A method for locating adhesion lesions in a rat model of endometriosis

Abstract

This invention relates to the field of image recognition technology, specifically to a method for locating adhesion lesions in a rat model of endometriosis. This application first adjusts the clustering distance based on color differences and positional inertia between saturation connected domains to obtain connected domain clusters. Since these clusters contain ectopic glands, the clusters need to be screened to obtain the ectopic glands. By utilizing the characteristic that columnar epithelial cells are distributed on the inner wall of the ectopic glands, the probability of each connected domain cluster being a gland is obtained, determining the likelihood that each cluster belongs to an ectopic gland. Because ectopic glands exhibit a concentrated distribution pattern, the agglomeration of connected domain clusters with higher gland probabilities is used to obtain the abnormal probability of these clusters, further determining the likelihood that each cluster belongs to an abnormal ectopic gland, thus improving the accuracy of lesion localization in the rat model of endometriosis.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a method for locating adhesion lesions in a rat model of endometriosis. Background Technology

[0002] A rat model of endometriosis is an experimental animal model established by mimicking the pathological features of human endometriosis, primarily used to study the pathogenesis and treatment of this disease. Common construction methods include autologous transplantation and allogeneic transplantation, where endometrial tissue from healthy rats is implanted into their abdominal cavity or other ectopic sites to form a pathological structure similar to human endometriotic lesions. This model can exhibit typical characteristics of endometriosis, such as the formation of ectopic lesions, inflammatory responses, pain behaviors, and fertility dysfunction, providing a reliable platform for a deeper understanding of the disease mechanism and the evaluation of the efficacy of new drugs.

[0003] Localizing lesions in a rat model of endometriosis allows for a more direct assessment of their distribution, size, and number, thus facilitating better teaching about the disease's pathological mechanisms and progression. Furthermore, precise lesion localization is crucial for histological analysis, molecular biological testing, and evaluating the sensitivity and specificity of novel diagnostic tools, providing key data support for basic research and clinical translation in endometriosis.

[0004] Hematoxylin-eosin (HE) staining is widely used in tissue section staining. Current techniques typically rely on optical microscopy to observe HE-stained sections and identify and locate ectopic lesions based on their typical histological features. However, because lesions in rat models of endometriosis coexist with surrounding tissues in the stained images, the lesion areas are not obvious, making it difficult to distinguish some lesion areas from normal tissues. Consequently, the lesion localization in rat models of endometriosis is inaccurate. Summary of the Invention

[0005] To address the technical problem of inaccurate lesion localization in rat models of endometriosis, where lesions coexist with surrounding tissues in stained images, resulting in indistinct lesion areas and difficulty in distinguishing some lesion areas from normal tissue, this application aims to provide a method for localizing adhesion lesions in rat models of endometriosis. The specific technical solution adopted is as follows:

[0006] This application proposes a method for locating adhesion lesions in a rat model of endometriosis, the method comprising:

[0007] Acquire stained section images of a rat model of endometriosis; determine all saturation connected components in each saturation interval based on the saturation distribution of the stained section images; obtain all grayscale connected components after grayscale conversion of the stained section images;

[0008] Based on the distance between saturation connected components and the saturation deviation of the corresponding saturation interval, the boundary probability of each saturation connected component is determined; based on the boundary probability, the positional distribution between each saturation connected component, and the relative saturation deviation, at least two connected component clusters are selected.

[0009] Region synthesis is performed based on the distribution and adjacency of each gray-level connected domain and the approximation of columnar features in morphology to determine all suspected columnar epithelial regions; based on the number of gray-level connected domains and the columnar approximation in the suspected columnar epithelial regions, the corresponding epithelial probability is determined.

[0010] Based on the boundary probability distribution in each connected domain cluster and the epithelial probability of adjacent suspected columnar epithelial regions, the corresponding ectopic gland probability is determined; based on the ectopic gland probability and the concentrated distribution of connected domain clusters, the anomaly probability of each connected domain cluster is determined.

[0011] Abnormal connected domain clusters are selected based on the abnormal probability; adhesion lesions in the rat model of endometriosis are located based on the abnormal connected domain clusters.

[0012] Furthermore, the process of obtaining the saturation connected component includes:

[0013] The stained slice image is converted into an HSV image; all local minima in the distribution histogram of the S channel corresponding to the HSV image are obtained; the distribution histogram of the S channel is divided into at least two saturation intervals with all local minima as intervals; all connected components formed by the pixels in each saturation interval are taken as saturation connected components.

[0014] Furthermore, the process of obtaining the boundary possibilities includes:

[0015] Each saturation-level connected component is sequentially designated as the target connected component; the saturation interval containing the target connected component is designated as the target interval; and other saturation intervals outside the target interval are designated as reference intervals.

