A method and device for evaluating the quality of HE digital pathology sections

Abstract

The application provides an HE digital pathological section quality evaluation method and device, acquires an HE digital pathological section to be evaluated, and performs scanning quality detection on the HE digital pathological section to obtain scanning quality judgment information; when the HE digital pathological section is determined to be a qualified section based on the scanning quality judgment information, performs section-level quality detection and pixel-level instance segmentation identification on the HE digital pathological section to obtain section-level quality attribute information and pixel-level instance attribute information; performs comprehensive calculation based on the section-level quality attribute information, the pixel-level instance attribute information, and a preset quality quantification rule to obtain a comprehensive quality quantification value of the HE digital pathological section, and generates a structured quality control report to form a closed loop of detection, interception, tracing, and improvement, thereby systematically improving section preparation quality, solving the problem of low efficiency of high-magnification dense block processing mode in pathological section quality detection, reducing the calculation overhead, and improving the overall efficiency of the quality control process.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and apparatus for evaluating the quality of HE digital pathological slides. Background Technology

[0002] Today, with the rapid development of digital pathology, whole-slide digital images have become a crucial data foundation for pathological diagnosis. The quality of pathological slides directly affects a physician's ability to identify key morphological features such as cell mitosis, apoptosis, and nuclear fragmentation, thus influencing diagnostic accuracy and patient treatment decisions. Given the continuously increasing workload in pathology, systematic quality control of pathological slides has become a vital step in ensuring diagnostic accuracy, reducing medical errors, and preventing liability incidents.

[0003] Currently, automated slice quality assessment typically employs a method of densely segmenting the entire slice image at high magnification, analyzing each image block individually, and then integrating the results to achieve an overall judgment. This method involves cropping the entire slice image into fixed-size small blocks and inputting them into a classification or segmentation model to identify whether quality defects such as defocusing and artifacts exist in each block. Finally, the analysis results of each block are summarized to obtain the quality control judgment for the entire slice.

[0004] The main drawbacks of the aforementioned technical solutions are as follows: Firstly, the paradigm of dense segmentation and block-by-block analysis at high magnification results in excessively long processing times and huge computational resource consumption for single slides, making it difficult to meet the real-time requirements of large-scale, high-throughput clinical practice. Secondly, existing solutions mostly focus on a single image analysis step after scanning, treating quality control as an independent image analysis task without integrating it into the overall pathological workflow for technical architecture design. There is a lack of a unified platform framework capable of interfacing with different slide preparation equipment and information systems and integrating multi-stage quality control algorithms. Simultaneously, there is a lack of joint modeling and traceability mechanisms for quality indicators related to the entire process, making it difficult to conduct multi-dimensional, refined quality evaluation of slides, thus restricting the overall efficiency and comprehensiveness of the quality control process. Summary of the Invention

[0005] This application provides a method and apparatus for evaluating the quality of HE digital pathological slides, forming a closed loop of detection, interception, traceability and improvement, thereby systematically improving the quality of slide preparation, solving the problem of low efficiency in high-magnification dense block processing during pathological slide quality detection, reducing computational overhead and improving the overall efficiency of the quality control process.

[0006] In a first aspect, embodiments of this application provide a method for evaluating the quality of HE digital pathology slides, the method comprising: Obtain the HE digital pathology slides to be evaluated, and perform scanning quality detection on the HE digital pathology slides to obtain scanning quality judgment information; When the HE digital pathology slide is determined to be a qualified slide based on the scan quality judgment information, slice-level quality detection and pixel-level instance segmentation and recognition are performed on the HE digital pathology slide to obtain slice-level quality attribute information and pixel-level instance attribute information. The comprehensive quality quantification value of the HE digital pathology slide is obtained by comprehensively calculating based on the slice-level quality attribute information, the pixel-level instance attribute information, and the preset quality quantification rules. Based on the slice-level quality attribute information, the pixel-level instance attribute information, and the comprehensive quality quantification value, a structured quality control report is generated; wherein, the structured quality control report includes detection status information for each quality control category, detailed information on each quantification deduction value, and improvement suggestions for the detected quality control category.

[0007] Secondly, embodiments of this application also provide a device for evaluating the quality of HE digital pathological slides, the device comprising: The scanning quality assessment module is used to acquire the HE digital pathology slides to be evaluated, and to perform scanning quality detection on the HE digital pathology slides to obtain scanning quality assessment information. The slide quality detection module is used to perform slide-level quality detection and pixel-level instance segmentation and recognition on the HE digital pathology slide when the HE digital pathology slide is determined to be a qualified slide based on the scan quality judgment information, so as to obtain slide-level quality attribute information and pixel-level instance attribute information. The slide quantification evaluation module is used to perform comprehensive calculations based on the slide-level quality attribute information, the pixel-level instance attribute information, and preset quality quantification rules to obtain the comprehensive quality quantification value of the HE digital pathology slide; The quality control report generation module is used to generate a structured quality control report based on the slice-level quality attribute information, the pixel-level instance attribute information, and the comprehensive quality quantification value; wherein, the structured quality control report includes detection status information of each quality control category, detailed information of each quantification deduction value, and improvement suggestions for the detected quality control category.

[0008] Thirdly, embodiments of this application also provide an electronic device, which includes: One or more processors; Storage device for storing one or more programs. When one or more programs are executed by one or more processors, the one or more processors implement the HE digital pathology slide quality evaluation method as described in any of the embodiments of this application.

[0009] Fourthly, embodiments of this application also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the HE digital pathology slide quality evaluation method as described in any of the embodiments of this application.

[0010] This application provides a method for evaluating the quality of HE digital pathology slides. The method involves acquiring the HE digital pathology slide to be evaluated and performing scanning quality detection to obtain scanning quality judgment information. If the HE digital pathology slide is determined to be a qualified slide based on the scanning quality judgment information, then the slide undergoes slice-level quality detection and pixel-level instance segmentation and recognition to obtain slice-level quality attribute information and pixel-level instance attribute information. Based on the slice-level quality attribute information, pixel-level instance attribute information, and preset quality quantification rules, a comprehensive quality quantification value for the HE digital pathology slide is obtained. Finally, based on the slice-level quality attribute information, pixel-level instance attribute information, and comprehensive quality quantification value, a structured quality control report is generated. This structured quality control report includes detection status information for each quality control category, detailed information on each quantification deduction value, and improvement suggestions for the detected quality control categories. The technical solution of this application constructs a two-stage quality control process of rapid scanning quality detection and systematic pre-diagnosis assessment. First, it intercepts severely defective slices and displays early warning prompts. Then, it performs dual quality analysis at the slice level and pixel level on qualified slices simultaneously. Finally, it generates a structured quality control report containing quantitative deduction details and improvement suggestions to trace the cause of quality defects. This forms a closed-loop mechanism of detection, interception, tracing, and improvement, which solves the problem of low efficiency of existing high-magnification dense block processing methods, reduces computational overhead, and improves the overall efficiency of the quality control process and the ability to systematically improve the quality of the slides. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in describing the embodiments are briefly introduced below. Obviously, the accompanying drawings described are only a portion of the embodiments to be described in this application, and not all of them. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort.

[0012] Figure 1 This is a flowchart illustrating a method for evaluating the quality of HE digital pathological slides provided in an embodiment of this application. Figure 2 This is a flowchart illustrating another HE digital pathology slide quality evaluation method provided in an embodiment of this application. Figure 3 A flowchart illustrating another HE digital pathology slide quality evaluation method provided in this application embodiment; Figure 4This is a schematic diagram of the structure of a HE digital pathology slide quality evaluation device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0014] Before introducing the technical solutions provided in this application, the application scenarios of the solutions can be explained first. HE digital pathology slides refer to the process of converting human or animal tissues into a digital image file of the entire slide after hematoxylin and eosin (HE) staining, using a digital scanning device. This image is the core basis for pathologists to conduct histological observation and disease diagnosis. This embodiment is applicable to various scenarios requiring quality evaluation of HE digital pathology slides. Currently, while quality evaluation methods based on high-magnification dense segmentation and block-by-block analysis of whole-slide images have achieved automated evaluation to some extent, traditional methods have significant limitations. In practical applications, due to the diverse types of quality defects in HE digital pathology slides—including serious scanning quality problems affecting diagnosis and various local defects introduced during slide preparation—and the need for a balance between detection efficiency and fine-grained localization in the quality evaluation process, traditional high-magnification dense segmentation methods struggle to achieve an effective balance between rapid screening and refined evaluation at the global level. This easily leads to problems such as excessively long processing times for single slides, huge computational resource consumption, and failure to promptly intercept severely defective slides. Therefore, there is an urgent need for a phased quality control method that can integrate rapid detection of scanning quality with systematic pre-diagnosis assessment, in order to improve the overall efficiency and comprehensiveness of HE digital pathology slide quality evaluation.

[0015] Example 1 Figure 1 This is a flowchart illustrating a method for evaluating the quality of HE digital pathological slides provided in an embodiment of this application. This embodiment is applicable to various situations requiring quality evaluation of HE digital pathological slides. The method can be executed by an HE digital pathological slide quality evaluation device, which can be implemented in the form of software and / or hardware. The hardware can be a controller, such as a mobile terminal, a PC, or a server.

[0016] like Figure 1 As shown, the HE digital pathology slide quality evaluation method provided in this embodiment of the invention includes the following steps: S110. Obtain the HE digital pathology slides to be evaluated, and perform scanning quality detection on the HE digital pathology slides to obtain scanning quality judgment information.

