Multimodal pathological image processing method, detection system and kit thereof

Abstract

The application provides a multimodal pathological image processing method, a detection system and a kit thereof. The method comprises the following steps: acquiring three digital pathological images of adjacent tissue sections of the same intraoperative lymph node sample; processing a cytokeratin 19 rapid immunohistochemical staining image and an epithelial membrane antigen rapid immunohistochemical staining image respectively, identifying a first target area meeting the cytokeratin 19 specific coloring feature and a second target area meeting the epithelial membrane antigen specific coloring feature; fusing the first target area and the second target area to determine a joint positive area, and performing spatial correlation and feature fusion between the joint positive area and an intraoperative frozen HE staining image to generate a fusion analysis result. The application solves the problem that the existing intraoperative frozen HE staining image is difficult to accurately identify a low-volume metastasis due to imaging artifacts and similar cell features, and the postoperative immunohistochemical detection is accurate but time-consuming and cannot meet the intraoperative time requirement.

Description

Technical Field

[0001] This invention relates to the field of medical testing technology, and in particular to a multimodal pathological image processing method, a detection system, and a reagent kit thereof. Background Technology

[0002] Endometrial cancer is a common malignant tumor that seriously threatens women's health, and its lymph node metastasis status is a key factor affecting patient prognosis and clinical staging. In recent years, with the refinement of pathological diagnostic criteria, low-volume metastases in lymph nodes (including micrometastases and isolated tumor cells) have been proven to have significant prognostic value and directly affect surgical staging and subsequent treatment decisions. Therefore, in radical hysterectomy for endometrial cancer, achieving accurate and timely diagnosis of lymph node metastases, especially low-volume metastases, is crucial for guiding surgeons in determining the extent of lymph node dissection and achieving individualized surgical procedures.

[0003] Currently, intraoperative frozen section combined with hematoxylin and eosin (HE) staining is a routine method for rapidly assessing lymph node metastasis. However, this technique has inherent limitations when processing images related to low-volume metastases: First, the energy instruments used in surgery can easily cause "burns" to the edges of lymph node tissue, resulting in blurred cell structures and reduced tissue identification in the acquired HE-stained images; second, lymph node tissue is highly susceptible to artifacts such as "tissue folding" and "tissue detachment" during rapid freezing preparation, leading to missing information or deformation in key areas of the image, potentially obscuring the imaging area of ​​small lesions; finally, scattered or atypical tumor cells in HE-stained images have pixel features highly similar to the inherent histiocytes and follicular dendritic cells within the lymph node, making accurate differentiation through simple morphological image analysis difficult. These factors collectively result in a severe deficiency in the ability to identify low-volume metastases based on the interpretation of intraoperative frozen HE-stained images. Although routine immunohistochemical testing of paraffin sections after surgery can provide highly specific molecular marker information, the process involves multiple steps such as fixation, embedding, sectioning, and staining, which takes several hours to several days and cannot meet the time requirements for rapid intraoperative diagnosis.

[0004] Therefore, there is an urgent need for an image analysis and processing technology that can be integrated with intraoperative freezing procedures, completed in a very short time, and significantly improve the ability to identify low-volume transfers. Summary of the Invention

[0005] The purpose of this invention is to provide a multimodal pathological image processing method, detection system and reagent kit to solve the problems of existing intraoperative frozen HE staining images being difficult to accurately identify low-volume metastatic lesions due to imaging artifacts and similar cell characteristics, and postoperative immunohistochemical detection being accurate but too time-consuming to meet the requirements of intraoperative timeliness.

[0006] To achieve the above objectives, the present invention is implemented as follows: In a first aspect, the present invention provides a multimodal pathological image processing method for assisting in the analysis of lymph node samples from endometrial cancer, the method comprising: Three digital pathological images of adjacent tissue sections of the same intraoperative lymph node sample were obtained. The three images were an intraoperative frozen HE staining image, a cytokeratin 19 rapid immunohistochemical staining image, and an epithelial membrane antigen rapid immunohistochemical staining image, respectively. The rapid immunohistochemical staining images of cytokeratin 19 and the rapid immunohistochemical staining images of epithelial membrane antigens were processed to identify a first target region that conforms to the specific staining characteristics of cytokeratin 19 and a second target region that conforms to the specific staining characteristics of epithelial membrane antigens. The first target region and the second target region are fused to determine the joint positive region, and the joint positive region is spatially correlated and feature-fused with the intraoperative frozen HE staining image to generate fusion analysis results.

[0007] Furthermore, in the fusion analysis step, when tissue folds, burns, or tissue detachment areas are present in the intraoperative frozen HE-stained image, the fusion analysis result uses the annotation information of the combined positive areas as the primary analytical basis. Furthermore, when tissue folds, burns, or tissue detachment areas are present in the intraoperative frozen HE-stained image, the system automatically increases the weight of the immunohistochemical signal and uses this as the primary judgment criterion. Specifically, an image quality scoring module is set up; when the HE image quality score is below a threshold, the fusion analysis result uses the annotation information of the combined positive areas as the primary analytical basis, achieving dynamic weight switching.

[0008] Furthermore, the identification of the first target region and the second target region involves inputting the rapid immunohistochemical staining images of cytokeratin 19 and epithelial membrane antigen into a trained deep learning model, which then outputs a first mask image corresponding to the first target region and a second mask image corresponding to the second target region.

[0009] Furthermore, the spatial association is achieved using at least one registration method based on slice numbering order, tissue contour matching, or manual marker points.

[0010] Secondly, a kit for rapid immunohistochemical diagnosis of lymph node metastasis in endometrial cancer is provided, the kit comprising: The container, and the following components individually packaged in the container: Enzyme-labeled antibody against cytokeratin 19; Enzyme-labeled antibody against epithelial membrane antigen; DAB colorimetric solution; Buffer and blocking agent for rapid immunohistochemical staining of frozen sections; The enzyme-labeled antibody against cytokeratin 19 and the enzyme-labeled antibody against epithelial membrane antigen are used to simultaneously stain adjacent tissue sections of the same lymph node sample for joint interpretation with intraoperative frozen HE-stained sections.

