Fuzzy detection method and device of histopathological section, storage medium and computer equipment

Abstract

The fuzzy detection method, apparatus, storage medium, and computer equipment for histopathological slides provided in this application generate segmentation prompts based on the position and category information of closed regions in the whole-slice image, driving a basic segmentation model to accurately segment the foreground mask of the tissue region, effectively eliminating interference from non-tissue regions. Based on this, and combining candidate fuzzy regions obtained from the fuzzy detection processing of the whole-slice image, the calculation between the foreground segmentation mask and the candidate fuzzy regions accurately detects fuzzy regions in the tissue portion of the whole-slice image, significantly improving the specificity and accuracy of fuzzy detection, reducing false positives and false negatives, and providing strong technical support for ensuring the quality of pathological diagnosis and improving diagnostic efficiency.

Description

Technical Field

[0001] This application relates to the field of digital pathology technology, and in particular to a method, apparatus, storage medium, and computer equipment for fuzzy detection of tissue pathology slides. Background Technology

[0002] In the field of digital pathology, the quality of tissue pathology slides is a crucial foundation for the accuracy of pathological diagnosis. Whole-slide images (WSI), as the core carrier of digital pathological diagnosis, directly affect the pathologist's observation and judgment of key pathological features such as cell morphology and tissue structure. However, during the preparation and scanning of pathology slides, various factors, such as knife marks, wrinkles, and air bubbles during slide preparation, focal length deviations and lens contamination of scanning equipment, and unevenness of the sample itself, can all lead to localized or overall blurred areas in the generated WSI. These blurred areas may obscure important pathological information, not only increasing the diagnostic difficulty and reading time for pathologists, but also potentially leading to misdiagnosis or missed diagnosis, posing potential risks to patient treatment. Therefore, accurate and efficient automatic detection of blurred areas in tissue pathology slide images is of significant practical importance for ensuring the quality of pathological diagnosis and improving diagnostic efficiency. Currently, traditional fuzz detection methods are mostly based on global features of images or simple local gradient information, which makes it difficult to accurately distinguish between real fuzz within tissue areas and interference from non-tissue areas (such as background, slide edges, etc.). This results in low detection specificity, making it easy to produce false positives or false negatives, and failing to meet the stringent requirements of accuracy and reliability in clinical pathological diagnosis. Summary of the Invention

[0003] The purpose of this application is to at least solve one of the above-mentioned technical defects, in particular the technical defect that the existing fuzzy detection method is difficult to accurately distinguish between real fuzziness in the tissue area and interference in the non-tissue area, resulting in low detection specificity and easy to produce false detection or missed detection.

[0004] This application provides a method for detecting fuzziness in histopathological sections, the method comprising:

[0005] Obtain whole-section images of pathological tissue;

[0006] Based on the location and category information of at least one closed region contained in the pathological whole slice image, segmentation prompt information is generated to drive the preset segmentation base model.

[0007] The pathological whole-section image and the segmentation prompt information are input into the segmentation base model to obtain the tissue region foreground mask output by the segmentation base model;

[0008] The pathological whole-section image is subjected to blur detection processing to obtain candidate blur regions;

[0009] Based on the foreground mask of the tissue region and the candidate blurred region, the blurred region of the tissue part in the whole pathological slice image is detected, and the detection result is obtained.

[0010] Optionally, the step of generating segmentation prompts for driving a preset segmentation model based on the location and category information of at least one closed region contained in the pathological whole-slice image includes:

[0011] The pathological whole slide image is preprocessed to determine at least one closed region contained in the pathological whole slide image;

[0012] Determine the location and category information of each closed region;

[0013] Based on the location and category information of each closed region, segmentation prompts are generated to drive the preset segmentation base model.

[0014] Optionally, the preprocessing of the whole pathological slide image to determine at least one closed region contained in the whole pathological slide image includes:

[0015] The pathological whole-section image was downsampled to obtain a low-resolution whole-section image;

[0016] The low-resolution full-slice image is segmented by saturation thresholding to obtain a binary mask of the initial foreground region;

[0017] Closed region detection is performed on the binary mask to obtain at least one closed region.

[0018] Optionally, the step of performing closed region detection on the binary mask to obtain at least one closed region includes:

[0019] Convert the binary mask into an 8-bit single-channel image;

[0020] The 8-bit single-channel image is closed region detection is performed using the contour retrieval mode RETR_TREE to obtain the contour line of at least one closed region, wherein the contour line preserves the hierarchical relationship between the outer contour and the internal hole.

[0021] Optionally, determining the location and category information of each closed region includes:

[0022] For each closed region:

[0023] The centroid of the closed region is calculated to obtain its coordinates, and these coordinates are used as the position information of the closed region.

[0024] Determine whether the closed region is close to the image edge of the pathological whole-section image;

[0025] If not, the closed region is determined to be an organizational region, and a first label is assigned to indicate that it is an organizational region, and the first label is used as the category information of the closed region.

[0026] If so, the closed region is determined to be a non-organizational region, and a second label is assigned to indicate that it is a non-organizational region, with the second label serving as the category information for the closed region.

[0027] Optionally, the step of calculating the centroid of the closed region to obtain the centroid coordinates of the closed region includes:

[0028] Calculate the geometric moments corresponding to the contour of the closed region to obtain the zeroth moment and the first moment;

[0029] Calculate the centroid coordinates of the closed region based on the zeroth moment and the first moment.

[0030] Optionally, determining whether the closed region is close to the image edge of the whole pathological slide image includes:

[0031] Determine the distance between the centroid coordinates of the closed region and the image boundary of the pathological whole-section image;

[0032] If the distance between the centroid coordinates and the left and right boundaries of the pathological whole slice image is less than a first preset percentage of the image width, or the distance between the centroid coordinates and the upper and lower boundaries of the pathological whole slice image is less than a second preset percentage of the image height, then it is determined that the closed region is close to the image edge of the pathological whole slice image.

[0033] Otherwise, it is determined that the closed region is not close to the image edge of the pathological whole slice image.

[0034] Optionally, the step of performing blur detection processing on the whole pathological slide image to obtain candidate blurred regions includes:

[0035] The pathological whole-section image is converted to grayscale space to obtain a grayscale image;

[0036] The grayscale image is subjected to a Laplacian transform to obtain candidate blurred regions.