[0017] HE digital pathology slides refer to whole-slide digital image files generated by digital scanning equipment after hematoxylin-eosin staining. Scan quality assessment information is data used to indicate whether the HE digital pathology slide meets the conditions for further processing.

[0018] Specifically, HE digital pathology slides can be loaded for subsequent analysis. Based on this, by identifying whether there are serious defects in the HE digital pathology slides that lead to extremely poor slide quality or affect diagnosis on low-magnification thumbnails, scanning quality judgment information is finally generated to indicate whether the HE digital pathology slide meets the conditions for continued processing.

[0019] In this embodiment, optionally, the specific implementation of obtaining the HE digital pathology slides to be evaluated may include the following steps: (1) Obtain the original image file of HE digital pathology slide and identify the encapsulation format type of the original image file.

[0020] The original image file refers to the undecoded raw data file of HE digital pathology slides, directly generated by a digital slide scanning device and stored on a storage medium. The encapsulation format type refers to the data organization structure and encoding standard category used by the original image file, identifying the type of digital pathology image format to which the file belongs.

[0021] Specifically, the raw image data generated directly by the scanning device and not yet decoded can be read from the storage medium for subsequent parsing and processing. Based on this, by parsing the format identification information in the file header and data structure of the raw image file, the data organization structure and encoding standard category adopted by the raw image file can be determined, thereby obtaining the encapsulation format type used to guide the selection of subsequent decoding tools.

[0022] (2) Based on the encapsulation format type, select the target read adapter that matches the encapsulation format type from multiple candidate read adapters.

[0023] Here, a candidate read adapter refers to a pre-built set of decoding tool components used to parse different container format types. A target read adapter refers to a specific decoding tool component selected from the candidate read adapters that matches the container format type.

[0024] Specifically, based on the obtained container format type of the original image file, the container format type is used as the matching criterion. The format range supported by each of the pre-built candidate reading adapters is compared one by one, and finally a specific decoding tool that is compatible with the container format type is selected as the target reading adapter.

[0025] (3) Call the target reading adapter to decode the original image file and load the decoded image data into memory as HE digital pathology slides to be evaluated.

[0026] Specifically, a target read adapter can be used to parse and restore the image data stored in the original image file according to specific encoding rules to obtain pixel data that can be directly used in subsequent image analysis operations. Based on this, the decoded pixel data can be loaded into memory as HE digital pathology slides to be evaluated.

[0027] S120. When the HE digital pathology slide is determined to be a qualified slide based on the scanning quality judgment information, slice-level quality detection and pixel-level instance segmentation and recognition are performed on the HE digital pathology slide to obtain slice-level quality attribute information and pixel-level instance attribute information.

[0028] Slice-level quality inspection is a process that identifies and judges the quality attributes of the entire slice as a whole. Slice-level quality attribute information is a set of status data that records whether each slice-level quality control category has been detected at the slice level.

[0029] Pixel-level instance segmentation and recognition involves detecting the location and classifying each quality defect target in the sliced ​​image pixel by pixel. Pixel-level instance attribute information is a dataset recording the location distribution and coverage area of ​​instances of each pixel-level quality control category within the slice.

[0030] Specifically, the scan quality assessment information generated during the scan quality inspection stage serves as the basis for conditional judgment. When this information indicates that the current HE digital pathology slide meets the conditions for continued processing, it is considered a qualified slide that can pass the scan quality control stage and enter the subsequent pre-diagnostic quality evaluation process. Under this premise, the image data of the HE digital pathology slide is loaded into memory, and two parallel quality analysis operations—slice-level quality inspection and pixel-level instance segmentation and recognition—are performed simultaneously. Slice-level quality inspection identifies the overall quality attributes of the HE digital pathology slide, determining whether the entire slide has quality categories reflecting the overall condition of the slide, such as tremor marks, staining depth, or thickness abnormalities. Pixel-level instance segmentation and recognition detects and classifies various quality defect targets pixel by pixel on the HE digital pathology slide image, locating quality defect instances that may appear at any location on the slide, such as knife marks, cracks, fissures, folds, bubbles, and incomplete sections.

[0031] Through the above-described slice-level quality detection process, slice-level quality attribute information is obtained. This information records whether each slice-level quality control category is detected at the HE digital pathology slide level. Through the above-described pixel-level instance segmentation and recognition process, pixel-level instance attribute information is obtained. This information records the location distribution and coverage area of ​​instances of each pixel-level quality control category within the HE digital pathology slide. These two parallel quality control detection modules comprehensively characterize the quality status of the HE digital pathology slide from both overall attribute and local instance dimensions.

[0032] In this embodiment, optionally, when the HE digital pathology slide is determined to be an unqualified slide based on the scan quality judgment information, a quality control early warning message is displayed.

[0033] The quality control early warning information refers to the warning notification generated when an HE digital pathology slide is determined to be unqualified, serving as a reminder for technicians to intervene manually. The quality control early warning information includes at least the scan quality assessment result, a description of the defect type, and processing suggestions, including suggestions for rescanning and re-preparation.

[0034] The scan quality assessment result refers to the specific category name of the scan quality control category to which the HE digital pathology slide belongs, as recorded in the scan quality assessment information. The defect type description refers to a textual explanation of the nature of the specific quality defect that caused the slide to be deemed unacceptable. The processing recommendations refer to a set of suggestions provided to guide technicians in taking subsequent actions for the currently unacceptable slide. The rescan recommendation refers to a suggestion in the processing recommendations instructing the technician to re-perform a digitization scan of the original slide corresponding to the slide. The re-slide preparation recommendation refers to a suggestion in the processing recommendations instructing the technician to re-prepare the tissue sample corresponding to the slide.

[0035] Specifically, when the scan quality control category indicated by the scan quality assessment information generated during the scan quality inspection stage is deemed unqualified, the HE digital pathology slide is identified as having a serious defect that results in extremely poor slide preparation quality or affects diagnosis. In this case, a quality control warning message is triggered to promptly notify relevant personnel that the slide has failed the scan quality control process and requires manual intervention. This quality control warning message contains at least three categories of content: the scan quality assessment result, which indicates the specific scan quality control category to which the slide belongs; the defect type description, which explains the nature of the specific quality defect that caused the unqualified slide; and the handling suggestions, which provide actionable follow-up instructions. These handling suggestions specifically include a rescanning suggestion to re-perform a digital scan on the original slide corresponding to the slide, and a re-slide preparation suggestion to re-prepare the tissue sample corresponding to the slide, allowing relevant personnel to choose the appropriate handling method based on the actual situation.

[0036] S130. Based on slice-level quality attribute information, pixel-level instance attribute information, and preset quality quantification rules, a comprehensive quality quantification value for HE digital pathology slices is obtained through comprehensive calculation.

[0037] Among them, the preset quality quantification rules are a set of pre-defined mapping rules used to convert the detection status and coverage of each quality control category into quantified deduction values.

[0038] The comprehensive quality quantification value is a quantitative value used to characterize the overall quality level of the HE digital pathology slide.

[0039] Specifically, the detection status of each slice-level quality control category recorded in the slice-level quality attribute information and the detection status and area ratio of each pixel-level quality control category recorded in the pixel-level instance attribute information can be used as input. Based on the quantification deduction mapping relationship set for different quality control categories and their severity in the preset quality quantification rules, the first quantification deduction value corresponding to each slice-level quality control category and the second quantification deduction value corresponding to each pixel-level quality control category are determined respectively. Then, all the first quantification deduction values ​​and all the second quantification deduction values ​​are accumulated. Through this accumulation operation, a comprehensive quality quantification value used to characterize the overall quality level of the HE digital pathology slide is obtained.

[0040] S140. Generate a structured quality control report based on slice-level quality attribute information, pixel-level instance attribute information, and comprehensive quality quantification value.

[0041] A structured quality control report is a structured document generated based on quality evaluation results, containing various indicators and evaluation information related to slice quality. The structured quality control report includes detection status information for each quality control category, detailed information on each quantitative deduction value, and improvement suggestions for the detected quality control categories.

[0042] The detection status information for each quality control category refers to the status data of whether each category in the documentary-level and pixel-level quality control categories is detected in HE digital pathology slides. The detailed information for each quantitative deduction value refers to the specific numerical value of the first or second quantitative deduction value corresponding to each detected quality control category, along with detailed data on its category. The improvement suggestion information refers to the specific improvement measures text matched based on a preset standard knowledge base for each detected quality control category, used to guide technicians in optimizing slide preparation operations.

[0043] Specifically, the quality attribute information output from the slice-level quality inspection stage and the pixel-level instance attribute information output from the pixel-level instance segmentation and recognition stage can be used as the data sources for quality control results. The comprehensive quality quantification value obtained from the comprehensive calculation stage serves as the quantitative representation of the overall quality level. These three types of information are organized and arranged according to a preset document structure to form a structured quality evaluation document. In this structured quality control report, the detection status information for each quality control category records whether each category in the slice-level and pixel-level quality control categories is detected in the HE digital pathology slide. The detailed information for each quantified deduction value records the specific deduction value and its category for each detected quality control category. The improvement suggestion information matches corresponding improvement measures from a preset standard knowledge base for each detected quality control category, guiding the optimization of slide preparation operations. Therefore, this structured quality control report not only presents the quantitative results of the quality evaluation but also provides actionable improvement guidance.

