Pathological section scanner real-time tissue region perception and adaptive acquisition parameter control method and device and medium

Abstract

The application provides a pathological section scanner real-time tissue area perception and adaptive acquisition parameter control method and device and medium, aiming at the problems of block-by-block parameter adjustment jitter and imaging fault in tissue mutation in full-section scanning, the application extracts tissue density, staining intensity and thickness estimation values in a block-by-block real-time manner during scanning; a continuous preset number of blocks confirmation judgment lag strategy is introduced to identify the tissue type; according to the tissue type, a mapping table is used to determine the exposure time, gain, Z sampling layer number, scanning speed and ISP parameter group of the next block, and the transition is smoothly limited according to a preset step length; meanwhile, a tissue distribution map is maintained, and the unscanned block is predicted based on spatial topology and the acquisition parameters are pre-adjusted before entering the boundary. The application is mainly used for high-resolution digital scanning of pathological sections, and can effectively avoid brightness jitter and focal fault, and improve the scanning efficiency.

Description

Technical Field

[0001] This invention relates to the field of digital scanning technology for pathological slide microscopic images, and in particular to a method, device, and medium for real-time tissue region sensing and adaptive acquisition parameter control of a pathological slide scanner. Background Technology

[0002] Pathological slide scanners perform high-resolution, tile-by-tile scans of tissue samples on glass slides. By stitching together the images from each tile, a complete digital image of the slide is generated. These scanners are widely used in digital pathology diagnosis, remote pathology analysis, and AI-assisted diagnosis. The distribution of tissue on a glass slide is typically uneven: some areas are densely packed with tissue, rich in cells, and have complex structures; other areas are sparsely packed; and some areas are completely blank.

[0003] Currently, there are two main methods for parameter selection in pathology slide scanners: The first method is fixed-parameter scanning. The scanner uses the same exposure time, Z-axis sampling layer number, and scanning speed for all tiles during the full-slide scan. Because the tissue distribution on the slide is uneven, blank areas do not require high exposure and multiple Z-sampling layers, while dense areas require longer exposure times and denser Z-sampling layers to cover the depth of field. The fixed-parameter method cannot accommodate the needs of different areas, resulting in wasted scanning time in blank areas and insufficient image quality in dense areas.

[0004] The second method involves pre-scan parameter configuration. Before scanning begins, the scanner selects a set of acquisition parameters based on the sample type (e.g., routine paraffin sections, cell smears, frozen sections). However, once scanning starts, all tiles use the same set of parameters. This method's parameter decision granularity is at the level of the entire slide or the entire sample, and it cannot handle the differences in optimal parameters between different regions within the same slide.

[0005] The two methods described above share the following common problems: blank areas (no tissue coverage) typically occupy 30% to 60% of the slide area. Using the same exposure and Z-sampling parameters for these areas as for tissue areas results in a significant waste of scanning time. When scanning moves from blank areas to tissue areas, if the parameters are not adjusted in time, the first few frames may be overexposed or underexposed, affecting the stitching quality. In dense tissue areas, due to insufficient exposure time and Z-sampling layers, the image clarity and detail retention are insufficient to meet diagnostic requirements. Summary of the Invention

[0006] This invention provides a method, device, and medium for real-time tissue region perception and adaptive acquisition parameter control in a pathological slide scanner. Existing pathological slide scanning schemes, due to their use of globally fixed acquisition parameters, cannot adapt to the uneven tissue distribution within the slide, resulting in low scanning efficiency in blank areas and insufficient image quality in dense areas. Furthermore, simple tile-by-tile independent parameter adjustment easily leads to frequent parameter fluctuations, brightness and focal tomography at tissue abrupt changes, and bottlenecks in the preprocessing data pipeline.

[0007] The core technology of this invention is to determine the tissue type by extracting tissue density, staining intensity and thickness estimation indicators in real time on a tile-by-tile basis during the whole-slice scanning process. It introduces a determination lag strategy for confirmation of a number of consecutive preset tiles and a differential step size limit to smoothly control the five-dimensional acquisition parameters. Based on the two-dimensional spatial topology weighted prediction of the tissue distribution map, the acquisition parameters are pre-adjusted in advance before crossing the tissue boundary.

[0008] In a first aspect, the present invention provides a method for real-time tissue region sensing and adaptive acquisition parameter control of a pathological slide scanner, the method comprising the following steps: During the full-slice scanning process, tissue feature indicators of the current tile image are extracted in real time, including tissue density, staining intensity and thickness estimate. The organization type of the current tile is determined based on the organization characteristic indicators. A determination lag strategy is introduced, which confirms the change of organization type when a preset number of N tiles are determined to be of the same organization type. N is a positive integer greater than 1. Based on the confirmed tissue type, query the preset parameter mapping table to determine the target acquisition parameters for the next tile. The target acquisition parameters include exposure time, gain, number of Z sampling layers, scan speed, and image signal processing (ISP) parameter group. Based on a preset step size limit, the current acquisition parameters are smoothly transitioned to the target acquisition parameters, and the scanner is controlled to perform the next tile image acquisition according to the adjusted acquisition parameters. Maintain and record the tissue distribution map of each tile tissue type. Based on the spatial distribution of tissue types of scanned tiles in the tissue distribution map, predict the tissue types of unscanned tiles. When a tissue type abrupt boundary is predicted, pre-adjust the acquisition parameters before entering the boundary tile.

[0009] Furthermore, tissue density is calculated by adaptive thresholding to determine the proportion of foreground tissue pixels in the total pixels of the current tile. Brightfield images are segmented using Otsu thresholding, while fluorescence images are segmented using the background mean plus three times the standard deviation thresholding. The staining intensity is determined by calculating the average gray value of the foreground region or the mean value of a specific channel. The thickness estimate is determined based on the full width at half maximum (FWHM) of the focal plane sharpness curve obtained during focusing. When the FWHM is greater than the first threshold, it is determined to be a thick tissue; when the FWHM is less than the second threshold, it is determined to be a thin tissue; and when the FWHM is between the second threshold and the first threshold, it is determined to be of medium thickness. When the current tile skips Z-sampling and does not obtain a focal plane sharpness curve, the thickness extrapolation prediction value of the adjacent scanned tiles in the tissue distribution map or the preset default value is used as the thickness estimate.

