A multi-organ joint diagnosis system for female reproductive system

Abstract

The application provides a multi-organ joint diagnosis system for the female reproductive system, comprising: a tissue area extraction module that extracts the tissue area in each slide image and performs contour detection and instance segmentation on the tissue area of each slide image; a tissue attribution determination module that determines the attribution of each slide based on the contour detection and instance segmentation results, and determines the corresponding relationship between the sampling site number and the slide number; an image slicing module that analyzes each slide according to the sampling site number and the slide number to determine a combined slide, and performs image slicing processing on the combined slide; and a joint diagnosis module that collects the image areas of the same sampling site according to the tissue attribution determination results and the image slicing processing results, and performs independent pathological diagnosis on each sampling site image area based on an AI diagnosis model. The application realizes intelligent classification and splitting of the slide and tissue images, independent analysis of each sampling site, and avoids misdiagnosis and missed diagnosis caused by model misuse.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and artificial intelligence-assisted diagnosis, and in particular to a multi-organ combined diagnostic system for the female reproductive system. Background Technology

[0002] Currently, with the rapid development of digital pathology, the application of artificial intelligence in pathological image-assisted diagnosis is gradually maturing, especially in the field of female reproductive system-related diseases (such as cervical cancer, endometrial cancer, and ovarian tumors). Deep learning-based models are already capable of detecting, grading, and conducting preliminary screening of tumor areas. In gynecological pathology examinations, to comprehensively assess reproductive system lesions, pathologists often employ a "multi-site combined sampling" strategy—for example, in a total hysterectomy, in addition to the main body of the uterus, tissues from connecting sites such as the isthmus of the fallopian tubes and the ovarian ligaments need to be collected simultaneously. Due to the small size of the tissue blocks (usually 3-5 mm), in practice, tissues from different sampling sites (such as the uterine body, fallopian tubes, and ovaries) are often embedded side-by-side in the same paraffin block, forming a mixed slide containing tissues from multiple anatomical sites after sectioning and staining. This routine clinical practice results in a single digital pathology image potentially containing multiple tissue types, such as the myometrium, fallopian tube mucosa, and ovarian cortex, and existing AI diagnostic systems do not yet have the intelligent processing capabilities for such mixed slides.

[0003] Traditional AI pathology systems often use single slides or sections as the smallest diagnostic unit, and the common problem of "multiple tissue sites sharing the same slide" in actual pathological sampling procedures has not yet been systematically addressed. In the sampling of tissues from the female reproductive system, pathologists often collect a complete uterus (or part of the uterus) along with surrounding tissues such as the fallopian tubes and ovaries. During slide preparation, to save slides or for easier observation, doctors may embed tissue blocks from different anatomical sites onto the same slide for sectioning. This results in a single pathological image containing tissue regions from multiple different sampling sites. Without proper identification and separation, this directly impacts the diagnostic accuracy of the AI ​​model.

[0004] Therefore, in order to overcome the above-mentioned technical problems, the present invention provides a multi-organ combined diagnostic system for the female reproductive system. Summary of the Invention

[0005] This invention provides a multi-organ combined diagnostic system for the female reproductive system. It differs from traditional diagnostic processes based on slides or single sections by using a diagnostic process design mechanism based on pathological sampling sites. The system proposes using "sampling site" as the basic diagnostic unit, integrating multiple slides from the same anatomical location to achieve structured summarization and unified diagnostic output. An intelligent tissue attribution judgment method based on sampling information and numbering rules is employed: using the sampling names and slide numbering rules registered in the pathology information system, intelligent classification and segmentation of slides and tissue images are achieved, providing a basis for subsequent model selection and inference. A slide image segmentation method proposed for multi-site shared slides facilitates the automatic identification and segmentation of tissue regions belonging to different anatomical locations on shared slides. A diagnostic trigger mechanism using AI models is employed: after dividing tissue images by location, each is input into the corresponding anatomical location's AI model for independent analysis, avoiding misdiagnosis and missed diagnosis caused by model mixing.

[0006] A multi-organ combined diagnostic system for the female reproductive system includes:

[0007] The tissue region extraction module is used to extract tissue regions from each slide image and perform contour detection and instance segmentation on the tissue regions of each slide image.

[0008] The tissue attribution determination module is used to determine the tissue attribution of each slide based on the contour detection and instance segmentation results, and to determine the correspondence between the sampling site number and the slide number.

