Digital slice quality control method and device, electronic equipment and storage medium

Abstract

The invention relates to the field of digital pathology, and provides a digital slice quality control method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a highest magnification tile of a digital slice to be subjected to quality control; based on a trained pathological defect segmentation model, carrying out pathological defect identification and segmentation on the highest-magnification tile to obtain a highest-magnification mask graph; performing pixel-level statistics on the highest-magnification mask image to obtain a single-tile quality control result, and summarizing the single-tile quality control result to obtain a global quality control result of the digital slice; and based on the highest multiplying power mask graph, reconstructing a pyramid mask graph containing multiple multiplying power hierarchies by adopting a plurality of parallel processes, and performing rechecking based on the global quality control result and the pyramid mask graph. According to the digital slice quality control method and device, the electronic equipment and the storage medium provided by the invention, the efficiency bottleneck problem of the quality control process is effectively solved through the technical innovation of reconstructing the pyramid mask graph by the parallel process.

Description

Technical Field

[0001] This invention relates to the field of digital pathology technology, and in particular to a digital slide quality control method, apparatus, electronic device, and storage medium. Background Technology

[0002] In the field of digital pathology, the quality of digital slides is a key factor in ensuring the reliability of downstream diagnostic, teaching, and research tasks. With the widespread adoption of whole-slide scanning technology, the scale of digital slide generation has grown exponentially, with tens of thousands of high-resolution tiles produced in a single scan. However, digital slides are susceptible to quality defects such as creases, blemishes, pen marks, air bubbles, and blurred areas during scanning, transmission, and storage. These defects significantly reduce the accuracy of subsequent analyses. Therefore, establishing efficient and accurate slide quality control techniques has become a core issue that urgently needs to be addressed in the field of digital pathology.

[0003] Currently, quality control of digital tile tiling mainly relies on two methods: manual evaluation and AI-assisted inspection. Traditional manual evaluation requires professionals to check each tile's quality indicators one by one. While this method provides intuitive judgment, it is extremely inefficient; evaluating a single batch of 50,000 tiles can take several hours, and it is easily influenced by subjective experience, making it difficult to standardize quality control. Some studies have attempted to introduce AI models to assist quality control, but these also have significant technical limitations—AI models can only be trained and predicted for tiles at specific magnifications (e.g., 20x), and cannot adapt to the multi-magnification characteristics of digital tiles (e.g., 5x, 10x, 20x, 40x). Furthermore, generating quality control results for specific magnifications cannot construct a complete pyramid diagram (i.e., a standard storage format containing multi-magnification levels), making it impossible to verify the consistency of quality control results through multi-magnification scaling during manual review. Summary of the Invention

[0004] This invention provides a digital slicing quality control method, apparatus, electronic device, and storage medium to address the shortcomings of existing technologies, such as low efficiency and high subjectivity of manual evaluation, which makes it difficult to meet the quality control requirements of large-scale digital slicing, and the low reliability of AI-assisted quality control due to its applicability only to specific magnifications. This invention provides a digital slice quality control method, comprising: Obtain the highest magnification tile of the digital slice to be quality controlled; Based on the trained pathological defect segmentation model, the highest magnification tile is used to identify and segment pathological defects to obtain the highest magnification mask image. Pixel-level statistics are performed on the highest magnification mask to obtain the single-tile quality control results. The single-tile quality control results are then summarized to obtain the global quality control results of the digital slice. Based on the highest magnification mask image, multiple parallel processes are used to reconstruct a pyramid mask image containing multiple magnification levels, and the results are reviewed based on the global quality control results and the pyramid mask image.

[0005] According to the digital slicing quality control method provided by the present invention, the step of reconstructing a pyramid mask image containing multiple magnification levels based on the highest magnification mask image using multiple parallel processes includes: Multiple parallel processes repeatedly execute the reconstruction step until tiles of all scaling levels are generated, resulting in the pyramid mask map. The reconstruction step includes: Based on the maximum row and column index of the current high-magnification level tile, determine the total number of tiles in the current high-magnification level tile, and the initial high-magnification level tile is the highest magnification mask image; Based on the total number of tiles and the number of processes, the tile range of the current high-magnification level tiles that each process needs to process is calculated; Each process stitches together the tiles within the tile range and downsamples them to the next low-magnification level to generate the next low-magnification level tile. After all processes generate the next low-magnification level tile, they synchronize the data.