[0011] Thirdly, a rapid intraoperative immunohistochemical detection system for lymph node metastasis in endometrial cancer is provided, including: The slide preparation module is used to prepare adjacent intraoperative frozen HE-stained sections, cytokeratin 19 rapid immunohistochemical staining sections, and epithelial membrane antigen rapid immunohistochemical staining sections from the same lymph node sample obtained during surgery. The image acquisition module is used to acquire digital pathological images of the three slides; The image analysis module is used to automatically identify and label the brownish-yellow cytoplasmic region of cytokeratin 19 and the brownish-yellow cell membrane region of epithelial membrane antigen in the digital pathological image based on a deep learning algorithm; and, The report generation module is used to integrate the results of automatic annotation with the image features of the intraoperative frozen HE-stained sections to generate an auxiliary diagnostic report, which indicates whether metastatic lesions exist and their types.

[0012] Fourthly, the present invention also provides a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method described in the first aspect.

[0013] The beneficial effects of this invention are as follows: The multimodal pathological image processing method of the present invention acquires intraoperative frozen HE staining images, cytokeratin 19 rapid immunohistochemical staining images, and epithelial membrane antigen rapid immunohistochemical staining images of adjacent tissue sections of the same lymph node sample. It then identifies a first target region and a second target region that conform to their respective specific staining characteristics from the latter two types of images, and fuses them to determine a joint positive region. This joint positive region is then spatially correlated and feature-fused with the intraoperative frozen HE staining image to generate a fusion analysis result. This solves the technical problem that intraoperative frozen HE staining images are difficult to accurately identify low-volume metastatic lesions due to imaging artifacts and similar cell characteristics, while accurate postoperative immunohistochemical detection is too time-consuming to meet the timeliness requirements of intraoperative procedures. Specifically, this invention introduces rapid immunohistochemical staining images of two complementary epithelial markers, cytokeratin 19 and epithelial membrane antigen, to provide specific signals at the molecular level that are not affected by imaging artifacts such as tissue folding and burns. Then, by fusing the positive regions of the two markers, it effectively avoids missed detections that may occur due to the heterogeneity of tumor cell antigen expression of a single marker. Finally, by spatially associating and feature-fusing the fused combined positive region with intraoperative frozen HE staining images, it not only retains the morphological information of traditional HE images, but also highlights suspicious lesion areas in an objective and quantitative manner, providing multimodal fused and feature-enhanced image data for subsequent auxiliary analysis. Thus, it significantly improves the ability to identify low-volume metastases while maintaining intraoperative speed.

[0014] This invention achieves a balance between high-precision multimodal image analysis and intraoperative timeliness by strictly controlling the total time consumption of the entire image processing workflow to within 33 minutes. Specifically, after acquiring three digital pathology images, the invention sequentially completes target region identification, joint positive region fusion, and spatial correlation and feature fusion with HE images. The entire process undergoes targeted lightweight design and parallel processing to ensure that the entire workflow time window from image acquisition to fusion analysis result output matches the conventional time limit for waiting for frozen pathology reports during clinical surgery. Therefore, this method significantly improves the ability to identify low-volume metastases without increasing the waiting burden for intraoperative decision-making, enabling high-precision multimodal image analysis results to be truly applied to intraoperative auxiliary analysis scenarios, thereby solving the technical problem of existing solutions where accurate detection is too time-consuming and cannot meet intraoperative timeliness requirements.

[0015] Furthermore, this invention automatically calculates the geometric parameters (such as maximum diameter and pixel area) of the combined positive region after the fusion analysis step, and generates corresponding metastatic lesion type classification suggestions (macrometastasis, micrometastasis, or isolated tumor cells) based on preset thresholds (such as 2 mm, 0.2 mm, and 200 cells). These quantitative information and classification suggestions are presented in a structured form, providing direct and clear auxiliary basis for subsequent clinical decision-making. This not only solves the problem of accurately identifying low-volume metastatic lesions in intraoperative frozen HE images due to imaging artifacts and similar cell characteristics, but also further transforms the identification results into quantitative parameters that conform to clinical staging standards, reducing diagnostic discrepancies caused by inconsistent interpretation standards and improving the consistency and clinical reference value of the auxiliary analysis results. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of a multimodal pathological image processing method for assisting in the analysis of lymph node samples of endometrial cancer according to an embodiment of the present invention; Figure 2 This is an example image of micrometastases detected by intraoperative frozen section HE combined with rapid immunohistochemistry based on an embodiment of the present invention; Figure 3 An example image of isolated tumor cells detected by intraoperative frozen section HE combined with rapid immunohistochemistry based on an embodiment of the present invention; Figure 4 This is a schematic flowchart of a multimodal pathological image processing method for assisting in the analysis of lymph node samples of endometrial cancer according to another embodiment of the present invention; Figure 5 Example image of lymph node metastases observed based on traditional intraoperative frozen section HE staining; Figure 6 A schematic structural block diagram of an intraoperative rapid immunohistochemical detection system for lymph node metastasis of endometrial cancer according to an embodiment of the present invention; Figure 7 This is a topology diagram of a computer-readable storage medium disclosed in this invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, it should be noted that these embodiments are not intended to limit the present invention. Equivalent changes or substitutions in function, method, or structure made by those skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0018] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0019] Example 1: like Figure 1As shown, this embodiment provides a multimodal pathological image processing method (hereinafter referred to as "image processing method" or "method") for assisting in the analysis of endometrial cancer lymph node samples. This method is applied to an intraoperative rapid immunohistochemical detection system for endometrial cancer lymph node metastasis, a kit for intraoperative rapid immunohistochemical diagnosis of endometrial cancer lymph node metastasis, or a multimodal pathological image processing system for assisting in the analysis of endometrial cancer lymph node samples. The method includes: Step 102. Acquire three digital pathological images of adjacent tissue sections from the same intraoperative lymph node sample. These three images are an intraoperative frozen section with HE staining, a cytokeratin 19 rapid immunohistochemical staining image, and an epithelial membrane antigen rapid immunohistochemical staining image, respectively. That is, by preparing adjacent tissue sections of the same lymph node sample into three different stained sections and digitizing them, a high spatial correspondence of the tissue structures reflected in the three images is ensured, laying a data foundation for subsequent multimodal image fusion and joint analysis. Thus, the three images originate from adjacent sections of the same lymph node sample, enabling precise spatial correspondence between the morphological and molecular marker information presented by different staining methods, avoiding registration errors caused by differences in section position, thereby ensuring the accuracy and reliability of subsequent feature fusion. When spatially associating the joint positive region with the intraoperative frozen section with HE staining, at least one of the following methods is used for registration: alignment based on section number order, automatic matching based on tissue contour, or alignment based on manual marker points. These methods do not rely on complex algorithms to achieve precise spatial correspondence of multimodal images.