[0044] This application provides a method for evaluating the quality of HE digital pathology slides. The method involves acquiring the HE digital pathology slide to be evaluated and performing scanning quality detection to obtain scanning quality judgment information. If the HE digital pathology slide is determined to be a qualified slide based on the scanning quality judgment information, then the slide undergoes slice-level quality detection and pixel-level instance segmentation and recognition to obtain slice-level quality attribute information and pixel-level instance attribute information. Based on the slice-level quality attribute information, pixel-level instance attribute information, and preset quality quantification rules, a comprehensive quality quantification value for the HE digital pathology slide is obtained. Finally, based on the slice-level quality attribute information, pixel-level instance attribute information, and comprehensive quality quantification value, a structured quality control report is generated. This structured quality control report includes detection status information for each quality control category, detailed information on each quantification deduction value, and improvement suggestions for the detected quality control categories. The technical solution of this application constructs a two-stage quality control process of rapid scanning quality detection and systematic pre-diagnosis assessment. First, it intercepts severely defective slices and displays early warning prompts. Then, it performs dual quality analysis at the slice level and pixel level on qualified slices simultaneously. Finally, it generates a structured quality control report containing quantitative deduction details and improvement suggestions to trace the cause of quality defects. This forms a closed-loop mechanism of detection, interception, tracing, and improvement, which solves the problem of low efficiency of existing high-magnification dense block processing methods, reduces computational overhead, and improves the overall efficiency of the quality control process and the ability to systematically improve the quality of the slides.

[0045] After completing the quality evaluation of HE digital pathology slides and obtaining a comprehensive quality quantification value and a structured quality feedback document, in order to correlate the image-level quality evaluation results with production information in the slide preparation and scanning process, and to achieve traceability from quality issues to the production process, the method further includes the following steps: S1. Obtain the structured metadata generated during the scanning process of HE digital pathology slides.

[0046] Among them, structured metadata refers to a set of production process information associated with HE digital pathology slides and organized according to a preset data structure. This set of information includes data items such as slide batch information, scanning equipment parameters, and timestamp information used to trace the source of the slides.

[0047] Specifically, during the process of digitizing HE digital pathology slides via scanning equipment, various production process information related to the scanning operation is collected and extracted simultaneously. This information is then organized into structured metadata according to a preset data structure for subsequent traceability and query. This structured metadata includes data items such as slide batch information, scanning equipment parameters, and timestamp information used to identify the slide source and production conditions.

[0048] S2. Establish the association mapping relationship between HE digital pathology slides and their corresponding structured metadata.

[0049] The association mapping relationship refers to the corresponding association established between HE digital pathology slides and their corresponding structured metadata, which is used to index all metadata information generated during the production scanning process when obtaining slide image data.

[0050] Specifically, the unique identifier of an HE digital pathology slide can be paired and bound with the association key information in the structured metadata, so that all the structured metadata generated during the scanning process can be indexed through the HE digital pathology slide, thereby realizing the correspondence between image data and production process information.

[0051] S3. Based on the comprehensive quality quantification value and the associated mapping relationship, determine the source information of quality defects in HE digital pathology slides from the structured metadata, and associate the source information of quality defects with the structured quality feedback document.

[0052] Among them, the information on the source of quality defects refers to the identification information of the specific production process that caused the quality defect, which is located from the structured metadata based on the type of quality defect reflected by the comprehensive quality quantification value and the associated mapping relationship.

[0053] Specifically, based on the quality level represented by the comprehensive quality quantification value, and combined with the structured metadata corresponding to the HE digital pathology slides obtained through the association mapping relationship, the production links that may lead to quality defects and match the quality level can be located from the data items such as slide batch information and scanning equipment parameters contained in the structured metadata. This yields information on the source links of quality defects. The identified source links of quality defects are then associated with the structured quality feedback document, so that the structured quality feedback document contains the location information of the defect links while recording the quality evaluation results.

[0054] The above solution, after obtaining the comprehensive quality quantification value and structured quality feedback document, further acquires the structured metadata generated during the scanning process and establishes an association mapping with the slices. This allows the quality evaluation results to no longer be isolated at the image level, but to trace and locate the source of the defects from the structured metadata based on the quality defects reflected by the comprehensive quality quantification value and the association mapping relationship. The information of the source link is then fed back to the structured quality feedback document, thereby realizing a closed-loop traceability from image quality problems to the film production or scanning production process, providing data support for the source investigation and targeted improvement of quality defects.

[0055] Example 2 Figure 2This is a schematic diagram of a method for evaluating the quality of HE digital pathological slides provided in this application embodiment. Based on the foregoing embodiments, this embodiment will provide a more detailed description of steps S110 and S120. For specific implementation details, please refer to the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0056] like Figure 2 As shown, the method specifically includes the following steps: S210. Obtain the HE digital pathology slides to be evaluated.

[0057] S220. Load the pyramid layer of HE digital pathology slides and extract low-magnification thumbnails from the pyramid layer.

[0058] In this context, the pyramid layer refers to the multi-resolution image data structure used in the digital storage of HE digital pathology slides, consisting of multiple image layers ranging from low to high resolution. The low-magnification thumbnail refers to a lower-resolution image layer extracted from the pyramid layer, used for a quick overview of the entire slide.

[0059] Specifically, the multi-resolution image data structure of HE digital pathology slides stored in the storage medium can be read into memory as a whole. Based on this, a low-resolution image level can be selected from the multiple image levels contained in the pyramid layer and extracted as a low-magnification thumbnail. This low-magnification thumbnail is used to complete the preliminary analysis of the whole slide in subsequent processing with low computational overhead.

[0060] S230. Input the low-magnification thumbnail into the tissue block detection model to identify the effective tissue region in the HE digital pathology slide and obtain the tissue region location information.

[0061] The tissue block detection model is a pre-trained deep learning model used to locate effective tissue regions in slides on low-magnification thumbnails. An effective tissue region refers to the pixel area in an HE digital pathology slide that actually contains tissue samples, distinguished from the blank background area without tissue. Tissue region location information refers to the coordinate position data of the effective tissue region in the pyramid layer of the HE digital pathology slide.

[0062] Specifically, the extracted low-magnification thumbnail can be used as input data for a pre-trained tissue block detection model. The model performs forward inference calculations on the low-magnification thumbnail to identify the location of the effective tissue region in the HE digital pathology slide. Finally, the coordinate data of the effective tissue region in the HE digital pathology slide is output as the tissue region location information.

[0063] S240. Based on the location information of the tissue region, determine the cropped image containing the effective tissue region in the pyramid layer.

[0064] The cropped image is an image patch containing the effective tissue region that is cropped from the corresponding layer of the pyramid layer based on the location information of the tissue region.

[0065] Specifically, based on the coordinates of the effective tissue region indicated by the tissue region location information in the HE digital pathology slide, an appropriate level can be selected from multiple resolution levels of the pyramid layer. The tissue region location information can be mapped to the selected level, and an image block that completely contains the effective tissue region can be cropped at that level with the effective tissue region as the center. This image block is the cropped image.

[0066] In this embodiment, optionally, the specific implementation of determining the cropped image containing the effective tissue region in the pyramid layer based on the tissue region location information may include the following steps: S2401. Map the location information of the tissue region to the highest resolution layer of the pyramid layer to obtain the original width and original height of the effective tissue region at the highest resolution.

[0067] Specifically, the coordinates of the effective tissue region indicated by the tissue region location information on the low-magnification thumbnail can be used as a basis to transform the coordinates to the highest resolution level of the pyramid layer, thereby obtaining the original width and original height of the effective tissue region at the highest resolution level, which can be used for subsequent selection of target layers.

[0068] S2402. Select the target layer from the pyramid layer based on the original width and original height.

[0069] The target layer refers to a specific resolution level selected from multiple resolution levels of the pyramid layer. The resolution of the target layer ensures that the maximum side length of the effective organization region on the target layer is close to the preset cropping size. The preset cropping size is a pre-set target size value used to crop the image. This size guides the selection of the target layer so that the size of the cropped image patch meets the input requirements of the subsequent classification model.

[0070] Specifically, the original width and height of the effective organization area at the highest resolution level can be used as the selection criteria. The resolution scaling factor of each level in the pyramid layer is applied to the original width and height respectively. The maximum side length of the effective organization area at each level is calculated, and the layer that makes the maximum side length closest to the preset clipping size is selected as the target layer.

[0071] S2403. Map the location information of the tissue region to the target layer, and crop an image block centered on the effective tissue region and completely containing the effective tissue region from the target layer as the cropped image.

[0072] Specifically, the coordinates of the effective tissue region indicated by the tissue region location information are transformed from the original layer to the resolution layer of the target layer. Then, the image is cropped on the target layer with the effective tissue region defined by the mapped coordinates as the center, so that the cropped image block completely contains the effective tissue region on the target layer, and the image block is used as the cropped image.

[0073] S250. Input the cropped image into the scanning quality control classification model to obtain the region recognition result of the effective organization area.

[0074] The scan quality control classification model is a pre-trained classification model used to identify the scan quality category of tissue blocks in cropped images. The region identification result refers to the scan quality category judgment result of the scan quality control classification model for a single cropped image, indicating the category to which the tissue block belongs.