[0006] According to the digital tile quality control method provided by the present invention, the step of calculating the tile range of the current high-magnification level tiles to be processed by each process based on the total number of tiles and the number of processes includes: The total number of tiles is evenly distributed among the processes to obtain the average tile allocation and the number of remaining tiles for each process; Based on the process number and average tile allocation, and the minimum value between the process number and the remaining number of tiles, the starting tile index to be processed by each process is determined. Based on the starting tile index, the average tile allocation, and the size of the number and the remaining tile quantity, the termination tile index to be processed by each process is determined, and the termination tile index is protected against out-of-bounds errors.

[0007] According to the digital tile quality control method provided by the present invention, each process downsamples the tiles within the tile range to the next low-magnification level after splicing the tiles, generating the next low-magnification level tiles, including: Each process determines the coordinates of a preset number of sub-tiles for the current high magnification level based on the index of each tile within the tile range; Based on the coordinates, the preset number of sub-tiles are sequentially spliced ​​together, and then downsampled to the next low-magnification level to generate the next low-magnification level tile.

[0008] According to the digital slicing quality control method provided by the present invention, obtaining the highest magnification tile of the digital slice to be quality controlled includes: A quality control task is generated based on a quality control request carrying a digital slice to be quality controlled, and the quality control task is written into a message queue. Based on the queue management strategy, the current quality control task is extracted from the message queue, and the digital slice corresponding to the current quality control task is uploaded to the quality control module so that the quality control module can obtain the highest magnification tile of the digital slice to be quality controlled.

[0009] According to the digital tile quality control method provided by the present invention, the step of performing pixel-level statistics on the highest magnification mask image to obtain single-tile quality control results includes: Based on the number of pixels of various defects in the highest magnification mask image and the total number of pixels of the highest magnification tile, the regional proportion of various defects in the single tile quality control result is determined.

[0010] The digital slice quality control method provided by the present invention further includes: Based on the number of tile reads and writes, it is determined that a solid-state drive (SSD) will be used for multiple parallel process reconstructions, and the pyramid mask image will be stored on the SSD. The statistical information of the pyramid mask image and the global quality control results will be written into the database. Every preset period, the currently stored pyramid mask image, statistical information, and global quality control results are synchronized to the backup module, which includes a mechanical hard disk.

[0011] The present invention also provides a digital slice quality control device, comprising: The quality control module is used to acquire the highest magnification tile of the digital slice to be quality controlled; based on the trained pathological defect segmentation model, the highest magnification tile is used to identify and segment pathological defects to obtain a highest magnification mask image; pixel-level statistics are performed on the highest magnification mask image to obtain the single-tile quality control result; the single-tile quality control results are summarized to obtain the global quality control result of the digital slice. The high-performance computing storage server module is used to reconstruct a pyramid mask map containing multiple multi-levels using multiple parallel processes based on the highest multiplier mask map, and to perform verification based on the global quality control results and the pyramid mask map.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the digital slice quality control method as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the digital slice quality control method as described above.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the digital slice quality control method as described above.

[0015] The digital tile quality control method, apparatus, electronic device, and storage medium provided by this invention fully utilize the detailed features of images at the highest resolution by directly segmenting pathological defects on tiles at the highest magnification, avoiding the inconsistency in judgment caused by differences in features at different magnifications in existing technologies. Single-tile quality control results based on pixel-level statistics ensure the precision of detection, while the aggregation of global quality control results guarantees the comprehensiveness of the overall evaluation. This invention, through the technological innovation of reconstructing pyramid mask images via parallel processes, significantly reduces the pyramid image construction time of more than 30 minutes required in the traditional single-threaded processing mode, effectively solving the efficiency bottleneck problem of the quality control process. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts of the digital slice quality control method provided by the present invention.

[0018] Figure 2 This is the second flowchart of the digital slice quality control method provided by the present invention.

[0019] Figure 3 This is a schematic diagram of comparative experimental data provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the structure of the digital slice quality control device provided by the present invention.

[0021] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Existing digital slicing quality control systems are mainly divided into two categories: "human-driven" and "AI-assisted." Their core technical solutions share common limitations: In the quality control and evaluation process: the manual-led approach relies on professionals to visually inspect the tiles for defects such as creases and blemishes, requiring confirmation of tile quality on a per-tile basis, and lacks automated analysis capabilities; the AI-assisted approach, while introducing image segmentation models (such as U-Net), is only trained for a single fixed magnification (such as 20x), and can only process tile images at that magnification to generate a single-magnification quality control mask, failing to cover the full magnification levels of digital tiles; if multiple magnifications need to be adapted, multiple models need to be trained separately, and there is no collaborative mechanism between the models, making it impossible to guarantee the consistency of multi-magnification prediction results for the same region.

