Method and system for automatically identifying primordial follicles in ovary microsection image

Abstract

The invention discloses a method and a system for automatically identifying primordial follicles in an ovary microscopic section image, which are used for realizing automatic identification and counting of primordial follicles in the microscopic image by carrying out image binaryzation, expansion corrosion processing, edge detection, circular feature extraction and texture feature analysis on the ovary microscopic section image dyed under a microscope. According to the method, a local binary pattern (LBP) is adopted to extract texture features, and primordial follicles in a microscopic slice image are effectively distinguished by combining edge detection and a hollow circular screening algorithm; the method has the characteristics of high processing speed, high precision and strong adaptability, and can be widely applied to ovarian slice analysis, assisted reproductive medicine and research of ovarian function related diseases.

Description

Technical Field

[0001] The invention relates to the field of medical image processing, and in particular to a method and system for automatically identifying primordial follicles in ovarian microscopic slice images. Background Art

[0002] The primordial follicle is the most primary follicular structure in the female ovary. It is a tiny tissue structure composed of a primary oocyte and a layer of flat granulosa cells surrounding it. The number of primordial follicles reaches a peak when a woman is born, and then gradually depletes with age. Therefore, the number of primordial follicles is considered an important indicator of ovarian reserve function, and their accurate identification and counting are of great significance in fertility assessment, ovarian function prediction, and diagnosis of ovarian-related diseases (such as premature ovarian failure and polycystic ovary syndrome). Currently, the identification of primordial follicles is mainly accomplished by observing tissue sections under a microscope. In actual operation, doctors need to stain tissue sections (such as hematoxylin-eosin staining), and then observe and identify specific primordial follicle structures under a microscope. However, this traditional manual identification method has the following significant problems:

[0003] 1. Huge workload. Microscopic images are usually high-resolution, and a single slice may contain many cells or tissue structures. Doctors need to screen and count target follicles one by one, which is not only time-consuming and labor-intensive, but also takes a lot of time in batch analysis of multiple slices.

[0004] 2. It is highly subjective and the recognition accuracy is limited. The morphological characteristics of primordial follicles (such as cell size, shape, staining intensity, etc.) may vary to a certain extent in different samples and slice preparation processes, which can easily lead to inconsistent recognition results between observers or by the same observer at different time points. In addition to the target follicles, the microscopic images also contain other follicle types (such as primary follicles, secondary follicles, etc.), interstitial cells and other tissue structures. How to effectively separate the primordial follicles from the background is a technical challenge.

[0005] 3. Low data processing efficiency. With the widespread application of high-resolution microscopic image data, manual methods have been unable to cope with current image processing needs, especially when large-scale samples need to be processed in scientific research, the problem of low efficiency of manual methods is particularly prominent.

[0006] In response to the above problems, developing an automated identification and counting method for primordial follicle cells with the help of image processing technology can significantly improve efficiency and accuracy. By introducing advanced image processing algorithms, primordial follicles in microscopic images can be quickly and objectively identified, reducing doctors' repetitive work while providing stable and reliable results. Summary of the invention

[0007] In view of the problems in the prior art that the screening of primordial follicles is labor-intensive and highly subjective, the purpose of the present invention is to provide a method and system for automatically identifying primordial follicles in ovaries, so as to at least partially solve the above problems.

[0008] According to the different texture structures of follicles at different stages, the corresponding image processing algorithm can effectively identify follicles at different stages, and mark and count the primordial follicles for doctors' reference. The specific processing steps are as follows.

[0009] Step s1: Binarize the stained slice image under a microscope to obtain a binary image.

[0010] After binarization of the stained slice image under the microscope, a binary image can be obtained to extract the significant feature areas, including the outer granulosa cells and the cell nucleus. The specific implementation method is as follows: the stained slice image is collected through a digital microscope to obtain a high-resolution image containing the oocyte, the cell nucleus and the surrounding granulosa cells. The image format is an RGB three-channel color image. The RGB color image is converted into a grayscale image. Through grayscale processing, the data dimension is reduced while retaining the main information of the cell structure, so that the cell nucleus and granulosa cells appear as lower grayscale value areas in the grayscale image.

[0011] The Otsu threshold segmentation method is used to automatically calculate the optimal threshold T based on the image histogram to separate the foreground (target area, such as cell nuclei and granulosa cells) from the background (other tissues or blank areas). It marks the area with pixel grayscale values ​​lower than the threshold T as the foreground, and the area higher than T as the background, generating a binary image: in the binary image, the outer granulosa cell area appears as a connected area surrounding the oocyte, while the cell nucleus appears as a circular or nearly circular area with a darker center. Through the binarization process, these features are significantly retained, laying the foundation for subsequent regional segmentation and classification.