[0075] Specifically, a cropped image can be used as input to a pre-trained scanning quality control classification model. This model performs a discriminant analysis on the scanning quality category of the effective tissue region in the cropped image, and finally outputs the scanning quality category determination result of the effective tissue region. This determination result is the region recognition result.

[0076] S260. Based on the region identification results, a voting decision is made to determine the scanning quality control category corresponding to the HE digital pathology slide, and scanning quality judgment information is generated according to the scanning quality control category.

[0077] Among them, the scanning quality control category refers to the predefined classification labels used to judge the scanning quality of HE digital pathology slides, including blurring and defocusing, scanning splicing problems, solid color slides, severe contamination, and normal.

[0078] Specifically, when an HE digital pathology slide contains multiple effective tissue regions, the region identification results corresponding to each effective tissue region are obtained, and the frequency of each scan quality control category appearing in all region identification results is counted. The scan quality control category with the highest frequency is determined as the scan quality control category corresponding to the HE digital pathology slide. Then, scan quality judgment information is generated based on the scan quality control category. This scan quality judgment information is used to indicate whether the HE digital pathology slide meets the conditions for continued processing.

[0079] After obtaining the region identification results corresponding to each effective tissue region through the scanning quality control classification model, since a single HE digital pathology slide may contain multiple discrete effective tissue regions with varying tissue origins and imaging conditions, this embodiment uses a voting decision based on the region identification results to determine the scanning quality control category corresponding to the HE digital pathology slide. This includes: when the HE digital pathology slide contains multiple effective tissue regions, obtaining the region identification results corresponding to each effective tissue region; counting the frequency of each scanning quality control category in all region identification results; and determining the scanning quality control category with the highest frequency as the scanning quality control category corresponding to the HE digital pathology slide. This scheme integrates the independent identification results of multiple tissue regions through a voting mechanism, effectively reducing the interference of single-region identification noise on slide-level classification decisions and improving the robustness and accuracy of scanning quality control category determination.

[0080] S270. When the HE digital pathology slide is determined to be a qualified slide based on the scanning quality judgment information, slice-level quality detection and pixel-level instance segmentation and recognition are performed on the HE digital pathology slide to obtain slice-level quality attribute information and pixel-level instance attribute information.

[0081] In this embodiment, optionally, the specific implementation method for obtaining slice-level quality attribute information by performing slice-level quality detection on HE digital pathology slices may include the following steps: (1) Load the pyramid layer of HE digital pathology slides and extract low-magnification thumbnails from the pyramid layer. Input the low-magnification thumbnails into the tissue block detection model to identify the effective tissue regions in HE digital pathology slides and obtain the tissue region location information. Based on the tissue region location information, determine the cropped image containing the effective tissue region in the pyramid layer.

[0082] Specifically, the pyramid layer of the HE digital pathology slide is loaded, and a low-magnification thumbnail is extracted from it. This low-magnification thumbnail is then input into the tissue block detection model to identify the effective tissue regions in the HE digital pathology slide and obtain the location information of the tissue regions. Based on this location information, a cropped image containing the effective tissue regions is determined in the pyramid layer. This process involves quickly detecting and locating the tissue regions on the low-magnification thumbnail and then mapping it back to the pyramid layer for precise cropping to obtain a cropped image containing the effective tissue regions for use by the subsequent slice-level quality control classification model.

[0083] (2) Input the cropped image into the slice-level quality control classification model to obtain the regional multi-label classification results of the effective organization area.

[0084] Among them, the slice-level quality control classification model is a pre-trained multi-label classification model used to simultaneously distinguish multiple slice-level quality control categories for effective tissue regions in cropped images.

[0085] Among them, the regional multi-label classification result refers to the multi-label classification vector output by the slice-level quality control classification model for a cropped image of a single effective tissue region. The regional multi-label classification result is used to indicate whether each slice-level quality control category exists in the effective tissue region.

[0086] Specifically, the cropped image can be used as input to a pre-trained slice-level quality control classification model. This model performs multi-label classification on the effective tissue regions in the cropped image and determines whether the effective tissue region exists in each of the multiple slice-level quality control categories. Finally, the output is a multi-label classification result indicating whether the effective tissue region has a positive judgment in each slice-level quality control category.

[0087] (3) Make decisions based on the regional multi-label classification results corresponding to each effective tissue region to determine the slice-level quality attribute information of HE digital pathological slices.

[0088] Specifically, when an HE digital pathology slide contains multiple effective tissue regions, the multi-label classification results corresponding to each effective tissue region are obtained. For each slide-level quality control category, the presence or absence of the category is determined by combining the multi-label classification results of all effective tissue regions. Finally, the detection status of the slide-level quality control category at the slide level is determined. The slide-level detection status of all slide-level quality control categories together constitutes the slide-level quality attribute information of the HE digital pathology slide.

[0089] The above-described scheme, when performing slice-level quality inspection on HE digital pathology slides, reuses the same tissue region localization and cropping process as the scanning quality control module in slice-level quality inspection. This allows the slice-level quality control classification model to directly obtain cropped images focused on effective tissue regions as input, avoiding redundant calculations of invalid background regions across the entire slice. Simultaneously, by combining rapid tissue detection on low-magnification thumbnails with precise cropping of pyramid layers, the computational overhead of image loading and preprocessing is significantly reduced while ensuring the quality of the input images to the classification model. Furthermore, comprehensive decision-making based on the multi-label classification results of each effective tissue region integrates the quality information of multiple tissue blocks on the slice, avoiding slice-level misjudgments caused by local slide preparation abnormalities in a single tissue block. This improves the accuracy and robustness of slice-level quality attribute information determination.

[0090] In this embodiment, optionally, the specific implementation method for performing pixel-level instance segmentation and recognition on HE digital pathology slides to obtain pixel-level instance attribute information may include the following steps: (1) Load the pyramid layer of HE digital pathology slides and determine the target layer for instance segmentation inference based on the preset physical resolution parameters.

[0091] The preset physical resolution parameter refers to the micrometer-per-pixel value pre-set to determine the position of the pyramid layer in which instance segmentation inference should be performed. The target layer refers to the specific resolution layer selected from the pyramid layer according to the preset physical resolution parameter for performing the sliding window clipping operation.

[0092] Specifically, the multi-resolution image data structure of HE digital pathology slides in the storage medium can be read into memory as a whole. Based on this, the layer that is closest to the preset physical resolution parameter is selected from multiple resolution layers of the pyramid layer as the target layer for instance segmentation inference, so as to ensure that subsequent inference operations are performed at an appropriate image resolution.

[0093] (2) On the target layer, perform sliding window cropping with preset window size and preset overlap step size to generate cropping coordinates of multiple image blocks covering the entire area of ​​HE digital pathology slide.

[0094] The preset window size refers to the pre-defined size of the square image patch used for cropping on the target layer. The preset overlap step size refers to the pre-defined overlap pixel distance between adjacent cropping windows.

[0095] The cropping coordinates of multiple image blocks refer to the set of position coordinates of each image block on the target layer after the sliding window is cropped according to the preset window size and preset overlap step size.

[0096] Specifically, a preset window size can be used as the size of the cropping frame, and a preset overlap step size can be used as the movement step distance between adjacent cropping frames. The cropping frames are moved row by row and column by column on the target layer in the order from top to bottom and from left to right to traverse and crop the entire area of ​​the target layer, generating multiple cropping position coordinates that cover the entire area of ​​the HE digital pathology slide. These coordinates are the cropping coordinates of multiple image blocks.

[0097] (3) Based on each cropping coordinate, extract the corresponding image blocks from the target layer and construct the image queue to be inferred.

[0098] The image queue to be inferred refers to the sequence of image blocks that are waiting to be inferred by the input instance segmentation model after the corresponding image blocks are extracted from the target layer based on each cropping coordinate.

[0099] Specifically, using the cropping coordinates of the generated multiple image patches as position indices, pixel data within the regions defined by each coordinate is read on the target layer to form image patches that correspond one-to-one with the cropping coordinates. These image patches are then organized into a queue of images to be inferred according to the extraction order.

[0100] (4) Multiple instance segmentation model instances are called in parallel to perform instance segmentation inference on each image block in the image queue to be inferred, and the block-level instance segmentation results corresponding to each image block are obtained.

[0101] Among them, the block-level instance segmentation result refers to the category and location information of each quality defect instance contained in the image block after the instance segmentation model infers on a single image block.

[0102] Specifically, multiple inference instances loaded with the same instance segmentation model weights can be launched simultaneously. These inference instances take different image blocks from the image queue to be inferred and execute instance segmentation inference calculations synchronously. After inference, each image block outputs the category and location information of various quality defect instances contained within it, which serves as the block-level instance segmentation result corresponding to that image block.

[0103] (5) Map the block-level instance segmentation results to the highest resolution layer of the pyramid layer, and perform deduplication and merging processing on the mapped instances in the coordinate system of the highest resolution layer to obtain the pixel-level instance attribute information of HE digital pathology slides.

[0104] Specifically, the category and location information of each quality defect instance contained in the block-level instance segmentation result corresponding to each image block are transformed from the coordinate system of the target layer to the coordinate system of the highest resolution layer of the pyramid layer, so that the location information of all instances is unified in the same coordinate reference system. Then, the mapped instances are deduplicated and merged in the coordinate system of the highest resolution layer. That is, for the same instance that is repeatedly detected across the image block boundary, it is merged into a complete instance and redundant detection is eliminated, and finally the pixel-level instance attribute information of the HE digital pathology slide is obtained.