[0010] Furthermore, tissue types include blank background, sparse tissue, normal tissue, dense tissue, and darkly stained tissue; If the tissue density is less than 5%, it is considered a blank background. If 5% ≤ tissue density < 30%, it is judged as sparse tissue; If 30% ≤ tissue density ≤ 70%, it is considered normal tissue; If the tissue density is >70%, it is considered dense tissue; When the staining intensity is greater than the staining threshold At that time, tissue density was used to determine whether it was a deeply stained tissue; among which, the staining threshold was... The foreground region average gray level is calibrated based on the 80th percentile of the standard dark stained sample image, and adaptive fine-tuning is performed during the operation by statistically analyzing the mean foreground gray level through a sliding window; the value of N ranges from 2 to 4.

[0011] Furthermore, the acquisition parameters corresponding to each tissue type in the parameter mapping table are as follows: Blank background corresponds to an exposure time of 0.5 times the baseline value, low gain, skip Z sampling layers, scan speed of 2 times the baseline value, and standard ISP parameter set; The sparse tissue corresponds to an exposure time of 1 times the baseline value, standard gain, 2 Z-sampling layers, a scan speed of 1.5 times the baseline value, and standard ISP parameter sets; For normal tissue, the exposure time is 1 times the baseline value, the standard gain is 5 Z-sampling layers, the scan speed is 1 times the baseline value, and the standard ISP parameter set is used. Dense tissue corresponds to an exposure time 1.5 times the baseline value, high gain, 8 Z-sampling layers, a scan speed 0.7 times the baseline value, and a high dynamic range ISP parameter set; For deeply stained tissues, the exposure time is 0.7 times the baseline value, with low gain, 5 Z-sampling layers, 1 times the baseline value, and high dynamic range ISP parameter group. When the current tile simultaneously meets the criteria for both dense tissue and dark stained tissue, parameter conflict arbitration is performed: the exposure time and gain are based on the parameters corresponding to dark stained tissue, the number of Z sampling layers is based on the parameters corresponding to dense tissue, the scanning speed is taken as the lower of the two, and the ISP parameter group adopts the high dynamic range ISP parameter group. During smooth transition adjustments, the exposure time is adjusted according to the formula. Adjustments, among which and These are the exposure times for the nth and (n+1)th tiles, respectively. For the target exposure ratio, This means restricting x to Within the range; Gain according to formula Adjustments, among which For the current gain, For the target gain, level 1 corresponds to 6dB; The high dynamic range ISP parameter set, compared to the standard ISP parameter set, includes: using compressed gamma curves, enabling local tone mapping, using adaptive white balance, and enhancing noise reduction intensity.

[0012] Furthermore, each tile of the tissue distribution map corresponds to one pixel, which is initialized to a blank background or based on the foreground segmentation results of the pre-scanned panoramic image before the scan begins; Predicting the organization type of an unscanned tile involves: for an unscanned tile (i,j), obtaining its 8 spatially scanned neighboring tiles, assigning a weight of 1.0 to orthogonal neighbors and a weight of 0.7 to diagonal neighbors, calculating the weighted votes for each organization type, and using the highest vote as the predicted type; if there are no scanned tiles in the 8 neighbors, then the organization type that appears most frequently among the 3 most recent scanned tiles in the scanning order is selected. When an unscanned tile is predicted to have a different tissue type than the current tile, it is marked as a boundary tile. Pre-adjustment acquisition parameters are performed on the tile one tile before entering the boundary tile.

[0013] Furthermore, when it is determined that the organization type has changed, a resource requirement notification message is sent to the scanner's preprocessing scheduling module through a shared memory message queue so that the preprocessing scheduling module can pre-allocate processing resources. The fields of the resource demand notification message include message type identifier, next region organization type, next tile number, expected resource change amount, and timestamp; Before entering a densely organized area, the expected resource change is positive to increase CPU core binding and GPU memory reservation quota; before entering a blank background area, the expected resource change is negative to release the resource quota for immediate processing tasks; no messages are sent if the organization type has not changed.

[0014] Furthermore, it also includes exception handling: Real-time detection of tissue type switching frequency; when the number of tissue type changes exceeds 3 times within 10 consecutive tiles, lock the current collection parameter to the collection parameter corresponding to the most recent stable tissue type and pause adaptive adjustment. When the tissue density is detected to be zero and the staining intensity is greater than the staining threshold, it is determined to be an abnormal feature extraction, and the acquisition parameters of the previous tile are used. When the organization type predicted by the organization distribution map deviates from the actual organization type by more than 50%, the organization distribution map is reset.

[0015] Secondly, the present invention provides a real-time tissue region sensing and adaptive acquisition parameter control device for a pathological slide scanner, comprising: The tissue feature analysis module is used to extract tissue feature indicators of the current tile image in real time during the whole-slice scanning process. The tissue feature indicators include tissue density, staining intensity and thickness estimate. The organization type determination module is used to determine the organization type of the current tile based on organization characteristic indicators. It introduces a determination lag strategy, which confirms the organization type change when a preset number of N tiles are determined to be the same organization type, where N is a positive integer greater than 1. The parameter decision module is used to query a preset parameter mapping table based on the confirmed tissue type to determine the target acquisition parameters for the next watt. The target acquisition parameters include exposure time, gain, number of Z sampling layers, scanning speed, and image signal processing (ISP) parameter group. The parameter execution module is used to smoothly transition the current acquisition parameters to the target acquisition parameters based on a preset step size limit, and control the scanner to perform the next tile image acquisition according to the adjusted acquisition parameters; The tissue distribution prediction module is used to maintain and record the tissue distribution map of each tile's tissue type. Based on the spatial distribution of tissue types in the scanned tiles in the tissue distribution map, it predicts the tissue types of unscanned tiles. When a tissue type abrupt boundary is predicted, the acquisition parameters are pre-adjusted before entering the boundary tile.

[0016] Thirdly, the present invention provides 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 the above-described method for real-time tissue region sensing and adaptive acquisition parameter control of a pathological slide scanner.

[0017] Fourthly, the present invention provides a readable storage medium storing a computer program, which, when executed by a processor, implements the real-time tissue region sensing and adaptive acquisition parameter control method for pathological slide scanners as described above.

[0018] The main contributions and innovations of this invention are as follows: 1. This invention avoids wasting acquisition time in unorganized areas by automatically applying a combination of parameters to blank / background areas, such as halving the exposure, skipping Z-axis sampling, and doubling the scanning speed. At the same time, it appropriately speeds up sparse areas and slows down dense areas to ensure quality, thereby significantly reducing the overall scanning time while maintaining image quality.