[0024] The pyramid construction process adopts a single-threaded serial processing mode, with the workflow being "reading single-ratio tiles → downsampling to generate low-ratio tiles → writing them to storage one by one". The construction of a pyramid with 50,000 tiles took more than 30 minutes, becoming the core bottleneck of the quality control process. Moreover, there is no standardized storage and traceability mechanism after the construction is completed, making it difficult to correlate the quality control results with the tile information.

[0025] To address the aforementioned issues, this invention proposes a digital slice quality control method. This method first obtains the highest magnification tile of the digital slice to be quality controlled. Then, based on a trained pathological defect segmentation model, pathological defects are identified and segmented on the highest magnification tile to obtain a highest magnification mask image. Pixel-level statistics are performed on the highest magnification mask image to obtain single-tile quality control results. These single-tile quality control results are then summarized to obtain the global quality control results of the digital slice. Based on the highest magnification mask image, multiple parallel processes are used to reconstruct a pyramid mask image containing multiple magnification levels, and the results are reviewed based on the global quality control results and the pyramid mask image.

[0026] The digital tile quality control method provided in this invention directly segments pathological defects in the highest magnification tiles, fully utilizing the image detail features at the highest resolution and avoiding the inconsistency in judgment caused by differences in features at different magnifications in existing technologies. Single-tile quality control results based on pixel-level statistics ensure the precision of detection, while the aggregation of global quality control results guarantees the comprehensiveness of the overall evaluation. This solution, through the innovative technology of parallel process reconstruction of the pyramid mask image, significantly reduces the pyramid image construction time of more than 30 minutes required in the traditional single-threaded processing mode, effectively solving the efficiency bottleneck problem of the quality control process.

[0027] Furthermore, this solution achieves true multi-magnification consistency quality control by reconstructing multi-level pyramid mask maps from the highest magnification mask map downwards, solving the problem of inconsistent prediction results at different magnifications in the same area in existing technologies. This top-down reconstruction method not only ensures logical consistency between magnification levels but also allows for verification of quality control results through multi-magnification scaling during manual review, significantly improving the accuracy and reliability of the review.

[0028] This invention can be applied to scenarios requiring digital slice quality control. The executing entity of this method can be an electronic device such as a terminal device, computer, server, server cluster, or specially designed digital slice quality control equipment, or a digital slice quality control device installed within such an electronic device. This digital slice quality control device can be implemented through software, hardware, or a combination of both. In the description of this invention, "multiple" means two or more, unless otherwise explicitly specified.

[0029] Figure 1 This is one of the flowcharts illustrating the digital slice quality control method provided by this invention. Figure 2 This is the second flowchart of the digital slice quality control method provided by this invention. Please refer to [link / reference]. Figure 1 and Figure 2 The method includes the following steps: Step 110: Obtain the highest magnification tile of the digital slice to be quality controlled.

[0030] Specifically, digital slides awaiting quality control refer to digital pathological images generated from whole-slide scanning that require quality assessment (such as checking for defects like stains, creases, and blurring). These slides can be TMAP (Tissue MicroArray Pathology Scanning Slide) pathological scan slides. Digital slides contain multi-magnification tile indexing information. The highest magnification tile refers to a small image patch (e.g., 256×256 or 512×512 pixels) segmented at the highest resolution level (e.g., 40x) of the digital pathological slide, containing the finest tissue structure information.

[0031] Obtaining the highest magnification tile can be achieved by reading from a storage system, dynamically decompressing, or memory mapping. In some embodiments, obtaining the highest magnification tile of a digital slice to be quality controlled includes: Step 111: Generate a quality control task based on a quality control request carrying a digital slice to be quality controlled, and write the quality control task into a message queue; Step 112: Extract the current quality control task from the message queue based on the queue management strategy, and upload the digital slice corresponding to the current quality control task to the quality control module so that the quality control module can obtain the highest magnification tile of the digital slice to be quality controlled.

[0032] Specifically, considering that multiple batches of slice quality control tasks are prone to conflict, this embodiment uses a task distribution module to design a task distribution and message queue mechanism to achieve orderly scheduling and avoid efficiency loss caused by resource contention.

[0033] The quality control request carries a digital slice to be quality controlled. After receiving the quality control request from the quality control module, the task distribution module automatically generates a unique "task processing ID" and returns it to the quality control module for subsequent task tracking and result tracing.

[0034] Write the quality control tasks (including task ID, TMAP path, and priority) corresponding to the TMAP slice to a message queue (such as RabbitMQ) and use a "first-in, first-out + priority adjustment" strategy to ensure that high-urgency tasks are processed first.