[0012] Step s2: Perform expansion and corrosion processing on the binary image to fill the disconnected areas of the granule cells and retain the true boundaries of the granule cells.

[0013] After the binary image generation is completed, due to the uneven staining of the microscopic sections and the changes in the illumination of the microscope imaging, the granulosa cells in the outer layer of the primordial follicle may be broken or gapped. This phenomenon will affect the subsequent accurate identification of the entire area of ​​the follicle. Therefore, the present invention uses an expansion operation to deal with this problem, using the image after binary processing as input, where the disconnected area of ​​the granulosa cells is shown as a gap between the small background area and the foreground area. In order to adapt to the annular distribution of the granulosa cells in the outer layer of the follicle, the circular structural element is selected to effectively fill the disconnected area. The radius of the structural element is dynamically set according to the resolution of the microscopic image. The recommended size is 3 to 5 pixels to ensure that no additional noise is introduced while filling the gap. The expansion operation expands the boundaries of the foreground pixels and converts the background pixels in contact with the structural element into foreground pixels, thereby filling the disconnected area. The circular structural element with an appropriate radius is used to expand the image to connect the broken parts of the edge of the granulosa cells. After the expansion process, the excess part is removed by the corrosion operation, and only the real boundary of the granulosa cells is retained.

[0014] Step s3: Apply an edge detection algorithm to detect boundary information and extract a circular boundary area as a candidate circular boundary area; the candidate circular boundary area includes a potential follicle area, and the potential follicle area includes a circular area surrounded by cell nuclei and a circular area surrounded by granulosa cells.

[0015] Use Canny edge detection or Sobel operator to extract the boundary information in the image. This step can help identify possible circular areas in the binary image, whether hollow or solid. Input the binary image after dilation processing, output the intensity map of all boundaries in the binary image, and mark the boundaries of the foreground area in the image. And because the follicles are circular or quasi-circular, after obtaining the boundary information, the circular boundary information is extracted, which is the potential follicle area. The potential follicle area includes the circular area surrounded by the cell nucleus and the circular area surrounded by the granulosa cells.

[0016] Step s4: the candidate circular boundary regions detected in step s3 are screened for hollow circle features and circularity features. The circular region surrounded by cell nuclei presents a feature of a solid circle, and the circular region surrounded by granular cells presents a feature of a solid middle cell nucleus portion and a hollow circle for the rest of the portion. This step filters out interference from the cell nucleus region and obtains a circular region representing the granular cells.

[0017] There are two cases of the circular boundary area obtained in step s3. The first is the outer granulosa cells of the follicle, which is also the target area; the second is the area where the cell nucleus is located. If it is the area where the cell nucleus is located, this area belongs to the noise interference area and needs to be filtered; the third is the interference of other circular areas. According to the texture characteristics of the outer granulosa cells and the cell nucleus, the cell nucleus as a whole is a solid dark area, and the inner part of the circular boundary of the outer granulosa cells contains a dark cell nucleus area, but also contains a certain area of ​​light-colored areas. From the perspective of the image after binarization, the cell nucleus is a solid circle, and the central cell nucleus area of ​​the outer granulosa cells is solid, and the outer layer of the cell nucleus is a hollow area. According to the different characteristics of this image, the following steps can be used to screen out the target area corresponding to the outer granulosa cells.

[0018] Step S41: using an inner filling algorithm to perform hollow feature screening on the candidate circular boundary region to obtain a hollow circular region; using the image inner filling algorithm to fill each candidate region, and calculating the area difference between the regions before and after filling, if the ratio exceeds a threshold value T, the value range of T can be set to 2-3, then the region is defined as a hollow circle, and this step filters out the interference of the solid cell nucleus region;

[0019] By detecting the areas in the edge image with complete outer circles and hollow interiors, we can identify hollow circular areas. Use an in-image filling algorithm (such as Flood-Fill) to fill each candidate area and calculate the area difference between the areas before and after filling. If the border of a circular area is very regular and hollow, the area after filling will increase significantly, reflecting that its interior is divided into hollow structures. Calculate the ratio of the area of ​​the filled area to the area of ​​the original area. If the ratio exceeds a certain threshold T, according to the structural characteristics of the follicular cells, the proportion of the hollow part is 2-3 times the proportion of the area of ​​the cell nucleus area; setting T to a threshold of 2-3 has the best effect, preferably T can be set to 2-3, and the area is defined as a hollow circle. For the cell nucleus area, the area after filling is slightly different from the area of ​​the original area, and its ratio does not meet the range of the threshold T, so it can be filtered.