[0019] 2. This invention automatically increases the exposure time and Z-axis sampling layer number for dense areas with a tissue density exceeding 70%, ensuring that deep cell structures and complex tissue morphologies are fully captured, avoiding the problems of underexposure or insufficient depth of focus coverage in dense areas caused by fixed parameters.

[0020] 3. This invention filters out misjudgments caused by single-frame noise by using a hysteresis strategy, and achieves smooth parameter transition by limiting the exposure time to ±20% per watt and the gain to ±1 level per watt, ensuring the consistency of image brightness and focus between adjacent tiles, and effectively eliminating brightness breaks and focus jumps at the stitching boundary.

[0021] 4. This invention automatically reduces the exposure time and gain for deeply stained areas with staining intensity exceeding the threshold, and switches to the high dynamic range (ISP) parameter group to protect the detail information in the bright areas from overexposure and saturation, thereby improving the imaging quality of special stained samples.

[0022] 5. This invention ensures that different tiles of the same tissue type use similar parameter configurations by mapping the tissue type to the five-dimensional acquisition parameters, making the brightness, contrast, and color consistency of the whole image significantly better than the fixed parameter scheme.

[0023] 6. This invention sends a resource requirement notification to the resource scheduling module when the organization type changes, pre-allocating CPU and GPU processing resources in dense areas to avoid pipeline blockage caused by sudden increases in data volume, thus ensuring continuous and stable operation of scanning and image processing.

[0024] Details of one or more embodiments of the present invention are set forth in the following drawings and description, so that other features, objects and advantages of the invention will be more readily understood. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the tissue type-collection parameter mapping matrix according to an embodiment of the present invention; Figure 2 This is a schematic diagram of tissue region boundary prediction and parameter pre-adjustment according to an embodiment of the present invention; Figure 3This is a schematic diagram of the architecture and data path of the tissue-aware adaptive acquisition system according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the main process of tissue-aware adaptive data acquisition according to an embodiment of the present invention. Detailed Implementation

[0026] 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.

[0027] 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.

[0028] like Figure 3 As shown, the present invention provides a real-time tissue region perception and adaptive acquisition parameter control device for a pathological slide scanner, which includes five mutually cooperating proprietary functional modules: a tissue feature analysis module, a tissue type determination module, a parameter decision module, a parameter execution module, and a tissue distribution prediction module.

[0029] The input of the tissue feature analysis module is connected to the scanner's camera acquisition unit, used to extract the tissue feature indicators of the current tile in real time after acquiring images of each tile during the full-scan process. The output of the tissue feature analysis module is connected to the input of the tissue type determination module. The output of the tissue type determination module is connected to the input of the parameter decision module. The output of the parameter decision module is connected to the input of the parameter execution module. The output of the parameter execution module is connected to the camera's exposure and gain control interface, the Z-axis motor drive interface, the stage motion controller, and the ISP processing unit, respectively. The tissue distribution prediction module receives the output of the tissue type determination module, maintains the tissue distribution map, and sends a pre-adjustment signal to the parameter decision module when a tissue type boundary is detected.

[0030] The modules described above can be implemented using embedded processors, FPGAs, ASICs, or other programmable logic devices, or they can run as software modules within the scanner's control computer. Data communication between the modules is achieved through the scanner's internal bus or shared memory.

[0031] like Figure 4 As shown, the main process of real-time organization area perception and adaptive acquisition parameter control of the present invention includes a cyclic execution process of steps S201 to S207. The following describes each technical step in detail according to the logical order of the main process.

[0032] I. Tile Image Acquisition and Real-time Extraction of Tissue Features In step S201, the scanner's camera acquires the current tile image at the current stage coordinate position. In step S202, the tissue feature analysis module extracts multidimensional tissue feature indicators in real time from the acquired current tile image. The extracted feature indicators include tissue density, staining intensity, and thickness estimate.

[0033] 1.1 Tissue density calculation Tissue density is defined as the proportion of foreground tissue pixels in the total pixels of the current tile image after adaptive thresholding, with a value ranging from 0 to 100%. The specific calculation method is as follows: For bright-field imaging scenarios (such as H&E staining and IHC staining), the Otsu global adaptive thresholding algorithm is used to obtain the foreground tissue segmentation mask. Pixels with gray values ​​higher than the threshold are identified as foreground (tissue) pixels, and pixels with gray values ​​lower than the threshold are identified as background pixels. For fluorescence imaging scenarios, the mean gray value of the background region plus three times the standard deviation (background + ...) is used. A threshold value is used as the segmentation threshold, and pixels with fluorescence intensity higher than this threshold are classified as foreground pixels. The tissue density is calculated by dividing the total number of foreground pixels by the total number of pixels in the current tile image. The threshold segmentation algorithm is executed in real time as a lightweight operator after each tile image is acquired, without occupying an additional acquisition frame period.

[0034] 1.2 Calculation of staining intensity Staining intensity is defined as the average gray value of the foreground region, reflecting the staining depth or fluorescence intensity of the tissue area. For bright-field 8-bit images, the value ranges from 0 to 255. Specifically, based on the foreground tissue segmentation mask, the arithmetic mean of the gray values ​​of all foreground pixels within the mask is calculated. For multi-channel images, the mean of a specific channel can also be used as the staining intensity.

[0035] 1.3 Thickness Estimation Methods Thickness estimation is based on the focal plane sharpness curve acquired during focusing. Before acquiring each tile, the scanner performs autofocus; the Z-axis motor moves the objective lens along the Z-direction, acquiring images at multiple Z-positions and calculating image sharpness values ​​(such as evaluation metrics based on Laplacian variance or gradient energy) to form a focal plane sharpness curve. This curve exhibits a single-peak shape, with the peak position corresponding to the optimal focal plane position.

[0036] The full width at half maximum (FWHM) of the sharpness curve is positively correlated with tissue thickness: thicker tissues have a greater depth of focus along the optical axis, resulting in a wider peak width on the sharpness curve; thinner tissues have a narrower sharpness curve. The specific criteria are: FWHM greater than a first threshold indicates thick tissue; FWHM less than a second threshold indicates thin tissue; and FWHM between the second and first thresholds indicates medium thickness.