[0016] The average distance between the centroids of all saturated connected regions in each reference interval and the centroid of the target connected region is negatively correlated to determine the boundary contribution index of each reference interval to the target connected region.

[0017] The difference between the maximum S-channel value of each reference interval and the maximum S-channel value of the target interval is taken as the saturation interval deviation for each reference interval.

[0018] Based on the product of the boundary contribution index and the saturation interval deviation, the boundary influence parameter of each reference interval on the target connected domain is determined; the cumulative value of the boundary influence parameters of all reference intervals corresponding to the target connected domain is normalized to determine the boundary probability of the target connected domain.

[0019] Furthermore, the process of obtaining the connected component clusters includes:

[0020] Based on the extension of the boundary minimum probability within the adjacent regions of the target connected component, the central connected component of the target connected component is determined; other saturated connected components outside the target connected component are used as comparison connected components.

[0021] The saturation deviation value of each contrasting connected component is determined based on the difference between the mean S-channel value of all pixels in each contrasting connected component and the mean S-channel value of all pixels in the target connected component.

[0022] The distance between the centroid of the target connected component and the centroid of each compared connected component is taken as the first comparison distance of each compared connected component; the distance between the centroid of the central connected component and the centroid of each compared connected component is taken as the second comparison distance of each compared connected component.

[0023] Calculate the difference between the second contrast distance and the first contrast distance, and use the product of the difference and the boundary probability as the boundary weighting coefficient; calculate the sum of the boundary weighting coefficient and the first contrast distance; use the product of the sum and the saturation deviation value as the clustering distance between the target connected component and each contrast connected component;

[0024] Calculate the clustering distance between every two saturated connected components; based on all clustering distances, perform cluster analysis on all saturated connected components using the DBSCAN clustering algorithm to obtain at least two connected component clusters.

[0025] Furthermore, the process of obtaining the central connected domain includes:

[0026] The reference connected region with the smallest boundary probability among each saturated connected region and all its adjacent saturated connected regions is taken as the corresponding minimum connected region.

[0027] Using the target connected region as the seed point, the acquisition of the minimum connected region as the growth condition, and the minimum connected region as the new seed point, the region is grown until the resulting grown region does not change. The minimum connected region obtained at the end of the region growth process is then used as the central connected region of the target connected region.

[0028] Furthermore, the process of obtaining the suspected columnar epithelial region includes:

[0029] The ratio between the width and length of the minimum bounding rectangle of each gray-level connected region is negatively correlated to determine the columnarity of each gray-level connected region; gray-level connected regions with columnarity greater than a preset columnar threshold are designated as columnar connected regions.

[0030] Merge all adjacent columnar connected regions to obtain all suspected columnar epithelial regions.

[0031] Furthermore, the process of obtaining the epithelial probability includes:

[0032] The product of the mean columnarity of all columnar connected regions in the suspected columnar epithelial region and the total number of columnar connected regions is normalized to determine the corresponding epithelial probability.

[0033] Furthermore, the process of obtaining the probability of ectopic glands includes:

[0034] The epithelial probability of the nearest suspected columnar epithelial region to each saturation connected region is used as the adjacency reference probability of each saturation connected region.

[0035] The product of the normalized boundary probability of each saturated connected region and the adjacent reference probability is used as the local reference probability.

[0036] The probability of the corresponding ectopic gland is determined by summing the local reference probabilities of all saturated connected components in each connected component cluster.

[0037] Furthermore, the process of obtaining the anomaly probability includes:

[0038] Each connected component cluster is then designated as the target cluster; other connected component clusters not included in the target clusters are designated as comparison clusters.

[0039] The minimum distance between all pixels in each comparison cluster and all pixels in the target cluster is taken as the inter-cluster distance of each comparison cluster.

[0040] The product of the negative correlation normalized value of the inter-cluster distance and the ectopic gland probability of the corresponding contrastive cluster is used as the reference concentration of each contrastive cluster.

[0041] The product of the mean of the reference concentration of all comparative clusters and the probability of ectopic glands in the target cluster is taken as the anomaly probability of the target cluster.

[0042] Furthermore, the process of obtaining the abnormal connected component clusters includes:

[0043] Clusters of connected components with an anomaly probability greater than a preset anomaly threshold are designated as abnormal connected component clusters.