[0009] The image segmentation module is used to analyze each glass slide according to the sampling part number and the glass slide number, identify the combined glass slides in each glass slide, and perform image segmentation processing on the combined glass slides.

[0010] The joint diagnosis module is used to summarize the image regions of the same sampling site based on the tissue attribution determination results and the image segmentation processing results, and to perform independent pathological diagnosis on the image regions corresponding to each sampling site based on the AI ​​diagnosis model, and output a diagnosis report.

[0011] Preferably, a multi-organ combined diagnostic system for the female reproductive system further includes:

[0012] The slide pretreatment module is used for:

[0013] Before extracting the tissue regions from each slide image, read the slide and patient information, which includes: the patient described on each slide and the preliminary recorded sampling site label;

[0014] A preliminary mapping structure of "patient-site-slide" is constructed based on slides and patient information;

[0015] The initial preprocessing of each glass slide is completed based on the preliminary mapping structure.

[0016] Preferably, a multi-organ combined diagnostic system for the female reproductive system includes a tissue region extraction module, comprising:

[0017] The tissue region detection unit is used to process the slide image corresponding to each slide based on the tissue region segmentation network and extract the tissue region contained in the slide image.

[0018] The contour detection unit is used to perform contour detection and instance segmentation on the extracted tissue regions to determine the contour information of each tissue region and the corresponding instance.

[0019] Preferably, a multi-organ combined diagnostic system for the female reproductive system includes a tissue attribution determination module, comprising:

[0020] The slide analysis unit is used to analyze the contour detection results and instance segmentation results of each slide image to extract the material description field and slide number of each slide.

[0021] The numbering unit is used to number the sampling location of each glass slide based on the sampling description field, thereby obtaining the sampling location number;

[0022] The slide assignment unit is used to assign slides to each slide image according to the number sequence of the sampling site, and to determine the correspondence between the slide number and the sampling site number.

[0023] Preferably, a multi-organ combined diagnostic system for the female reproductive system includes an image slicing module comprising:

[0024] The assembled glass slide positioning unit is used to read the correspondence between the glass slide number and the sampling part number, and to position the assembled glass slide in each glass slide according to the correspondence.

[0025] The processing unit is used to perform standardized rotation on the composite glass slide to obtain the target composite glass slide;

[0026] The tissue region separation unit is used to extract and segment tissue regions from the target composite slide image, separating independent tissue regions in the target composite slide to form tissue candidate regions;

[0027] The central axis determination unit is used to select and analyze the target tissue region in the tissue candidate region to determine the central axis of the target composite slide image;

[0028] Classification unit, used for:

[0029] Extract the centroid coordinates of each tissue region in the candidate tissue region, compare the centroid coordinates of each tissue region with the central axis, and classify the sampling location of each tissue region according to the comparison results;

[0030] Based on the classification results, the tissue region images on each side are cropped and labeled for extraction to determine the image set of each sampling site.

[0031] Preferably, in a multi-organ combined diagnostic system for the female reproductive system, the midline determination unit selects and analyzes a target tissue region from the tissue candidate region to determine the midline of the target composite slide image. The specific working process includes:

[0032] S1: Select the two largest tissue regions from the tissue candidate region as the preliminary judgment objects for the left part and the right part. The tissue candidate region includes n tissue regions.

[0033] S2: Collect the coordinates of the first centroid of the object on the left side for preliminary judgment, and at the same time, collect the coordinates of the second centroid of the object on the right side;

[0034] S3: Determine whether the preliminary judgment objects on the left and right sides are symmetrical in the target composite slide image;

[0035] If the objects initially judged on the left and right are symmetrical in the target composite slide image, then calculate the target median value of the x-coordinate between the first and second centroid coordinates, and use the target median value as the central axis of the target composite slide image;

[0036] S4: If the preliminary judgment object on the left side and the preliminary judgment object on the right side are asymmetrical in the target composite slide image, repeat steps S1-S4 in the remaining n-2 tissue regions in the tissue candidate region until the central axis of the target composite slide image is obtained.

[0037] Preferably, a multi-organ combined diagnostic system for the female reproductive system includes a combined diagnostic module comprising:

[0038] The summarization unit is used to read the tissue attribution determination results and image segmentation processing results, determine the set of image regions of the same sampling site, and summarize the set of image regions of the same sampling site to obtain the image region corresponding to each sampling site.

[0039] The AI ​​diagnostic model training unit is used to acquire sample data from each sampling site and pre-train the sample data from each sampling site to obtain the AI ​​diagnostic model.