[0020] Step s42: Calculate and screen the circularity of the hollow circular areas to obtain the circular areas representing the candidate granule cells; calculate the circularity of each hollow circular area, and set the range of the circularity value to be within the threshold interval T1, and the range of T1 is [0.8-1].

[0021] The shape of the follicle cell is round or oval, and its circularity value is relatively high. The circularity is calculated for each candidate area.

[0022] Where A is the area of ​​the hollow circular region, and P is the perimeter of the hollow circular region. A circularity value close to 1 indicates that the region is circular, and regions with circularity values ​​close to 1 are screened out. Preferably, the circularity value range can be set to be within the threshold interval T1, and the structural characteristics of the follicular cells are close to circular. The recognition effect is best when the T1 value range is set to [0.8-1]. This step can reduce the interference of other background noises with lower circularity in the image.

[0023] Step s5: The area in the color image corresponding to the circular area representing the granule cells identified in step S4 is taken as the region of interest (ROI).

[0024] The circular area surrounded by the representative granulosa cells selected in step s4 is the region of interest (ROI). At this time, solid areas such as stromal cells will be excluded, and follicular cells will be retained as the region of interest. For the retained region of interest, it is necessary to confirm again whether it is a primitive follicular cell. The edge position area of ​​the binarized image is corresponded to the color map, and the hollow circular area of ​​the granulosa cells corresponding to the color map is used as the region of interest, which contains a large number of texture features.

[0025] Step s6: Extracting texture features of the region of interest and granulosa cell layer thickness features; the granulosa cell layer of the primordial follicle is flat and single-layered, the granulosa cell layer of the primary follicle is cubic and granular, the thickness of the granulosa cell layer of the primary follicle is greater than that of the primordial follicle, and the granulosa cell layer thickness feature is introduced to distinguish primordial follicles from primary follicles to improve the recognition rate of primordial follicles;

[0026] The hollow circular region (ROI) screened out in step s4 is retained as a candidate follicle cell region. In order to further identify the follicle types (primordial follicles, primary follicles, secondary follicles, mature follicles) in these regions, it is necessary to classify based on the significant features of the follicle structure. There are differences between the texture structures of follicles at different stages. The thickness of the granulosa cell layer of the primordial follicle is flat and monolayer, the oocyte has the smallest diameter, and the shape feature is close to a circle. The outer granulosa cell layer has a simple texture. The stained microscopic section images are collected to ensure that all follicle types (primordial, primary, secondary, and mature follicles) are included. The boundaries and categories of each follicle are first manually marked, and a marked image is generated, and a training set is generated for training the model. Extract key features from each region of interest (ROI) for input to a machine learning classifier or a deep learning model.

[0027] Step s61: Texture feature extraction. The granulosa cell layer of the follicle and other regions are distinguished by texture analysis. Local binary pattern (LBP) features can be used to extract corresponding texture features. The primordial follicles are distinguished from primary follicles, secondary follicles and mature follicles based on their different texture features.

[0028] Step s62: extract the ratio of the area of ​​the connected domain of the granular cell layer in the region of interest to the circumference as the granular cell layer thickness feature.

[0029] In the primary follicle, a zona pellucida begins to appear around the oocyte, which is composed of glycoproteins and is lightly stained. The outer layer of the oocyte is wrapped by a single layer of cubic granules, and the thickness of the granulosa cell layer is thicker than that of the primordial follicle. In the primordial follicle, the oocyte is surrounded by a single layer of flat follicle cells. These are the two biggest differences between the primary follicle and the primordial follicle. The texture structure of the primordial follicle cells and the primary follicle is also the most similar, which can easily cause misclassification by machine recognition. Therefore, in addition to extracting texture features for judgment, it is also possible to make judgments based on the thickness characteristics of the granulosa cell layer between the two.

[0030] First, the connected domain of the granulosa cell layer is extracted. The granulosa cell layer is the outermost part of the region of interest (ROI) and is located outside the oocyte. The binary image of the granulosa cell layer is processed to obtain all connected regions. All connected regions are detected in the binary image of the granulosa cell layer, and the noise connected domains with too small areas are removed (the area threshold is set, such as 10 pixels), and the ratio of the connected domain area to the circumference is calculated.