[0044] This application has the following beneficial effects:

[0045] In the process of locating endometriosis in a rat model, since glands are the most specific and distinctive structures in the lesions, this invention focuses on acquiring the location of glandular structures. Because uneven staining during HE staining can prevent the complete segmentation of individual glandular structures, this invention first obtains several saturation-level connected domains using the S-channel. The clustering distance is adjusted based on the color differences and positional inertia between these domains to obtain connected domain clusters. Since these clusters contain ectopic glands, they need to be screened to identify them. The probability of each connected domain cluster containing glands is obtained by analyzing the columnar epithelial cells distributed on the inner wall of the ectopic glands, thus determining the likelihood that each cluster belongs to an ectopic gland. Because ectopic glands exhibit a concentrated distribution pattern, the anomalous probability of connected domain clusters with higher gland probabilities is obtained through clustering, further determining the likelihood that these clusters belong to abnormal ectopic glands. Thus, this invention improves the accuracy of lesion localization in a rat model of endometriosis by annotating the locations of potential lesions in stained slide images through the abnormal probability of connected component clusters. Attached Figure Description

[0046] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of a method for locating adhesion lesions in a rat model of endometriosis provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the morphological structure of columnar epithelial cells provided in one embodiment of the present invention. Detailed Implementation

[0049] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for locating adhesion lesions in a rat model of endometriosis according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment, and specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0051] The following describes in detail, with reference to the accompanying drawings, a specific scheme for locating adhesion lesions in a rat model of endometriosis provided by the present invention.

[0052] This application provides a method for locating adhesion lesions in a rat model of endometriosis. Please refer to [link to relevant documentation]. Figure 1 The diagram illustrates a flowchart of a method for locating adhesion lesions in a rat model of endometriosis according to an embodiment of the present invention. The method includes:

[0053] Step S101: Collect stained section images of the rat model of endometriosis; determine all saturation connected components under each saturation interval based on the saturation distribution of the stained section images; obtain all grayscale connected components after grayscale conversion of the stained section images.

[0054] In one specific implementation of this invention, a rat model of endometriosis is first created. Specifically, cut endometrial tissue is applied to a subcutaneous fascia rich in blood vessels. The rats are then fed normally until the model is confirmed to be established, thus obtaining the rat model of endometriosis. Subsequently, HE-stained sections are prepared, and images of the stained sections from the rat model of endometriosis are acquired using a microscope.

[0055] Endometriosis is caused by the shedding of endometrial tissue under certain pathological conditions and its migration to external sites such as the abdominal cavity and ovaries via retrograde menstruation, lymphatic or blood circulation. This results in endometrial tissue growing outside the uterine cavity and myometrium. Therefore, the lesions of endometriosis are structurally similar to normal endometrial tissue. Endometrial tissue includes glandular structures and stromal tissue. When locating lesions in a rat model of endometriosis, glandular structures should be identified, as they are the most specific and characteristic structures in the lesions, directly reflecting the presence of endometrial tissue. Under HE staining, ectopic glands appear as light pink cytoplasm, while they are usually surrounded by fibrotic stromal tissue, which appears as dark pink fibers. This contrast in saturation allows for differentiation between ectopic glands and ectopic stroma based on color. Therefore, preferably, in some possible implementations of this invention, the process of obtaining the saturation connected region includes:

[0056] The stained slice image is converted into an HSV image. All local minima are obtained from the distribution histogram of the S-channel of the HSV image. The distribution histogram of the S-channel is divided into at least two saturation intervals, with each local minima as an interval. All connected components formed by pixels within each saturation interval are considered as saturation connected components. The existence of local minima clearly reveals different saturation characteristics; therefore, saturation intervals are divided using local minima. It should be noted that the S-channel corresponds to the saturation value, and subsequent S-channel values ​​correspond to saturation values; these are standard concepts in HSV images and will not be elaborated further here.

[0057] In one specific implementation of this invention, the stained slide image is converted to grayscale to obtain a grayscale image of the slide; connected component analysis is performed on the grayscale image of the slide to obtain all grayscale connected components obtained in this embodiment of the invention; the grayscale connected components are used for subsequent determination of columnar epithelial regions, which will not be further elaborated here.

[0058] Step S102: Determine the boundary probability of each saturation connected component based on the distance between them and the saturation deviation of the corresponding saturation interval; based on the boundary probability, the positional distribution between each saturation connected component, and the relative saturation deviation, select at least two connected component clusters.

[0059] Because the components in ectopic glands have different affinities for dyes, the region of the ectopic gland is divided into blocky regions with small differences in saturation. That is, the interior of the ectopic gland is divided into several saturation connected regions. Therefore, it is necessary to cluster the saturation connected regions to obtain the location of the entire ectopic gland.

[0060] When clustering saturation connected components, the clustering distance is mainly adjusted based on the color of the saturation connected components. However, during clustering, the color difference (i.e., the saturation difference) of saturation connected components located at the boundary of ectopic glands and those located at the center of ectopic glands relative to saturation connected components of the same stromal tissue is the same or similar. However, their Euclidean distances to saturation connected components of the same stromal tissue differ greatly. Therefore, the clustering distance cannot be adjusted solely based on color difference. It is especially necessary to distinguish between saturation connected components at the boundary of ectopic glands and stromal tissue, so that saturation connected components belonging to ectopic glands can be clustered into one class, making the obtained ectopic glands more complete. Therefore, the boundary probability of saturation connected components is calculated.