[0020] The intraoperative frozen section HE-stained images, cytokeratin 19 rapid immunohistochemical staining images, and epithelial membrane antigen rapid immunohistochemical staining images were derived from digital scanning results of intraoperative frozen HE-stained sections, cytokeratin 19 rapid immunohistochemical staining sections, and epithelial membrane antigen rapid immunohistochemical staining sections prepared simultaneously from adjacent tissue sections of the same lymph node sample. This definition clarifies that the physical sections corresponding to the three images were prepared synchronously within the same time window using the same frozen section parameters. This synchronous preparation process ensures a high degree of consistency among the three sections in terms of tissue morphology preservation, section thickness, and staining background, minimizing image variables introduced by batch differences in slide preparation. This allows subsequent image analysis to be performed under uniform imaging conditions, improving the stability and comparability of multimodal image fusion results.

[0021] Furthermore, the rapid immunohistochemical staining images of cytokeratin 19 and epithelial membrane antigen were obtained from slides prepared using a rapid immunohistochemical staining process that employed enzyme-labeled primary antibodies followed by incubation with an inhibitor, primary antibody, enhancement solution, and DAB development. This process utilizes direct labeling with enzyme-labeled primary antibodies, eliminating the cumbersome steps of secondary antibody incubation and multiple washing in traditional immunohistochemistry. Simultaneously, the inhibitor reduces background interference, and the enhancement solution amplifies signal intensity, allowing the entire staining process to be completed within minutes. This rapid staining process, while ensuring antibody specific binding and clear color development, compresses the traditional immunohistochemical process, which takes several hours, to the minute level. It enables molecular-level specific labeling to be completed simultaneously with intraoperative frozen HE staining, thus providing highly specific molecular marker information for image analysis without extending the overall detection time. The rapid immunohistochemical staining process includes: The slides were fixed using 10% neutral formalin buffer for 30 to 60 seconds. This fixation step is crucial in rapid immunohistochemical staining, as it quickly preserves tissue antigenicity, fixes cell morphology, and prevents antigen diffusion or degradation. Using 10% neutral formalin buffer as the fixative and strictly controlling the fixation time to 30 to 60 seconds strikes a balance between rapid fixation and antigen preservation. Experimental results have shown that this fixation condition effectively fixes epithelial membrane antigens and cytokeratin 19 antigens in a very short time, while avoiding excessively long fixation times that could inhibit subsequent antibody binding and staining. This ensures signal intensity and background clarity in the stained image, providing high-quality image input for subsequent target region identification.

[0022] Step 104. Process the rapid immunohistochemical staining images of cytokeratin 19 and the rapid immunohistochemical staining images of epithelial membrane antigens respectively to identify the first target region that matches the specific staining characteristics of cytokeratin 19 and the second target region that matches the specific staining characteristics of epithelial membrane antigens.

[0023] It should be understood that the identification process in this embodiment is based on the specific localization patterns of two antibodies in tumor cells: cytokeratin 19 positivity is manifested as brownish-yellow staining of the cytoplasm, and epithelial membrane antigen positivity is manifested as brownish-yellow staining of the cell membrane. By identifying these two staining features respectively, it is possible to accurately locate potential tumor cells and their distribution areas at the molecular level. This transforms the subjective experience of pathologists in interpreting cytoplasmic and cell membrane brownish-yellow staining into a quantifiable image feature recognition process, thereby achieving objective and automated localization of suspicious areas. This effectively avoids missed detections due to human fatigue, experience differences, or visual omissions, and provides a precise spatial coordinate basis for subsequent multimodal fusion analysis.

[0024] Specifically, identifying the first and second target regions involves inputting rapid immunohistochemical staining images of cytokeratin 19 and epithelial membrane antigens into a trained deep learning model. The model then outputs a first mask image corresponding to the first target region and a second mask image corresponding to the second target region. This deep learning model is pre-trained with a large number of labeled immunohistochemical staining images, learning and mastering typical image feature patterns of cytokeratin 19 cytoplasmic staining and epithelial membrane antigen cell membrane staining, including color thresholds, morphological structures, texture features, and rules for distinguishing background interference. In practical applications, the model outputs mask images with pixel-level precision, clearly defining the boundaries of each positive region. In this way, the deep learning model can automatically capture subtle feature differences that are difficult for the human eye to quantify, and can still stably identify positive signals in complex backgrounds (such as tissue folds and burn areas), significantly improving the accuracy and anti-interference ability of target region identification. Simultaneously, the output in mask image form retains complete spatial morphological information, providing accurate segmentation data for subsequent joint positive region fusion and spatial registration with HE images, effectively supporting the overall performance of multimodal image analysis.

[0025] In another embodiment, the steps for identifying the first and second target regions do not employ a deep learning model. Instead, they are based on a rule-based algorithm using preset color thresholds and connected component analysis. Specifically, this includes: extracting brownish-yellow pixel regions from the cytokeratin 19 image (segmented using HSV or RGB color space thresholds); extracting cell membrane ring structure features from the epithelial membrane antigen image; and determining positive regions using rules such as chromaticity range, connected component area thresholds (excluding noise), roundness, or cell membrane integrity. Subsequently, logical operations (intersection or union) are performed on the regions identified by the two rules to determine the joint positive region.

[0026] Step 106. Merge the first target region and the second target region to determine the joint positive region, and spatially correlate and feature-fuse the joint positive region with the intraoperative frozen HE staining image to generate the fusion analysis result.