[0035] The system monitors the resource idle status (CPU / GPU utilization, disk I / O) of the AI ​​quality control module and high-performance computing storage server in real time. When the service is idle, it extracts a task from the message queue as the current quality control task, uploads the TMAP slice corresponding to the current quality control task to the AI ​​quality control module, and synchronizes the task ID and processing requirements.

[0036] The hardware configuration requirements for the task distribution module are as follows: ≥16 CPU cores to ensure efficient task scheduling and queue management; ≥1TB high-capacity hard disk for storing TMAP slices and task logs to be processed.

[0037] The AI ​​quality control module receives TMAP slices uploaded by the task distribution module, automatically parses their magnification levels (such as 5x, 10x, 20x, 40x), and extracts the highest magnification tile (such as 40x). Since the highest magnification contains the richest defect features, it can be adapted to low magnification through downsampling to ensure prediction consistency.

[0038] This embodiment uses "task ID tracing + priority queue" to achieve orderly scheduling of multiple batches of quality control tasks, avoid resource contention, and ensure the stability of the system in high-concurrency scenarios.

[0039] Step 120: Based on the trained pathological defect segmentation model, identify and segment pathological defects in the highest magnification tiles to obtain the highest magnification mask image.

[0040] Specifically, once the AI ​​quality control module obtains the highest magnification tile, it can perform high-precision defect prediction. The pathological defect segmentation model trained here is a deep learning model capable of identifying quality defect regions in the tile and outputting pixel-level segmentation results, such as the PPLiteSeg lightweight semantic segmentation model. The PPLiteSeg model employs a "lightweight backbone + pyramid pooling module," which, while ensuring segmentation accuracy, improves inference speed by 3 times compared to the traditional U-Net, meeting the needs of large-scale tile processing. The "highest magnification prediction" strategy avoids the need for multi-model training.

[0041] Load the trained PPLiteSeg pathological defect segmentation model (optimized for 5 types of defects: creases, stains, pen marks, bubbles, and blur), predict each tile at the highest magnification, and generate the corresponding quality control mask map, i.e., the highest magnification mask map. Different colors in the mask map mark different defect areas.

[0042] Step 130: Perform pixel-level statistics on the highest magnification mask to obtain the single-tile quality control results. Summarize the single-tile quality control results to obtain the global quality control results of the digital slice.

[0043] Specifically, pixel-level statistics refer to the quantitative analysis of defective pixels in the mask image, such as the proportion of defective area and the classification of defect types. The quality control results of a single tile can be judged as qualified or unqualified based on preset thresholds. The quality control results of all tiles are integrated to generate an overall quality report for the slice, i.e., global quality control results, such as the pass rate and a heatmap of major defect distribution.

[0044] In some embodiments, pixel-level statistics are performed on the highest magnification mask to obtain single-tile quality control results, including: determining the regional proportion of various defects in the single-tile quality control results based on the number of pixels of various defects in the highest magnification mask and the total number of pixels of the highest magnification tile.

[0045] Specifically, the quality control results for a single tile can include the percentage of areas with various defects. This can be expressed by the formula:

[0046] in, The percentage of a certain type of defect in the region. This represents the number of pixels in the mask image for this type of defect. This represents the total number of pixels in the tile.

[0047] It should be noted that steps 110-130 can be executed by the AI ​​quality control module. The hardware configuration requirements for the AI ​​quality control module are: CPU cores ≥ 16 (to ensure parallel reading and preprocessing of tiles); high read / write performance SSD ≥ 8000MB / s (to reduce tile reading latency and improve prediction efficiency).

[0048] The AI ​​quality control module packages all quality control mask images, single-tile statistical results, and global quality control results at the highest magnification into a "TMAP quality control package," which is then uploaded to the high-performance computing storage server via TMAP interface 1, along with a task ID for association.

[0049] Step 140: Based on the highest magnification mask image, a pyramid mask image containing multiple magnification levels is reconstructed using multiple parallel processes, and the results are reviewed based on the global quality control results and the pyramid mask image.

[0050] Specifically, this step can be performed by a high-performance computing storage server module. Considering that the pyramid construction process in existing technologies typically employs a single-threaded processing mode, building a pyramid from 50,000 tiles takes over 30 minutes. This excessive time consumption leads to low overall efficiency in the quality control process, becoming a core bottleneck. Furthermore, there is no standardized storage and traceability mechanism after construction, making it difficult to correlate quality control results with tile information.

[0051] In this embodiment, multiple processes are used in parallel computation based on MPI (Message Passing Interface) to complete the pyramid reconstruction of the quality control mask map. The pyramid construction process can be accelerated by assigning tiles to different processes, with each process responsible for constructing a portion of the pyramid.