[0037] Optionally, the step of detecting blurred regions of the tissue portion in the whole pathological slice image based on the foreground mask of the tissue region and the candidate blurred regions, and obtaining detection results, includes:

[0038] Upsample the foreground mask of the tissue region to obtain a foreground segmentation mask with the same resolution as the whole pathological slice image;

[0039] Calculate the difference image between the foreground segmentation mask and the candidate blurred region;

[0040] Threshold segmentation is performed on the differential image to obtain a blurred region mask of the tissue portion in the pathological whole slice image, and the blurred region mask is used as the detection result.

[0041] Optionally, the method further includes:

[0042] The blurred region mask is subjected to contour extraction and smoothing to obtain a contour curve;

[0043] Output or upload the contour curve and the detection results.

[0044] This application also provides a device for detecting fuzziness in histopathological sections, comprising:

[0045] The image acquisition module is used to acquire images of whole pathological slides;

[0046] The prompt generation module is used to generate segmentation prompt information to drive a preset segmentation base model based on the location and category information of at least one closed region contained in the pathological whole slice image.

[0047] The region segmentation module is used to input the pathological whole slice image and the segmentation prompt information into the segmentation base model to obtain the tissue region foreground mask output by the segmentation base model;

[0048] The first detection module is used to perform blur detection processing on the pathological whole slice image to obtain candidate blurred regions;

[0049] The second detection module is used to detect blurred regions of the tissue portion in the pathological whole-section image based on the foreground mask of the tissue region and the candidate blurred regions, and to obtain the detection results.

[0050] This application also provides a computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the fuzzy detection method for histopathological sections as described in any of the above embodiments.

[0051] This application also provides a computer device, including: one or more processors, and memory;

[0052] The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the fuzzy detection method for histopathological sections as described in any of the above embodiments.

[0053] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0054] The fuzzy detection method, apparatus, storage medium, and computer equipment for histopathological slides provided in this application generate segmentation prompts based on the position and category information of closed regions in the whole-slice image, driving a basic segmentation model to accurately segment the foreground mask of the tissue region, effectively eliminating interference from non-tissue regions. Based on this, and combining candidate fuzzy regions obtained from the fuzzy detection processing of the whole-slice image, the calculation between the foreground segmentation mask and the candidate fuzzy regions accurately detects fuzzy regions in the tissue portion of the whole-slice image, significantly improving the specificity and accuracy of fuzzy detection, reducing false positives and false negatives, and providing strong technical support for ensuring the quality of pathological diagnosis and improving diagnostic efficiency. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A schematic diagram illustrating the process of blur detection in histopathological slide images provided by existing technology;

[0057] Figure 2 A flowchart of a threshold-based foreground blur detection method for histopathological sections provided by existing technology;

[0058] Figure 3 A flowchart illustrating a method for fuzzy detection of histopathological sections provided in this application embodiment;

[0059] Figure 4 Comparison chart of marker pen handwriting interference detection results provided in the embodiments of this application;

[0060] Figure 5 A comparison diagram of low-contrast background and glue overflow interference detection results provided for embodiments of this application;

[0061] Figure 6 Comparison of shadow edge interference detection results provided in the embodiments of this application;

[0062] Figure 7 A flowchart of a method for detecting blurred tissue pathology sections based on a segmentation model is provided for embodiments of this application;

[0063] Figure 8 A schematic diagram of the structure of a fuzzy detection device for histopathological sections provided in an embodiment of this application;

[0064] Figure 9 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0066] Blur detection in histopathological slide images is a crucial step in slide preparation quality control. Accurately filtering blurred areas in the foreground tissue is a prerequisite and foundation for subsequent precise quality control. Existing methods segment the foreground region based on thresholding and then combine the Laplacian transform results to locate blurred regions, as follows: Figure 1 , Figure 2 As shown, Figure 1 This is a schematic diagram illustrating the process of blur detection in histopathological slide images using existing technology. Figure 2 The flowchart of the existing method for detecting blurred foreground in histopathological sections based on threshold segmentation is as follows:

[0067] (1) Blurred areas are out-of-focus areas, which usually lose details and texture details and exhibit smooth features. Therefore, the color space of the whole slice image is first converted to obtain a grayscale image, and then the Laplacian transform is performed to obtain smooth areas as candidate blurred areas.

[0068] (2) Threshold segmentation is performed on the whole slice image in saturation space to obtain the foreground region containing the tissue;

[0069] (3) Perform difference calculation between candidate blurred regions and foreground regions, and perform threshold segmentation to obtain the blurred region detection results of the tissue part.

[0070] As can be seen from the above process, existing threshold-based foreground blur detection methods for histopathological slide images have insufficient accuracy in tissue region identification, leading to inaccurate blur detection. In H&E-stained pathological slides, cell nuclei are typically blue-purple, while the cytoplasm and extracellular matrix are mostly pink to magenta, making the tissue region appear as a brightly saturated area. Therefore, existing methods segment highly saturated regions based on a set threshold in the saturation space and treat them as tissue foreground regions for blur detection. However, in actual slide preparation and scanning, various complex situations often arise, leading to misidentification of many non-tissue areas. These include marker marks left by doctors on the slide surface, adhesive overflow during mounting, black edges caused by uneven lighting during scanning, and low-contrast backgrounds. These areas may exhibit high saturation and are thus misidentified as tissue foreground during the threshold segmentation stage. Simple threshold segmentation methods cannot effectively filter out these complex structures, resulting in poor foreground segmentation performance. Furthermore, these areas are mostly out of focus during scanning, further leading to misidentification as blurred areas and severely interfering with subsequent quality assessment results. Therefore, foreground segmentation strategies that rely solely on saturation thresholds are ill-suited to complex pathological image scenarios. There is an urgent need for a foreground tissue extraction and fuzzy detection method with stronger semantic understanding and structural recognition capabilities to improve detection accuracy and system robustness.

[0071] Furthermore, with the accumulation of massive amounts of data and the maturity of large-scale network training methods, the Segment Anything Model (SAM), a basic segmentation model with significant generalization capabilities, has been widely applied. Further lightweight SAM models exhibit even higher resource utilization and inference efficiency. A fully pre-trained SAM demonstrates significant generalization in downstream tasks, requiring no retraining or fine-tuning. It can achieve accurate pixel-level target segmentation relying solely on simple cue information (such as the point coordinates of the segmentation target, its bounding rectangle, etc.). However, manually labeling pathological slide images with cue information is not only time-consuming and labor-intensive but also difficult to implement in real-world scenarios. Therefore, an automated cue information generation method without human intervention is needed to drive efficient inference in the precise tissue segmentation and blurred region detection processes of SAM. The details are as follows:

[0072] In one embodiment, such as Figure 3 As shown, Figure 3 A flowchart illustrating a method for fuzzy detection of tissue pathology sections provided in this application embodiment; this application provides a method for fuzzy detection of tissue pathology sections, the method including:

[0073] S110: Obtain whole pathological slide images.