[0027] The first target region originates from the cytoplasmic positive staining of cytokeratin 19, and the second target region originates from the cell membrane positive staining of epithelial membrane antigens. The combined positive region formed by the fusion of these two regions integrates the molecular information of two complementary epithelial markers, providing a more comprehensive reflection of the presence and distribution of tumor cells. Subsequently, this combined positive region is spatially correlated with intraoperative frozen HE-stained images to precisely align the molecular-level positive signals with the tissue morphological background. Furthermore, feature fusion is used to superimpose the information from both modalities. Thus, on the one hand, dual-marker fusion effectively overcomes the potential for missed detections due to the heterogeneity of tumor cell antigen expression caused by single-marker fusion (e.g., ...). Figure 3 Three positive cell nests were detected in b of EMA. Figure 3 (The complementary phenomenon of CK19 detecting only one positive cell nest in c) On the other hand, spatial correlation and fusion of molecular marker information and HE morphological information not only preserves the complete presentation of tissue structure in traditional HE images, but also highlights suspicious areas in a visual way, providing a more informative and feature-rich multimodal fusion image for subsequent analysis, thereby solving the problem that simple HE images are difficult to accurately identify due to imaging artifacts and similar cell features in the existing technology.

[0028] Furthermore, in the fusion analysis step, when tissue folds, burns, or tissue detachment areas are present in the intraoperative frozen HE-stained images, the fusion analysis results are primarily based on the annotation information of the combined positive areas. This rule is designed based on in-depth observation of actual clinical cases: Figure 2 Among the multiple missed diagnosis cases shown (such as...) Figure 2 Burns of d, e, and f in the text Figure 2 The tissue folding of g, h, and i in the text Figure 2 j, k, l and Figure 3 In HE-stained images, due to preparation artifacts, key areas of cellular structure become blurred, information is lost, or deformation occurs, making morphological identification extremely difficult or even impossible. However, in immunohistochemical staining images of adjacent sections of the same lymph node sample, the specific staining signals of cytokeratin 19 and epithelial membrane antigens are not affected by the same artifacts, and the presence of tumor cells can still be clearly shown. Therefore, this processing rule introduces a confidence assessment mechanism for multimodal image fusion analysis: when the quality of HE image data decreases due to preparation artifacts, the system automatically shifts the focus of analysis to the immunohistochemical image signals that are less affected by artifacts, using the annotation information of the combined positive areas as the main basis. This improves the robustness and reliability of the entire image analysis method under complex clinical sample conditions, ensuring that valuable auxiliary analysis results can still be provided based on molecular marker information even when the HE image quality is poor, thus effectively overcoming the inherent limitation of traditional methods that are highly dependent on preparation quality.

[0029] In another implementation, a suspected metastatic area is identified when either marker (cytokeratin 19 or epithelial membrane antigen) reaches a preset positive threshold, without requiring double positivity. Specifically, a first trigger condition is set when the area of ​​positive cytokeratin 19 is greater than a preset area threshold A, and a second trigger condition is when the number of positive epithelial membrane antigen cells is greater than a preset cell number threshold B; a "suspected metastatic indication" is output when either condition is met.

[0030] like Figure 4 As shown, the method in this embodiment further includes: Step 108. Output the fusion analysis results, which include at least a fusion visualization image containing joint positive region annotation information and quantitative analysis information calculated based on the joint positive regions.

[0031] Quantitative analysis information includes the maximum diameter of the combined positive region, pixel area, or classification suggestions based on preset thresholds to categorize metastatic lesions. The maximum diameter reflects the linear size of the positive region, the pixel area reflects its overall range, and the classification suggestions categorize metastatic lesions according to preset thresholds. This embodiment can convert the maximum diameter of metastatic lesions manually measured by pathologists and the number of cells manually determined into quantitative parameters automatically calculated and output by the system, achieving standardization and automation of the assessment process. The automatic calculation of the maximum diameter and pixel area eliminates human measurement errors, while the classification suggestions directly connect the quantitative results with clinical staging standards, providing structured and interpretable auxiliary information for subsequent decision-making.

[0032] Fusion visualization overlays combined positive areas onto intraoperative frozen HE-stained images, presenting molecular marker signals and tissue morphology background in a single view. Quantitative analysis transforms the geometric and morphological characteristics of combined positive areas into objective numerical values. Fusion visualization visually displays the spatial location and distribution of suspicious lesions, reducing the cognitive burden on the interpreter of spatial comparison and memorization of multiple slides. Quantitative analysis provides repeatable and comparable objective indicators. The combination of these two approaches ensures both intuitiveness and accuracy in the analysis results, effectively supporting subsequent clinical decision support.

[0033] In the above embodiments, the classification suggestion information for classifying metastatic lesion types based on preset thresholds includes: When the maximum diameter of the combined positive region is greater than 2 mm, a macro-metastasis classification suggestion is output; when the maximum diameter of the combined positive region is between 0.2 mm and 2 mm and / or the number of tumor cells within the region is greater than 200, a micro-metastasis classification suggestion is output; when the maximum diameter of the combined positive region is no greater than 0.2 mm and the number of tumor cells within the region is no greater than 200, an isolated tumor cell classification suggestion is output. That is, clinically accepted pathological staging standards are directly embedded into the output layer of the image analysis algorithm, ensuring seamless integration of the system's output classification suggestions with clinical diagnostic and treatment guidelines. Figure 2 Case B12 (maximum diameter 0.5 mm) in category a can be automatically classified as a micrometastasis recommendation. Figure 3 Case B17 (maximum diameter <0.2mm, cell count <200) can be automatically classified as an isolated tumor cell, providing a direct and clear reference for subsequent clinical decision-making.

[0034] In the above embodiment, the total processing time from acquiring three digital pathology images to outputting the fusion analysis result does not exceed 33 minutes. This time window ensures that the method of this embodiment can complete the entire image processing and analysis process within the normal time limit of waiting for the pathology report during surgery. Converting the total physical process time from sample reception to report issuance into the total algorithmic processing time from image acquisition to result output demonstrates that the method of this embodiment satisfies the requirements of high-precision recognition without sacrificing the timeliness of intraoperative operations, truly achieving a balance between speed and accuracy, and solving the defect that excessively long accurate detection time cannot meet the requirements of intraoperative timeliness.

[0035] In one embodiment of the present invention, time constraints are set for each processing step: slide preparation should not exceed 10 minutes, rapid immunohistochemical staining should not exceed 10 minutes, image analysis should not exceed 5 minutes, and report generation should not exceed 5 minutes. If any step exceeds the time limit, it will automatically downgrade to HE staining combined with single biomarker analysis mode to ensure that the total time is controllable.