[0061] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the boundary possibilities includes:

[0062] Each saturated connected component is sequentially designated as the target connected component; the saturation interval in which the target connected component is located is designated as the target interval; other saturation intervals outside the target interval are designated as reference intervals; the average distance between the centroids of all saturated connected components in each reference interval and the centroid of the target connected component is negatively correlated and mapped to determine the boundary contribution index of each reference interval to the target connected component.

[0063] For each reference interval, the smaller the relative distance between each saturation connected region and the target connected region, the more likely that the saturation connected regions closer to the target connected region are located within the corresponding reference interval. Since the saturation difference between ectopic glands and ectopic stroma is significant, the boundary contribution index is relatively large. Furthermore, the greater the difference between the saturation characteristics in the corresponding reference region and the target connected region, the higher the probability that the target region is located at the boundary. Based on this characteristic, the difference between the maximum S-channel value of each reference interval and the maximum S-channel value of the target interval is further used as the saturation interval deviation for each reference interval. Then, the boundary influence parameter of each reference interval on the target connected region is determined by multiplying the boundary contribution index and the saturation interval deviation. The cumulative value of the boundary influence parameters of all reference intervals corresponding to the target connected region is normalized to determine the boundary probability of the target connected region.

[0064] In one specific implementation of this invention, the process of obtaining boundary possibilities is expressed by the following formula: Among them, D k A represents the boundary probability of the target connected region k; k S represents the number of reference intervals corresponding to the target connected component k; k S represents the maximum value of the S channel within the saturation interval corresponding to the target connected component k; k,aB represents the maximum value of the S-channel in the a-th reference interval corresponding to the target connected component k; || represents the absolute value sign; a d represents the number of saturated connected components in the a-th reference interval corresponding to the target connected component k; k,a,b Let be the distance between the centroid of the target connected region k and the centroid of the b-th saturation connected region in the corresponding a-th reference interval; exp() is an exponential function with the natural constant as the base. The boundary contribution index of the corresponding a-th reference interval to the target connected region k; is the boundary influence parameter of the corresponding a-th reference interval on the target connected region k; Norm() is the linear normalization function; further, the boundary probability of all saturated connected regions is determined according to the boundary probability acquisition process of the target connected region.

[0065] The clustering distance between saturation connected components is further adjusted based on boundary probability, that is, the boundary probability is used as the degree of adjustment. Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining connected component clusters includes:

[0066] Based on the extension of the minimum boundary probability among the adjacent regions of the target connected region, the central connected region of the target connected region is determined. The process of obtaining the central connected region includes: taking the reference connected region with the minimum boundary probability among each saturated connected region and all its adjacent saturated connected regions as the corresponding minimum connected region; using the target connected region as the seed point, using the acquisition of the minimum connected region as the growth condition, and using the minimum connected region as the new seed point, performing region growth until the obtained grown region does not change, and taking the final minimum connected region obtained in the region growth process as the central connected region of the target connected region. That is, for the target connected region, the saturated connected region with the minimum boundary probability among the target connected region itself and all its adjacent saturated connected regions is obtained, and the iteration continues to obtain the saturated connected region with the minimum boundary probability among all its adjacent saturated connected regions, until the minimum saturated connected region does not change, the iteration stops, and the minimum saturated connected region is recorded as the central connected region of the target connected region; by determining the central connected region, it is possible to avoid saturated connected regions at the boundary being assigned to other clusters when determining the cluster distance in the subsequent process.

[0067] Furthermore, other saturated connected regions outside the target connected region are used as contrast connected regions. Based on the difference between the mean S-channel values ​​of all pixels in each contrast connected region and the mean S-channel values ​​of all pixels in the target connected region, the saturation deviation value of each contrast connected region is determined. The smaller the saturation deviation value, the more likely they belong to the same region, and the smaller the corresponding clustering distance should be.

[0068] The distance between the centroid of the target connected component and the centroid of each contrast connected component is taken as the first contrast distance of each contrast connected component; the distance between the centroid of the central connected component and the centroid of each contrast connected component is taken as the second contrast distance of each contrast connected component; the difference between the second contrast distance and the first contrast distance is calculated, and the product of the difference and the boundary probability is taken as the boundary weighting coefficient. For a target connected component, if it is located at the boundary and the compared connected components do not belong to the same type of region, then theoretically the corresponding clustering distance should be larger; conversely, if the compared connected components belong to the same type of region, then the corresponding clustering distance should be relatively smaller. Therefore, for a target connected component with a high probability of being a boundary, its distance to the compared connected components of other types of regions is usually smaller. So, in order to avoid being classified into other connected components, it is necessary to adjust the distance by combining the difference between the second and first comparison distances. The larger the second comparison distance is relative to the first comparison distance, the greater the probability of the target boundary, that is, the larger the boundary weighting coefficient, which indicates that the situation is more consistent. At this time, it is necessary to increase the physical distance to a certain extent, so that the compared connected components and the target connected component are classified into different clusters. Therefore, the sum of the boundary weighting coefficient and the first comparison distance is further calculated. The product of the sum and the saturation deviation value is used as the clustering distance between the target connected component and each compared connected component. After adding the boundary weighting distance and the first comparison distance, the probability of the connected component located at the boundary being classified into other types of regions is reduced, making the obtained clustering distance more reasonable.