[0074] In this step, the whole-slice pathology image is typically a high-resolution digital image obtained by scanning glass pathology slides with a digital pathology scanner. It contains complete morphological information of the tissue slide, with a resolution of hundreds of thousands or even millions of pixels, capable of meticulously displaying the structural features of tissue cells. Therefore, when performing fuzzy detection on the tissue pathology slides, this application can first acquire the whole-slice pathology image. This image can be acquired directly from the image file output by the pathology scanner, or retrieved from a hospital's Picture Archiving and Communication System (PACS) or other medical image storage and transmission systems; no limitation is imposed here. This image serves as the raw data basis for subsequent tissue region segmentation and fuzzy detection, and its quality directly affects the accuracy of subsequent processing.

[0075] S120: Based on the location and category information of at least one closed region contained in the pathological whole slice image, generate segmentation prompt information to drive the preset segmentation base model.

[0076] In this step, after obtaining the pathological whole slice image through S110, this application can generate segmentation prompt information to drive the preset segmentation basic model based on the location information and category information of at least one closed region contained in the pathological whole slice image. In this way, blur detection of the pathological whole slice image can be automatically performed without manual intervention.

[0077] Understandably, the generation of segmentation prompts is a crucial bridge connecting whole-slice pathological images with the underlying segmentation model. This application, by automatically identifying closed regions in the image and determining their location and category, provides precise and targeted guidance to the underlying segmentation model, enabling it to focus more on the actual tissue regions when processing complex pathological images, thereby improving the accuracy and efficiency of subsequent tissue region segmentation. This method of generating prompts based on the image's own features avoids the subjectivity and tediousness of manual annotation, making the entire fuzzy detection process more automated and intelligent.

[0078] S130: Input the whole pathological slice image and segmentation prompts into the segmentation baseline model to obtain the foreground mask of the tissue region output by the segmentation baseline model.

[0079] In this step, after generating segmentation prompt information to drive the preset segmentation base model through S120, this application can input the pathological whole slice image and segmentation prompt information into the segmentation base model, so as to utilize the powerful semantic understanding and pixel-level segmentation capabilities of the segmentation base model to accurately extract the tissue region in the pathological whole slice image.

[0080] Specifically, the segmentation model in this application, upon receiving a whole-section pathological image and segmentation prompts, can perform in-depth analysis and processing of the image based on the location and category features of the tissue regions indicated by the prompts. For example, the segmentation model can focus on potential tissue regions guided by the segmentation prompts, capturing key visual features such as texture, shape, and color of the tissue regions through its internal feature extraction network. Combined with prior knowledge from the prompts, it performs pixel-level classification and outputs a foreground mask of the tissue regions corresponding to the resolution of the whole-section pathological image. In this mask, pixels belonging to the tissue regions are marked as foreground (e.g., represented by 1), while pixels not belonging to the tissue regions are marked as background (e.g., represented by 0), thus clearly outlining the contours and extent of the tissue regions and laying a solid foundation for subsequent blurry region detection.

[0081] S140: Perform blur detection processing on the whole pathological slide image to obtain candidate blur regions.

[0082] In this step, after acquiring the whole pathological slide image and completing the segmentation of the foreground mask of the tissue region, in order to identify possible blurred regions in the image, this application can perform blur detection processing on the whole pathological slide image to obtain candidate blurred regions.

[0083] Understandably, blurred areas in whole-section pathology images typically exhibit characteristics such as loss of image detail, unclear texture information, and blurred edges. These features directly affect pathologists' observation and diagnosis of tissue morphology. Therefore, blur detection processing of whole-section pathology images is a crucial step in locating areas with potential quality problems.

[0084] Specifically, this application can achieve preliminary screening of candidate blurred regions using existing blur detection algorithms or by combining image texture characteristics and sharpness features. For example, in addition to the Laplacian transform method mentioned above, this application can also use gradient-based methods, such as the Sobel operator and the Prewitt operator, to determine the sharpness of a region by calculating the rate of change of pixel grayscale values ​​in the image; or it can use frequency domain analysis methods, where the absence of high-frequency components after converting the image to the frequency domain usually corresponds to blurred regions, and candidate blurred regions can be identified by detecting the distribution of high-frequency energy. Through these processes, all regions that may have blurred problems in the whole pathological slice image can be preliminarily screened out, providing a foundation for subsequent accurate detection by combining tissue region foreground masks.

[0085] S150: Based on the foreground mask of the tissue region and candidate blurred regions, the blurred regions of the tissue part in the whole pathological slice image are detected and the detection results are obtained.

[0086] In this step, the foreground mask of the tissue region output by the segmentation base model is obtained through S130, and the candidate blurred region is obtained through S140. Based on the foreground mask of the tissue region and the candidate blurred region, this application can detect the blurred region of the tissue part in the whole pathological slice image and obtain the detection result.

[0087] Specifically, this application performs a logical AND operation between the tissue region foreground mask and the candidate blurred region, retaining only the portion that simultaneously belongs to both the tissue region foreground mask and the candidate blurred region. This effectively excludes the portion of the candidate blurred region located outside the tissue region, resulting in a blurred region containing only the tissue portion. This blurred detection method based on precise tissue region segmentation significantly reduces false detections of blurred regions caused by misidentifying non-tissue regions as tissue regions, greatly improving the accuracy and reliability of the blurred detection results. The detection results can be output in image form, such as outlining the blurred region of the tissue portion with a specific color or marker on the original whole pathological slide image, or in data form, such as recording the coordinate range and area of ​​the blurred region for subsequent quality assessment and diagnostic reference of the pathological slide.