[0037] In this embodiment, the first threshold is set to 8 μm, and the second threshold is set to 3 μm. The first and second thresholds are obtained through factory calibration of typical thick and thin tissue samples: the focal plane sharpness curves of known thick tissue samples (such as kidney parenchyma, with a slice thickness of about 5 μm) and known thin tissue samples (such as skin epidermis, with a slice thickness of about 2 μm) are collected respectively. The 10% quantile of the FWHM distribution of the thick tissue sample is taken as the first threshold (to ensure that about 90% of the thick tissue FWHM is greater than this threshold, avoiding the omission of thick tissue). The 90% quantile of the FWHM distribution of the thin tissue sample is taken as the second threshold (to ensure that about 90% of the thin tissue FWHM is less than this threshold, avoiding the misclassification of thin tissue as medium thickness).

[0038] 1.4 Thickness estimation processing when Z-sampling is skipped Thickness estimation relies on the focal plane sharpness curve obtained during focusing, while the Z-sampling layer number is a controlled parameter. When a tile is identified as a blank area, Z-axis sampling is skipped to accelerate scanning, making it impossible to directly obtain the measured focal plane sharpness curve of the current tile. To address the cyclic dependency between thickness estimation and the Z-sampling controlled parameter, this embodiment uses the following processing logic: when Z-axis sampling is skipped, the thickness estimate of the current tile is directly obtained by extrapolating the thickness from adjacent scanned tiles in the tissue distribution map; if there are no adjacent reference tiles (e.g., at the starting boundary of a slice scan), a preset default value of "thin" is assigned. Since the identification condition for blank areas depends only on tissue density (less than 5%), the prediction backsliding of thickness estimation will not logically interfere with the determination of blank areas. When the scan re-enters the tissue area from a blank area, the thickness estimate predicted by the distribution map can provide a reference for subsequent Z-sampling layer number decisions.

[0039] II. Organization Type Classification and Determination Rules In step S203, the tissue type determination module determines the current tile as the corresponding tissue type according to the extracted tissue feature indicators and preset determination rules.

[0040] 2.1 Density Level 4 Determination Based on tissue density, the tiles are divided into four levels: tissue density less than 5% is classified as blank / background type, typically in blank areas or areas without tissue coverage on a slide; tissue density greater than or equal to 5% and less than 30% is classified as sparse tissue type, typically in adipose tissue or loose connective tissue; tissue density greater than or equal to 30% and less than or equal to 70% is classified as normal tissue type, typically in conventional solid tissue; and tissue density greater than 70% is classified as dense tissue type, typically in tumor areas or areas with dense cells.

[0041] 2.2 Determination of Dark Staining When the staining intensity is greater than the preset staining threshold When the tissue is classified as deeply stained, it is typically found in areas with densely packed deeply stained nuclei. The determination of a deeply stained tissue type is independent of density determination; at any tissue density, as long as the staining intensity exceeds [a certain threshold], it is considered a deep-stained tissue type. All of these trigger the dark staining determination.

[0042] staining threshold The determination method is divided into two stages: factory calibration and on-site adaptive fine-tuning. During factory calibration, images are acquired from standard deeply stained samples (such as DAB-developed immunohistochemically positive samples), and the 80% quantile of the average gray level of the foreground region is taken as the baseline. In this embodiment, 180 is used for bright-field 8-bit images. During field operation, the system adaptively fine-tunes by statistically analyzing the average foreground hue of the current field of view through a sliding window: when the average foreground grayscale value within the sliding window continuously exceeds... At that time, The staining intensity was dynamically increased by 5% to accommodate differences in staining intensity between different batches of slides.

[0043] 2.3 Determining the Lag Strategy To prevent frequent changes in tissue type between discrete tiles due to local intercellular gaps or image noise, a decision lag strategy is introduced in the decision process: the system sets a continuous preset number of... Each tile serves as a confirmation window, only when consecutive tiles are used. The system only officially confirms the organization type change and triggers parameter adjustments when all tiles are determined to be of the same new organization type. Accidental mutations in a single tile do not trigger parameter resets.

[0044] It is a positive integer greater than 1, ranging from 2 to 4. In this embodiment... Take 3. The selection is based on the physical matching of the width of the transition zone of typical pathological tissue (about 2 to 3 tile spans) and the acquisition frame rate of 30fps - the spatial distance corresponding to the lag of 3 tiles just covers the transition zone, which not only filters out the false switching caused by single tile noise, but also ensures that parameter convergence is completed within the third tile after entering the new region. Values ​​that are too small (e.g.) This can cause frequent parameter fluctuations and inconsistent brightness between adjacent tiles, affecting the splicing quality. Values ​​that are too large (e.g.) This can lead to delayed response, and the failure to adjust the parameters of the first few tiles in time after entering a new organizational region can cause a decline in quality.

[0045] 2.4 Parameter Conflict Arbitration Rules When the same tile simultaneously meets the criteria of "dense tissue" (tissue density greater than 70%) and "deeply stained tissue" (staining intensity greater than 70%), When determining the acquisition parameters for the two tissue types, there is a conflict—dense tissues require increased exposure to ensure image brightness, while deeply stained tissues require decreased exposure to prevent overexposure. The parameter decision module determines the acquisition parameters according to the following multi-dimensional arbitration rules: Exposure time and gain were adjusted based on the parameters for deeply stained tissues. The rationale is that overexposure leads to irrecoverable loss of bright area information, which is a hard constraint, while underexposure can be compensated for by subsequent gain adjustments.

[0046] The number of Z-sampling layers is set to 8, based on the dense tissue parameters. The reason is that dense tissues are usually thicker and require more Z-sampling layers to cover the depth of field, while deep staining does not affect tissue thickness.

[0047] The scanning speed is set to the slower of the two (i.e., 0.7 times the baseline for dense tissues) to ensure stage movement stability and exposure time margin.

[0048] The ISP parameter group is uniformly switched to the high dynamic range parameter group. Both types require this parameter group and there is no conflict.

[0049] III. Parameter Mapping Table and Smooth Transition 3.1 Parameter Mapping Table The parameter decision module queries a preset parameter mapping table based on the confirmed tissue type to determine the five-dimensional acquisition parameters (exposure time, gain, number of Z-sampling layers, scan speed, and ISP parameter group) for the next watt. The acquisition parameters corresponding to each tissue type are shown in Table 1. Table 1. Organization Type - Data Acquisition Parameter Mapping Table

[0050] The parameter adjustment principles are as follows: Halve the exposure for blank / background areas, skip Z-sampling, and double the speed to pass through blank areas as quickly as possible without sacrificing tissue quality; use standard exposure, low Z-sampling (2 layers are enough to cover thin tissue), and 1.5 times the speed for sparse tissue; use standard parameters for normal tissue; increase exposure by 50% for dense tissue, use 8 Z-sampling layers (to cover the depth of focus range of thick tissue), and reduce the speed to 0.7 times to ensure image quality; reduce exposure by 30% for deeply stained tissue (to prevent overexposure), reduce gain (to reduce noise), use standard Z-sampling, and switch the ISP to the high dynamic range parameter group.