[0040] The pathology diagnosis unit is used to perform independent pathological diagnosis on the image regions corresponding to each sampling site based on the AI ​​diagnosis model, and output a diagnosis report based on the diagnosis results.

[0041] Preferably, in a multi-organ combined diagnostic system for the female reproductive system, the pathology diagnostic unit outputs a diagnostic report based on the diagnostic results, including:

[0042] The report generation subunit is used to read the diagnostic results of each sampling site and perform unified structural fusion of the diagnostic results of each sampling site to generate the corresponding diagnostic report.

[0043] Preferably, a multi-organ combined diagnostic system for the female reproductive system includes a report generation subunit comprising:

[0044] Embedded subunits, used for:

[0045] Generate a visual slice distribution map and an AI decision-making heatmap based on the diagnostic results;

[0046] Embed the visual slice distribution map and the AI ​​decision heat map in the diagnostic report;

[0047] The report reading subunit is used to retrieve the slice distribution map and AI decision heat map corresponding to the sampling site when reading the diagnostic report, based on the viewing requirements.

[0048] Preferably, a multi-organ combined diagnostic system for the female reproductive system further includes: a review and editing module.

[0049] It is used to review diagnostic reports and, when there are misjudged areas, to correct, supplement, or annotate the diagnostic reports. At the same time, it reassigns the image to the misjudged areas.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] This invention distinguishes itself from traditional diagnostic processes based on slides or single sections by designing a diagnostic workflow based on the pathological sampling site. It is the first to propose using the "sampling site" as the basic diagnostic unit, integrating multiple slides from the same anatomical location to achieve structured summarization and unified diagnostic output. An intelligent tissue attribution judgment method based on sampling information and numbering rules is employed: using the sampling names and slide numbering rules registered in the pathology information system, intelligent classification and segmentation of slides and tissue images are achieved, providing a basis for subsequent model selection and inference. A slide image segmentation method proposed for multi-site shared slides facilitates the automatic identification and segmentation of tissue regions belonging to different anatomical locations on shared slides. A diagnostic trigger mechanism using AI models is employed: after dividing tissue images by location, each is input into the corresponding anatomical location's AI model for independent analysis, avoiding misdiagnosis and missed diagnosis caused by model mixing.

[0052] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is a structural diagram of a multi-organ combined diagnostic system for the female reproductive system according to an embodiment of the present invention;

[0056] Figure 2 This is a structural diagram of a tissue region extraction module in a multi-organ combined diagnostic system for the female reproductive system, as described in an embodiment of the present invention.

[0057] Figure 3 This is a structural diagram of a tissue attribution determination module in a multi-organ combined diagnostic system for the female reproductive system, as described in an embodiment of the present invention.

[0058] Figure 4 This is an image segmentation structure diagram of a multi-organ combined diagnostic system for the female reproductive system in an embodiment of the present invention;

[0059] Figure 5 This is a structural diagram of a combined diagnostic module in a multi-organ combined diagnostic system for the female reproductive system, as described in an embodiment of the present invention.

[0060] Figure 6 This is a thermal overview diagram of a multi-organ combined diagnostic system for the female reproductive system in an embodiment of the present invention;

[0061] Figure 7 This is an example diagram of tissue diagnosis in a multi-organ combined diagnostic system for the female reproductive system according to an embodiment of the present invention. Detailed Implementation

[0062] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0063] Example 1:

[0064] This embodiment provides a multi-organ combined diagnostic system for the female reproductive system, such as... Figure 1 As shown, it includes:

[0065] The tissue region extraction module is used to extract tissue regions from each slide image and perform contour detection and instance segmentation on the tissue regions of each slide image.

[0066] After the whole slide image (WSI) for each slide is input into the system, it first passes through the tissue region detection module. This module uses a deep learning-based tissue region segmentation network to accurately extract regions containing tissue from the entire image, filtering out background, blank areas, and contaminated areas.

[0067] Subsequently, the system performs contour detection and instance segmentation on the extracted tissue regions, and outputs the contour information of each continuous tissue block (such as embedded blocks A, B, C, etc.) as the basic unit for subsequent organ attribution determination.

[0068] The tissue attribution determination module is used to determine the tissue attribution of each slide based on the contour detection and instance segmentation results, and to determine the correspondence between the sampling site number and the slide number.