[0088] In one specific implementation, this application can effectively identify tissue regions under various complex conditions, thereby achieving accurate fuzzy region detection and removing various false alarms. Specific results are as follows: Figures 4 to 6 As shown, Figure 4 , 5 The first row of images in Figure 6 shows the foreground region segmentation mask, and the second row of images shows the final blurred region detection results. Figure 4 The image shows a comparison of marker pen handwriting interference detection results provided in this application embodiment. Marker pen handwriting is typically located on a coverslip, within the out-of-focus range of the scanning lens. Threshold-based segmentation methods cannot distinguish the handwriting from the foreground in saturation space, resulting in large areas of blurred detection and false alarms. This application, however, can accurately segment the tissue region at the center of the slice, thereby removing the false alarms caused by the handwriting. Figure 5 The image provided in this application provides a comparison of detection results for low-contrast background and glue overflow interference, showing the false alarms caused by the blurred areas due to the low-contrast background and glue overflow. Figure 6 This is a comparison image of shadow edge interference detection results provided in the embodiments of this application. Figure 6 This paper demonstrates how uneven lighting causes shadow edges to interfere with blur detection. Methods based on saturation threshold segmentation cannot effectively eliminate the interference in these areas. However, this application utilizes the image semantic understanding and generalization capabilities of SAM to effectively eliminate the interference and complete the task of accurate and efficient blur region detection.

[0089] In the above embodiments, segmentation prompts are generated based on the location and category information of closed regions in the whole pathological slide image. This drives the basic segmentation model to accurately segment the foreground mask of the tissue region, effectively eliminating interference from non-tissue regions. Furthermore, by combining candidate blurred regions obtained from the blur detection processing of the whole pathological slide image, and through the operation of the foreground segmentation mask with the candidate blurred regions, blurred regions of the tissue portion in the whole pathological slide image can be accurately detected. This significantly improves the specificity and accuracy of blur detection, reduces false positives and false negatives, and provides strong technical support for ensuring the quality of pathological diagnosis and improving diagnostic efficiency.

[0090] In one embodiment, S120 generates segmentation prompts for driving a preset segmentation model based on the location and category information of at least one closed region contained in the pathological whole-slice image, which may include:

[0091] S121: Preprocess the pathological whole slide image to determine at least one closed region contained in the pathological whole slide image.

[0092] S122: Determine the location and category information of each closed region.

[0093] S123: Based on the location and category information of each closed region, generate segmentation prompt information to drive the preset segmentation base model.

[0094] In this embodiment, when generating segmentation prompts to drive the preset segmentation model, the pathological whole-slice image first needs to be preprocessed to accurately identify various closed regions in the image. The preprocessing process can include image denoising, contrast enhancement, and edge detection. For example, denoising algorithms such as Gaussian filtering are used to remove random noise from the image, preventing noise from interfering with the identification of closed regions; histogram equalization and other methods are used to enhance image contrast, making the boundaries of different regions clearer and facilitating subsequent edge detection. Edge detection can employ classic algorithms such as the Canny operator, which identifies potential region boundaries by calculating changes in image gradients, thus initially outlining the contours of closed regions. After these preprocessing steps, relatively complete and clear closed region candidates can be extracted from complex pathological whole-slice images.

[0095] After identifying at least one closed region, it is necessary to further clarify the location and category information of each closed region. Location information can be represented by obtaining the coordinates of the circumscribed rectangle, the coordinates of the center point, or the coordinate set of boundary pixels of the closed region. For example, a coordinate system can be established with the top-left corner of the image as the origin, recording the coordinates of the top-left and bottom-right corners of the circumscribed rectangle of each closed region to accurately locate its position in the image. Determining the category information requires combining the characteristics of the whole pathological slide image with prior knowledge to determine whether the closed region belongs to a tissue region or a non-tissue region. Since tissue regions in H&E-stained pathological slides typically have specific color and texture characteristics, category determination can be made by analyzing the color parameters (such as RGB values, HSV values) and texture features (such as gray-level co-occurrence matrix, entropy value, etc.) of the closed region. For example, calculating the average saturation value of pixels within the closed region; if it is significantly higher than the average saturation of the background region, and the texture features conform to the distribution pattern of tissue cells, then the closed region can be preliminarily classified as a tissue region; otherwise, it is classified as a non-tissue region.

[0096] After obtaining the location and category information of each closed region, this application can generate segmentation prompts based on this information. For closed regions identified as tissue regions, corresponding prompts are generated according to their location information, such as using the coordinates of the center point of the closed region as a point prompt, or the coordinates of the circumscribed rectangle as a bounding box prompt. Considering that there may be multiple independent closed regions in a pathological tissue region, corresponding segmentation prompts need to be generated for closed regions of each tissue region category. These segmentation prompts will serve as guidance to drive the preset segmentation base model (such as the lightweight SAM model) to accurately locate and segment the tissue region in the whole pathological slide image. For closed regions that are not tissue region categories, they are excluded when generating segmentation prompts to avoid unnecessary processing of these regions by the segmentation base model, thereby improving segmentation efficiency and accuracy.

[0097] In this way, the generated segmentation hints can accurately guide the segmentation model to focus on the actual tissue regions, laying a good foundation for the subsequent acquisition of the tissue region foreground mask.

[0098] In one embodiment, such as Figure 7 As shown, Figure 7 A flowchart of a method for blur detection of histopathological slides based on a segmentation model provided in this application embodiment; S121 preprocesses the whole pathological slide image to determine at least one closed region contained in the whole pathological slide image, which may include:

[0099] S1211: Downsample the pathological whole slice image to obtain a low-resolution whole slice image.

[0100] S1212: Perform saturation thresholding on the low-resolution full-slice image to obtain a binary mask for the initial foreground region.

[0101] S1213: Perform closed region detection on the binary mask to obtain at least one closed region.

[0102] In this embodiment, when determining at least one closed region contained in a pathological whole-slice image, this application can first perform downsampling processing on the pathological whole-slice image to obtain a low-resolution whole-slice image. This is because pathological whole-slice images typically have extremely high resolution, and directly processing them would result in a huge computational load and memory consumption, affecting processing efficiency. By downsampling, the image resolution can be significantly reduced while preserving the overall image structure and key features, thereby reducing the computational burden of subsequent processing steps and improving the algorithm's running speed. The downsampling ratio can be adjusted according to actual needs and hardware conditions, for example, reducing the resolution of the original image to 1 / 16 or 1 / 32 of its original value, achieving efficient processing while ensuring sufficient detail.