[0051] 3.2 Differences in ISP parameter groups Table 2 Definitions of ISP parameter group differences

[0052] 3.3 Smooth Parameter Transition In step S204, the parameter execution module performs a smooth transition adjustment from the current acquisition parameters to the target acquisition parameters according to the preset step size limit formula.

[0053] Exposure time is adjusted using the following formula:

[0054] in, Indicates the first The actual exposure time of each tile Indicates the first The exposure time is set for each tile. This indicates the target scaling factor given by the parameter mapping table. Indicates the numerical value Strictly limited to Within the range. That is, the rate of change of exposure time per step between adjacent tiles is strictly limited to a certain range per tile. Within.

[0055] The gain is adjusted according to the following formula:

[0056] in, For the current gain, For the next watt gain, The target gain for the mapping table is 1 dB, with level 1 corresponding to 6 dB. This means the gain variation between adjacent tiles is limited to per tile. files.

[0057] The number of Z-sampling layers is a discrete parameter (2 layers, 5 layers, 8 layers, or skipped). After confirming the type change, the layer setting can be switched directly without any smoothing limitations. The target switching value for the scanning speed is sent to the stage motion controller, which automatically achieves smooth acceleration and deceleration transitions during physical motion.

[0058] The technical effect of the aforementioned step size limitation is that per watt The exposure step size allows for gradual changes in exposure between adjacent tiles (exposure and signal intensity have an approximately linear relationship; a 20% change in exposure corresponds to approximately a 20% change in signal intensity). Combined with decision lag and stitching fusion, this avoids abrupt changes in visible brightness; each tile The gain step size corresponds to a smooth transition of noise changes.

[0059] IV. Organizational Distribution Map Maintenance and Boundary Prediction In step S206, the current tile image data is output to the image stitching pipeline, and the tissue distribution map is updated simultaneously. In step S207, the tissue distribution prediction module performs two-dimensional spatial topological weighted prediction and boundary pre-adjustment of tissue features in unscanned areas based on the updated tissue distribution map.

[0060] 4.1 Initialization and Maintenance of Organizational Distribution Map The tissue distribution map is a low-resolution grid map, with each tile corresponding to a single pixel. The pixel value records the identified tissue type for the corresponding tile. Before scanning begins, the tissue distribution map is initialized to "blank / background". If the scanner is equipped with a pre-scan panoramic camera, the foreground coarse segmentation mask of the pre-scan panoramic image can be used to initialize the tissue distribution map, improving the prediction accuracy of the first row of tiles. During scanning, whenever the tissue type of a tile is confirmed (after confirmation by a decision hysteresis strategy), the corresponding pixel in the tissue distribution map is updated to that type.

[0061] 4.2 Two-Dimensional Spatial Topology Weighted Prediction Method like Figure 2 As shown, the tile to be scanned The algorithm retrieves the set of 8 neighboring tiles (top, bottom, left, right, and four diagonal directions) that have been scanned in spatial geometry. Orthogonal neighboring tiles (top, bottom, left, right) are assigned a weight of 1.0, and diagonal neighboring tiles are assigned a weight of 0.7. The weighted votes for each organization type are accumulated, and the organization type with the highest weighted vote is taken as the predicted type for the unscanned tile. If there is no scanned data within the 8 neighbors (e.g., scanning the first row along a unidirectional raster), the algorithm backtracks by taking the organization type with the highest frequency among the three most recent scanned tiles along the scanning timeline as the predicted value.

[0062] This two-dimensional spatial topology-weighted prediction method replaces simple unidirectional nearest neighbor extrapolation. Its advantage lies in the fact that for S-shaped or zigzag scanning paths, extrapolation based solely on preceding tiles in the scanning sequence may lead to misjudgments at the turning boundaries of the scanning path (because the scanning direction turns 180 degrees at this point, and the spatial position of the preceding tile may be far from the current target tile). The two-dimensional spatial topology-weighted prediction method utilizes the spatial distribution information of scanned tiles around the target tile, enabling it to correctly handle boundary predictions under various scanning paths.

[0063] 4.3 Boundary Detection and Pre-adjustment Strategy When the type of unscanned tile predicted in the tissue distribution map differs from the current tile type, the system marks the unscanned tile as a mutation boundary tile. One tile before entering this boundary tile, the parameter decision module performs advance parameter look-ahead adjustments according to a step-size constraint formula: if the next tile is predicted to be a blank area, the exposure time and Z-sampling layer number are reduced in advance to avoid overexposure in the first frame after entering the blank area; if the next tile is predicted to be dense tissue, the exposure time and Z-sampling layer number are increased in advance to avoid underexposure in the first frame after entering the dense area. The pre-adjustment amount is constrained by the step-size constraint and is executed in conjunction with the normal parameter adjustment process.

[0064] V. Integration with Pre-processing Production Line Scheduling When the organization type is confirmed to have changed in step S203, the system asynchronously sends a resource requirement notification message to the general preprocessing resource scheduling module of the scanner's host computer via a shared memory message queue, so that the preprocessing scheduling module can pre-allocate processing resources. The preprocessing scheduling module is responsible for allocating CPU, GPU, and memory resources according to task priority, and is not the subject matter of this invention; this invention only sends resource requirement notifications to it.

[0065] The message data structure is shown in Table 3, containing the following five fields: message type identifier (msg_type, data type uint8, fixed at 0x01 indicating resource demand notification), next area organization type (tissue_type, data type uint8, values ​​0=blank, 1=sparse, 2=normal, 3=dense, 4=darkly colored), and next tile number (next_tile_id, data type uint32, representing the row number). Total number of columns The unique index of the column number), the expected resource change (resource_delta, data type int8, value range) to (positive values ​​are the percentage increase in quota), and message timestamp (timestamp_us, data type uint64, microsecond precision).