[0052] The high-performance computing storage server module receives the "TMAP quality control package" uploaded by the AI ​​quality control module, automatically decompresses it to obtain the highest magnification mask image and statistical results, and associates the task ID through TMAP interface 2 to ensure data traceability.

[0053] In some possible implementations, step 140 specifically includes: Multiple parallel processes repeatedly execute the reconstruction steps until tiles of all scaling levels are generated, resulting in a pyramid mask map. The reconstruction steps include: Step 141: Based on the maximum row and column index of the current high-magnification level tile, determine the total number of tiles in the current high-magnification level tile. The initial high-magnification level tile is the highest magnification mask image. Step 142: Based on the total number of tiles and the number of processes, calculate the tile range of the current high-magnification level tiles that each process needs to process; Step 143: Each process stitches together the tiles within the tile range and downsamples them to the next low-magnification level to generate the next low-magnification level tile. After all processes generate the next low-magnification level tile, they synchronize the data.

[0054] Specifically, the reconstruction steps are described using an example with 15 parallel processes, where the current high-multiplier level is the highest multiplier level at 40x, and the next low-multiplier level is 20x.

[0055] Calculate the maximum row and column indices (xMax, yMax) of the highest magnification mask image and then calculate the total number of tiles. And broadcast the statistics to all processes (processes 0-14). (The total number of tiles is...) .

[0056] In step 142, the tile range of the current high-magnification level tiles that each process needs to process can be determined based on the total number of tiles and the number of processes. Specifically: the total number of tiles is evenly distributed among the processes to obtain the average tile allocation and the number of remaining tiles for each process; based on the process number, the average tile allocation, and the minimum value between the process number and the number of remaining tiles, the starting tile index that each process needs to process is determined; based on the starting tile index, the average tile allocation, and the size of the process number and the number of remaining tiles, the ending tile index that each process needs to process is determined, and boundary crossing prevention is performed on the ending tile index.

[0057] Expressed as a formula: ; ; ; ; right Implement boundary crossing prevention measures: .

[0058] in, 15 represents the total number of tiles and the number of processes. For the average tile distribution, The number of remaining tiles. For the starting tile index, For the process number, To terminate the tile index.

[0059] In step 143, regarding tile splicing and downsampling, each process determines the coordinates of a preset number of sub-tiles at the current high-magnification level based on the index of each tile within the tile range. Based on these coordinates, the preset number of sub-tiles are sequentially spliced ​​together, and then downsampled to the next low-magnification level to generate the next low-magnification level tile. After all processes generate the next low-magnification level tile, data consistency is ensured through an MPI synchronization mechanism.

[0060] The preset quantity here can be 4, meaning 4 sub-tiles of the current high-magnification level are combined to form a single tile of the next low-magnification level. Each process is based on index. Read the four sub-tiles of the current high magnification level ((2x,2y), (2x+1,2y), (2x,2y+1), (2x+1,2y+1)), stitch them together in the order of "top, bottom, left, and right", and downsample them to the target magnification size (e.g., 40x → 20x) to generate the next low magnification level tile.

[0061] Understandably, after generating a 20x level mask based on a 40x level mask, steps 141-143 can be repeated to generate 10x and 5x level mask images, thus obtaining a pyramid mask image containing multiple level magnifications.

[0062] Based on this, the method also includes: determining the use of solid-state drives for multiple parallel process reconstruction based on the number of tile reads and writes, storing the pyramid mask image to the solid-state drive, and writing the statistical information of the pyramid mask image and the global quality control results into the database; Every preset period, the currently stored pyramid mask image, statistical information, and global quality control results are synchronized to the backup module, which includes a mechanical hard drive.

[0063] Specifically, the computation process of reconstructing a pyramid mask map with multiple multiplier levels using multiple parallel processes can be implemented using solid-state drives.

[0064] Tile read count calculation: When the pyramid is constructed, the first... layer( From 1 to The tile needs to be read. The total number of reads for the four sub-tiles of the layer is: in, For the first Total number of roof tiles For the highest multiplier (the first) Total number of (layers) of tiles.

[0065] Tile write count calculation: 1st The number of times the layer tile was written was the first Layer The total number of writes is:

[0066] Speedup ratio calculation: when When large enough, ,but Since the number of read operations far exceeds the number of write operations, hardware selection should prioritize read performance.

[0067] Under experimental conditions of an Intel(R) Core(TM) i7-14700 CPU, an NVMe PVC10 SK hynix 512GB SSD, and a Seagate BarraCuda 2TB 7200RPM 256MB (ST2000DM008) HDD, a parallel algorithm was used to compare storage media. The experiment showed that the speedup ratio S_15≈43.6, which means that the parallel computing speed using the SSD is about 43.6 times faster than that using the HDD. Figure 3 This is a diagram illustrating the experimental data for comparison.