[0069] Instead of relying on deep learning to identify the organ to which a tissue belongs, the system combines the sampling site information recorded in the Pathology Information System (LIS) and uses a rule-driven approach to classify each slide and image into its corresponding sampling site. The system first parses the sampling description field and number information of each slide, such as standard names like "uterus," "ovary," and "left fallopian tube," and classifies them into the corresponding anatomical site set. For complex names (such as "uterus and left adnexa" or "uterus + ovary + fallopian tube"), the system automatically determines, based on naming conventions, that the slide may contain tissues from multiple organs.

[0070] Subsequently, the system further refines the slide classification by combining the numbering sequence of the sampling sites. For example, in a pathology case, five tissue samples were taken, with specimens 1 and 2 labeled "uterus", specimen 3 "ovary", specimen 4 "fallopian tube", and specimen 5 "uterus and adnexa". The system can classify slides numbered 1 and 2 as "uterus", 3 as "ovary", 4 as "fallopian tube", and 5 as containing multiple tissue sources.

[0071] For cases where a single slide contains tissues from multiple sampling sites, this invention introduces an image-level segmentation algorithm to divide the tissue regions. The system first uses traditional image processing methods or semi-automatic graphic annotation tools (such as boundary tracking, threshold segmentation, and connectivity analysis) to extract the contours of multiple independent tissue blocks on the slide. It then combines information such as numbering sequence, slice order, and tissue morphology to assist in inferring their affiliation, thereby disassembling a single slide into multiple image regions and assigning them to sets such as "uterus," "fallopian tube," or "ovary." This process ensures that even when the same slide contains tissues from mixed sources, the AI ​​models for each region can still receive accurate, single-source image input.

[0072] This rule-driven and image-assisted hybrid tissue attribution scheme balances automation with the interpretability of actual pathological workflows, greatly avoiding uncontrollable judgment errors caused by misuse of deep models, while improving the accuracy of anatomical structure attribution and laying an accurate data foundation for subsequent AI diagnosis.

[0073] The image segmentation module is used to analyze each glass slide according to the sampling part number and the glass slide number, identify the combined glass slides in each glass slide, and perform image segmentation processing on the combined glass slides.

[0074] When processing slides with "multiple tissue sites on the same slide", this invention further introduces an image-level segmentation algorithm based on spatial distribution and tissue centroid judgment, which is used to accurately classify multiple tissue regions mixed in a slide to the corresponding anatomical sites.

[0075] The joint diagnosis module is used to summarize the image regions of the same sampling site based on the tissue attribution determination results and the image segmentation processing results, and to perform independent pathological diagnosis on the image regions corresponding to each sampling site based on the AI ​​diagnosis model, and output a diagnosis report.

[0076] After determining tissue affiliation and segmenting the images, the system aggregates all image regions belonging to the same sampling site (which may come from different slides) as the complete image input for that site. The system has pre-trained dedicated AI diagnostic models (such as endometrial cancer identification models and ovarian tumor classification models) for anatomical sites such as the uterus, ovary, and fallopian tubes, with each model optimized for its organ-specific pathological characteristics.

[0077] The model can accept inputs such as multi-image joint inputs, multi-patch feature fusion, or multi-modal mechanisms to fuse tissue morphology with other metadata (such as age, medical history, etc.) to improve diagnostic accuracy and context awareness. The output is a structured diagnostic result, including lesion type, grade, presence of cancer, and lesion location heatmap.

[0078] The beneficial effects of the above technical solution are as follows: By designing a diagnostic process based on the pathological sampling site, which differs from the traditional diagnostic process based on slides or single sections, this invention proposes for the first time to use the "sampling site" as the basic diagnostic unit, integrating multiple slides from the same anatomical location to achieve structured summarization and unified diagnostic output. An intelligent tissue attribution judgment method based on sampling information and numbering rules: Through the sampling name (e.g., "uterus and left adnexa") and slide numbering rules registered in the pathology information system, intelligent classification and segmentation of slides and tissue images are achieved, providing a basis for subsequent model selection and inference. The proposed slide image segmentation method for multi-site shared slides facilitates the automatic identification and segmentation of tissue regions belonging to different anatomical locations on shared slides. A diagnostic triggering mechanism using AI models: After dividing tissue images by location, each image is input into the corresponding anatomical location's AI model for independent analysis, avoiding misdiagnosis and missed diagnosis caused by model mixing.