[0066] Table 3 Data Structure of Resource Request Notification Messages

[0067] The resource scheduling strategy is as follows: When the scanned field of view is about to enter a densely organized area, the system sends a notification with a positive resource_delta value (e.g., +40), driving the host computer scheduler to pre-bind more CPU cores and increase the GPU memory reservation quota to ensure the real-time processing of massive multi-layer Z-image data in dense areas; when the scanned field of view is about to enter a blank area, a notification with a negative resource_delta value is sent, releasing real-time processing resources for background large image stitching and pyramid generation tasks; when the organization type remains stable, the system suppresses message sending to avoid frequent resource adjustments. After receiving the message, the preprocessing scheduling module adjusts resource allocation according to its own scheduling strategy; this invention does not depend on its specific implementation.

[0068] VI. Abnormal Handling The system executes the anomaly detection and self-healing protection mechanism in parallel during the loop from steps S202 to S207. The anomaly types and handling strategies are shown in Table 4. Table 4. Exception Types and Handling Strategies

[0069] Frequent tissue type switching: When the tissue type changes more than 3 times within 10 consecutive tiles, a lock protection is triggered—the current collected parameters are locked to the parameters corresponding to the most recent stable tissue type, adaptive adjustment is paused, and it will automatically resume after scanning the fragment area.

[0070] Feature extraction anomaly: When a tissue density of zero is detected and the staining intensity is greater than the staining threshold (logical contradiction state), it is determined to be a feature extraction anomaly, marked as an abnormal tile, and the scanning continues using the acquisition parameters of the previous tile.

[0071] Parameter switching failure: When the ISP parameter group is switched from standard to high dynamic range (or vice versa), if the ISP processing unit returns a driver error, it will automatically fall back to the standard parameter group, generate a log alarm, and not interrupt the scan.

[0072] Organizational distribution map anomaly: When the cumulative deviation between the organization type predicted by the topology and the actual organization type is greater than 50%, the organizational distribution map is forcibly cleared and reset, and the prediction network is reinitialized with the current tile.

[0073] Z-axis motion exceeds limit: When the number of Z-sampling layers corresponding to dense tissue multiplied by the Z-axis step size (0.5 μm) exceeds the maximum Z-axis travel (50 μm), the number of Z-sampling layers is limited to the maximum allowable Z-axis travel. In this embodiment, the 8 Z-sampling layers corresponding to dense tissue in the parameter mapping table (covering a 4 μm depth of focus range) are much smaller than this limit, and this anomaly is usually not triggered, thus it is set as a system protection measure.

[0074] VII. Examples The technical solution and its effects of the present invention will be further described below with reference to specific embodiments. All embodiments are based on the following unified baseline conditions: The baseline frame rate is 30fps, the baseline exposure time is 33ms (1 / 30s), the baseline gain is 100 (ISO equivalent), the Z-axis step size is 0.5μm, the maximum Z-axis travel is 50μm, the first threshold is 8μm, the second threshold is 3μm, and the staining threshold is... =180 (8-bit grayscale), number of lagging tiles N=3.

[0075] Example 1: Adaptive scanning of H&E staining with large blank areas In this embodiment, a 26mm × 76mm HE-stained paraffin section was scanned using a 2× objective lens, and the effective tissue coverage on the slide was 40 × 2048 pixels.

[0076] The scan begins in the blank area at the edge of the slide. The tissue feature analysis module calculates that the tissue density of the current tile is less than 5%, and the tissue type determination module identifies it as a blank background type. The parameter decision module queries the parameter mapping table to determine the target parameters as follows: exposure time 16.5ms (0.5 times the baseline), low gain (ISO64), skip Z sampling, scan speed 2 times the baseline (due to the shortened exposure time, the acquisition time per tile is halved, and the frame rate remains at 30fps), and standard ISP parameter set. The parameter execution module directly executes these parameters to pass through the blank area as quickly as possible.

[0077] Upon scanning tile 15, the tissue density increased to 35% (entering the range of normal parenchymal tissue). The system triggered a type switch after three consecutive tile confirmations. Exposure time was smoothly increased over four tiles with a step size constraint of no more than +20% per tile: 16.5ms → → → →33.0ms. The number of Z-sampling layers was set to 5, and the scanning speed was smoothly reduced to 1 times the baseline.

[0078] Tumor cell infiltration was detected in tile 45, with a tissue density of 75%, which the system classified as dense tissue. The exposure time was smoothly transitioned to 49.5 ms over three tiles (33.0 → 0). → →49.5ms), Z-sampling was increased to 8 layers, and the scan speed was reduced to 0.7 times the baseline. To ensure that the exposure plus readout time does not exceed the frame period, the acquisition frame rate was adjusted accordingly to approximately 18fps to accommodate long exposure readout.

[0079] Results: Scanning time for blank areas is reduced by 50%, and image quality in dense areas is improved (Z-layer coverage is more complete). The total scan time for the entire film is reduced from 18 minutes in fixed parameter mode to approximately 12 minutes, and the image quality in dense areas is superior to that in the fixed parameter scheme.

[0080] Example 2: Topological prediction and look-ahead pre-adjustment of tissue mutation boundaries In this embodiment, during the scanning process of Embodiment 1, when scanning to tile number 44 (currently normal tissue, exposure time 33ms, Z-sampling 5 layers), the tissue distribution prediction module calculates based on the scanned neighboring data and predicts that tile number 45 is about to enter the dense tissue boundary.

[0081] After acquiring the image of tile 44, the system immediately initiated look-ahead pre-adjustment: before entering tile 45, the target exposure time for the next tile was increased to 39.6ms (33ms × 1.2 = 39.6ms, with the pre-adjustment amount not exceeding the +20% step size limit), and the Z-sampling layer number was switched to 7 layers in advance. When the stage moved to tile 45 and performed acquisition, the measured tissue density was 78%, confirming the entry into dense tissue. The parameter decision module further fine-tuned the exposure time from 39.6ms to 47.5ms (again subject to the +20% step size limit), and the Z-sampling was determined to be 8 layers. Due to the pre-adjustment with a step size of +20%, the first frame image of tile 45 was sufficiently exposed, and the sharpness and depth of focus fully met the standards.

[0082] Results: No quality fluctuations at the boundaries. Without pre-adjustment, the first frame of tile #45 was captured using normal organizational parameters (33ms baseline exposure, 5-layer Z-sampling). Insufficient exposure and Z-sampling may result in underexposure and focal plane breaks, requiring rescanning.

[0083] Example 3: Adaptive control of parameters in deeply stained tumor regions A strongly positive IHC immunohistochemically analyzed slide developed with DAB was scanned using a 40× objective lens. When scanning reached tile number 60, the tissue feature analysis module calculated the average grayscale of the foreground region to be 195, exceeding the staining threshold. =180.