[0031] Step a: Calculate the area of ​​the connected domain (A): Count the area of ​​the connected domain of the granular cell layer. The area of ​​the connected domain is the total number of all white pixels (value 1) in its binary image. Use a connected domain detection algorithm in the binary image (such as OpenCV's cv2.connectedComponents or Scikit-Image's label function) to mark all connected areas. For each connected domain, count the total number of white pixels:

[0032] Step b: Calculate the circumference (P) of the connected domain: Use boundary detection (such as the Canny algorithm) to extract the outer boundary of the granule cell layer and calculate the length of the boundary. If the boundary pixels are distributed regularly and approximately continuous, the number of boundary pixels in the image can be used as an estimate of the circumference.

[0033] Step c: Calculate the ratio V of the total area A to the circumference P. The thickness of the granulosa cell layer of the primary follicle is greater than that of the primordial follicle. The follicle cells whose area to circumference ratio V is in the range of 1-3 are identified as potential primordial follicle cells.

[0034] Step s7: Fusion and classification of the texture features of the region of interest and the thickness features of the granular cell layer.

[0035] The texture features of the region of interest extracted in step s6 and the thickness features of the granulosa cell layer are fused to obtain fused features, and the fused features are used to classify the follicular cells. The classifier can use a support vector machine (SVM). The feature vector is input into the SVM model to output the follicle category.

[0036] Step s8: Primordial follicle cell counting and labeling.

[0037] The identified primordial follicle areas are numbered and counted one by one, and the number and location of the follicles are output. In addition, the follicle areas at different stages in the image can be distinguished with different colors and marked for the doctor's reference, which saves a lot of statistical time for the doctor and can play a role in assisting judgment.

[0038] Beneficial effects: 1. Efficiency improvement: The use of automated image processing technology can greatly reduce the time for manual screening and counting, significantly improve the processing efficiency of microscopic section images, and meet the needs of large-scale sample analysis.

[0039] 2. High recognition accuracy: Through the processing flow of steps s1-s8, primordial follicles are effectively distinguished from primary follicles and background interference areas, reducing errors in manual recognition. The structural characteristics and texture characteristics of primordial follicles are fully considered, and the extraction of granulosa cell layer features is introduced and texture features are integrated for recognition, thereby improving recognition accuracy. The primordial follicles in the image can be effectively marked through the complete processing flow.

[0040] 3. Strong adaptability: The system algorithm is specifically optimized for the staining characteristics and follicle features of microscopic images, and can adapt to different slice preparation conditions and microscope settings, with wide applicability.

[0041] 4. Objective and consistent results: It avoids the subjectivity and errors in manual observation, ensures the stability and consistency of follicle identification and counting results, and facilitates data comparison across time and researchers.

[0042] 5. Clinical and scientific research value: This invention provides strong technical support for ovarian reserve function assessment, premature ovarian failure diagnosis, polycystic ovary syndrome research, etc., and promotes the development of assisted reproductive medicine and basic research.

[0043] 6. Technological innovation: For the first time, edge detection, hollow circle screening, LBP texture analysis and granulosa cell layer thickness characteristics are combined for the identification and classification of primordial follicles in microscopic images, solving the technical bottleneck of traditional manual methods that are difficult to identify small target areas and have low efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 Schematic diagram of various stages of follicle cells;

[0045] Figure 2 This is a microscopic section of the ovary;

[0046] Figure 3 Flow chart of the method for identifying primordial follicles in ovarian microscopic sections;

[0047] Figure 4 A diagram of the automatic identification system of primordial follicles in ovarian microscopic section images. DETAILED DESCRIPTION

[0048] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] It should be noted that, in the description of the present invention, the directions or positional relationships indicated by the terms "up", "down", "left", "right", "front", "back", etc. are descriptions of the structure of the present invention based on the accompanying drawings, and are only for the convenience of describing the present invention, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0050] The "first" and "second" in this technical solution are only used to distinguish the names of the same or similar structures, or corresponding structures with similar functions, and are not an arrangement of the importance of these structures, nor do they have a ranking, comparison of size, or other meanings.

[0051] In addition, unless otherwise clearly specified and limited, the terms "installation" and "connection" should be understood in a broad sense. For example, the connection can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be a connection between the two structures. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood based on the overall idea of ​​the present invention and the specific context of the present solution.