[0079] Example 2:

[0080] Based on Example 1, this example provides a multi-organ combined diagnostic system for the female reproductive system, which also includes:

[0081] The slide pretreatment module is used for:

[0082] Before extracting the tissue regions from each slide image, read the slide and patient information, which includes: the patient described on each slide and the preliminary recorded sampling site label;

[0083] A preliminary mapping structure of "patient-site-slide" is constructed based on slides and patient information;

[0084] The initial preprocessing of each glass slide is completed based on the preliminary mapping structure.

[0085] In this embodiment, the slide and patient information provided by the pathology information system or scanning platform are first read, including the patient to whom each slide belongs, the preliminary recorded sampling site label (such as "uterine body", "right ovary", etc.), and a preliminary "patient-site-slide" mapping structure is constructed. Considering that slide records in reality are often incomplete, mislabeled, or mixed, this step is only used as a coarse screening and cannot be directly used as input for the AI ​​model.

[0086] The working principle and beneficial effects of the above technical solution are as follows: First, the slide and patient information, including patient and sampling site labels, are read to construct a preliminary mapping structure of "patient-site-slide". Then, each slide is preprocessed according to the preliminary mapping structure to associate the slide, patient and sampling site, making the data relationship clear and organized. Through preprocessing, the accuracy and efficiency of each slide processing are effectively improved.

[0087] Example 3:

[0088] Based on Example 1, this example provides a multi-organ combined diagnostic system for the female reproductive system, such as... Figure 2 As shown, the tissue region extraction module includes:

[0089] The tissue region detection unit is used to process the slide image corresponding to each slide based on the tissue region segmentation network and extract the tissue region contained in the slide image.

[0090] The contour detection unit is used to perform contour detection and instance segmentation on the extracted tissue regions to determine the contour information of each tissue region and the corresponding instance.

[0091] The beneficial effect of the above technical solution is that it provides effective data support for subsequent organizational attribution determination.

[0092] Example 4:

[0093] Based on Example 1, this example provides a multi-organ combined diagnostic system for the female reproductive system, such as... Figure 3 As shown, the organization affiliation determination module includes:

[0094] The slide analysis unit is used to analyze the contour detection results and instance segmentation results of each slide image to extract the material description field and slide number of each slide.

[0095] The numbering unit is used to number the sampling location of each glass slide based on the sampling description field, thereby obtaining the sampling location number;

[0096] The slide assignment unit is used to assign slides to each slide image according to the number sequence of the sampling site, determining the correspondence between the slide number and the sampling site number. For example, in a pathology case, five tissue samples were taken, with specimens 1 and 2 labeled "uterus", specimen 3 "ovary", specimen 4 "fallopian tube", and specimen 5 "uterus and adnexa". The system can then assign slides numbered 1 and 2 as "uterus", 3 as "ovary", 4 as "fallopian tube", and 5 as containing multiple tissue sources.

[0097] The beneficial effect of the above technical solution is that it effectively ensures the accuracy of determining the correspondence between the slide number and the sampling location number.

[0098] Example 5:

[0099] Based on Example 1, this example provides a multi-organ combined diagnostic system for the female reproductive system, including an image segmentation module, such as... Figure 4 As shown, it includes:

[0100] The assembled glass slide positioning unit is used to read the correspondence between the glass slide number and the sampling part number, and to position the assembled glass slide in each glass slide according to the correspondence.

[0101] The processing unit is used to perform standardized rotation on the composite glass slide to obtain the target composite glass slide;

[0102] The tissue region separation unit is used to extract and segment tissue regions from the target composite slide image, separating independent tissue regions in the target composite slide to form tissue candidate regions;

[0103] The central axis determination unit is used to select and analyze the target tissue region in the tissue candidate region to determine the central axis of the target composite slide image;

[0104] Classification unit, used for:

[0105] Extract the centroid coordinates of each tissue region in the candidate tissue region, compare the centroid coordinates of each tissue region with the central axis, and classify the sampling location of each tissue region according to the comparison results;

[0106] Based on the classification results, the tissue region images on each side are cropped and labeled for extraction to determine the image set of each sampling site.

[0107] In this embodiment, the central axis determination unit selects and analyzes target tissue regions within the tissue candidate region to determine the central axis of the target composite slide image. The specific working process includes: S1: Selecting the two largest tissue regions from the tissue candidate region as the preliminary judgment objects for the left and right parts, wherein the tissue candidate region includes n tissue regions; S2: Acquiring the first centroid coordinates of the preliminary judgment object for the left part, and simultaneously acquiring the second centroid coordinates of the right part; S3: Determining whether the preliminary judgment objects for the left and right parts are symmetrical in the target composite slide image; if the preliminary judgment objects for the left and right parts are symmetrical in the target composite slide image, then calculating the target median of the x-coordinates between the first and second centroid coordinates, and using the target median as the central axis of the target composite slide image; S4: If the preliminary judgment objects for the left and right parts are asymmetrical in the target composite slide image, repeating steps S1-S4 in the remaining n-2 tissue regions within the tissue candidate region until the central axis of the target composite slide image is obtained.