[0036] In a specific embodiment of the present invention, the aforementioned multimodal pathological image processing method was applied to real clinical samples to verify its effectiveness in identifying lymph node metastases in endometrial cancer. This embodiment included 18 patients with endometrial cancer undergoing surgical treatment, aged between 47 and 78 years. All patients met the following criteria: (1) pathologically confirmed endometrial cancer; (2) planned to undergo systematic lymph node dissection or sentinel lymph node biopsy; and (3) intraoperative rapid pathological evaluation of lymph nodes was required. A total of 104 lymph nodes were acquired for image data acquisition in this embodiment. The specific implementation process is as follows: For each lymph node sample, after bisecting along its long axis, half of the tissue is used to prepare three adjacent consecutive frozen sections. These sections are then subjected to intraoperative frozen section hematoxylin and eosin (HE) staining, rapid immunohistochemical staining for cytokeratin 19, and rapid immunohistochemical staining for epithelial membrane antigens, respectively. The three stained physical sections are then digitized using a whole-slide scanner to obtain three corresponding digital pathological images—namely, the intraoperative frozen section hematoxylin and eosin (HE) staining image, the cytokeratin 19 rapid immunohistochemical staining image, and the epithelial membrane antigen rapid immunohistochemical staining image. These three images constitute a set of multimodal image data, which serves as input for subsequent image processing steps.

[0037] The specific implementation process is as follows: 1. Sample pretreatment The lymph nodes obtained during the surgery were immediately sent for testing. The surgeon who harvested the tissue split the lymph node along its long axis, with one half of the tissue used for rapid intraoperative testing in this embodiment, and the other half retained for routine postoperative pathological examination (as the gold standard for diagnosis).

[0038] 2. Intraoperative frozen section preparation Tissue blocks for intraoperative testing were placed on a sample holder containing OCT embedding medium and rapidly frozen in a cryostat (-20°C to -26°C). The frozen tissue blocks were then cut into 4 μm thick serial sections and attached to glass slides. Three adjacent serial sections were prepared simultaneously for each sample: the first for intraoperative frozen HE staining, the second for rapid immunohistochemical staining of cytokeratin 19, and the third for rapid immunohistochemical staining of epithelial membrane antigens.

[0039] 3. Intraoperative frozen section HE staining The sections were fixed in fixative for 1 minute, rinsed slightly with tap water, and then stained with hematoxylin (1-2 minutes), blue, and eosin (2-5 seconds). They were then dehydrated with gradient alcohols, cleared with xylene, and finally mounted with neutral resin.

[0040] 4. Rapid immunohistochemical staining Two slides used for rapid immunohistochemistry were fixed in 10% neutral formalin buffer for 30-60 seconds and rinsed with running water. The slides were then placed in a humidified chamber preheated to 40-42°C, and the following reagents were added and incubated sequentially: blocking agent (30 seconds), enzyme-labeled primary antibody (for CK19 or EMA, 3-5 minutes), enhancement solution (3-5 minutes), and DAB chromogenic solution (2-3 minutes). Finally, hematoxylin was used for counterstaining (10-30 seconds), followed by rinsing with TBS and mounting. The enzyme-labeled primary antibodies used included anti-epithelial membrane antigen antibody and anti-cytokeratin 19 antibody. After staining, the three physical slides were digitized using a whole-slide scanner to obtain three corresponding digital pathological images—an intraoperative frozen HE staining image, a cytokeratin 19 rapid immunohistochemical staining image, and an epithelial membrane antigen rapid immunohistochemical staining image—forming a set of multimodal image data as input for subsequent image processing steps.

[0041] 5. Image Data Analysis and Result Generation Based on the aforementioned multimodal image data, image analysis is performed according to the following preset rules to generate analysis results: (1) Positive signal localization rules: Cytokeratin 19 positive is manifested as brownish-yellow staining of cytoplasm; epithelial membrane antigen positive is manifested as brownish-yellow staining of cell membrane.

[0042] (2) Classification rules for metastatic lesions: (a) Macro-transfer: The maximum diameter of the largest transfer lesion is >2 mm.

[0043] (b) Micrometastases: The largest diameter of the largest metastatic lesion is 0.2 mm to 2 mm, and / or the number of tumor cells is >200.

[0044] (c) Isolated tumor cells: The largest diameter of the largest metastatic lesion is ≤0.2mm and the number of tumor cells is ≤200.

[0045] Based on the above analysis, structured image analysis results are generated. The entire process, from receiving the data to completing the analysis, takes 20-33 minutes (median time 25 minutes), and the analysis results are used to assist in subsequent decision-making.

[0046] 6. Compare with the gold standard for testing. The remaining half of the lymph node tissue was used for routine postoperative pathological "hyperstaging" examination. Paraffin sections were prepared and subjected to HE and broad-spectrum cytokeratin immunohistochemical staining. The results served as the final diagnostic gold standard for evaluating the accuracy of the method in this embodiment of the invention. Specifically, after preparing paraffin blocks from the remaining contralateral half of the lymph node tissue, serial sections were prepared at 200-micrometer intervals, with a total of 5 sections taken from each block. One section underwent broad-spectrum cytokeratin immunohistochemical staining, while the other four sections underwent routine HE staining. Pathologists performed comprehensive microscopic examination of all sections, and these results served as the final diagnostic basis.

[0047]

[0048] (1) Analysis of diagnostic accuracy Using postoperative routine pathological hyperstaging results as the gold standard, as shown in Table 1, the diagnostic performance of the method (multimodal image analysis) in this embodiment is as follows: ① For all metastatic lesions: the total number detected was highly consistent with the gold standard.

[0049] ② For low-volume metastatic lesions (micrometastases and isolated tumor cells): the recognition sensitivity reached 100% (7 / 7), and the specificity was 100%, while the sensitivity of traditional intraoperative frozen section HE staining for low-volume metastases was only 28.57% (2 / 7). The formulas for calculating the sensitivity and specificity of intraoperative frozen section rapid pathological diagnosis before and after combining IR-IHC detection results are as follows: Sensitivity = (Number of true positive cases / Total number of positive cases) × 100% Specificity = (Number of true negative cases / Total number of negative cases) × 100%.