[0103] Next, this application can perform saturation thresholding on the obtained low-resolution full-slice image to obtain a binary mask for the initial foreground region. In H&E-stained pathological sections, tissue areas usually contain a large amount of staining agent, resulting in a significant difference in saturation channels between them and background areas (such as slides, blank areas, etc.). By setting an appropriate saturation threshold, such as between 10 and 30, this application can identify pixels in the image with saturation higher than the threshold as possible tissue areas (foreground), and pixels with saturation lower than the threshold as background areas. Specifically, this application can first convert the low-resolution full-slice image from the RGB color space to the HSV color space and extract the saturation (S) channel. Then, by analyzing the histogram distribution of the saturation channel, a suitable threshold is selected (e.g., using Otsu's automatic thresholding algorithm or setting a fixed threshold based on experience), and the saturation image is converted into a binary image, where foreground pixels are represented by 1 and background pixels by 0, thereby obtaining a binary mask for the initial foreground region.

[0104] Finally, this application can perform closed region detection on the binary mask to obtain at least one closed region. The initial binary mask of the foreground region may contain some discrete noise points or incomplete regions, which need to be processed by morphological operations (such as erosion, dilation, opening, closing, etc.) to eliminate noise and connect adjacent foreground regions to form complete closed regions. For example, first perform a closing operation on the binary mask to fill small holes inside the region, and then perform an opening operation to remove small burrs at the edges. After processing, the foreground regions in the binary mask are scanned and labeled using a connected component analysis algorithm (such as a region labeling algorithm based on 8 or 4 neighborhoods). Each labeled connected component is a closed region. In this way, at least one closed region that may contain tissue is initially identified from the whole pathological slice image, laying the foundation for further determination of its location and category information.

[0105] In one embodiment, S1213 performs closed region detection on the binary mask to obtain at least one closed region, which may include:

[0106] S12131: Convert the binary mask into an 8-bit single-channel image.

[0107] S12132: The 8-bit single-channel image is closed region detection is performed using the contour retrieval mode RETR_TREE to obtain the contour line of at least one closed region, wherein the contour line retains the hierarchical relationship between the outer contour and the internal hole.

[0108] In this embodiment, when performing closed region detection on a binary mask, the binary mask can first be converted into an 8-bit single-channel image. This is because subsequent contour retrieval algorithms typically process single-channel grayscale images. Converting the binary mask (usually with pixel values ​​of 0 or 1) to 8-bit format (pixel value range of 0-255) ensures that the image data format meets the algorithm requirements, facilitating subsequent contour extraction and analysis. During the conversion process, foreground pixels (value of 1) in the original binary mask can be mapped to 255, while background pixels (value of 0) remain unchanged, thus forming a single-channel image with high contrast, which is beneficial for accurate contour recognition.

[0109] Subsequently, this application can use the RETR_TREE contour retrieval mode to detect closed regions in an 8-bit single-channel image to obtain the contour lines of at least one closed region. The RETR_TREE mode is a retrieval method capable of establishing hierarchical relationships between contours. It can not only detect all contours in an image but also record the parent-child relationships between contours, i.e., the hierarchical structure between the outer contour and the contours of the pores it contains. For example, in a pathological whole-section image, the outer contour of a tissue region may contain multiple small contours formed by intercellular spaces or cavities. The RETR_TREE mode can clearly capture this hierarchical relationship, marking the outer contour as the parent contour and the internal pore contours as child contours. Preserving this hierarchical relationship is crucial for accurately determining the integrity and internal structure of closed regions. It helps to eliminate misjudgments caused by internal pores when determining the location and category information of closed regions, ensuring that the extracted closed regions truly reflect the morphological characteristics of the tissue.

[0110] Through the above steps, the contour lines of all closed regions can be accurately extracted from the processed binary mask, providing detailed contour data support for subsequent determination of the location and category information of each closed region.

[0111] In one embodiment, determining the location and category information of each closed region in S122 may include:

[0112] For each closed region:

[0113] S1221: Calculate the centroid of the closed region to obtain the centroid coordinates of the closed region, and use the centroid coordinates as the position information of the closed region.

[0114] S1222: Determine whether the closed region is close to the image edge of the pathological whole slice image; if not, execute S1223; if yes, execute S1224.

[0115] S1223: Determine that the closed region is an organization region, and assign a first label to indicate that it is an organization region, and use the first label as the category information of the closed region.

[0116] S1224: Determine that the closed region is a non-organizational region and assign a second label to indicate that it is a non-organizational region, and use the second label as the category information of the closed region.

[0117] In this embodiment, the centroid coordinates are selected as a key indicator when determining the location information of each closed region. Specifically, as follows: Figure 7As shown, for each closed region, this application accurately locates the region's position in the pathological whole-slice image by calculating its centroid coordinates. The centroid coordinates are calculated by weighted averaging of the coordinates of all pixels within the closed region, which comprehensively reflects the overall position of the region. Compared to the coordinates of the circumscribed rectangle or the center point, the centroid coordinates better represent the geometric center of the region, providing a reliable positional basis for generating accurate segmentation prompts.

[0118] In determining the category information, this application focuses on the relative positional relationship between closed regions and image edges. This is based on common phenomena during pathology slide preparation and scanning: typically, the actual tissue sample is placed in the central area of ​​the slide to avoid edge interference, while closed regions close to the image edges are more likely to be artifacts generated during scanning, reflections from the slide edges, or other non-tissue components. Therefore, as... Figure 7 As shown, for each closed region, this application first determines whether it is close to the edge of the whole pathological slide image. The judgment criterion can be implemented by setting an edge threshold distance. For example, if any part of the closed region is less than the distance to the image edge (e.g., 50 pixels), then the region is considered to be close to the edge. If the judgment result is negative, that is, the closed region is far from the image edge, then it is judged as a tissue region and assigned a first label, such as 1, as its category information; if the judgment result is positive, that is, the closed region is close to the image edge, then it is judged as a non-tissue region and assigned a second label, such as 0.

[0119] This location-based category judgment method can quickly and effectively screen out potential tissue regions, eliminate obvious edge non-tissue interference, and provide clear category guidance for the subsequent generation of segmentation prompts.

[0120] In one embodiment, calculating the centroid of the closed region in step S1221 to obtain the centroid coordinates of the closed region may include:

[0121] S12211: Calculate the geometric moments corresponding to the contour of the closed region to obtain the zeroth moment and the first moment.