[0105] The above scheme, when performing pixel-level instance segmentation and recognition on HE digital pathology slides, reduces the amount of image data to be processed by determining the target layer based on preset physical resolution parameters while ensuring instance segmentation accuracy. By using a sliding window to crop and cover the entire slice area, it ensures that no quality defect instances that may appear at any location are missed. At the same time, it uses parallel invocation of multiple instance segmentation model instances to synchronously infer the image blocks in the image queue to be inferred, which significantly shortens the inference time of the entire slice. Furthermore, by mapping the block-level instance segmentation results to the highest resolution layer and performing deduplication and merging processing, it eliminates redundant information caused by repeated detection across window boundaries. Ultimately, it significantly improves the processing efficiency of pixel-level instance segmentation and recognition while ensuring detection recall and localization accuracy.

[0106] S280. Based on slice-level quality attribute information, pixel-level instance attribute information, and preset quality quantification rules, a comprehensive quality quantification value for HE digital pathology slices is obtained through comprehensive calculation.

[0107] The technical solution of this application embodiment, when performing scanning quality detection on HE digital pathology slides, loads a pyramid layer and extracts low-magnification thumbnails during the scanning quality control stage. At low resolution, it uses a tissue block detection model to quickly locate the effective tissue area, significantly reducing the processing overhead of invalid background areas across the entire slide. Then, based on the tissue area location information, it accurately crops image blocks containing the effective tissue area from the pyramid layer, enabling the scanning quality control classification model to directly focus on the tissue area for quality judgment, avoiding interference from background areas. At the same time, when the slide contains multiple tissue blocks, it uses a voting decision mechanism to fuse the recognition results of each area, effectively suppressing the impact of single-area recognition noise on the slide-level judgment results. Thus, while ensuring the accuracy of scanning quality control classification, it significantly reduces the computational overhead and inference time of this stage.

[0108] Example 3 Figure 3 This is a schematic diagram of a method for evaluating the quality of HE digital pathological slides provided in this application embodiment. Based on the foregoing embodiments, this embodiment provides a more detailed description of step S130. For specific implementation details, please refer to the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0109] like Figure 3 As shown, the method specifically includes the following steps: S310. Obtain the HE digital pathology slides to be evaluated, and perform scanning quality detection on the HE digital pathology slides to obtain scanning quality judgment information.

[0110] S320. When the HE digital pathology slide is determined to be a qualified slide based on the scanning quality judgment information, slice-level quality detection and pixel-level instance segmentation and recognition are performed on the HE digital pathology slide to obtain slice-level quality attribute information and pixel-level instance attribute information.

[0111] S330. Based on the chip-level quality attribute information, determine the first detection status corresponding to each chip-level quality control category.

[0112] Among them, each slice-level quality control category refers to a predefined set of quality control items used to evaluate the overall quality status of HE digital pathology slices, including tremor scar category, eosin staining depth category, eosin staining light category, hematoxylin staining depth category, hematoxylin staining light category, slice thickness category, and slice thickness unevenness category.

[0113] Among these, the "vibration mark" category refers to the quality control item category of regular texture defects formed on the tissue surface due to vibration of tools or equipment during slide preparation. The "deep eosin staining" category refers to the quality control item category of eosin staining exceeding the appropriate range, resulting in excessively dark cytoplasm. The "light eosin staining" category refers to the quality control item category of eosin staining below the appropriate range, resulting in excessively light cytoplasm. The "deep hematoxylin staining" category refers to the quality control item category of hematoxylin staining exceeding the appropriate range, resulting in excessively dark nuclei. The "light hematoxylin staining" category refers to the quality control item category of hematoxylin staining below the appropriate range, resulting in excessively light nuclei. The "thick slide" category refers to the quality control item category of slide thickness exceeding the appropriate range, resulting in tissue overlap. The "uneven slide thickness" category refers to the quality control item category of quality control item category of uneven thickness in different areas within the slide, resulting in significant differences in image clarity.

[0114] The first detection status refers to the status indicator data indicating whether each slice-level quality control category has been detected in HE digital pathology slides.

[0115] Specifically, the slice-level quality attribute information obtained during the slice-level quality detection stage can be used as the data source. This slice-level quality attribute information records the status data of whether each slice-level quality control category is detected at the HE digital pathology slide level. By parsing and extracting the slice-level quality attribute information, the corresponding detection status indicator value is obtained for each slice-level quality control category, thereby obtaining the first detection status corresponding to each slice-level quality control category. This first detection status is used to characterize whether the corresponding slice-level quality control category exists in the HE digital pathology slide.

[0116] S340. Based on the first detection status, determine the first quantitative deduction value corresponding to each slice-level quality control category from the slice-level quality quantification sub-rules.

[0117] Among them, the slice-level quality quantification sub-rule refers to the mapping rule part used to convert the detection status of each slice-level quality control category into the corresponding quantification deduction value.

[0118] The first quantitative deduction value refers to the quantitative deduction value assigned to each detected slice-level quality control category according to the slice-level quality quantification sub-rule.

[0119] Specifically, the first detection status corresponding to each slice-level quality control category is used as the matching basis. For each slice-level quality control category, the detection status of the category is compared with the deduction conditions preset in the slice-level quality quantification sub-rule. When the detection status indicates that the category has been detected, the quantification deduction value of the category is obtained from the slice-level quality quantification sub-rule as the first quantification deduction value. If it has not been detected, the corresponding first quantification deduction value is zero.

[0120] In this embodiment, optionally, the specific implementation method for determining the first quantitative deduction value corresponding to each slice-level quality control category from the slice-level quality quantification sub-rules based on the first detection status may include: Obtain the first detection status corresponding to each of the following categories: tremor mark category, eosin staining dark category, eosin staining light category, hematoxylin staining dark category, hematoxylin staining light category, section thickness category, and section thickness unevenness category; for any detected section-level quality control category, find the first quantitative deduction value corresponding to the current section-level quality control category from the section-level quality quantification sub-rules based on the first detection status corresponding to the current section-level quality control category.

[0121] Specifically, using slice-level quality attribute information as the data source, the detection status indication data at the slice level is extracted for each of the aforementioned slice-level quality control categories to obtain the status determination result of whether each category is detected. Based on this, for any current slice-level quality control category whose detection status indication is detected, the current slice-level quality control category is used as the query object, and the first detection status corresponding to the current slice-level quality control category is used as the condition for triggering the search operation. The quantification deduction value matching the current slice-level quality control category is searched from the pre-set slice-level quality quantification sub-rules, and this value is used as the first quantification deduction value corresponding to the current slice-level quality control category. For slice-level quality control categories that are not detected, no search operation is required, and their corresponding first quantification deduction value is considered to be zero.

[0122] S350. Based on the pixel-level instance attribute information, determine the second detection status and instance area ratio corresponding to each pixel-level quality control category.

[0123] Among them, each pixel-level quality control category refers to a predefined set of quality control item categories used to evaluate local quality defects in HE digital pathology slides, including knife mark category, crack category, fissure category, bubble category, fold category, and incomplete slide category.

[0124] Among these, the following quality control items are categorized as follows: Blade Mark Category: Linear mechanical damage defects formed on the tissue during the microtome preparation process due to blade breakage or improper cutting. Crack Category: Through-crack defects formed within the tissue during preparation or processing due to mechanical stress or improper dehydration. Fissure Category: Small gaps or separation surfaces appearing within the tissue in the section. Bubble Category: Circular or near-circular blank areas formed during the mounting process due to residual gas between the sealing agent and the tissue. Fold Category: Overlapping areas formed during the unfolding or mounting process due to tissue shrinkage or folding. Incomplete Section Category: Sections that fail to completely contain the entire area of ​​the tissue to be examined.

[0125] The second detection status refers to the status indicator data indicating whether each pixel-level quality control category has been detected in the HE digital pathology slide. The instance area ratio refers to the percentage of the effective tissue area occupied by the detected instances of each pixel-level quality control category in the HE digital pathology slide.

[0126] Specifically, the pixel-level instance attribute information obtained in the pixel-level instance segmentation and recognition stage can be used as the data source. From this information, the status indication data of whether each pixel-level quality control category has been detected can be extracted as the second detection status. At the same time, for each detected pixel-level quality control category, the area ratio data of each instance in the effective tissue area of ​​the HE digital pathology slide can be extracted from this information as the instance area ratio.

[0127] S360. Based on the second detection status and instance area ratio, determine the second quantization deduction value corresponding to each pixel-level quality control category from the pixel-level quality quantization sub-rules.

[0128] Among them, the pixel-level quality quantization sub-rule refers to the mapping rule part used to convert the detection status and instance area ratio of each pixel-level quality control category into the corresponding quantization deduction value.

[0129] The second quantization deduction value refers to the quantization deduction value assigned to each detected pixel-level quality control category according to the pixel-level quality quantization sub-rule.

[0130] Specifically, the second detection status and instance area ratio corresponding to each pixel-level quality control category can be used as the matching basis. For each detected pixel-level quality control category, the first-level quantization deduction value of the category is found from the pixel-level quality quantization sub-rules based on its second detection status as the basic quantization deduction value. Then, the corresponding second-level quantization deduction value is found from the pixel-level quality quantization sub-rules based on the preset ratio range to which its instance area ratio belongs as the step quantization deduction value. The two are added together to obtain the second quantization deduction value of the category.