[0050] (2) Detection effect on difficult cases ① Results of intraoperative frozen section HE imaging alone: ​​Metastatic lesions were identified in 21 lymph node images. Among them, 19 were macrometastases, in which the tumor tissue was irregularly distributed in nest-like patterns, destroying the original structure of the lymph nodes, and the maximum diameter of the metastatic lesions was >2mm (e.g., Figure 5 (As shown in a). Two micrometastases were observed. In the image, the tumor tissue was scattered in small nests under the lymph node capsule. The lymph node structure was not significantly destroyed. The maximum diameter of the metastatic lesions was no greater than 2 mm and greater than 0.2 mm (e.g., ...). Figure 5 (as shown in b in the text).

[0051] ② Results of multimodal image analysis using the method of this embodiment: When analyzing the aforementioned multimodal image data using the method of this embodiment, a total of 26 lymph node images were identified with clear brown-yellow staining of cytokeratin 19 (cytoplasm) and / or epithelial membrane antigen (cell membrane), indicating metastatic lesions. The method of this embodiment successfully identified all 4 micrometastases and 1 isolated tumor cell lesion that were missed by intraoperative frozen section HE analysis alone. Specific examples are as follows: Case B12 (micrometastasis): In the HE section, tumor cells were arranged in elongated, glandular patterns within the lymphoid tissue. The maximum diameter of the metastatic lesion was approximately 0.5 mm (e.g., Figure 2 (as shown in a); In the analysis results, EMA (as shown in a) in IR-IHC slices Figure 2 (as shown in b) and CK19 (as shown in b) Figure 2 (as shown in c) all show clear brown-colored areas.

[0052] Case B13 (micrometastases with burns): HE images show suspicious lesions with a maximum diameter of approximately 0.2 mm in the metastatic lesions due to interference from burns (e.g., Figure 2 (as shown in d); in the analysis results, in the IR-IHC labeled EMA (such as... Figure 2 (as shown in e) and CK19 (as shown in e) Figure 2 In the image shown in f), in addition to the tumor cell nests that can be observed in the HE image, brown-stained tumor cell regions were also identified in the tissue with severe burns on the side, and the maximum diameter of the metastatic lesions increased to about 0.6 mm.

[0053] Case B13 (micrometastasis with tissue folding): HE images showed only a few atypical tumor cells with crush marks under the lymph node capsule, morphologically easily confused with subcapsular histiocytes. The maximum diameter of the metastatic lesion was less than 0.2 mm (e.g., Figure 2 (as shown in g); In the analysis results, IR-IHC labeled EMA (such as g) Figure 2 (as shown in h) and CK19 (as shown in h) Figure 2 In the image shown in i), multifocal brown-stained nests of tumor cells were identified under the lymph node capsule, with the largest foci having a maximum diameter of approximately 0.3 mm.

[0054] Case B17 (micrometastasis, with tissue folding and desquamation): HE image shows a small number of atypical cells near the folded area; the maximum diameter of the metastatic lesion is approximately 0.3 mm (e.g., Figure 2 (As shown in j); In the analysis results, brown-stained tumor cells were identified in the EMA images of IR-IHC not only near the tissue folding area, but also observed in the tissue folding area (as shown in j). Figure 2 As shown in k), in IR-IHC-labeled CK19 images, only scattered brown-stained tumor cells were identified due to tissue detachment (as shown in k). Figure 2 (as shown in l in the image).

[0055] Case B17 (isolated tumor cells): Due to artifacts such as folding and detachment on HE images, tumor cells were virtually unidentifiable (e.g., Figure 3 (as shown in a) In the analysis results, the EMA image of IR-IHC showed three isolated positive cell nests, and the CK19 image showed one positive cell nest (as shown in a) Figure 3 As shown in b and c), the maximum diameter of the tumor cells was less than 0.2 mm, and the number of simulated cells was less than 200. The combined use of the two markers demonstrated significant complementary value in this case: epithelial membrane antigen identified more tumor cell nests, while cytokeratin 19 provided further confirmation. Thus, the two markers can identify tumor cells with different antigen expression profiles, and their combined use can minimize missed diagnoses due to heterogeneity in tumor antigen expression.

[0056] These results demonstrate that the method of this embodiment can effectively overcome the diagnostic difficulties caused by tissue folding, burns, and detachment in HE images, and achieve accurate judgment by using specific antigen staining.

[0057] It should be noted that in the examination of 104 lymph node images, normal structures within the lymph nodes, such as vascular endothelial cells and stromal cells, did not exhibit non-specific brownish-yellow staining from cytokeratin 19 or epithelial membrane antigens. This indicates that the antibody combination and staining system used in this embodiment have high specificity and can effectively distinguish metastatic tumor cells from the normal components of lymph nodes.

[0058] (3) Analyze timeliness The total time from receiving the image to completing the analysis was 20 to 33 minutes, with a median time of 25 minutes. This demonstrates that the method of this embodiment fully meets the real-time requirements of clinical surgery while maintaining high recognition sensitivity.

[0059] In summary, the effectiveness of the image analysis method in this embodiment has been fully verified. By simultaneously preparing HE and rapid immunohistochemical sections and performing multimodal image analysis, this method can significantly improve the intraoperative identification sensitivity of low-volume lymph node metastases in endometrial cancer from 28.57% to 100% compared to the traditional method. Moreover, the entire process can be completed within the conventional surgical waiting time (approximately 30 minutes), providing reliable technical support for solving the clinical challenges of accurate intraoperative staging and immediate decision-making.

[0060] Building upon the aforementioned dual labeling of cytokeratin 19 and epithelial membrane antigen, an optimized combination of a third epithelial-specific marker can be further incorporated. For example, a combination of cytokeratin 19, epithelial membrane antigen, and Ber-EP4 antibody can be used. The implementation involves simultaneously preparing four sections from consecutive sections of the same lymph node sample, and performing intraoperative frozen section HE staining, CK19, EMA, and Ber-EP4 rapid immunohistochemical staining, respectively. Based on the digital images of these four sections, when any epithelial marker (CK19, EMA, or Ber-EP4) exhibits specific staining characteristics in the image, combined with the morphological features of the HE image, it can be identified as a positive area. Thus, this embodiment addresses the high heterogeneity of tumor cell antigen expression (some cells may only express some markers), with the triple markers forming a wider "detection network," theoretically increasing detection sensitivity to near the theoretical limit and further reducing the risk of missed diagnoses due to antigen loss.