[0122] S12212: Calculate the centroid coordinates of the closed region based on the zeroth moment and the first moment.

[0123] In this embodiment, the calculation of the centroid coordinates is based on the geometric moments of the closed region's contour. Geometric moments are important parameters describing the shape characteristics of a region, among which the zeroth and first moments are the basis for calculating the centroid coordinates.

[0124] Specifically, for the contour of a closed region, its zeroth-order moment m00 represents the area-related information of the region, which can be obtained by summing the gray values ​​of all pixels within the contour (in this case, a binary image, with the foreground pixel gray value being 255 and the background gray value being 0), i.e., m00 = ΣΣI(x,y), where I(x,y) is the pixel value of the image at coordinates (x,y). The first-order moments m10 and m01 represent the mass moments of the region in the x and y directions, respectively, calculated as m10 = ΣΣx*I(x,y) and m01 = ΣΣy*I(x,y).

[0125] After obtaining the zeroth moment m00, the first moment m10, and m01, the centroid coordinates (Cx, Cy) of the closed region can be calculated using the formulas Cx=m10 / m00 and Cy=m01 / m00. This calculation method based on geometric moments allows for the precise determination of the centroid position of the closed region, ensuring the accuracy of the positional information and providing reliable data support for the subsequent generation of segmentation prompts based on the centroid.

[0126] In one embodiment, determining whether the closed region is close to the image edge of the whole pathological slide image in step S1222 may include:

[0127] S12221: Determine the distance between the centroid coordinates of the closed region and the image boundary of the pathological whole slice image.

[0128] S12222: If the distance between the centroid coordinates and the left and right boundaries of the pathological whole slice image is less than a first preset percentage of the image width, or the distance between the centroid coordinates and the upper and lower boundaries of the pathological whole slice image is less than a second preset percentage of the image height, then it is determined that the closed region is close to the image edge of the pathological whole slice image.

[0129] S12223: Otherwise, it is determined that the closed region is not close to the image edge of the pathological whole slice image.

[0130] In this embodiment, the distance between the centroid coordinates and the image boundary is the core criterion for determining whether a closed region is close to the image edge. Specifically, this application first calculates the distance between the centroid coordinates (Cx, Cy) of the closed region and the left and right boundaries, as well as the top and bottom boundaries, of the pathological whole-slice image. Assuming the width of the pathological whole-slice image is W and the height is H, then the distance between the centroid Cx and the left boundary is Cx, and the distance to the right boundary is W-Cx; the distance between the centroid Cy and the top boundary is Cy, and the distance to the bottom boundary is H-Cy.

[0131] Next, this application can set a first preset percentage and a second preset percentage as judgment thresholds. The first preset percentage refers to the image width, for example, it can be set to 5%, that is, 5% of the image width W; the second preset percentage refers to the image height, for example, it can also be set to 5%, that is, 5% of the image height H. If the distance (Cx) between the centroid coordinate Cx and the left boundary is less than W×5%, or the distance (W - Cx) between the centroid coordinate Cx and the right boundary is less than W×5%, it indicates that the closed region is close to the image edge in the horizontal direction. Similarly, if the distance (Cy) between the centroid coordinate Cy and the upper boundary is less than H×5%, or the distance (H - Cy) between the centroid coordinate Cy and the lower boundary is less than H×5%, it indicates that the closed region is close to the image edge in the vertical direction. As long as any of the above conditions are met, it is determined that the closed region is close to the image edge of the pathological whole slide image. Conversely, if the distance between the centroid coordinate and all boundaries is greater than or equal to the corresponding preset percentage, it is determined that the closed region is not close to the image edge.

[0132] By setting a dynamic threshold based on the image size ratio, it is possible to adapt to pathological whole-section images of different resolutions, ensuring the accuracy and universality of edge judgment and effectively eliminating misjudgments caused by differences in image size.

[0133] In one embodiment, performing blur detection processing on the whole pathological slide image in S140 to obtain candidate blurred regions may include:

[0134] S141: Convert the pathological whole slice image to grayscale space to obtain a grayscale image.

[0135] S142: Perform a Laplacian transform on the grayscale image to obtain candidate blurred regions.

[0136] In this embodiment, when performing blur detection processing on the whole pathological slide image to obtain candidate blurred regions, such as... Figure 7 As shown, the first step is to convert the whole pathological slide image to grayscale space to obtain a grayscale image. Whole pathological slide images are typically stored in RGB color mode, while blur detection algorithms often analyze images based on brightness information. Converting a color image to grayscale simplifies data processing while preserving the image's main structural features. During the conversion, a weighted average method can be used, for example, by converting the pixel values ​​of the three RGB channels to a single grayscale value using the formula Gray = 0.299*R + 0.587*G + 0.114*B, where R, G, and B are the pixel values ​​of the red, green, and blue channels, respectively. This step results in a grayscale image that more clearly reflects the texture and edge information of the tissue, providing suitable input for subsequent Laplacian transforms.

[0137] Next, this application can perform a Laplacian transform on the grayscale image to obtain candidate blurred regions. The Laplacian transform is a commonly used second-order differential operator, which is very sensitive to edge and detail changes in an image. Sharp regions in an image typically contain rich high-frequency information and will produce larger response values ​​after the Laplacian transform; while blurred regions, due to the loss of high-frequency information, will have relatively smaller response values ​​after the Laplacian transform. Specifically, this application can use a 3x3 Laplacian operator template (such as [[0,1,0],[1,-4,1],[0,1,0]]) to perform convolution operations on the grayscale image. After convolution, the absolute value of the pixel value in the image reflects the edge sharpness at that location. By setting an appropriate threshold, regions with absolute pixel values ​​less than the threshold after the Laplacian transform are identified as candidate blurred regions. For example, the threshold can be determined by analyzing the histogram distribution of the Laplacian transform results, selecting the optimal segmentation point that can distinguish between sharp regions and potential blurred regions.

[0138] In this way, through Laplacian transform and threshold segmentation, candidate regions that may have blurring problems can be initially screened from the grayscale image, providing target regions for subsequent accurate detection and analysis.

[0139] In one embodiment, S150, based on the foreground mask of the tissue region and the candidate blurred regions, detects blurred regions of the tissue portion in the whole pathological slide image and obtains detection results, which may include:

[0140] S151: Upsample the foreground mask of the tissue region to obtain a foreground segmentation mask with the same resolution as the whole pathological slice image.