[0052] Example:

[0053] like Figure 1 and Figure 2As shown in the figure, the primordial follicle (Primordial Follicle), which contains the oocyte (Oocyte) and the outer granulosa cells. The oocyte has a nucleus, and the nucleus is relatively large, located in the center of the oocyte, round or oval, and stained darker (basophilic). The oocyte is wrapped by a single layer of flat follicle cells, which are close to the surface of the oocyte and arranged tightly. There is no zona pellucida around the oocyte. There is no obvious basement membrane or follicular cavity (Antrum) in the oocyte. The next stage of the growth of the primordial follicle is the primary follicle. How to accurately distinguish between the primordial follicle and the primary follicle is a very important factor affecting the recognition rate of the primordial follicle. In the primary follicle, the zona pellucida begins to appear around the oocyte, which is composed of glycoproteins and is stained lighter. The outer layer of the oocyte is surrounded by a single layer of cubic granules, and the thickness of the granulosa cell layer is thicker than that of the primordial follicle, while the oocyte in the primordial follicle is surrounded by a single layer of flat follicle cells. These are the two biggest differences between primary follicles and primordial follicles. Figure 2 As shown in the figure, the thickness of the granulosa cell layer of the primary follicle is significantly greater than that of the primordial follicle. The mature follicle has multiple layers of granulosa cell layers, the granulosa cells are closely arranged, the oocyte has the largest diameter, and there is a distinct crescent-shaped follicular cavity and corona radiata structure.

[0054] According to the different texture structures of follicles at different stages, the corresponding image processing algorithm can effectively identify follicles at different stages, and mark and count the primordial follicles for doctors' reference. The specific processing steps are as follows.

[0055] Step s1: binarize the stained slice image under a microscope to obtain a binary image.

[0056] After binarization of the stained slice image under the microscope, a binary image can be obtained to extract the significant feature areas, including the outer granulosa cells and the cell nucleus. The specific implementation method is as follows: the stained slice image is collected through a digital microscope to obtain a high-resolution image containing the oocyte, the cell nucleus and the surrounding granulosa cells. The image format is an RGB three-channel color image. The RGB color image is converted into a grayscale image. Through grayscale processing, the data dimension is reduced while retaining the main information of the cell structure, so that the cell nucleus and granulosa cells appear as lower grayscale value areas in the grayscale image.

[0057] The Otsu threshold segmentation method is used to automatically calculate the optimal threshold T based on the image histogram to separate the foreground (target area, such as cell nuclei and granulosa cells) from the background (other tissues or blank areas). It marks the area with pixel grayscale values ​​lower than the threshold T as the foreground, and the area higher than T as the background, generating a binary image: in the binary image, the outer granulosa cell area appears as a connected area surrounding the oocyte, while the cell nucleus appears as a circular or nearly circular area with a darker center. Through the binarization process, these features are significantly retained, laying the foundation for subsequent regional segmentation and classification.

[0058] Step s2: Perform expansion and corrosion processing on the binary image to fill the disconnected areas of the granule cells and retain the true boundaries of the granule cells.

[0059] After the binary image generation is completed, due to the uneven staining of the microscopic sections and the changes in the illumination of the microscope imaging, the granulosa cells in the outer layer of the primordial follicle may be broken or gapped. This phenomenon will affect the subsequent accurate identification of the entire area of ​​the follicle. Therefore, the present invention uses an expansion operation to deal with this problem, using the image after binary processing as input, where the disconnected area of ​​the granulosa cells is shown as a gap between the small background area and the foreground area. In order to adapt to the annular distribution of the granulosa cells in the outer layer of the follicle, the circular structural element is selected to effectively fill the disconnected area. The radius of the structural element is dynamically set according to the resolution of the microscopic image. The recommended size is 3 to 5 pixels to ensure that no additional noise is introduced while filling the gap. The expansion operation expands the boundaries of the foreground pixels and converts the background pixels in contact with the structural element into foreground pixels, thereby filling the disconnected area. The circular structural element with an appropriate radius is used to expand the image to connect the broken parts of the edge of the granulosa cells. After the expansion process, the excess part is removed by the corrosion operation, and only the real boundary of the granulosa cells is retained.

[0060] Step s3: Apply an edge detection algorithm to detect boundary information and extract a circular boundary area as a candidate circular boundary area; the candidate circular boundary area includes a potential follicle area, and the potential follicle area includes a circular area surrounded by cell nuclei and a circular area surrounded by granulosa cells.