[0069] In one specific implementation of this invention, the process of obtaining the clustering distance is expressed by the formula: F k,h =|G k -G k，h |×(H k,h +(H′ k,h -H k,h )×D k ); where F k,h G is the clustering distance between the target connected component k and its corresponding h-th comparison connected component; k G is the mean of the S-channel values ​​of all pixels within the target connected component k; k,h For the h-th compared connected component corresponding to the target connected component k; D k H represents the boundary probability of the target connected region k; k，h H′ is the distance between the centroid of the target connected component k and the centroid of the corresponding h-th comparison connected component, which is also the first comparison distance; kLet H' be the distance between the centroid of the central connected component of the target connected component k and the centroid of the corresponding h-th comparison connected component, which is also the second comparison distance; it should be noted that all distances in this embodiment of the invention are calculated using the Euclidean distance calculation method; (H′ k，h -H k，h )×D k The boundary weighting coefficients between the target connected component k and the corresponding h-th comparison connected component are given.

[0070] Further, the clustering distance between every two saturated connected components is calculated; based on all clustering distances, the DBSCAN clustering algorithm is used to perform clustering analysis on all saturated connected components to obtain at least two connected component clusters; it should be noted that the DBSCAN clustering algorithm is a well-known technique in the art, and other clustering methods, such as the k-means clustering algorithm, can also be used depending on the specific implementation environment, without further limitation or elaboration here.

[0071] Step S103: Based on the distribution and adjacency of each gray-level connected domain and the approximation of columnar features in morphology, perform region synthesis to determine all suspected columnar epithelial regions; based on the number of gray-level connected domains and the columnar approximation in the suspected columnar epithelial regions, determine the corresponding epithelial probability.

[0072] Each connected component cluster represents a complete regional tissue structure. Some connected component clusters belong to ectopic glands, so it is necessary to screen for ectopic glands from the connected component clusters, i.e., calculate the gland probability of each connected component cluster. Since ectopic glands are usually surrounded by columnar epithelial cells distributed along the inner wall of the glandular lumen, and columnar epithelial cells are columnar or rectangular, the aspect ratio of the minimum bounding rectangle of columnar epithelial cells is relatively large. Furthermore, because ectopic glands are usually surrounded by columnar epithelial cells with a continuous arrangement, regions that may contain columnar epithelial cells are identified in the image. Then, based on the surrounding nature of the columnar epithelial cells around the ectopic gland boundary, the gland probability of the analyzed region is determined. Please refer to [link to relevant documentation]. Figure 2 It shows a schematic diagram of the morphological structure of columnar epithelial cells provided in an embodiment of the present invention. Figure 2 In this cell, the columnar epithelial cells are all slender columnar in shape and are densely distributed in a continuous manner when they are closely attached.

[0073] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the suspected columnar epithelial region includes:

[0074] The columnarity of each gray-level connected region is determined by negatively mapping the ratio of its width to its length to the minimum bounding rectangle. Gray-level connected regions with a columnarity greater than a preset columnarity threshold are designated as columnar connected regions. All adjacent columnar connected regions are then merged to obtain all suspected columnar epithelial regions. Firstly, columnar epithelial cells exhibit a columnar shape, approximating a rectangle with a longer longer side. Therefore, after drawing a minimum bounding rectangle, the length is longer than the width, resulting in a higher calculated columnarity, which increases the probability that the corresponding gray-level connected region belongs to a columnar epithelial cell. Furthermore, columnar epithelial cells are arranged continuously; therefore, all adjacent columnar connected regions are further merged to determine the corresponding suspected columnar epithelial regions that may belong to the epithelial cell area. It should be noted that the width is the shorter side of the minimum bounding rectangle, and the length is the longer side.

[0075] In one specific implementation of this invention, the process of obtaining the columnar degree is expressed by the formula: Among them, Z r L1 represents the columnarity of the r-th gray-level connected region. r L2 is the width of the minimum bounding rectangle of the r-th gray-level connected region. r Let be the length of the minimum bounding rectangle of the r-th gray-level connected region. In one specific implementation of this invention, the preset columnar threshold is set to 0.7, which can be adjusted according to the specific implementation environment.