[0068] Furthermore, considering that quality control results and slice information are not stored in a standardized manner, there is no basis for subsequent tracing of defect causes or review of historical data. This embodiment of the invention also establishes a standardized quality control result storage and periodic backup mechanism to ensure that quality control information is traceable and not lost, while also associating it with slice metadata for easy subsequent traceability.

[0069] In this embodiment, the reconstructed pyramid tiles are stored in a designated SSD directory. At the same time, statistical information such as the number of pyramid levels, the number of tiles at each scaling factor, and the storage path are collected and written to a MySQL database along with the AI ​​quality control statistical results (single tile quality control results and global quality control results). The fields include "task_id, tmap_path, pyramid_levels, defect_ratio (crease / stain / pen mark / bubble / blur), storage_path, and create_time".

[0070] A backup command is triggered at a preset period (e.g., every hour) to synchronize all currently stored quality control tiles and MySQL statistics to the backup module, ensuring that no data is lost.

[0071] The backup module is responsible for the scheduled backup and long-term storage of quality control data. The backup module receives synchronization instructions from the high-performance computing storage server and reads the quality control tiles (pyramid diagram) and MySQL statistics from the server periodically (e.g., hourly). It adopts an "incremental backup" strategy (backing up only newly added / modified data) to reduce the amount of data transmission.

[0072] Synchronized tile data is stored on a high-capacity hard drive and organized in a directory structure of "task_id / multiplier level / tile file" to ensure clear data classification; MySQL statistics are backed up as SQL files and stored in association with the tile data.

[0073] After backup is complete, the system automatically verifies data integrity by calculating the tile MD5 value and comparing it with the source data. If the verification fails, a retry mechanism is triggered, and if the verification fails after three retries, an alert is sent to the administrator terminal.

[0074] The hardware configuration requirements for the backup module are as follows: a large-capacity mechanical hard drive ≥12TB (to meet long-term backup storage needs); and a CPU core count ≥16 (to ensure the efficiency of file comparison and data transfer in incremental backups).

[0075] As can be seen from the descriptions of the above embodiments, the digital slice quality control method provided by the embodiments of the present invention has the following technical effects: 1. Significantly improved quality control efficiency: Through "PPLiteSeg model prediction + MPI parallel reconstruction", the AI ​​prediction + pyramid reconstruction + review preparation of 50,000 tiles only takes 1 minute and 30 seconds, which is 8 times more efficient than the existing single-process solution (12 minutes) and more than 40 times more efficient than the manual solution, meeting the needs of large-scale tile processing.

[0076] 2. Improved prediction consistency and credibility: AI prediction is performed only on the highest magnification slices, and the low magnification mask image is reconstructed by downsampling to avoid prediction conflicts caused by multi-model training. The consistency of defect judgment for different magnifications in the same area reaches 99.2%, and the credibility of quality control results is significantly improved.

[0077] 3. Improved verification capabilities: Generates a complete quality control mask pyramid, supports online zoom preview (from 5x to 40x), and allows manual verification of defect performance at different magnifications, improving verification accuracy to 98.5% and reducing missed screening issues.

[0078] 4. Reduced resource waste: The efficient pre-processing quality control can quickly screen out substandard slices (such as those with a defect rate exceeding the threshold), preventing them from flowing into downstream tasks. This reduces computing power and time resource waste by more than 90%, ensuring the reliability of downstream systems.

[0079] 5. Data security and traceability: Standardized MySQL storage (associated with task ID, defect percentage, and storage path) and a scheduled backup mechanism ensure that quality control data is traceable and not lost, and the historical data query response time is ≤1 second.

[0080] 6. Hardware compatibility optimization: By verifying that "the number of reads is much greater than the number of writes", the hardware selection priority is clarified (SSD read performance > write performance). Experiments verify that this configuration can maximize the efficiency of parallel processing and reduce hardware procurement costs.

[0081] The digital slice quality control device provided by the present invention is described below. The digital slice quality control device described below can be referred to in correspondence with the digital slice quality control method described above.

[0082] Figure 4 This is a schematic diagram of the structure of the digital slice quality control device provided by the present invention, as shown below. Figure 4 As shown, a digital slice quality control device is provided, comprising: The quality control module 410 is used to acquire the highest magnification tile of the digital slice to be quality controlled; based on the trained pathological defect segmentation model, the highest magnification tile is used to identify and segment pathological defects to obtain a highest magnification mask image; pixel-level statistics are performed on the highest magnification mask image to obtain the single-tile quality control result; the single-tile quality control results are summarized to obtain the global quality control result of the digital slice. The high-performance computing storage server module 420 is used to reconstruct a pyramid mask map containing multiple multi-levels based on the highest multiplier mask map using multiple parallel processes, and to perform verification based on the global quality control results and the pyramid mask map.

