. For complex, computationally intensive preprocessing operations (such as color normalization algorithms with specific styles), an FPGA acceleration card is designed and inserted into the storage server. The preprocessing pipeline utilizes the FPGA for hardware acceleration, significantly improving single-processing throughput.

[0055] Example 3 This embodiment also provides an electronic device, see reference. Figure 5 It includes a memory 404 and a processor 402, wherein the memory 404 stores a computer program and the processor 402 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0056] Specifically, the processor 402 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0057] Memory 404 may include a mass storage device for data or instructions. For example, and not limitingly, memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 404 may include removable or non-removable (or fixed) media. Where appropriate, memory 404 may be internal or external to a data processing device. In a particular embodiment, memory 404 is non-volatile memory. In a particular embodiment, memory 404 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0058] The memory 404 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 402.

[0059] The processor 402 reads and executes computer program instructions stored in the memory 404 to implement any of the AI-oriented pathological whole slide image storage methods or AI-oriented pathological whole slide image data supply methods in the above embodiments.

[0060] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408, wherein the transmission device 406 is connected to the processor 402, and the input / output device 408 is connected to the processor 402.

[0061] The transmission device 406 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 406 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0062] The input / output device 408 is used to input or output information. In this embodiment, the input information may be a whole pathological slide image, and the output information may be the preprocessing result of an image patch, etc.

[0063] Optionally, in this embodiment, the processor 402 can be configured to perform the following steps via a computer program: Acquire pathological whole-slice images with at least one multi-layer pyramid structure, and divide each pyramid level of the pathological whole-slice image into multiple image blocks and store them in the storage engine; For each image block in the storage engine, a first-level localization index, a second-level prediction index, and a third-level semantic index are constructed. The first-level localization index is based on the location information of the image block in the storage engine, the second-level prediction index is based on the access probability of the image block for each type of AI prediction task, and the third-level semantic index is based on the semantic tags of the pathological tissue content of the image block itself.

[0064] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0065] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0066] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figure 5 Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0067] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for storing whole-slice images of pathological tissue for AI processing, characterized in that, Includes the following steps: Acquire pathological whole-slice images with at least one multi-layer pyramid structure, and divide each pyramid level of the pathological whole-slice image into multiple image blocks and store them in the storage engine; For each image block in the storage engine, a first-level localization index, a second-level prediction index, and a third-level semantic index are constructed. The first-level localization index is based on the location information of the image block in the storage engine, the second-level prediction index is based on the access probability of the image block for each type of AI prediction task, and the third-level semantic index is based on the semantic tags of the pathological tissue content of the image block itself.

2. The method for storing whole pathological slide images for AI processing according to claim 1, characterized in that, A target image block size is preset for each pyramid level image, and the corresponding pyramid level image is divided into multiple image blocks based on the preset target image block size.

3. The method for storing whole pathological slide images for AI processing according to claim 1, characterized in that, Each image block is assigned a unique block ID, and a mapping relationship between the block ID and the corresponding image block location information is constructed as a first-level index. The location information includes the pathological whole slice image to which it belongs, the pyramid level to which it belongs, the spatial coordinates, and the data storage address.

4. The method for storing whole pathological slide images for AI processing according to claim 1, characterized in that, Historical access data for each image block in historical AI prediction tasks is obtained, and an access probability prediction model is pre-trained based on the historical access data. The access probability prediction model takes the block ID of the image block, the type of AI prediction task, and the semantic label of the pathological tissue content of the image block as input, and outputs the access probability of the image block for each type of AI prediction task as a secondary prediction index.

5. A method for storing pathological whole-slice images for AI processing according to claim 1, characterized in that, The pathological tissue region within each image block is obtained using a tissue region segmentation model. Then, the pathological tissue content of the pathological tissue region is identified using a tissue recognition model through semantic analysis. Based on the pathological tissue content, semantic labels are constructed as a three-level semantic index for the corresponding image block.

6. A method for supplying pathological whole-section image data for AI processing, characterized in that, include: The target AI prediction task and corresponding preprocessing process obtained from the client are sent to the server, which is the storage engine built in Implementation Example 1; The server retrieves image blocks with valid semantics based on the three-level semantic index of each image block to form a valid image block set; based on the two-level prediction index of each image block in the valid image block set, it retrieves image blocks with a probability of being accessed greater than a set threshold for the target AI prediction task to form a high-probability image block set; based on the one-level localization index of each image block in the high-probability image block set, it retrieves the original data of the corresponding image block in the storage engine to form a predicted image block set. The server performs preprocessing on each image block in the set of predicted image blocks based on the preprocessing process of the target AI prediction task, and returns the preprocessing results to the client.

7. The method for supplying pathological whole-section image data for AI processing according to claim 6, characterized in that, The client uses a structured language to compile the target AI prediction task and the corresponding preprocessing process to obtain a structured data requirement descriptor, and sends the structured data requirement descriptor to the server via the data query API.

8. A method for supplying pathological whole-section image data for AI processing according to claim 6, characterized in that, Preprocessing results are obtained by preprocessing each image patch in the predicted image patch set based on pluggable image processing operators.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to execute a method for storing pathological whole slide images for AI processing as described in any one of claims 1-5 or a method for supplying pathological whole slide image data for AI processing as described in any one of claims 6-8.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which, when executed by a processor, implements a method for storing pathological whole-slice images for AI processing as described in any one of claims 1-5, or a method for supplying pathological whole-slice image data for AI processing as described in any one of claims 6-8.

Images