[0061] Use Canny edge detection or Sobel operator to extract the boundary information in the image. This step can help identify possible circular areas in the binary image, whether hollow or solid. Input the binary image after dilation processing, output the intensity map of all boundaries in the binary image, and mark the boundaries of the foreground area in the image. And because the follicles are circular or quasi-circular, after obtaining the boundary information, the circular boundary information is extracted, which is the potential follicle area. The potential follicle area includes the circular area surrounded by the cell nucleus and the circular area surrounded by the granulosa cells.

[0062] Step s4: the candidate circular boundary regions detected in step s3 are screened for hollow circle features and circularity features. The circular region surrounded by cell nuclei presents a feature of a solid circle, and the circular region surrounded by granular cells presents a feature of a solid middle cell nucleus portion and a hollow circle for the rest of the portion. This step filters out interference from the cell nucleus region and obtains a circular region representing the granular cells.

[0063] There are two cases of the circular boundary area obtained in step s3. The first is the outer granulosa cells of the follicle, which is also the target area; the second is the area where the cell nucleus is located. If it is the area where the cell nucleus is located, this area belongs to the noise interference area and needs to be filtered; the third is the interference of other circular areas. According to the texture characteristics of the outer granulosa cells and the cell nucleus, the cell nucleus as a whole is a solid dark area, and the inner part of the circular boundary of the outer granulosa cells contains a dark cell nucleus area, but also contains a certain area of ​​light-colored areas. From the perspective of the image after binarization, the cell nucleus is a solid circle, and the central cell nucleus area of ​​the outer granulosa cells is solid, and the outer layer of the cell nucleus is a hollow area. According to the different characteristics of this image, the following steps can be used to screen out the target area corresponding to the outer granulosa cells.

[0064] Step S41: using an inner filling algorithm to perform hollow feature screening on the candidate circular boundary region to obtain a hollow circular region; using the image inner filling algorithm to fill each candidate region, and calculating the area difference between the regions before and after filling, if the ratio exceeds a threshold value T, the value range of T can be set to 2-3, then the region is defined as a hollow circle, and this step filters out the interference of the solid cell nucleus region;

[0065] By detecting the areas in the edge image with complete outer circles and hollow interiors, we can identify hollow circular areas. Use an in-image filling algorithm (such as Flood-Fill) to fill each candidate area and calculate the area difference between the areas before and after filling. If the border of a circular area is very regular and hollow, the area after filling will increase significantly, reflecting that its interior is divided into hollow structures. Calculate the ratio of the area of ​​the filled area to the area of ​​the original area. If the ratio exceeds a certain threshold T, according to the structural characteristics of the follicular cells, the proportion of the hollow part is 2-3 times the proportion of the area of ​​the cell nucleus area; setting T to a threshold of 2-3 has the best effect, preferably T can be set to 2-3, and the area is defined as a hollow circle. For the cell nucleus area, the area after filling is slightly different from the area of ​​the original area, and its ratio does not meet the range of the threshold T, so it can be filtered.

[0066] Step s42: Calculate and screen the circularity of the hollow circular areas to obtain the circular areas representing the candidate granule cells; calculate the circularity of each hollow circular area, and set the range of the circularity value to be within the threshold interval T1, and the range of T1 is [0.8-1].

[0067] The shape of the follicle cell is round or oval, and its circularity value is relatively high. The circularity is calculated for each candidate area.

[0068] Where A is the area of ​​the hollow circular region, and P is the perimeter of the hollow circular region. A circularity value close to 1 indicates that the region is circular, and regions with circularity values ​​close to 1 are screened out. Preferably, the circularity value range can be set to be within the threshold interval T1, and the structural characteristics of the follicular cells are close to circular. The recognition effect is best when the T1 value range is set to [0.8-1]. This step can reduce the interference of other background noises with lower circularity in the image.

[0069] Step s5: The area in the color image corresponding to the circular area representing the granule cells identified in step S4 is taken as the region of interest (ROI).

[0070] The circular area surrounded by the representative granulosa cells selected in step s4 is the region of interest (ROI). At this time, solid areas such as stromal cells will be excluded, and follicular cells will be retained as the region of interest. For the retained region of interest, it is necessary to confirm again whether it is a primitive follicular cell. The edge position area of ​​the binarized image is corresponded to the color map, and the hollow circular area of ​​the granulosa cells corresponding to the color map is used as the region of interest, which contains a large number of texture features.