[0076] Furthermore, based on the continuous arrangement of columnar epithelial cells, the reliability of the obtained suspected columnar epithelial regions is measured, that is, the probability that they belong to true columnar epithelial regions is verified, thereby better assigning weights for subsequent gland probability calculations. Preferably, in some possible implementations of this invention, the process of obtaining the epithelial probability is expressed by the following formula: The process of obtaining the epithelial probability includes:

[0077] The product of the mean columnarity of all columnar connected regions in a suspected columnar epithelial region and the total number of columnar connected regions is normalized to determine the corresponding epithelial probability. For each suspected columnar epithelial region, the greater the overall columnarity of its columnar connected regions, the more its local shape features match the shape of epithelial cells; and the greater the number of columnar connected regions, the denser and more continuous the epithelial cells, and the more likely they are to conform to the characteristic of continuous arrangement of epithelial cells. Therefore, by combining the mean columnarity of all columnar connected regions with the total number of columnar connected regions through a product, the epithelial probability representing a true columnar epithelial region is determined. The purpose of normalization is to allow the epithelial probability to play a better role in subsequent weighting.

[0078] In one specific implementation of this invention, the process of obtaining the epithelial probability is expressed by the formula:

[0079] Among them, Q q The epithelial probability of the qth suspected columnar epithelial region; M is the mean of the columnarity of all columnar connected regions in the q-th suspected columnar epithelial region; q represents the total number of columnar connected regions in the q-th suspected columnar epithelial region; Norm() is the linear normalization function.

[0080] Step S104: Determine the corresponding ectopic gland probability based on the boundary probability distribution in each connected domain cluster and the epithelial probability of adjacent suspected columnar epithelial regions; determine the anomaly probability of each connected domain cluster based on the ectopic gland probability and the concentrated distribution of connected domain clusters.

[0081] Since ectopic glands are typically surrounded by columnar epithelial cells distributed along the inner wall of the glandular lumens, the boundary of the ectopic gland is very close to the epithelial cells. Therefore, saturation connected regions with higher boundary probability are also very close to the epithelial cells. Consequently, the probability of epithelial tissue in the suspected columnar epithelial region closest to the saturation connected region with higher boundary probability within the connected region cluster is higher, and the probability of the connected region cluster containing the gland is also higher. Based on this logic, the process of obtaining the ectopic gland probability for each connected region cluster includes:

[0082] The epithelial probability of the nearest suspected columnar epithelial region to each saturation connected region is used as the adjacency reference probability of each saturation connected region. The product of the normalized boundary probability of each saturation connected region and the adjacency reference probability is used as the local reference probability. The probability of ectopic glands is determined by summing the local reference probabilities of all saturation connected regions in each connected region cluster. Based on the characteristic that the higher the epithelial probability of the nearest suspected columnar epithelial region to the saturation connected region with the higher the boundary probability within the connected region cluster, the higher the gland probability of the connected region cluster, the higher the overall local reference probability of each saturation connected region for each connected region cluster. This indicates that columnar epithelial cells are more likely to be distributed on the edge of the corresponding location feature, which is more consistent with the characteristics of ectopic glands, i.e., the probability of ectopic glands is higher.

[0083] In one specific implementation of this invention, the process of obtaining the probability of ectopic glands is expressed by the following formula: Among them, P c U represents the probability of ectopic glands in the c-th connected component cluster; c D represents the number of saturation connected components in the c-th connected component cluster;c，u Let Q' be the boundary probability of the u-th saturated connected component in the c-th connected component cluster; softmax() is the activation normalization function, which makes the sum of all normalized values ​​equal to 1, that is, the sum of all normalized boundary probabilities equal to 1. Combined with the accumulation sign, this makes the calculated ectopic gland probability more robust; c，u Let be the epithelial probability of the nearest suspected columnar epithelial region in the u-th saturation connected region of the c-th connected region cluster.

[0084] In cases of endometriosis, endometrial tissue sheds under certain pathological conditions and migrates to locations outside the uterus, such as the abdominal cavity and ovaries, via retrograde menstruation, lymphatic or blood circulation. Ectopic glands typically concentrate in specific areas, exhibiting a clustered distribution. Therefore, the presence or absence of abnormalities can be determined by whether the ectopic glands exhibit this clustered distribution characteristic. Preferably, in some possible implementations of this invention, the process of obtaining the abnormality probability includes:

[0085] Each connected component cluster is sequentially designated as the target cluster; other connected component clusters outside the target clusters are designated as comparison clusters; the minimum distance between all pixels in each comparison cluster and all pixels in the target cluster is taken as the inter-cluster distance of each comparison cluster. For each connected component cluster, the smaller the overall inter-cluster distance between it and each comparison cluster, the more consistent the phenomenon of ectopic glands exhibiting a concentrated distribution is with the cluster, and the greater the probability of anomalies. Therefore, the negative correlation normalized value of the inter-cluster distance is further correlated with the ectopic gland distance of the corresponding comparison cluster. The product of the probabilities of glands is used as the reference concentration level for each comparative cluster. A higher reference concentration level indicates a greater likelihood of matching the characteristics of ectopic glands. The probability of ectopic glands, on the other hand, represents the likelihood of belonging to ectopic glands, i.e., the degree of abnormality. Therefore, the product of the mean of the reference concentration levels of all comparative clusters and the probability of ectopic glands in the target cluster is used as the abnormality probability of the target cluster, with the probability of ectopic glands as the weight and combined with the overall magnitude of the reference concentration level. A higher abnormality probability indicates a greater probability of belonging to ectopic glands, and a greater likelihood of matching the characteristics of the corresponding glandular lesions.

[0086] In one specific implementation of this invention, the process of obtaining the anomaly probability is expressed by the formula: Among them, T c P represents the anomaly probability of the c-th connected component cluster; c V represents the probability of ectopic glands in the c-th connected component cluster; c X represents the number of comparative clusters for the c-th connected component cluster; c,vP is the inter-cluster distance between the c-th connected component cluster and the v-th contrastive cluster; c,v The probability of ectopic glands in the v-th contrastive cluster of the c-th connected component cluster; softmax(exp(-X c，v ))×P c，v is the reference concentration of the vth comparative cluster of the Cth connected component cluster; exp() is the exponential function with the natural constant as the base; softmax() is the activation normalization function.

[0087] Step S105: Select abnormal connected domain clusters based on the abnormal probability; locate the adhesion lesions in the rat model of endometriosis based on the abnormal connected domain clusters.

[0088] Since a higher abnormality probability corresponds to a higher probability of belonging to ectopic glands, and a more consistent characteristic with the corresponding glandular lesions, it is further necessary to screen out abnormal connected component clusters to locate adhesion lesions in the rat model of endometriosis. Specifically, connected component clusters with an abnormality probability greater than a preset abnormality threshold are designated as abnormal connected component clusters. In one specific implementation of this invention, the preset abnormality threshold is set to 0.67, which can be adjusted according to the specific implementation environment. After screening out the connected component clusters, the minimum bounding rectangle of the connected component clusters is obtained to obtain the corresponding abnormality annotation borders. These abnormality annotation borders are then labeled or segmented to locate the adhesion lesions in the rat model of endometriosis.

[0089] In summary, this application first adjusts the clustering distance based on the color difference and positional inertia frame between saturation connected components to obtain connected component clusters. Since these clusters contain ectopic glands, they need to be screened to identify the ectopic glands. By utilizing the characteristic of columnar epithelial cells distributed on the inner wall of ectopic glands, the probability of each connected component cluster containing glands is obtained, determining the likelihood that each cluster belongs to an ectopic gland. Because ectopic glands exhibit a concentrated distribution pattern, the anomalous probability of connected component clusters with higher gland probabilities is obtained through clustering, further determining the likelihood that the cluster belongs to an abnormal ectopic gland. Thus, this invention improves the accuracy of lesion localization in a rat model of endometriosis by marking potential lesion locations in stained slide images using the anomalous probability of connected component clusters.

[0090] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for locating adhesion lesions in a rat model of endometriosis, characterized in that, The method includes: Acquire stained section images of a rat model of endometriosis; determine all saturation connected components in each saturation interval based on the saturation distribution of the stained section images; obtain all grayscale connected components after grayscale conversion of the stained section images; Based on the distance between saturation connected components and the saturation deviation of the corresponding saturation interval, the boundary probability of each saturation connected component is determined; based on the boundary probability, the positional distribution between each saturation connected component, and the relative saturation deviation, at least two connected component clusters are selected. Region synthesis is performed based on the distribution and adjacency of each gray-level connected domain and the approximation of columnar features in morphology to determine all suspected columnar epithelial regions; based on the number of gray-level connected domains and the columnar approximation in the suspected columnar epithelial regions, the corresponding epithelial probability is determined. Based on the boundary probability distribution in each connected domain cluster and the epithelial probability of adjacent suspected columnar epithelial regions, the corresponding ectopic gland probability is determined; based on the ectopic gland probability and the concentrated distribution of connected domain clusters, the anomaly probability of each connected domain cluster is determined. Abnormal connected domain clusters are selected based on the abnormal probability; adhesion lesions in the rat model of endometriosis are located based on the abnormal connected domain clusters. The process of obtaining the probability of ectopic glands includes: The epithelial probability of the nearest suspected columnar epithelial region to each saturation connected region is used as the adjacency reference probability of each saturation connected region. The product of the normalized boundary probability of each saturated connected region and the adjacent reference probability is used as the local reference probability. The probability of the corresponding ectopic gland is determined by summing the local reference probabilities of all saturated connected components in each connected component cluster. The process of obtaining the anomaly probability includes: Each connected component cluster is then designated as the target cluster; other connected component clusters not included in the target clusters are designated as comparison clusters. The minimum distance between all pixels in each comparison cluster and all pixels in the target cluster is taken as the inter-cluster distance of each comparison cluster. The product of the negative correlation normalized value of the inter-cluster distance and the ectopic gland probability of the corresponding contrastive cluster is used as the reference concentration of each contrastive cluster. The product of the mean of the reference concentration of all contrast clusters and the probability of ectopic glands in the target cluster is taken as the anomaly probability of the target cluster. The process of obtaining the abnormal connected component clusters includes: Clusters of connected components with an anomaly probability greater than a preset anomaly threshold are designated as abnormal connected component clusters.