[0141] S152: Calculate the difference image between the foreground segmentation mask and the candidate blurred region.

[0142] S153: Threshold segmentation is performed on the differential image to obtain a blurred region mask of the tissue portion in the pathological whole slice image, and the blurred region mask is used as the detection result.

[0143] In this embodiment, after obtaining the foreground mask and candidate blurred regions of the tissue region, it is necessary to further accurately detect the blurred regions of the tissue portion in the whole pathological slide image. Firstly, considering that the foreground mask of the tissue region may have used a lower resolution in the preprocessing stage to improve computational efficiency, this application can upsample the foreground mask of the tissue region to ensure that its resolution is consistent with the original whole pathological slide image. Upsampling methods can employ bilinear interpolation, bicubic interpolation, or nearest neighbor interpolation, selecting an interpolation algorithm that better preserves edge information to ensure that the upsampled foreground segmentation mask accurately corresponds to the tissue region position in the original image. Through upsampling, each pixel of the foreground segmentation mask can correspond one-to-one with a pixel in the whole pathological slide image, laying the foundation for subsequent region calculations.

[0144] Next, this application can calculate the difference image between the foreground segmentation mask and the candidate blurred regions. Here, "difference" is not a simple subtraction operation, but rather refers to obtaining the intersection region of the two. Specifically, in the foreground segmentation mask, regions with a pixel value of 1 (or other values ​​representing the foreground) represent tissue regions, and regions with a pixel value of 1 in the candidate blurred regions represent potential blurred regions. By performing a logical AND operation between the foreground segmentation mask and the candidate blurred regions, the regions with a pixel value of 1 in the resulting image (i.e., the difference image) are those that belong to both tissue regions and candidate blurred regions. The purpose of this step is to filter out blurred regions located within tissue from the candidate blurred regions, eliminating pseudo-blurred regions that may be located in the background or non-tissue regions, thereby narrowing the detection range and improving the targeting and accuracy of blur detection.

[0145] Finally, this application can perform thresholding on the difference image to obtain a mask of blurred regions in the tissue portion of the pathological whole-slice image. Since the difference image is already binarized (assuming both the foreground segmentation mask and the candidate blurred regions are binary images), the thresholding at this point is mainly to further confirm and extract effective blurred regions. For example, a minimum region area threshold can be set, and connected regions in the difference image with an area smaller than this threshold are considered noise or small artifacts and are removed, while larger connected regions are retained as the final blurred regions of the tissue portion. The binary image obtained after thresholding is the blurred region mask, where the foreground pixels (with a value of 1) mark the blurred positions within the tissue region. Using this blurred region mask as the detection result can intuitively show the blurred distribution of the tissue portion in the pathological whole-slice image, providing a clear reference for pathological diagnosis.

[0146] In one embodiment, the method may further include:

[0147] S154: Perform contour extraction and smoothing on the blurred region mask to obtain a contour curve.

[0148] S155: Output or upload the contour curve and the detection results.

[0149] In this embodiment, after obtaining the blurred region mask, to more intuitively display the shape and boundary of the blurred region, this application can perform contour extraction and smoothing processing on the blurred region mask to obtain a contour curve. Contour extraction can be achieved through edge detection algorithms, such as using the Canny edge detection operator to process the blurred region mask and identify the boundary pixels of the blurred region. These boundary pixels usually constitute a series of discrete coordinate points, which may be jagged or discontinuous. Therefore, it is necessary to smooth the extracted contour to eliminate noise interference and obtain a more regular curve shape. Smoothing processing can employ methods such as polynomial fitting, Bézier curve fitting, or moving average filtering. For example, the Douglas-Peucker algorithm can be used to simplify the contour points, retain key feature points, and remove redundant detail points, making the contour curve smoother and more natural. The contour curve obtained after contour extraction and smoothing processing can clearly outline the boundary shape of the blurred region within the organization area, providing a more accurate geometric description for subsequent analysis and display.

[0150] After obtaining the contour curve and detection results (i.e., the blurred region mask), this application can output or upload this information. Output methods can include visual display on a local display device, such as overlaying the contour curve onto the original pathological whole-slice image, marking the boundaries of the blurred regions with a specific color (e.g., red), and displaying statistical information such as the area and location coordinates of the blurred regions for easy viewing by pathologists. Uploading methods can involve sending the contour curve data, blurred region mask data, and related statistical information to the hospital's pathology information system (PACS) or a cloud server via a network, achieving centralized data management and sharing. Furthermore, the detection results can be output in the form of a structured report, including key information such as the number of blurred regions, the total area percentage, and the location and size of the largest blurred region, providing quantitative reference for the pathological diagnosis process and assisting doctors in making more accurate judgments. Through output or upload steps, the detection results of this application can be effectively integrated into the actual workflow of pathological diagnosis, improving diagnostic efficiency and accuracy.

[0151] The following describes the fuzzy detection device for histopathological slides provided in the embodiments of this application. The fuzzy detection device for histopathological slides described below can be referred to in correspondence with the fuzzy detection method for histopathological slides described above.

[0152] In one embodiment, such as Figure 8 As shown, Figure 8This is a schematic diagram of a blur detection device for tissue pathology slides provided in an embodiment of this application. This application also provides a blur detection device for tissue pathology slides, which may include an image acquisition module 210, a prompt generation module 220, a region segmentation module 230, a first detection module 240, and a second detection module 250, specifically including the following:

[0153] Image acquisition module 210 is used to acquire images of whole pathological slides.

[0154] The prompt generation module 220 is used to generate segmentation prompt information to drive a preset segmentation base model based on the location and category information of at least one closed region contained in the pathological whole slice image.

[0155] The region segmentation module 230 is used to input the pathological whole slice image and the segmentation prompt information into the segmentation base model to obtain the tissue region foreground mask output by the segmentation base model.

[0156] The first detection module 240 is used to perform blur detection processing on the pathological whole slice image to obtain candidate blur regions.

[0157] The second detection module 250 is used to detect the blurred regions of the tissue portion in the pathological whole slice image based on the foreground mask of the tissue region and the candidate blurred regions, and obtain the detection results.