[0108] The above describes the process: First, the system accurately identifies which slides are "combined slides" (i.e., slides containing tissues from multiple sampling sites) by using the correspondence between the sampling site numbers and slide numbers recorded in the pathology information system. Then, the slide image is standardized and rotated to ensure uniform slide orientation, guaranteeing that the algorithm correctly utilizes the physician's default tissue embedding habits (i.e., the first tissue is usually placed on the left side of the slide, and the second on the right).

[0109] After completing rotational normalization, the system performs preprocessing and tissue region extraction on the image. It uses tissue segmentation algorithms (such as traditional image processing methods based on color thresholding, edge detection, and morphological processing) to separate all independent tissue regions on the slide, forming a set of tissue candidate regions.

[0110] The system then selects the two largest tissue blocks from all segmented tissue regions as preliminary left and right location determination objects. The system calculates the centroid coordinates of these two tissue blocks and determines whether they are located on the left and right sides of the image respectively: if the left-right symmetry condition is met (i.e., one on the left and one on the right), the system directly calculates the midline of the line connecting their centroids (with the median of the x-coordinate as the dividing line) as the basis for slide segmentation; if the two largest tissue blocks fall on the same side of the image, the system will continue to search for the third and fourth largest tissue blocks in turn, until a pair of tissue blocks that are clearly located on the left and right sides are found, and the midline of the segmentation is recalculated accordingly.

[0111] Once the central axis is identified, the system performs centroid coordinate analysis on all tissue blocks: those with their centroid falling to the left of the central axis are automatically classified as "left-side sampling sites"; those with their centroid to the right of the central axis are classified as "right-side sampling sites." The tissue images from each side are then cropped or annotated and assigned to the corresponding site image set. The entire process does not rely on deep learning inference but instead uses positional relationships and tissue morphology to achieve highly interpretable and logically clear image-level tissue segmentation and attribution determination.

[0112] This method is particularly suitable for "uterus + adnexa" type slices commonly found in female reproductive system pathology. By combining image geometry and numbering rules, it effectively solves the problem of unclear tissue attribution in composite slices, laying an accurate foundation for subsequent use of AI models based on location.

[0113] Example 6:

[0114] Based on Example 1, this example provides a multi-organ combined diagnostic system for the female reproductive system, such as... Figure 5 As shown, the combined diagnostic module includes:

[0115] The summarization unit is used to read the tissue attribution determination results and image segmentation processing results, determine the set of image regions of the same sampling site, and summarize the set of image regions of the same sampling site to obtain the image region corresponding to each sampling site.

[0116] The AI ​​diagnostic model training unit is used to acquire sample data from each sampling site and pre-train the sample data from each sampling site to obtain the AI ​​diagnostic model.

[0117] The pathology diagnosis unit is used to perform independent pathological diagnosis on the image regions corresponding to each sampling site based on the AI ​​diagnosis model, and output a diagnosis report based on the diagnosis results.

[0118] In this embodiment, the pathological diagnosis unit outputs a diagnosis report based on the diagnosis results, including: a report generation subunit, which reads the diagnosis results of each sampling site and performs unified structural fusion of the diagnosis results of each sampling site to generate a corresponding diagnosis report.

[0119] Based on the hospital's HIS / LIS interface standards, the system outputs a structured diagnostic result document that conforms to the electronic pathology report specifications. This document clearly lists all identified sampling sites and their AI analysis suggestions. For example, entries such as "Uterine body: Complex endometrial hyperplasia with dysplasia; Ovary: No obvious lesions; Fallopian tube: Mild chronic inflammation" will be output in standard diagnostic language for doctors to quickly review. Furthermore, each diagnostic suggestion includes a key image annotation area generated by AI-assisted analysis, which may include boundary contours, classification heatmaps, and lesion area overlay information, providing pathologists with intuitive visual references.

[0120] In this embodiment, after completing the AI ​​model inference for the images corresponding to all sampling sites, the system will perform a unified structured fusion output of the diagnostic results from multiple sampling sites. The fusion logic is based on anatomical sites, combining the independently processed diagnostic suggestion information for each site (including classification diagnostic labels, corresponding risk scores, key area heatmaps, etc.) into a complete site-specific report result.