[0061] In the above embodiments, a rapid multiplex fluorescent immunohistochemical staining protocol can be used in the rapid immunohistochemical staining step. Specifically, enzyme-labeled or directly fluorescently labeled antibodies against cytokeratin 19 and epithelial membrane antigens, respectively labeled with different fluorescent groups (such as Cy3 and FITC), are used. Simultaneous incubation and staining are performed on the same slide, allowing simultaneous observation of CK19 (e.g., red fluorescence) and EMA (e.g., green fluorescence) signals on the same slide in a single staining process. Image acquisition using a fluorescence microscope or digital pathology scanning system clearly distinguishes the two signals and their co-localization (the appearance of yellow merged signals). In this way, the image analysis of two slides is integrated into a single slide image, eliminating comparison errors caused by subtle differences in slide location, making the expression relationship of the two markers on the same cell readily apparent, and improving the speed and spatial accuracy of interpretation.

[0062] Example 2: This embodiment provides a kit for rapid immunohistochemical diagnosis of lymph node metastasis in endometrial cancer surgery. The kit includes a container and the following components individually packaged in the container: an enzyme-labeled antibody against cytokeratin 19; an enzyme-labeled antibody against epithelial membrane antigen; DAB chromogenic solution; and buffer and blocking agent for rapid immunohistochemical staining of frozen sections. The enzyme-labeled antibodies against cytokeratin 19 and epithelial membrane antigen are used to simultaneously stain adjacent tissue sections of the same lymph node sample for joint interpretation with intraoperative frozen HE-stained sections.

[0063] This kit integrates a specially optimized anti-cytokeratin 19 enzyme-labeled antibody, an anti-epithelial membrane antigen enzyme-labeled antibody, and a precisely matched DAB colorimetric system, buffer, and blocking agent into a standardized package, providing a specific, stable, and ready-to-use core material guarantee for the aforementioned intraoperative rapid immunohistochemical detection method. All components in this kit are optimized for rapid frozen section staining, eliminating the need for users to screen and optimize antibodies and reagent systems themselves, significantly reducing the operational threshold and technical variability, and ensuring the consistency and reliability of test results across different operators and laboratories. Furthermore, two of the enzyme-labeled antibodies are specifically configured for simultaneous staining of adjacent sections of the same sample, allowing kit-based immunohistochemical staining to be completed in parallel with traditional intraoperative frozen section hematoxylin and eosin (HE) staining within the same timeframe, thus naturally supporting the diagnostic logic of joint interpretation of HE sections and dual-labeled IHC sections. Furthermore, by providing ready-to-use key reagents, the kit in this embodiment enables pathology departments to perform highly sensitive detection efficiently and stably, thereby transforming the high sensitivity and specificity of intraoperative diagnosis of low-volume metastases into routine and reproducible clinical services, ultimately providing surgeons with immediate and accurate decision-making basis and effectively avoiding secondary surgeries.

[0064] It should be noted that the kit for rapid immunohistochemical diagnosis of lymph node metastasis in endometrial cancer in this embodiment involves the same scheme or principle as that described in Example 1, and the same or similar contents will not be described in detail.

[0065] Example 3: by Figure 6 This embodiment provides an intraoperative rapid immunohistochemical detection system 600 for lymph node metastasis in endometrial cancer, comprising: a slide preparation module 601, used to prepare adjacent intraoperative frozen HE-stained slides, cytokeratin 19 rapid immunohistochemical stained slides, and epithelial membrane antigen rapid immunohistochemical stained slides from the same lymph node sample obtained intraoperatively; an image acquisition module 602, used to acquire digital pathological images of the three slides; an image analysis module 603, used to automatically identify and label the cytokeratin 19 cytoplasmic brown-yellow stained area and the epithelial membrane antigen cell membrane brown-yellow stained area in the digital pathological images according to a deep learning algorithm; and a report generation module 604, used to fuse the automatically labeled results with the image features of the intraoperative frozen HE-stained slides to generate an auxiliary diagnostic report, which indicates whether metastatic lesions exist and their types.

[0066] The intraoperative rapid immunohistochemical detection system 600 in this embodiment achieves a leap from traditional manual microscopic examination to intelligent assisted diagnosis by digitizing pathological slides and introducing artificial intelligence analysis. The digital pathological image refers to the high-resolution, full-view digital image of the three slides obtained by a full-slide scanner, which can be analyzed by computer. The deep learning algorithm is a convolutional neural network model trained on a large number of labeled images of metastatic lymph node slides from endometrial cancer. It can autonomously learn and recognize complex image patterns corresponding to CK19 cytoplasmic staining and EMA cell membrane staining.

[0067] It should be understood that the system 600 in this embodiment automatically identifies and annotates specifically stained areas through the image analysis module 603, significantly reducing the reliance of diagnostic results on the pathologist's personal experience and subjective interpretation, and improving the consistency and reproducibility of diagnostic results from different operators and at different time points. The system 600 can perform rapid, parallel, and automated analysis of digital slides, completing the initial screening and annotation of the entire slide within minutes, assisting physicians in focusing their attention on suspicious areas, thereby significantly shortening the overall interpretation time and highly aligning with the urgency requirements of intraoperative diagnosis. The report generation module 604 uses an algorithm to fuse the morphological features of HE slides with the molecular expression information from two immunohistochemical methods to comprehensively generate diagnostic suggestions. This integrated analysis of multimodal information helps to more accurately identify small, atypical, or metastatic lesions located in artifactive areas of slide preparation that are easily overlooked under a single modality, becoming a powerful decision support tool for pathologists. Ultimately, the system 600 solidifies the technical advantages of the detection method described in Example 1 and the reagent kit described in Example 2 into an efficient and intelligent analysis process, providing a practical and integrated solution for accurate intraoperative diagnosis.

[0068] It should be noted that the scheme or principle involved in the system 600 of this embodiment is the same as that described in Embodiment 1, and the same or similar contents will not be described in detail.