[0071] Step s6: Extracting texture features of the region of interest and granulosa cell layer thickness features; the granulosa cell layer of the primordial follicle is flat and single-layered, the granulosa cell layer of the primary follicle is cubic and granular, the thickness of the granulosa cell layer of the primary follicle is greater than that of the primordial follicle, and the granulosa cell layer thickness feature is introduced to distinguish primordial follicles from primary follicles to improve the recognition rate of primordial follicles;

[0072] The hollow circular region (ROI) screened out in step s4 is retained as a candidate follicle cell region. In order to further identify the follicle types (primordial follicles, primary follicles, secondary follicles, mature follicles) in these regions, it is necessary to classify based on the significant features of the follicle structure. There are differences between the texture structures of follicles at different stages. The thickness of the granulosa cell layer of the primordial follicle is flat and monolayer, the oocyte has the smallest diameter, and the shape feature is close to a circle. The outer granulosa cell layer has a simple texture. The stained microscopic section images are collected to ensure that all follicle types (primordial, primary, secondary, and mature follicles) are included. The boundaries and categories of each follicle are first manually marked, and a marked image is generated, and a training set is generated for training the model. Extract key features from each region of interest (ROI) for input to a machine learning classifier or a deep learning model.

[0073] Step s61: Texture feature extraction. The granulosa cell layer of the follicle and other regions are distinguished by texture analysis. Local binary pattern (LBP) features can be used to extract corresponding texture features. The primordial follicles are distinguished from primary follicles, secondary follicles and mature follicles based on their different texture features.

[0074] Step s62: extract the ratio of the area of ​​the connected domain of the granular cell layer in the region of interest to the circumference as the granular cell layer thickness feature.

[0075] In the primary follicle, a zona pellucida begins to appear around the oocyte, which is composed of glycoproteins and is lightly stained. The outer layer of the oocyte is wrapped by a single layer of cubic granules, and the thickness of the granulosa cell layer is thicker than that of the primordial follicle. In the primordial follicle, the oocyte is surrounded by a single layer of flat follicle cells. These are the two biggest differences between the primary follicle and the primordial follicle. The texture structure of the primordial follicle cells and the primary follicle is also the most similar, which can easily cause misclassification by machine recognition. Therefore, in addition to extracting texture features for judgment, it is also possible to make judgments based on the thickness characteristics of the granulosa cell layer between the two.

[0076] First, the connected domain of the granulosa cell layer is extracted. The granulosa cell layer is the outermost part of the region of interest (ROI) and is located outside the oocyte. The binary image of the granulosa cell layer is processed to obtain all connected regions. All connected regions are detected in the binary image of the granulosa cell layer, and the noise connected domains with too small areas are removed (the area threshold is set, such as 10 pixels), and the ratio of the connected domain area to the circumference is calculated.

[0077] Step a: Calculate the area of ​​the connected domain (A): Count the area of ​​the connected domain of the granular cell layer. The area of ​​the connected domain is the total number of all white pixels (value 1) in its binary image. Use a connected domain detection algorithm in the binary image (such as OpenCV's cv2.connectedComponents or Scikit-Image's label function) to mark all connected areas. For each connected domain, count the total number of white pixels:

[0078] Step b: Calculate the circumference (P) of the connected domain: Use boundary detection (such as the Canny algorithm) to extract the outer boundary of the granule cell layer and calculate the length of the boundary. If the boundary pixels are distributed regularly and approximately continuous, the number of boundary pixels in the image can be used as an estimate of the circumference.

[0079] Step c: Calculate the ratio V of the total area A to the circumference P. The thickness of the granulosa cell layer of the primary follicle is greater than that of the primordial follicle. The follicle cells whose area to circumference ratio V is in the range of 1-3 are identified as potential primordial follicle cells.

[0080] Step s7: Fusion and classification of the texture features of the region of interest and the thickness features of the granular cell layer.

[0081] The texture features of the region of interest extracted in step s6 and the thickness features of the granulosa cell layer are fused to obtain fused features, and the fused features are used to classify the follicular cells. The classifier can use a support vector machine (SVM). The feature vector is input into the SVM model to output the follicle category.

[0082] Step s8: Primordial follicle cell counting and labeling.

[0083] The identified primordial follicle areas are numbered and counted one by one, and the number and location of the follicles are output. In addition, the follicle areas at different stages in the image can be distinguished with different colors and marked for the doctor's reference, which saves a lot of statistical time for the doctor and can play a role in assisting judgment.