[0083] Based on the above embodiments, the high-performance computing storage server module is specifically used for: Multiple parallel processes repeatedly execute the reconstruction step until tiles of all scaling levels are generated, resulting in the pyramid mask map. The reconstruction step includes: Based on the maximum row and column index of the current high-magnification level tile, determine the total number of tiles in the current high-magnification level tile, and the initial high-magnification level tile is the highest magnification mask image; Based on the total number of tiles and the number of processes, the tile range of the current high-magnification level tiles that each process needs to process is calculated; Each process stitches together the tiles within the tile range and downsamples them to the next low-magnification level to generate the next low-magnification level tile. After all processes generate the next low-magnification level tile, they synchronize the data.

[0084] Based on the above embodiments, the high-performance computing storage server module is specifically used for: The total number of tiles is evenly distributed among the processes to obtain the average tile allocation and the number of remaining tiles for each process; Based on the process number and average tile allocation, and the minimum value between the process number and the remaining number of tiles, the starting tile index to be processed by each process is determined. Based on the starting tile index, the average tile allocation, and the size of the number and the remaining tile quantity, the termination tile index to be processed by each process is determined, and the termination tile index is protected against out-of-bounds errors.

[0085] Based on the above embodiments, the high-performance computing storage server module is specifically used for: Each process determines the coordinates of a preset number of sub-tiles for the current high magnification level based on the index of each tile within the tile range; Based on the coordinates, the preset number of sub-tiles are sequentially spliced ​​together, and then downsampled to the next low-magnification level to generate the next low-magnification level tile.

[0086] Based on the above embodiments, the device further includes a task distribution module, used for: A quality control task is generated based on a quality control request carrying a digital slice to be quality controlled, and the quality control task is written into a message queue. Based on the queue management strategy, the current quality control task is extracted from the message queue, and the digital slice corresponding to the current quality control task is uploaded to the quality control module so that the quality control module can obtain the highest magnification tile of the digital slice to be quality controlled.

[0087] Based on the above embodiments, the quality control module is specifically used for: Based on the number of pixels of various defects in the highest magnification mask image and the total number of pixels of the highest magnification tile, the regional proportion of various defects in the single tile quality control result is determined.

[0088] Based on the above embodiments, the high-performance computing storage server module is specifically used for: Based on the number of tile reads and writes, it is determined that a solid-state drive (SSD) will be used for multiple parallel process reconstructions, and the pyramid mask image will be stored on the SSD. The statistical information of the pyramid mask image and the global quality control results will be written into the database. Every preset period, the currently stored pyramid mask image, statistical information, and global quality control results are synchronized to the backup module, which includes a mechanical hard disk.

[0089] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a digital slice quality control method. This method includes: acquiring the highest magnification tile of the digital slice to be quality controlled; identifying and segmenting pathological defects in the highest magnification tile based on a trained pathological defect segmentation model to obtain a highest magnification mask image; performing pixel-level statistics on the highest magnification mask image to obtain single-tile quality control results; summarizing the single-tile quality control results to obtain the global quality control result of the digital slice; and reconstructing a pyramid mask image containing multiple magnification levels using multiple parallel processes based on the highest magnification mask image, and performing a review based on the global quality control result and the pyramid mask image.

[0090] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the digital slice quality control method provided by the above methods. The method includes: obtaining the highest magnification tile of the digital slice to be quality controlled; identifying and segmenting pathological defects in the highest magnification tile based on a trained pathological defect segmentation model to obtain a highest magnification mask image; performing pixel-level statistics on the highest magnification mask image to obtain a single-tile quality control result; summarizing the single-tile quality control results to obtain a global quality control result for the digital slice; and reconstructing a pyramid mask image containing multiple magnification levels using multiple parallel processes based on the highest magnification mask image, and performing a review based on the global quality control result and the pyramid mask image.

[0092] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the digital slice quality control method provided by the above methods. The method includes: obtaining the highest magnification tile of the digital slice to be quality controlled; identifying and segmenting pathological defects in the highest magnification tile based on a trained pathological defect segmentation model to obtain a highest magnification mask image; performing pixel-level statistics on the highest magnification mask image to obtain single-tile quality control results; summarizing the single-tile quality control results to obtain a global quality control result for the digital slice; and reconstructing a pyramid mask image containing multiple magnification levels using multiple parallel processes based on the highest magnification mask image, and performing a review based on the global quality control result and the pyramid mask image.