[0131] In this embodiment, optionally, the specific implementation method for determining the second quantization deduction value corresponding to each pixel-level quality control category from the pixel-level quality quantization sub-rules based on the second detection status and instance area ratio may include the following steps: S3601. Based on the second detection status corresponding to each pixel-level quality control category, find the basic quantization deduction value corresponding to each detected pixel-level quality control category from the pixel-level quality quantization sub-rules.

[0132] Among them, the basic quantization deduction value refers to the existence deduction benchmark value set for each pixel-level quality control category in the pixel-level quality quantization sub-rules, which is used to reflect the basic score that should be deducted once the quality control category is detected.

[0133] Specifically, the second detection status corresponding to each pixel-level quality control category can be used as the filtering basis to identify the detected category from all pixel-level quality control categories. Then, for each of these detected pixel-level quality control categories, the category can be used as the query key value to obtain the existence deduction benchmark value that matches the category from the pixel-level quality quantification sub-rules, which is used as the basic quantification deduction value for the category.

[0134] S3602. Based on the percentage of instance area of ​​each detected pixel-level quality control category in the effective tissue area of ​​the HE digital pathology slide, find the corresponding step quantification deduction value for each detected pixel-level quality control category from the pixel-level quality quantification sub-rules.

[0135] Among them, the stepped quantization deduction value refers to the segmented incremental deduction value set according to the instance area ratio of each detected pixel-level quality control category in the pixel-level quality quantization sub-rule. It is used to reflect the cumulative score that needs to be deducted for the quality control category due to the expansion of the coverage area.

[0136] Specifically, the instance area ratio of each detected pixel-level quality control category can be used as the query basis. The instance area ratio is matched with multiple ratio intervals pre-set for each quality control category in the pixel-level quality quantization sub-rule to determine the ratio interval to which the instance area ratio belongs. Then, the quantization deduction value corresponding to the ratio interval is obtained as the step quantization deduction value of the detected pixel-level quality control category.

[0137] In this embodiment, optionally, the specific implementation of finding the corresponding step quantization deduction value for each detected pixel-level quality control category from the pixel-level quality quantization sub-rules may include: When the detected pixel-level quality control category is knife mark, crack, fissure or bubble, if the instance area percentage belongs to the first percentage range, a first-level deduction value is assigned; if the instance area percentage belongs to the second percentage range, a second-level deduction value is assigned; if the instance area percentage belongs to the third percentage range, a third-level deduction value is assigned. When the detected pixel-level quality control category is the folded category, if the instance area percentage belongs to the fourth percentage range, a fourth-level deduction value is assigned; if the instance area percentage belongs to the fifth percentage range, a fifth-level deduction value is assigned; if the instance area percentage belongs to the sixth percentage range, a sixth-level deduction value is assigned; if the instance area percentage belongs to the seventh percentage range, a seventh-level deduction value is assigned. When the detected pixel-level quality control category is the incomplete slice category, if the instance area percentage belongs to the eighth percentage range, an eighth-level deduction value is assigned; if the instance area percentage belongs to the ninth percentage range, a ninth-level deduction value is assigned; if the instance area percentage belongs to the tenth percentage range, a tenth-level deduction value is assigned.

[0138] In this embodiment, specific rules are established for assigning corresponding tiered deduction values ​​based on the preset ranges in which the instance area percentage falls for different pixel-level quality control categories. For the four categories of knife marks, cracks, fissures, and bubbles, they share a common set of percentage range divisions and corresponding deduction standards: when the instance area percentage is between 0% and 2%, a first-tier deduction value of 2 points is assigned; when the percentage is between 3% and 6%, a second-tier deduction value of 5 points is assigned; and when the percentage is greater than 6.1%, a third-tier deduction value of 10 points is assigned. For the folding category, its percentage range and corresponding deduction values ​​are set independently: when the instance area percentage is between 0% and 5%, a fourth-tier deduction value of 1 point is assigned; when the percentage is between 5% and 15%, a fifth-tier deduction value of 3 points is assigned; when the percentage is between 15% and 30%, a sixth-tier deduction value of 5 points is assigned; and when the percentage is greater than 30%, a seventh-tier deduction value of 10 points is assigned. For the incomplete slice category, the rules are as follows: when the instance area percentage is between 0% and 10%, an eighth-tier deduction value of 1 point is assigned; when the percentage is between 10% and 20%, a ninth-tier deduction value of 2 points is assigned; when the percentage is between 20% and 30%, a tenth-tier deduction value of 3 points is assigned; and when the percentage is greater than 30%, a deduction value of 10 points is assigned. These tiered deduction values ​​are added to the basic quantitative deduction value of each category (i.e., 2.5 points deducted upon presence) to form the second quantitative deduction value for that pixel-level quality control category.

[0139] S3603. Add the base quantization deduction value and the step quantization deduction value corresponding to the same pixel-level quality control category to obtain the second quantization deduction value corresponding to the current pixel-level quality control category.

[0140] Specifically, for each detected pixel-level quality control category, the basic quantization deduction value of the category is summed with the step quantization deduction value of the category. The value obtained by this summation is the second quantization deduction value corresponding to the pixel-level quality control category.

[0141] S370. Accumulate all first-quantification deduction values ​​and all second-quantification deduction values ​​to obtain the comprehensive quality quantification value of the HE digital pathology slide.

[0142] Specifically, all the first quantitative deduction values ​​determined for each slice-level quality control category and all the second quantitative deduction values ​​determined for each pixel-level quality control category can be summed together. The numerical result obtained by this summation is used as the comprehensive quality quantification value of the HE digital pathology slice, which is used to characterize the overall quality level of the slice.

[0143] The technical solution of this application embodiment, when determining the comprehensive quality quantification value of HE digital pathology slides, extracts quantitative criteria such as the detection status and area ratio of each quality control category from slide-level quality attribute information and pixel-level instance attribute information, respectively. Based on two sets of pre-set independent quantification sub-rules at the slide and pixel levels, the corresponding first and second quantification deduction values ​​are determined respectively, realizing differentiated quantitative evaluation of different types of quality control problems. Among them, the slide-level quality control category determines the deduction based on whether it is detected or not, while the pixel-level quality control category further determines the step deduction based on the area ratio. Finally, the comprehensive quality quantification value is obtained by accumulating all deduction values. This quantification method takes into account both the existence and severity of quality control problems, so that the comprehensive quality quantification value can comprehensively and accurately reflect the overall quality level of HE digital pathology slides.

[0144] Example 4 Figure 4 This is a schematic diagram of a HE digital pathology slide quality evaluation device provided in an embodiment of this application. The device includes: The scanning quality assessment module 410 is used to acquire the HE digital pathology slide to be evaluated, and to perform scanning quality detection on the HE digital pathology slide to obtain scanning quality assessment information. The slide quality detection module 420 is used to perform slide-level quality detection and pixel-level instance segmentation and recognition on the HE digital pathology slide when the HE digital pathology slide is determined to be a qualified slide based on the scan quality judgment information, so as to obtain slide-level quality attribute information and pixel-level instance attribute information. The slide quantification evaluation module 430 is used to perform comprehensive calculations based on the slide-level quality attribute information, the pixel-level instance attribute information, and preset quality quantification rules to obtain the comprehensive quality quantification value of the HE digital pathology slide. The quality control report generation module 440 is used to generate a structured quality control report based on the slice-level quality attribute information, the pixel-level instance attribute information, and the comprehensive quality quantification value; wherein, the structured quality control report includes detection status information of each quality control category, detailed information of each quantification deduction value, and improvement suggestions for the detected quality control category.

[0145] This application provides a HE digital pathology slide quality evaluation device. When used, the device acquires the HE digital pathology slide to be evaluated and performs scanning quality detection to obtain scanning quality judgment information. Then, if the HE digital pathology slide is determined to be a qualified slide based on the scanning quality judgment information, it performs slide-level quality detection and pixel-level instance segmentation and recognition to obtain slide-level quality attribute information and pixel-level instance attribute information. Based on the slide-level quality attribute information, pixel-level instance attribute information, and preset quality quantification rules, a comprehensive quality quantification value for the HE digital pathology slide is obtained. Finally, based on the slide-level quality attribute information, pixel-level instance attribute information, and comprehensive quality quantification value, a structured quality control report is generated. The structured quality control report includes detection status information for each quality control category, detailed information on each quantification deduction value, and improvement suggestions for the detected quality control categories. The technical solution of this application constructs a two-stage quality control process of rapid scanning quality detection and systematic pre-diagnosis assessment. First, it intercepts severely defective slices and displays early warning prompts. Then, it performs dual quality analysis at the slice level and pixel level on qualified slices simultaneously. Finally, it generates a structured quality control report containing quantitative deduction details and improvement suggestions to trace the cause of quality defects. This forms a closed-loop mechanism of detection, interception, tracing, and improvement, which solves the problem of low efficiency of existing high-magnification dense block processing methods, reduces computational overhead, and improves the overall efficiency of the quality control process and the ability to systematically improve the quality of the slides.

[0146] Optionally, based on the above-mentioned device, the HE digital pathology slide quality evaluation device further includes: a quality control early warning module, used to display quality control early warning information when the HE digital pathology slide is determined to be an unqualified slide based on the scan quality judgment information; wherein the quality control early warning information includes at least the scan quality judgment result, the defect type description, and the processing suggestion, the processing suggestion including a rescan suggestion and a re-slide preparation suggestion.