[0084] A system for automatically identifying primordial follicles in ovarian microscopic slice images, such as Figure 4 As shown, it includes a memory storing executable program code and a processor coupled to the memory; wherein the processor calls the executable program code stored in the memory to execute the method steps disclosed in the above embodiment.

[0085] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A method for automatically identifying primordial follicles in ovarian microscopic section images, characterized in that: The following steps are involved: Step s1: binarizing the stained slice image under a microscope to obtain a binary image; Step s2: Perform expansion and corrosion processing on the binary image to fill the disconnected area of ​​the granule cells and retain the real boundary of the granule cells; Step s3: applying an edge detection algorithm to detect boundary information, and extracting a circular boundary area as a candidate circular boundary area; the candidate circular boundary area includes a potential follicle area, and the potential follicle area includes a circular area surrounded by cell nuclei and a circular area surrounded by granulosa cells; Step s4: the candidate circular boundary regions detected in step s3 are screened for hollow circle features and circularity features. The circular region surrounded by cell nuclei presents a feature of a solid circle, and the circular region surrounded by granular cells presents a feature of a solid middle cell nucleus portion and a hollow circle for the rest of the portion. This step filters out interference from the cell nucleus region and obtains a circular region representing the granular cells. Step s5: The area in the color image corresponding to the circular area representing the granule cells identified in step s4 is taken as the region of interest ROI; Step s6: Extracting texture features of the region of interest and granulosa cell layer thickness features; the granulosa cell layer of the primordial follicle is flat and single-layered, the granulosa cell layer of the primary follicle is cubic and granular, the thickness of the granulosa cell layer of the primary follicle is greater than that of the primordial follicle, and the granulosa cell layer thickness feature is introduced to distinguish primordial follicles from primary follicles to improve the recognition rate of primordial follicles; Step s7: Fusion and classification of the texture features of the region of interest and the thickness features of the granular cell layer; Step s8: Primordial follicle cell counting and labeling.

2. The method according to claim 1, characterized in that In the step s1, binarization is performed using the Otsu threshold segmentation method to separate the foreground from the background.

3. The method according to claim 2, characterized in that In step s2, a circular structural element is used to perform an expansion operation to fill the broken areas of the granular cells, and at the same time, an erosion operation is used to remove excess noise and retain the real boundary.

4. The method according to any one of claims 1 to 3, characterized in that In step s3, the boundary information of the foreground area is extracted using Canny edge detection or Sobel operator, and the circular boundary is screened out as a candidate circular boundary area.

5. The method according to claim 1, characterized in that In step s4, the hollow circular area is screened by the following sub-steps: Step s41: using an inner filling algorithm to perform hollow feature screening on the candidate circular boundary region to obtain a hollow circular region; using the image inner filling algorithm to fill each candidate region, and calculating the area difference between the regions before and after filling, if the ratio exceeds a threshold value T, the value range of T can be set to 2-3, then the region is defined as a hollow circle, and this step filters out the interference of the solid cell nucleus region; Step s42: calculating and screening the circularity of the hollow circular area to obtain the circular area surrounded by the candidate granule cells; The circularity is calculated for each hollow circular area, and the range of the circularity value is set to be within the threshold interval T1, and the range of T1 is 0.8-1.

6. The method according to claim 1, characterized in that In the step S6, the following sub-steps are performed: Step s61: Texture feature extraction, using local binary pattern LBP features to extract corresponding texture features, and distinguishing primordial follicles from primary follicles, secondary follicles and mature follicles based on their different texture features; Step s62: extract the ratio of the connected domain area of ​​the granular cell layer in the region of interest to the circumference as the granular cell layer thickness feature.

7. The method according to claim 6, characterized in that The step s62 specifically includes the following steps: Step a: Calculate the area A of the connected domain, count the area of ​​the connected domain of the granule cell layer, mark all connected areas, and count the total number of white pixels in each connected domain; Step b: Calculate the circumference P of the connected domain, use boundary detection to extract the outer boundary of the granule cell layer, calculate the length of the boundary, and use the number of boundary pixels in the image as an estimate of the circumference; Step c: Calculate the ratio V of the total area A to the circumference P. The thickness of the granulosa cell layer of the primary follicle is greater than that of the primordial follicle. The follicle cells whose area to circumference ratio V is in the range of 1-3 are identified as potential primordial follicle cells.

8. A system for automatically identifying primordial follicles in ovarian microscopic section images, characterized in that: Features: It comprises a memory storing executable program code and a processor coupled to the memory; wherein the processor calls the executable program code stored in the memory to execute the method according to any one of claims 1-7.

Images