[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for digital slide quality control, the method comprising: obtaining a digital image of a slide; and determining a quality metric for the digital image of the slide. The method comprises the following steps: obtaining a highest magnification tile of a digital slice to be quality controlled; performing pathological defect identification and segmentation on the highest magnification tile based on a trained pathological defect segmentation model to obtain a highest magnification mask image; performing pixel-level statistics on the highest magnification mask image to obtain a single-tile quality control result, and collecting the single-tile quality control results to obtain a global quality control result of the digital slice; based on the highest magnification mask image, reconstructing a pyramid mask image containing multiple magnification levels using multiple parallel processes, and performing review based on the global quality control result and the pyramid mask image.

2. The digital slide quality control method of claim 1, wherein, The step of reconstructing the pyramid mask image containing multiple magnification levels based on the highest magnification mask image comprises the following steps: multiple parallel processes repeatedly perform the reconstruction step until tiles of all magnification levels are generated to obtain the pyramid mask image, wherein the reconstruction step comprises the following steps: based on the maximum value of the row and column indexes of the current high magnification level tile, determining the total number of tiles of the current high magnification level tile, and the initial high magnification level tile being the highest magnification mask image; based on the total number of tiles and the number of processes, calculating the tile range of the current high magnification level tile to be processed by each process; each process performs tile stitching and downsampling to the next low magnification level to generate a next low magnification level tile, and all processes perform data synchronization after generating the next low magnification level tile.

3. The digital slide quality control method of claim 2, wherein, The step of calculating the tile range of the current high magnification level tile to be processed by each process based on the total number of tiles and the number of processes comprises the following steps: the total number of tiles is evenly distributed to each process to obtain an average tile allocation amount and a remaining tile number for each process; based on the number of each process and the average tile allocation amount, and the minimum value between the number and the remaining tile number, determining the starting tile index to be processed by each process; based on the starting tile index, the average tile allocation amount, and the size of the number and the remaining tile number, determining the termination tile index to be processed by each process, and performing anti-bounding box processing on the termination tile index.

4. The digital slide quality control method of claim 3, wherein, The step of each process performing tile stitching and downsampling to the next low magnification level to generate a next low magnification level tile comprises the following steps: each process determines the coordinates of a preset number of sub-tiles of the current high magnification level based on the index of each tile within the tile range; after sequentially stitching the preset number of sub-tiles based on the coordinates, downsample to the next low magnification level to generate a next low magnification level tile.

5. The digital slide quality control method of any one of claims 1 to 4, wherein, The step of obtaining the highest magnification tile of the digital slice to be quality controlled comprises the following steps: generating a quality control task based on a quality control request carrying the digital slice to be quality controlled, and writing the quality control task into a message queue; extracting the current quality control task from the message queue based on a queue management strategy, and uploading the digital slice corresponding to the current quality control task to the quality control module, so that the quality control module obtains the highest magnification tile of the digital slice to be quality controlled.

6. The digital slide quality control method of any one of claims 1 to 4, wherein, The step of performing pixel-level statistics on the highest magnification mask image to obtain a single-tile quality control result comprises the following steps: based on the number of pixels of each type of defect in the highest magnification mask image and the total number of pixels of the highest magnification tile, determining the area proportion of each type of defect in the single-tile quality control result.

7. The digital slide quality control method of any one of claims 1 to 4, wherein, The method further comprises the following steps: Determine to reconstruct using solid state disk for multiple parallel processes based on the number of tile reading and writing, and store the pyramid mask map to the solid state disk, write the statistical information of the pyramid mask map, the global quality control result to the database; Synchronize the currently stored pyramid mask map, statistical information, global quality control result to the backup module every preset period, and the backup module includes a mechanical hard disk.

8. A digital slide quality control device, comprising: Comprise: A quality control module for obtaining the highest magnification tile of the digital slice to be controlled; Based on the trained pathological defect segmentation model, the highest magnification tile is subjected to pathological defect identification and segmentation to obtain the highest magnification mask map; Pixel-level statistics are performed on the highest magnification mask map to obtain the single tile quality control result, and the single tile quality control result is summarized to obtain the global quality control result of the digital slice; A high-performance computing storage server module is used to reconstruct the pyramid mask map containing multiple magnification levels based on the highest magnification mask map using multiple parallel processes, and to review based on the global quality control result and the pyramid mask map.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor executes the computer program to realize the digital slice quality control method of any one of claims 1 to 7. 10.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the digital slice quality control method of any one of claims 1 to 7.

Images