[0147] Based on the above-mentioned device, optionally, a scan quality determination module 410 is used to load the pyramid layer of the HE digital pathology slide and extract a low-magnification thumbnail from the pyramid layer; input the low-magnification thumbnail into a tissue block detection model to identify the effective tissue region in the HE digital pathology slide and obtain tissue region location information; based on the tissue region location information, determine a cropped image containing the effective tissue region in the pyramid layer; input the cropped image into a scan quality control classification model to obtain the region identification result of the effective tissue region; perform a voting decision based on the region identification result to determine the scan quality control category corresponding to the HE digital pathology slide, and generate the scan quality determination information according to the scan quality control category.

[0148] Based on the above-mentioned device, optionally, the scanning quality determination module 410 is further configured to map the tissue region location information to the highest resolution layer of the pyramid layer to obtain the original width and original height of the effective tissue region at the highest resolution; select a target layer from the pyramid layer based on the original width and the original height; wherein the resolution of the target layer makes the maximum side length of the effective tissue region on the target layer close to a preset cropping size; map the tissue region location information to the target layer, and crop an image block centered on the effective tissue region and completely containing the effective tissue region on the target layer as the cropped image.

[0149] Based on the above-mentioned device, optionally, the slice quality detection module 420 includes: The slice-level quality assessment unit is specifically used to load the pyramid layer of the HE digital pathology slide and extract a low-magnification thumbnail from the pyramid layer; input the low-magnification thumbnail into a tissue block detection model to identify the effective tissue region in the HE digital pathology slide and obtain the tissue region location information; based on the tissue region location information, determine a cropped image containing the effective tissue region in the pyramid layer; input the cropped image into a slice-level quality control classification model to obtain the region multi-label classification result of the effective tissue region; wherein, the region multi-label classification result is used to indicate whether the effective tissue region exists in each slice-level quality control category; and make a decision based on the region multi-label classification result corresponding to each effective tissue region to determine the slice-level quality attribute information of the HE digital pathology slide.

[0150] A pixel-level attribute evaluation unit is specifically used to load the pyramid layer of the HE digital pathology slide and determine the target layer for instance segmentation inference based on preset physical resolution parameters. On the target layer, a sliding window is cropped with a preset window size and preset overlap step size to generate cropping coordinates for multiple image blocks covering the entire area of ​​the HE digital pathology slide. Based on each cropping coordinate, the corresponding image blocks are extracted from the target layer to construct a queue of images to be inferred. Multiple instance segmentation model instances are called in parallel to perform instance segmentation inference on each image block in the queue of images to be inferred, obtaining block-level instance segmentation results for each image block. Each block-level instance segmentation result is mapped to the highest resolution layer of the pyramid layer, and the mapped instances are deduplicated and merged in the coordinate system of the highest resolution layer to obtain pixel-level instance attribute information of the HE digital pathology slide.

[0151] Based on the above-mentioned device, optionally, the slice quantification evaluation module 430 includes: The first state determination unit specifically determines the first detection state corresponding to each piece-level quality control category based on the piece-level quality attribute information. The first deduction value quantification unit, specifically based on the first detection state, determines the first quantification deduction value corresponding to each slice-level quality control category from the slice-level quality quantification sub-rules; The second state determination unit specifically determines the second detection state and instance area ratio corresponding to each pixel-level quality control category based on the pixel-level instance attribute information. The second deduction value quantization unit specifically determines the second quantization deduction value corresponding to each pixel-level quality control category from the pixel-level quality quantization sub-rule based on the second detection state and the instance area ratio. The quantification value determination unit specifically accumulates all first quantification deduction values ​​and all second quantification deduction values ​​to obtain the comprehensive quality quantification value of the HE digital pathology slide.

[0152] Based on the above-mentioned device, optionally, the slice-level quality control categories include vibratory mark category, eosin staining dark category, eosin staining light category, hematoxylin staining dark category, hematoxylin staining light category, slice thickness category, and slice thickness unevenness category. The first deduction value quantification unit is specifically used to obtain the first detection status corresponding to each of the vibratory mark category, the eosin staining dark category, the eosin staining light category, the hematoxylin staining dark category, the hematoxylin staining light category, the slice thickness category, and the slice thickness unevenness category; for any detected slice-level quality control category, according to the first detection status corresponding to the current slice-level quality control category, the first quantification deduction value corresponding to the current slice-level quality control category is searched from the slice-level quality quantification sub-rules.

[0153] Based on the aforementioned device, optionally, the pixel-level quality control categories include knife mark categories, crack categories, fissure categories, bubble categories, fold categories, and incomplete section categories. The second deduction value quantization unit is specifically used to: search for the basic quantization deduction value corresponding to each detected pixel-level quality control category from the pixel-level quality quantization sub-rules based on the second detection state corresponding to each pixel-level quality control category; search for the step quantization deduction value corresponding to each detected pixel-level quality control category from the pixel-level quality quantization sub-rules based on the instance area ratio of each detected pixel-level quality control category instance in the effective tissue area of ​​the HE digital pathology slide; and add the basic quantization deduction value corresponding to the same pixel-level quality control category to the step quantization deduction value to obtain the second quantization deduction value corresponding to the current pixel-level quality control category.

[0154] Based on the above device, optionally, the second deduction value quantization unit is more specifically used to assign a first-step deduction value if the detected pixel-level quality control category is a knife mark category, crack category, fissure category, or bubble category, and if the instance area percentage belongs to a first percentage range; assign a second-step deduction value if the instance area percentage belongs to a second percentage range; assign a third-step deduction value if the instance area percentage belongs to a third percentage range; and assign a fourth-step deduction value if the detected pixel-level quality control category is a fold category and the instance area percentage belongs to a fourth percentage range. Value; if the instance area percentage belongs to the fifth percentage interval, a fifth-level deduction value is assigned; if the instance area percentage belongs to the sixth percentage interval, a sixth-level deduction value is assigned; if the instance area percentage belongs to the seventh percentage interval, a seventh-level deduction value is assigned; when the detected pixel-level quality control category is the incomplete slice category, if the instance area percentage belongs to the eighth percentage interval, an eighth-level deduction value is assigned; if the instance area percentage belongs to the ninth percentage interval, a ninth-level deduction value is assigned; if the instance area percentage belongs to the tenth percentage interval, a tenth-level deduction value is assigned.

[0155] The HE digital pathology slide quality evaluation device provided in this application embodiment can execute the HE digital pathology slide quality evaluation method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of the method execution.

[0156] It is worth noting that the various units and modules included in the above system are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this application.

[0157] Example 5 Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 A block diagram is shown of an exemplary electronic device 50 suitable for implementing embodiments of the present application. Figure 5 The electronic device 50 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0158] like Figure 5 As shown, the electronic device 50 is represented in the form of a general-purpose computing device. The components of the electronic device 50 may include, but are not limited to: one or more processors or processing units 501, system memory 502, and bus 503 connecting different system components (including system memory 502 and processing unit 501).

[0159] Bus 503 represents one or more of several bus architectures, including memory buses or memory electronics, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0160] Electronic device 50 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 50, including volatile and non-volatile media, removable and non-removable media.

[0161] System memory 502 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 504 and / or cache memory 505. Electronic device 50 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 506 may be used to read and write non-removable, non-volatile magnetic media (… Figure 5 Not shown (usually referred to as a hard drive). Although Figure 5 As not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a floppy disk) and an optical disc drive for reading and writing to a removable non-volatile optical disc (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 503 via one or more data media interfaces. Memory 502 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0162] A program / utility 508 having a set (at least one) of program modules 507 may be stored, for example, in memory 502. Such program modules 507 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 507 typically perform the functions and / or methods described in the embodiments of this application.

[0163] Electronic device 50 can also communicate with one or more external devices 509 (e.g., keyboard, pointing device, display 510, etc.), and with one or more devices that enable a user to interact with electronic device 50, and / or with any device that enables electronic device 50 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 511. Furthermore, electronic device 50 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 512. As shown, network adapter 512 communicates with other modules of electronic device 50 via bus 503. It should be understood that, although... Figure 5 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 50, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0164] The processing unit 501 executes various functional applications and page processing by running programs stored in the system memory 502, such as implementing the HE digital pathology slide quality evaluation method provided in the embodiments of this application.

[0165] Example 6 This application also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a method for evaluating the quality of HE digital pathology slides, the method comprising: Obtain the HE digital pathology slides to be evaluated, and perform scanning quality detection on the HE digital pathology slides to obtain scanning quality judgment information; When the HE digital pathology slide is determined to be a qualified slide based on the scan quality judgment information, slice-level quality detection and pixel-level instance segmentation and recognition are performed on the HE digital pathology slide to obtain slice-level quality attribute information and pixel-level instance attribute information. The comprehensive quality quantification value of the HE digital pathology slide is obtained by comprehensively calculating based on the slice-level quality attribute information, the pixel-level instance attribute information, and the preset quality quantification rules. Based on the slice-level quality attribute information, the pixel-level instance attribute information, and the comprehensive quality quantification value, a structured quality control report is generated; wherein, the structured quality control report includes detection status information for each quality control category, detailed information on each quantification deduction value, and improvement suggestions for the detected quality control category.