[0069] Example 4: This invention also provides a terminal device, which may include a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the above-described functionality. Figure 1 The various processes of the multimodal pathological image processing method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0070] Combination Figure 7 As shown, this embodiment also discloses a specific implementation of a computer-readable storage medium 700. This computer-readable storage medium 700 can be configured wholly or partially in a physical computer, server, cluster server, or data center.

[0071] In this embodiment, the computer-readable storage medium 700 stores computer program instructions 701. The computer program instructions 701 are read and executed by a processor 702 to perform the steps in the multimodal pathological image processing method disclosed in Embodiment 1.

[0072] Optionally, the computer-readable storage medium 700 can be configured as a server, and the server runs on a physical device used to build a private cloud, hybrid cloud, or public cloud. The computer-readable storage medium 700 can also be configured as random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0073] The computer-readable storage medium 700 is used to store a program, and the processor 702, upon receiving an execution instruction, executes the multimodal pathological image processing method disclosed in Embodiment 1.

[0074] Meanwhile, the processor 702 disclosed in this embodiment may be an integrated circuit chip with signal processing capabilities. The processor 702 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0075] The technical solution of the same part in the computer-readable storage medium 700 disclosed in this embodiment as in Embodiment 1 and / or Embodiment 2 is described in Embodiment 1 and / or Embodiment 2, and will not be repeated here.

[0076] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

[0077] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0078] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A multimodal pathological image processing method for assisting in the analysis of lymph node samples from endometrial cancer, characterized in that, The method includes: Three digital pathological images of adjacent tissue sections of the same intraoperative lymph node sample were obtained. The three digital pathological images were an intraoperative frozen HE staining image, a cytokeratin 19 rapid immunohistochemical staining image, and an epithelial membrane antigen rapid immunohistochemical staining image, respectively. The rapid immunohistochemical staining images of cytokeratin 19 and the rapid immunohistochemical staining images of epithelial membrane antigens were processed to identify a first target region that conforms to the specific staining characteristics of cytokeratin 19 and a second target region that conforms to the specific staining characteristics of epithelial membrane antigens. The first target region and the second target region are fused to determine the joint positive region, and the joint positive region is spatially correlated and feature-fused with the intraoperative frozen HE staining image to generate fusion analysis results; Specifically, the information from the two modalities is superimposed and presented by combining the combined positive region with the intraoperative frozen HE staining image through feature fusion. The identification of the first and second target regions is achieved using a deep learning model. When tissue folds, burns, or tissue detachment are present in the intraoperative frozen HE staining image, the weight of the immunohistochemical signal is automatically increased, and the annotation information of the combined positive area is used as the main basis for analysis.

2. The method of claim 1, wherein, Also includes: The fusion analysis results are output, which include at least a fused visualization image containing the joint positive region annotation information and quantitative analysis information calculated based on the joint positive region.

3. The method according to claim 2, characterized in that, The quantitative analysis information includes the maximum diameter of the combined positive region, pixel area, or classification suggestions for classifying metastatic lesion types based on preset thresholds.

4. The method according to claim 3, characterized in that, The classification suggestion information for classifying metastatic lesion types based on preset thresholds includes: When the maximum diameter of the combined positive region is greater than 2 mm, output macro-transfer classification suggestions; Micrometastasis classification suggestions are output when the maximum diameter of the combined positive area is 0.2 mm to 2 mm and / or the number of tumor cells in the area is greater than 200. When the maximum diameter of the combined positive region is no greater than 0.2 mm and the number of tumor cells in the region is no greater than 200, a classification suggestion for isolated tumor cells is output.

5. The method of claim 2, wherein, The total processing time from acquiring the three digital pathological images to outputting the fusion analysis results does not exceed 33 minutes.

6. The method of claim 1, wherein, The intraoperative frozen HE-stained images, cytokeratin 19 rapid immunohistochemical staining images, and epithelial membrane antigen rapid immunohistochemical staining images were obtained from the digital scanning results of intraoperative frozen HE-stained sections, cytokeratin 19 rapid immunohistochemical staining sections, and epithelial membrane antigen rapid immunohistochemical staining sections prepared simultaneously from adjacent tissue sections of the same lymph node sample.

7. The method of claim 1, wherein, The rapid immunohistochemical staining images of cytokeratin 19 and epithelial membrane antigen were obtained from sections prepared using a rapid immunohistochemical staining procedure involving enzyme-labeled primary antibody followed by incubation with an inhibitor, primary antibody, enhancement solution, and DAB development. The rapid immunohistochemical staining procedure included: Fix the sections using 10% neutral formalin buffer for 30 to 60 seconds.

8. The method according to claim 1, characterized in that, When the quantitative index of either the first target region or the second target region reaches a preset threshold, it is determined to be a suspected transfer region.

9. A kit for carrying out the method according to any one of claims 1 to 8, characterized in that, The kit includes: The container, and the following components individually packaged in the container: Enzyme-labeled antibody against cytokeratin 19; Enzyme-labeled antibody against epithelial membrane antigen; DAB colorimetric solution; Buffer and blocking agent for rapid immunohistochemical staining of frozen sections; The enzyme-labeled antibody against cytokeratin 19 and the enzyme-labeled antibody against epithelial membrane antigen are used to simultaneously stain adjacent tissue sections of the same lymph node sample for joint interpretation with intraoperative frozen HE-stained sections.

10. A rapid intraoperative immunohistochemical detection system for lymph node metastasis of endometrial cancer, used to implement the method according to any one of claims 1 to 8, characterized in that, include: The slide preparation module is used to prepare adjacent intraoperative frozen HE-stained sections, cytokeratin 19 rapid immunohistochemical staining sections, and epithelial membrane antigen rapid immunohistochemical staining sections from the same lymph node sample obtained during surgery. The image acquisition module is used to acquire digital pathological images of three slides; The image analysis module is used to automatically identify and label the brownish-yellow cytoplasmic region of cytokeratin 19 and the brownish-yellow cytoplasmic region of epithelial membrane antigen in the digital pathological image based on a deep learning algorithm. as well as, The report generation module is used to integrate the results of automatic annotation with the image features of the intraoperative frozen HE-stained sections to generate an auxiliary diagnostic report, which indicates whether metastatic lesions exist and their types.

Images