[0084] The tissue type determination module classifies the tissue as deeply stained tissue independently of the current tissue density (65%, within the normal range). After confirming three consecutive tiles, the system issues a parameter adjustment command: exposure time within two tiles is adjusted accordingly. The step size was reduced from 33ms to 23.1ms (33→ →23.1ms), the gain was reduced from ISO100 to ISO64 (approximately a 2 / 3 stop decrease), and the ISP parameter group was simultaneously switched to a high dynamic range parameter group that includes compressed gamma and local tone mapping. The number of Z-sampling layers remained unchanged at 5. The deeply stained lesion area covered a total of 20 consecutive tiles. The entire area image was acquired under low exposure and high dynamic range parameters, with no pixel highlight saturation overflow, and the cell nuclear structure was clearly defined.

[0085] Results: Darkly tinted areas are not overexposed, and details are well preserved. Without exposure reduction, darkly tinted areas will show obvious overexposure and saturation, and details in bright areas will be lost irreversibly.

[0086] Example 4: Frequent Switching Lock-in Protection for Fragmented Organizations When scanning a pathological slide with highly fragmented tissue structure, the scanning field of view alternates frequently between blank areas and sparse tissue areas due to the large number of tiny tissue debris scattered in the blank area of ​​the slide.

[0087] The system anomaly detection module detected that the tissue type change command was triggered 4 times within a consecutive time window of 10 tiles (exceeding the preset threshold of 3 times), determining that a high-frequency jitter anomaly had been triggered. The self-healing mechanism immediately forced the acquisition parameters to be locked to the sparse tissue parameters that were in a stable state the most recently (exposure 33ms, Z-sampling 2 layers, speed 1.5 times the baseline, frame rate maintained at 30fps), and paused type switching and parameter adaptation. After scanning over the debris-dense area at a constant speed, the system resumed the adaptive detection process.

[0088] Effect: Avoids image brightness fluctuations caused by frequent parameter switching; image brightness in fragmented areas is consistent and without significant fluctuations (brightness fluctuations are obvious when not locked).

[0089] Example 5: Smooth end-to-end transition from dense tissues to blank areas In Example 1, when scanning to tile number 78 (dense tissue, exposure time 49.5ms, Z-sampling 8 layers, speed 0.7 times the baseline), the tissue distribution map predicted that tile number 79 would soon leave the tissue area and enter a large blank background.

[0090] After the acquisition of tile number 78 was completed, the system initiated pre-adjustment, lowering the exposure to 39.6ms. (20% step size), Z-sampling was reduced from 8 layers to 5 layers. The measured tissue density of tile 79 was 2%, confirming entry into a blank background. The system continued to drive parameter decay with strict step size limits: the exposure time sequence was 39.6→31.7→25.3→20.3→16.5ms (smoothly reduced to 0.5 times the baseline over 4 tiles). The number of Z-sampling layers was reduced to 2 layers frame by frame with each step and eventually skipped. The stage speed was adjusted according to... By incrementally increasing the frame rate (accelerating through 3 tiles), the camera frame rate was restored to 30fps. The exposure between adjacent tiles showed a linear gradient, and the stitching completely eliminated the brightness discontinuity at the boundary of the field of view.

[0091] Effect: Dense The blank exit process exhibits a smooth parameter transition, with no brightness or focus gaps at the stitching boundaries, verifying the complete closed loop of the smooth parameter transition. If there is no smooth transition and the transition is abrupt (exposure 49.5ms → 16.5ms, exposure reduced to about 1 / 3), the brightness of the boundary tiles shows a significant jump, forming a visible gap after stitching.

[0092] 8. Electronic devices and computer-readable storage media The present invention also provides an electronic device including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to perform the various control steps in the above-described method embodiments. The electronic device may be an embedded controller of a pathology slide scanner, a host computer workstation, or a standalone image processing server.

[0093] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described control method. The computer-readable storage medium may be a read-only memory (ROM), random access memory (RAM), magnetic disk, optical disk, flash memory, or other storage media known in the art. Those skilled in the art will understand that all or part of the processes in the above method can be implemented using hardware related to computer program instructions.

[0094] 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.

[0095] The above embodiments are merely illustrative of several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. A method for real-time tissue region sensing and adaptive acquisition parameter control in a pathological slide scanner, characterized in that, Includes the following steps: During the full-slice scanning process, tissue feature indicators of the current tile image are extracted in real time, including tissue density, staining intensity and thickness estimate; The organizational type of the current tile is determined based on the organizational characteristic indicators. A determination lag strategy is introduced, which confirms the change of organizational type when a preset number of N tiles are determined to be of the same organizational type. N is a positive integer greater than 1. Based on the confirmed tissue type, a preset parameter mapping table is queried to determine the target acquisition parameters for the next tile. The target acquisition parameters include exposure time, gain, number of Z sampling layers, scanning speed, and image signal processing (ISP) parameter group. Based on a preset step size limit, the current acquisition parameters are smoothly transitioned to the target acquisition parameters, and the scanner is controlled to perform the next tile image acquisition according to the adjusted acquisition parameters. Maintain and record the tissue distribution map of each tile tissue type. Based on the spatial distribution of tissue types of scanned tiles in the tissue distribution map, predict the tissue types of unscanned tiles. When a tissue type abrupt boundary is predicted, pre-adjust the acquisition parameters before entering the boundary tile.

2. The method for real-time tissue region perception and adaptive acquisition parameter control of a pathological slide scanner as described in claim 1, characterized in that, The tissue density is calculated by adaptive threshold segmentation to determine the proportion of foreground tissue pixels in the total pixels of the current tile. Bright field images are segmented using Otsu thresholding, and fluorescence images are segmented using the background mean plus three times the standard deviation thresholding. The dyeing intensity is determined by calculating the average gray value of the foreground region or the mean value of a specific channel. The thickness estimate is determined based on the full width at half maximum (FWHM) of the focal plane sharpness curve obtained during focusing. When the FWHM is greater than a first threshold, it is determined to be a thick tissue; when the FWHM is less than a second threshold, it is determined to be a thin tissue; and when the FWHM is between the second threshold and the first threshold, it is determined to be of medium thickness. When the current tile skips Z-sampling and does not obtain a focal plane sharpness curve, the thickness extrapolation prediction value of the adjacent scanned tiles in the tissue distribution map or the preset default value is used as the thickness estimate.