2. The method for locating adhesion lesions in a rat model of endometriosis according to claim 1, characterized in that, The process of obtaining the saturation connected component includes: The stained slice image is converted into an HSV image; all local minima in the distribution histogram of the S channel corresponding to the HSV image are obtained; the distribution histogram of the S channel is divided into at least two saturation intervals with all local minima as intervals; all connected components formed by the pixels in each saturation interval are taken as saturation connected components.

3. The method for locating adhesion lesions in a rat model of endometriosis according to claim 2, characterized in that, The process of obtaining the boundary possibilities includes: Each saturation-limited connected component is sequentially designated as the target connected component; the saturation interval containing the target connected component is designated as the target interval; and other saturation intervals outside the target interval are designated as reference intervals. The average distance between the centroids of all saturated connected regions in each reference interval and the centroid of the target connected region is negatively correlated to determine the boundary contribution index of each reference interval to the target connected region. The difference between the maximum S-channel value of each reference interval and the maximum S-channel value of the target interval is taken as the saturation interval deviation for each reference interval. Based on the product of the boundary contribution index and the saturation interval deviation, the boundary influence parameter of each reference interval on the target connected domain is determined; the cumulative value of the boundary influence parameters of all reference intervals corresponding to the target connected domain is normalized to determine the boundary probability of the target connected domain.

4. The method for locating adhesion lesions in a rat model of endometriosis according to claim 3, characterized in that, The process of obtaining the connected component clusters includes: Based on the extension of the boundary minimum probability values ​​in the adjacent regions of the target connected component, the central connected component of the target connected component is determined; other saturated connected components outside the target connected component are used as comparison connected components. The saturation deviation value of each contrasting connected component is determined based on the difference between the mean S-channel value of all pixels in each contrasting connected component and the mean S-channel value of all pixels in the target connected component. The distance between the centroid of the target connected component and the centroid of each compared connected component is taken as the first comparison distance of each compared connected component; the distance between the centroid of the central connected component and the centroid of each compared connected component is taken as the second comparison distance of each compared connected component. Calculate the difference between the second contrast distance and the first contrast distance, and use the product of the difference and the boundary probability as the boundary weighting coefficient; calculate the sum of the boundary weighting coefficient and the first contrast distance; use the product of the sum and the saturation deviation value as the clustering distance between the target connected component and each contrast connected component; Calculate the clustering distance between every two saturated connected components; based on all clustering distances, perform cluster analysis on all saturated connected components using the DBSCAN clustering algorithm to obtain at least two connected component clusters.

5. The method for locating adhesion lesions in a rat model of endometriosis according to claim 4, characterized in that, The process of obtaining the central connected domain includes: The reference connected region with the smallest boundary probability among each saturated connected region and all its adjacent saturated connected regions is taken as the corresponding minimum connected region. Using the target connected region as the seed point, the acquisition of the minimum connected region as the growth condition, and the minimum connected region as the new seed point, the region is grown until the resulting grown region does not change. The minimum connected region obtained at the end of the region growth process is then used as the central connected region of the target connected region.

6. The method for locating adhesion lesions in a rat model of endometriosis according to claim 1, characterized in that, The process of obtaining the suspected columnar epithelial region includes: The ratio between the width and length of the minimum bounding rectangle of each gray-level connected region is negatively correlated to determine the columnarity of each gray-level connected region; gray-level connected regions with columnarity greater than a preset columnar threshold are designated as columnar connected regions. Merge all adjacent columnar connected regions to obtain all suspected columnar epithelial regions.

7. The method for locating adhesion lesions in a rat model of endometriosis according to claim 6, characterized in that, The process of obtaining the epithelial probability includes: The product of the mean columnarity of all columnar connected regions in the suspected columnar epithelial region and the total number of columnar connected regions is normalized to determine the corresponding epithelial probability.

Images