[0121] The beneficial effects of the above technical solution are: providing doctors with intuitive visual references and effectively ensuring the comprehensiveness, accuracy, and intelligence of diagnostic report generation.

[0122] Example 7:

[0123] Based on Example 6, this example provides a multi-organ combined diagnostic system for the female reproductive system, including a report generation subunit:

[0124] Embedded subunits, used for:

[0125] Generate a visual slice distribution map and an AI decision-making heatmap based on the diagnostic results;

[0126] Embed the visual slice distribution map and the AI ​​decision heat map in the diagnostic report;

[0127] The report reading subunit is used to retrieve the slice distribution map and AI decision heat map corresponding to the sampling site when reading the diagnostic report, based on the viewing requirements.

[0128] In this embodiment, to further enhance the clinical interpretability of the results, the system supports embedding a visual slide distribution map and an AI decision heatmap overview in the report. This map categorizes all slides according to their respective sampling sites, while highlighting key lesion images. When viewing the diagnostic results for any anatomical site, doctors can click to view all slide images related to that site and simultaneously browse the AI ​​diagnostic heatmap for that site. Specifically, for the display of results on "combined" slides (i.e., slides containing tissues from multiple sampling sites on the same slide), the system dynamically renders content based on the currently selected anatomical site: the combined slide image will only display the diagnostic results and heatmap area belonging to the current sampling site, while other tissue areas will be placed in a blue, inactive state to reduce their interference in the current view. This mechanism ensures that doctors can focus on the pathological manifestations of the relevant tissues when reviewing site by site and can switch to other sampling site views to obtain corresponding content, greatly improving interaction efficiency and content clarity.

[0129] In this embodiment, the thermal overview diagram is determined based on the sampling location, as follows: Figure 6 As shown, and in the example diagram of tissue diagnosis. Figure 7 As shown.

[0130] The above-mentioned composite slide results are displayed using a site-specific heatmap rendering mechanism: In the diagnostic results interface, the composite slide image can only display the AI ​​diagnostic results and heatmap of the tissue to which the currently selected sampling site belongs, while the images of other tissues are automatically set to blue and inactive, enhancing interpretability and interactive experience.

[0131] Example 8:

[0132] Based on Example 1, this example provides a multi-organ combined diagnostic system for the female reproductive system, and also includes: a review and editing module.

[0133] It is used to review diagnostic reports and, when there are misjudged areas, to correct, supplement, or annotate the diagnostic reports. At the same time, it reassigns the image to the misjudged areas.

[0134] The system integrates interfaces for manual review and editing, allowing pathologists to correct, supplement, or annotate AI-generated diagnostic results. It supports one-click marking of misjudged areas and reassignment of image attribution, ensuring that AI analysis is integrated into the actual workflow of pathologists, forming a complete clinical closed loop from automatic identification to manual review. Finally, the manually confirmed diagnostic results can be directly pushed to the hospital reporting system, achieving a fully automated closed loop from tissue identification, AI assistance, structured diagnosis to report publication.

[0135] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-organ joint diagnostic system for the female reproductive system, characterized by, The method comprises the following steps: a tissue region extraction module is used to extract tissue regions in each slide image and perform contour detection and instance segmentation on the tissue regions of each slide image; a tissue attribution determination module is used to determine the attribution of each slide based on the contour detection and instance segmentation results, and determine the correspondence between the tissue site number and the slide number; an image segmentation module is used to analyze each slide according to the tissue site number and the slide number, determine the presence of a combined slide in each slide, and perform image segmentation processing on the combined slide; a joint diagnosis module is used to aggregate image regions of the same tissue site based on the tissue attribution determination results and the image segmentation processing results, perform independent pathological diagnosis on the image regions corresponding to each tissue site based on an AI diagnosis model, and output a diagnosis report; The image segmentation module comprises: a combined slide positioning unit is used to read the correspondence between the slide number and the tissue site number, and position the combined slide in each slide according to the correspondence; a processing unit is used to standardize and rotate the combined slide to obtain a target combined slide; a tissue region separation unit is used to extract and segment the tissue regions of the target combined slide image, separate the independent tissue regions in the target combined slide, and form tissue candidate regions; a central axis determination unit is used to select a target tissue region in the tissue candidate region and analyze it to determine the central axis of the target combined slide image; a classification unit is used to: extract the center of gravity coordinates of each tissue region in the tissue candidate region, compare the center of gravity coordinates of each tissue region with the central axis, and classify the position of each tissue region according to the classification results; according to the classification results, the image regions on each side are cropped and labeled to determine the image set of each tissue site; in the central axis determination unit, the target tissue region is selected in the tissue candidate region and analyzed to determine the central axis of the target combined slide image, and the specific working process comprises: S1: In the tissue candidate region, the two largest tissue regions are selected as the left preliminary judgment object and the right preliminary judgment object, wherein the tissue candidate region comprises n tissue regions; S2: Collect the first center of gravity coordinates of the left preliminary judgment object, and collect the second center of gravity coordinates of the right preliminary judgment object; S3: Determine whether the left preliminary judgment object and the right preliminary judgment object are symmetrical in the target combined slide image; if the left preliminary judgment object and the right preliminary judgment object are symmetrical in the target combined slide image, then calculate the target median value of the x coordinates of the first center of gravity coordinates and the second center of gravity coordinates, and use the target median value as the central axis of the target combined slide image; S4: If the left preliminary judgment object and the right preliminary judgment object are not symmetrical in the target combined slide image, repeat steps S1-S4 in the remaining n-2 tissue regions in the tissue candidate region until the central axis of the target combined slide image is obtained.