3. The method for real-time tissue region perception and adaptive acquisition parameter control of a pathological slide scanner as described in claim 1, characterized in that, The tissue types include blank background, sparse tissue, normal tissue, dense tissue, and darkly stained tissue; If the tissue density is less than 5%, it is considered a blank background. If 5% ≤ tissue density < 30%, it is judged as sparse tissue; If 30% ≤ tissue density ≤ 70%, it is considered normal tissue; If the tissue density is >70%, it is considered dense tissue; When the staining intensity is greater than the staining threshold At that time, tissue density was used to determine whether it was a deeply stained tissue; among which, the staining threshold was... The foreground region average gray level is calibrated based on the 80% quantile of the standard dark-stained sample image, and adaptive fine-tuning is performed during operation by statistically analyzing the mean foreground gray level through a sliding window; the value of N ranges from 2 to 4.

4. The method for real-time tissue region perception and adaptive acquisition parameter control of a pathological section scanner as described in claim 1, characterized in that, The collection parameters corresponding to each tissue type in the parameter mapping table are as follows: Blank background corresponds to an exposure time of 0.5 times the baseline value, low gain, skip Z sampling layers, scan speed of 2 times the baseline value, and standard ISP parameter set; The sparse tissue corresponds to an exposure time of 1 times the baseline value, standard gain, 2 Z-sampling layers, a scan speed of 1.5 times the baseline value, and standard ISP parameter sets; For normal tissue, the exposure time is 1 times the baseline value, the standard gain is 5 Z-sampling layers, the scan speed is 1 times the baseline value, and the standard ISP parameter set is used. Dense tissue corresponds to an exposure time 1.5 times the baseline value, high gain, 8 Z-sampling layers, a scan speed 0.7 times the baseline value, and a high dynamic range ISP parameter set; For deeply stained tissues, the exposure time is 0.7 times the baseline value, with low gain, 5 Z-sampling layers, 1 times the baseline value, and high dynamic range ISP parameter group. When the current tile simultaneously meets the criteria for both dense tissue and dark stained tissue, parameter conflict arbitration is performed: the exposure time and gain are based on the parameters corresponding to dark stained tissue, the number of Z sampling layers is based on the parameters corresponding to dense tissue, the scanning speed is taken as the lower of the two, and the ISP parameter group adopts the high dynamic range ISP parameter group. In the smooth transition adjustment, the exposure time is calculated according to the formula. Adjustments, among which and These are the exposure times for the nth and (n+1)th tiles, respectively. For the target exposure ratio, This means restricting x to Within the range; Gain according to formula Adjustments, among which For the current gain, For the target gain, level 1 corresponds to 6dB; The high dynamic range ISP parameter set, compared to the standard ISP parameter set, includes: using compressed gamma curves, enabling local tone mapping, using adaptive white balance, and enhancing noise reduction intensity.

5. The method for real-time tissue region perception and adaptive acquisition parameter control of a pathological section scanner as described in claim 1, characterized in that, Each tile of the tissue distribution map corresponds to one pixel, and it is initialized to a blank background or based on the foreground segmentation results of the pre-scanned panoramic image before the scan begins. Predicting the organization type of an unscanned tile involves: for an unscanned tile (i,j), obtaining its 8 spatially scanned neighboring tiles, assigning a weight of 1.0 to orthogonal neighbors and a weight of 0.7 to diagonal neighbors, calculating the weighted votes for each organization type, and taking the one with the highest votes as the predicted type; If there are no scanned tiles in the 8-neighborhood, then the tissue type that appears most frequently in the 3 most recent scanned tiles in the scanning order is selected; When an unscanned tile is predicted to have a different tissue type than the current tile, it is marked as a boundary tile. The pre-adjustment acquisition parameters are then executed one tile before entering the boundary tile.

6. The method for real-time tissue region perception and adaptive acquisition parameter control of a pathological slide scanner as described in claim 1, characterized in that, When a change in organization type is detected, a resource requirement notification message is sent to the scanner's preprocessing scheduling module via a shared memory message queue, so that the preprocessing scheduling module can pre-allocate processing resources. The fields of the resource demand notification message include message type identifier, next region organization type, next tile number, expected resource change amount, and timestamp; Before entering a densely organized area, the expected resource change is positive to increase CPU core binding and GPU memory reservation quota; before entering a blank background area, the expected resource change is negative to release the resource quota for real-time processing tasks. No message is sent if the organization type has not changed.

7. The method for real-time tissue region perception and adaptive acquisition parameter control of a pathological section scanner as described in claim 1, characterized in that, It also includes exception handling: Real-time detection of tissue type switching frequency; when the number of tissue type changes exceeds 3 times within 10 consecutive tiles, lock the current collection parameter to the collection parameter corresponding to the most recent stable tissue type and pause adaptive adjustment. When the tissue density is detected to be zero and the staining intensity is greater than the staining threshold, it is determined to be an abnormal feature extraction, and the acquisition parameters of the previous tile are used. When the organization type predicted by the organization distribution map deviates from the actual organization type by more than 50%, the organization distribution map is reset.

8. A device for real-time tissue region sensing and adaptive acquisition parameter control for a pathological slide scanner, characterized in that, include: The tissue feature analysis module is used to extract tissue feature indicators of the current tile image in real time, tile by tile, during the whole-slice scanning process. The tissue feature indicators include tissue density, staining intensity and thickness estimate. The organization type determination module is used to determine the organization type of the current tile based on the organization characteristic indicators. A determination lag strategy is introduced, and when a preset number of N tiles are determined to be of the same organization type, the organization type change is confirmed, where N is a positive integer greater than 1. The parameter decision module is used to query a preset parameter mapping table based on the confirmed tissue type to determine the target acquisition parameters for the next tile. The target acquisition parameters include exposure time, gain, number of Z sampling layers, scanning speed, and image signal processing (ISP) parameter group. The parameter execution module is used to smoothly transition the current acquisition parameters to the target acquisition parameters based on a preset step size limit, and control the scanner to perform the next tile image acquisition according to the adjusted acquisition parameters; The tissue distribution prediction module is used to maintain and record the tissue distribution map of each tile's tissue type. Based on the spatial distribution of tissue types of scanned tiles in the tissue distribution map, it predicts the tissue types of unscanned tiles. When a tissue type abrupt boundary is predicted, the acquisition parameters are pre-adjusted before entering the boundary tile.

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 the real-time tissue region sensing and adaptive acquisition parameter control method for pathological slide scanners according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the real-time tissue region perception and adaptive acquisition parameter control method for pathological slide scanners as described in any one of claims 1 to 7.

Images