[0166] The computer storage medium in this application embodiment can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0167] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0168] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0169] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0170] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A method for evaluating the quality of HE digital pathological slides, characterized in that, The method includes: Obtain the HE digital pathology slides to be evaluated, and perform scanning quality detection on the HE digital pathology slides to obtain scanning quality judgment information; When the HE digital pathology slide is determined to be a qualified slide based on the scan quality judgment information, slice-level quality detection and pixel-level instance segmentation and recognition are performed on the HE digital pathology slide to obtain slice-level quality attribute information and pixel-level instance attribute information. The comprehensive quality quantification value of the HE digital pathology slide is obtained by comprehensively calculating based on the slice-level quality attribute information, the pixel-level instance attribute information, and the preset quality quantification rules. Based on the slice-level quality attribute information, the pixel-level instance attribute information, and the comprehensive quality quantification value, a structured quality control report is generated; wherein, the structured quality control report includes detection status information for each quality control category, detailed information on each quantification deduction value, and improvement suggestions for the detected quality control category.

2. The method according to claim 1, characterized in that, The method further includes: When the HE digital pathology slide is determined to be an unqualified slide based on the scan quality assessment information, a quality control early warning message is displayed; wherein, the quality control early warning message includes at least the scan quality assessment result, the defect type description, and the processing suggestion, which includes a rescan suggestion and a re-slide preparation suggestion.

3. The method according to claim 1, characterized in that, The step of performing scan quality detection on the HE digital pathology slides to obtain scan quality judgment information includes: Load the pyramid layer of the HE digital pathology slide and extract low-magnification thumbnails from the pyramid layer; The low-magnification thumbnail is input into the tissue block detection model to identify the effective tissue region in the HE digital pathology section and obtain the tissue region location information. Based on the location information of the tissue region, a cropped image containing the effective tissue region is determined in the pyramid layer; The cropped image is input into the scanning quality control classification model to obtain the region recognition result of the effective tissue region; Based on the region identification results, a voting decision is made to determine the scanning quality control category corresponding to the HE digital pathology slide, and the scanning quality judgment information is generated according to the scanning quality control category.

4. The method according to claim 3, characterized in that, The step of determining a cropped image containing the effective tissue region in the pyramid layer based on the tissue region location information includes: The location information of the tissue region is mapped to the highest resolution layer of the pyramid layer to obtain the original width and original height of the effective tissue region at the highest resolution. Based on the original width and the original height, a target layer is selected from the pyramid layer; wherein the resolution of the target layer is such that the maximum side length of the effective organization area on the target layer is close to the preset clipping size; The location information of the tissue region is mapped to the target layer, and an image block centered on the effective tissue region and completely containing the effective tissue region is cropped from the target layer as the cropped image.

5. The method according to claim 1, characterized in that, The step of performing slide-level quality testing on the HE digital pathology slides to obtain slide-level quality attribute information includes: Load the pyramid layer of the HE digital pathology slide and extract low-magnification thumbnails from the pyramid layer; The low-magnification thumbnail is input into the tissue block detection model to identify the effective tissue region in the HE digital pathology section and obtain the tissue region location information. Based on the location information of the tissue region, a cropped image containing the effective tissue region is determined in the pyramid layer; The cropped image is input into a slice-level quality control classification model to obtain the region multi-label classification result of the effective tissue region; wherein, the region multi-label classification result is used to indicate whether each slice-level quality control category exists in the effective tissue region; Decisions are made based on the regional multi-label classification results corresponding to each effective tissue region to determine the slice-level quality attribute information of the HE digital pathology slide.

6. The method according to claim 1, characterized in that, The step of performing pixel-level instance segmentation and recognition on the HE digital pathology slides to obtain pixel-level instance attribute information includes: Load the pyramid layer of the HE digital pathology slides, and determine the target layer for instance segmentation inference based on preset physical resolution parameters; On the target layer, a sliding window is cropped with a preset window size and a preset overlap step to generate cropping coordinates for multiple image blocks covering the entire area of ​​the HE digital pathology slide; Based on each cropping coordinate, the corresponding image blocks are extracted from the target layer to construct a queue of images to be inferred. Multiple instance segmentation model instances are invoked in parallel to perform instance segmentation inference on each image block in the image queue to be inferred, so as to obtain the block-level instance segmentation result corresponding to each image block; The block-level instance segmentation results are mapped to the highest resolution layer of the pyramid layer, and the mapped instances are deduplicated and merged in the coordinate system of the highest resolution layer to obtain the pixel-level instance attribute information of the HE digital pathology slide.

7. The method according to claim 1, characterized in that, The preset quality quantification rules include slice-level quality quantification sub-rules and pixel-level quality quantification sub-rules. The comprehensive quality quantification value of the HE digital pathology slide is obtained by comprehensively calculating based on the slice-level quality attribute information, the pixel-level instance attribute information, and the preset quality quantification rules, including: Based on the chip-level quality attribute information, determine the first detection status corresponding to each chip-level quality control category; Based on the first detection status, the first quantitative deduction value corresponding to each slice-level quality control category is determined from the slice-level quality quantification sub-rules; Based on the pixel-level instance attribute information, determine the second detection status and instance area ratio corresponding to each pixel-level quality control category; Based on the second detection status and the instance area ratio, the second quantization deduction value corresponding to each pixel-level quality control category is determined from the pixel-level quality quantization sub-rule; The comprehensive quality quantification value of the HE digital pathology slide is obtained by summing all the first quantification deduction values ​​and all the second quantification deduction values.

8. The method according to claim 7, characterized in that, The slice-level quality control categories include tremor mark category, eosin staining dark category, eosin staining light category, hematoxylin staining dark category, hematoxylin staining light category, slice thickness category, and slice thickness unevenness category; The step of determining the first quantitative deduction value corresponding to each piece-level quality control category from the piece-level quality quantification sub-rules based on the first detection status includes: Obtain the first detection status corresponding to each of the following categories: vibratory scar category, eosin staining depth category, eosin staining light category, hematoxylin staining depth category, hematoxylin staining light category, section thickness category, and section thickness unevenness category; For any detected slice-level quality control category, based on the first detection status corresponding to the current slice-level quality control category, the first quantitative deduction value corresponding to the current slice-level quality control category is searched from the slice-level quality quantification sub-rules.

9. The method according to claim 7, characterized in that, The pixel-level quality control categories include knife mark category, crack category, fissure category, bubble category, fold category, and incomplete slice category; the determination of the second quantization deduction value corresponding to each pixel-level quality control category from the pixel-level quality quantization sub-rules based on the second detection status and the instance area ratio includes: Based on the second detection status corresponding to each pixel-level quality control category, the basic quantization deduction value corresponding to each detected pixel-level quality control category is found from the pixel-level quality quantization sub-rules. Based on the percentage of instance area of ​​each detected pixel-level quality control category in the effective tissue area of ​​the HE digital pathology slide, the corresponding step quantification deduction value for each detected pixel-level quality control category is found from the pixel-level quality quantification sub-rules. The basic quantization deduction value corresponding to the same pixel-level quality control category is added to the stepped quantization deduction value to obtain the second quantization deduction value corresponding to the current pixel-level quality control category.

10. The method according to claim 9, characterized in that, The percentage of instance area in the effective tissue region of the HE digital pathology slide based on the detected pixel-level quality control category is used to find the corresponding step-level quantization deduction value for each detected pixel-level quality control category from the pixel-level quality quantization sub-rules, including: When the detected pixel-level quality control category is knife mark, crack, fissure or bubble, if the instance area percentage belongs to the first percentage range, a first-level deduction value is assigned; if the instance area percentage belongs to the second percentage range, a second-level deduction value is assigned; if the instance area percentage belongs to the third percentage range, a third-level deduction value is assigned. When the detected pixel-level quality control category is the folded category, if the instance area percentage belongs to the fourth percentage range, a fourth-level deduction value is assigned; if the instance area percentage belongs to the fifth percentage range, a fifth-level deduction value is assigned; if the instance area percentage belongs to the sixth percentage range, a sixth-level deduction value is assigned; if the instance area percentage belongs to the seventh percentage range, a seventh-level deduction value is assigned. When the detected pixel-level quality control category is the incomplete slice category, if the instance area percentage belongs to the eighth percentage range, then an eighth-level deduction value is assigned; if the instance area percentage belongs to the ninth percentage range, then a ninth-level deduction value is assigned; if the instance area percentage belongs to the tenth percentage range, then a tenth-level deduction value is assigned.

11. A device for evaluating the quality of HE digital pathological slides, characterized in that, The device includes: The scanning quality assessment module is used to acquire the HE digital pathology slides to be evaluated, and to perform scanning quality detection on the HE digital pathology slides to obtain scanning quality assessment information. The slide quality detection module is used to perform slide-level quality detection and pixel-level instance segmentation and recognition on the HE digital pathology slide when the HE digital pathology slide is determined to be a qualified slide based on the scan quality judgment information, so as to obtain slide-level quality attribute information and pixel-level instance attribute information. The slide quantification evaluation module is used to perform comprehensive calculations based on the slide-level quality attribute information, the pixel-level instance attribute information, and preset quality quantification rules to obtain the comprehensive quality quantification value of the HE digital pathology slide; The quality control report generation module is used to generate a structured quality control report based on the slice-level quality attribute information, the pixel-level instance attribute information, and the comprehensive quality quantification value; wherein, the structured quality control report includes detection status information of each quality control category, detailed information of each quantification deduction value, and improvement suggestions for the detected quality control category.

Images