[0158] In the above embodiments, segmentation prompts are generated based on the location and category information of closed regions in the whole pathological slide image. This drives the basic segmentation model to accurately segment the foreground mask of the tissue region, effectively eliminating interference from non-tissue regions. Furthermore, by combining candidate blurred regions obtained from the blur detection processing of the whole pathological slide image, and through the operation of the foreground segmentation mask with the candidate blurred regions, blurred regions of the tissue portion in the whole pathological slide image can be accurately detected. This significantly improves the specificity and accuracy of blur detection, reduces false positives and false negatives, and provides strong technical support for ensuring the quality of pathological diagnosis and improving diagnostic efficiency.

[0159] In one embodiment, this application also provides a computer-readable storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the fuzzy detection method for histopathological sections as described in any of the above embodiments.

[0160] In one embodiment, this application also provides a computer device, including: one or more processors, and memory.

[0161] The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the fuzzy detection method for histopathological sections as described in any of the above embodiments.

[0162] Indicatively, such as Figure 9 As shown, Figure 9 This is a schematic diagram of the internal structure of a computer device 300 provided in an embodiment of this application. The computer device 300 can be provided as a server. (Refer to...) Figure 9 The computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions executable by the processing component 302, such as application programs. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 302 is configured to execute instructions to perform the fuzzy detection method for histopathological slides according to any of the above embodiments.

[0163] The computer device 300 may also include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate on an operating system stored in memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0164] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0165] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0166] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0167] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fuzzy detection method for tissue pathology sections, characterized in that, The method includes: Obtain whole-section images of pathological tissue; Based on the location and category information of at least one closed region contained in the pathological whole slice image, segmentation prompt information is generated to drive the preset segmentation base model. The pathological whole-section image and the segmentation prompt information are input into the segmentation base model to obtain the tissue region foreground mask output by the segmentation base model; The pathological whole-section image is subjected to blur detection processing to obtain candidate blur regions; Based on the foreground mask of the tissue region and the candidate blurred region, the blurred region of the tissue part in the whole pathological slice image is detected, and the detection result is obtained.

2. The method according to claim 1, characterized in that, The step of generating segmentation prompts based on the location and category information of at least one closed region contained in the pathological whole-slice image to drive a preset segmentation model includes: The pathological whole slide image is preprocessed to determine at least one closed region contained in the pathological whole slide image; Determine the location and category information of each closed region; Based on the location and category information of each closed region, segmentation prompts are generated to drive the preset segmentation base model.

3. The method according to claim 2, characterized in that, The preprocessing of the whole pathological slide image to determine at least one closed region contained in the whole pathological slide image includes: The pathological whole-section image was downsampled to obtain a low-resolution whole-section image; The low-resolution full-slice image is segmented by saturation thresholding to obtain a binary mask of the initial foreground region; Closed region detection is performed on the binary mask to obtain at least one closed region.

4. The method according to claim 3, characterized in that, The step of performing closed region detection on the binary mask to obtain at least one closed region includes: Convert the binary mask into an 8-bit single-channel image; The 8-bit single-channel image is closed region detection is performed using the contour retrieval mode RETR_TREE to obtain the contour line of at least one closed region, wherein the contour line preserves the hierarchical relationship between the outer contour and the internal hole.

5. The method according to claim 2, characterized in that, The determination of the location and category information of each closed region includes: For each closed region: The centroid of the closed region is calculated to obtain its coordinates, and these coordinates are used as the position information of the closed region. Determine whether the closed region is close to the image edge of the pathological whole-section image; If not, the closed region is determined to be an organizational region, and a first label is assigned to indicate that it is an organizational region, and the first label is used as the category information of the closed region. If so, the closed region is determined to be a non-organizational region, and a second label is assigned to indicate that it is a non-organizational region, with the second label serving as the category information for the closed region.

6. The method according to claim 5, characterized in that, The process of calculating the centroid of the closed region to obtain its coordinates includes: Calculate the geometric moments corresponding to the contour of the closed region to obtain the zeroth moment and the first moment; Calculate the centroid coordinates of the closed region based on the zeroth moment and the first moment.

7. The method according to claim 5, characterized in that, The determination of whether the closed region is close to the image edge of the whole pathological slide image includes: Determine the distance between the centroid coordinates of the closed region and the image boundary of the pathological whole-section image; If the distance between the centroid coordinates and the left and right boundaries of the pathological whole slice image is less than a first preset percentage of the image width, or the distance between the centroid coordinates and the upper and lower boundaries of the pathological whole slice image is less than a second preset percentage of the image height, then it is determined that the closed region is close to the image edge of the pathological whole slice image. Otherwise, it is determined that the closed region is not close to the image edge of the pathological whole slice image.

8. The method according to claim 1, characterized in that, The blur detection processing of the whole pathological slide image to obtain candidate blurred regions includes: The pathological whole-section image is converted to grayscale space to obtain a grayscale image; The grayscale image is subjected to a Laplacian transform to obtain candidate blurred regions.

9. The method according to claim 1, characterized in that, The detection of blurred regions in the tissue portion of the whole pathological slide image based on the foreground mask of the tissue region and the candidate blurred regions, and obtaining the detection results, includes: Upsample the foreground mask of the tissue region to obtain a foreground segmentation mask with the same resolution as the whole pathological slice image; Calculate the difference image between the foreground segmentation mask and the candidate blurred region; Threshold segmentation is performed on the differential image to obtain a blurred region mask of the tissue portion in the pathological whole slice image, and the blurred region mask is used as the detection result.

10. The method according to claim 9, characterized in that, The method further includes: The blurred region mask is subjected to contour extraction and smoothing to obtain a contour curve; Output or upload the contour curve and the detection results.

11. A device for detecting fuzzy tissue pathology sections, characterized in that, include: The image acquisition module is used to acquire images of whole pathological slides; The prompt generation module is used to generate segmentation prompt information to drive a preset segmentation base model based on the location and category information of at least one closed region contained in the pathological whole slice image. The region segmentation module is used to input the pathological whole slice image and the segmentation prompt information into the segmentation base model to obtain the tissue region foreground mask output by the segmentation base model; The first detection module is used to perform blur detection processing on the pathological whole slice image to obtain candidate blurred regions; The second detection module is used to detect blurred regions of the tissue portion in the pathological whole-section image based on the foreground mask of the tissue region and the candidate blurred regions, and to obtain the detection results.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the fuzzy detection method for histopathological sections as described in any one of claims 1 to 10.

13. A computer device, characterized in that, include: One or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the fuzzy detection method for histopathological sections as described in any one of claims 1 to 10.

Images