2. The multi-organ combined diagnostic system for the female reproductive system according to claim 1, wherein Further comprising: a slide preprocessing module is used to: Before extracting the tissue area of each slide image, reading the slide and patient information, wherein the slide and patient information includes: the patient of each slide, the preliminary recorded biopsy site label; According to the slide and patient information, constructing a preliminary mapping structure of "patient-site-slide"; According to the preliminary mapping structure, completing the initial preprocessing of each slide.

3. The multi-organ combined diagnostic system for the female reproductive system according to claim 1, wherein The tissue area extraction module comprises: The tissue area detection unit is used for processing the slide image corresponding to each slide based on the tissue area segmentation network, and extracting the tissue area contained in the slide image; The contour detection unit is used for performing contour detection and instance segmentation on the extracted tissue area to determine the contour information of each tissue area and the instance corresponding to each tissue area.

4. The multi-organ combined diagnostic system for the female reproductive system according to claim 1, wherein The tissue attribution determination module comprises: The slide analysis unit is used for analyzing the biopsy description field and the slide number of each slide based on the contour detection result and the instance segmentation result of each slide image; The numbering unit is used for numbering the biopsy site of each slide based on the biopsy description field to obtain the biopsy site number; The slice attribution unit is used for performing slice attribution on each slide image according to the numbering sequence of the biopsy site to determine the corresponding relationship between the slide number and the biopsy site number.

5. The multi-organ combined diagnostic system for the female reproductive system according to claim 1, wherein The joint diagnosis module comprises: The summary unit is used for reading the tissue attribution determination result and the image slicing processing result, determining the image area set of the same biopsy site, and summarizing the image area set of the same biopsy site to obtain the image area corresponding to each biopsy site; The AI diagnosis model training unit is used for obtaining sample data of each biopsy site and pre-training the sample data of each biopsy site to obtain an AI diagnosis model; The pathological diagnosis unit is used for performing independent pathological diagnosis on the image area corresponding to each biopsy site based on the AI diagnosis model, and outputting a diagnosis report according to the diagnosis result.

6. The multi-organ combined diagnostic system for the female reproductive system according to claim 5, wherein In the pathological diagnosis unit, the diagnosis report is output according to the diagnosis result, comprising: The report generation subunit is used for reading the diagnosis result of each biopsy site and uniformly fusing the diagnosis result of each biopsy site to output a corresponding diagnosis report.

7. The multi-organ combined diagnostic system for the female reproductive system according to claim 6, characterized by The report generation subunit comprises: The embedding subunit is used for: Generating a visual slice distribution map and an AI decision heat overview map based on the diagnosis result; Embedding the visual slice distribution map and the AI decision heat overview map in the diagnosis report; The report reading subunit is used for reading the diagnosis report, and according to the viewing requirement, calling the slice distribution map and the AI decision heat overview map corresponding to the biopsy site.

8. The multi-organ combined diagnostic system for the female reproductive system according to claim 1, wherein Further comprising: The review editing module: is used for reviewing the diagnosis report, and when there is a misjudgment area, correcting, supplementing or remarking the diagnosis report, and reassigning the image attribution of the misjudgment area.

Images