Intelligent identification method for endometriosis lesions based on image processing

Abstract

The present application relates to the technical field of endometriosis lesion area identification, and particularly relates to an endometriosis lesion intelligent identification method based on image processing. The method acquires a target image in the uterus and a reference image in the uterus during uterine traction; according to the area coincidence of the target image and the reference image, a matching area of each area of the target image is acquired; according to the direction change of the line segment connected by adjacent edge pixel points, a suspected adhesion band area is acquired; according to the shape and position of the suspected adhesion band area in the matching area of the area, an adhesion band area in the target image is acquired; according to the gray scale distribution and shape of the area in the final image after the target image removes the adhesion band area, an endometriosis lesion area is acquired. The present application acquires the adhesion band area, effectively avoids the interference of the adhesion band area in the process of acquiring the endometriosis lesion area, and improves the accuracy of acquiring the endometriosis lesion area.

Description

Technical Field

[0001] This invention relates to the field of endometriosis lesion area recognition technology, specifically to an intelligent recognition method for endometriosis lesions based on image processing. Background Technology

[0002] Endometriosis is a common gynecological disease with a global incidence of approximately 10%, often accompanied by symptoms such as dysmenorrhea and infertility. Accurate analysis of endometriosis requires identifying the affected lesions. Current methods utilize medical imaging techniques to acquire images of the uterus, followed by grayscale analysis of different regions within these images to pinpoint the endometriosis lesions.

[0003] However, in reality, adhesions exist within the uterus, and these adhesions can interfere with the identification of endometriosis lesions. In other words, adhesions can easily be misidentified as endometriosis lesions, leading to an inability to accurately identify endometriosis lesions and affecting the accurate analysis of endometriosis. Summary of the Invention

[0004] To address the technical problem of uterine adhesions being misidentified as endometriosis lesions, thus hindering accurate identification of endometriosis lesions, this invention aims to provide an intelligent image processing-based method for identifying endometriosis lesions. The specific technical solution adopted is as follows:

[0005] This invention provides an intelligent identification method for endometriosis lesions based on image processing, the method comprising the following steps:

[0006] Acquire images inside the uterus as target images; acquire a preset number of consecutive reference frames inside the uterus during uterine traction.

[0007] Based on the overlap between each region in the target image and each region in each reference image, obtain the matching region of each region in the target image in each reference image;

[0008] Based on the change in direction of the line segments connecting adjacent edge pixels on the edge lines of each region in the target image and its matching region, the suspected adhesion zone regions in each region of the target image and its matching region are obtained.

[0009] Based on the shape and positional changes of suspected adhesion regions in the matching regions of each region in the target image, the adhesion regions in the target image are filtered out;

[0010] The image after removing the adhesion zone from the target image is used as the final image; the endometriosis lesion area is obtained based on the grayscale distribution and shape of each region in the final image.

[0011] Furthermore, the method for obtaining the matching region is as follows:

[0012] Take any region in the target image as the target region and for any reference image, overlap the target image with the reference image, obtain the overlap area between each region in the reference image and the target region, and use it as the reference area of ​​each region in the reference image.

[0013] For any region in the reference image, the ratio of the reference area of ​​that region to the total area of ​​that region is taken as the degree of overlap of that region.

[0014] The region corresponding to the greatest overlap in the reference image is taken as the matching region of the target region in the reference image.

[0015] Furthermore, the method for obtaining the suspected adhesion zone is as follows:

[0016] Based on the change in direction of the line segments connecting adjacent edge pixels on the edge lines of each region in the target image and its matching region, obtain the target edge pixels of each region in the target image and its matching region.

[0017] Based on the positional distribution of the target edge pixels, obtain the suspected adhesion zone region in each region of the target image and its matching region.

[0018] Furthermore, the method for obtaining the target edge pixels is as follows:

[0019] For each region in the target image and any edge pixel on the edge line of any region in its matching region, obtain the line segment connecting the edge pixel to its previous adjacent edge pixel as the first line segment, and obtain the line segment connecting the edge pixel to its next adjacent edge pixel as the second line segment.

[0020] The angle between the first line segment and the direction of the traction force on the uterus is taken as the first angle, and the angle between the second line segment and the direction of the traction force on the uterus is taken as the second angle.

[0021] Obtain the difference between the first included angle and the second included angle, and use it as the reference included angle for the edge pixel;

[0022] When the reference angle is within the preset angle range, the edge pixel is taken as the target edge pixel of the region.

[0023] Furthermore, the method for obtaining the suspected adhesion zone is as follows:

[0024] For each region in the target image and any target edge pixel in any region of its matching region, the distance between the target edge pixel and every target edge pixel in the region is obtained, and all of them are used as the first distance;

[0025] The target edge pixel is taken as the first designated point, and the target edge pixel corresponding to the smallest first distance is taken as the second designated point;

[0026] The straight lines drawn through the first and second designated points, respectively, along the direction of the traction force on the uterus, are both designated as target straight lines. Other target edge pixels in this region that lie on the target straight line passing through the first designated point are designated as fourth designated points; and other target edge pixels in this region that lie on the target straight line passing through the second designated point are designated as third designated points.

[0027] The region formed by connecting the first designated point, the second designated point, the third designated point, and the second designated point in sequence is considered as the suspected adhesion zone within that region.

[0028] Furthermore, the method for obtaining the adhesion zone is as follows:

[0029] For the k-th region in the target image, the matching regions of the k-th region are arranged according to the acquisition order of the reference images to obtain the matching region sequence of the k-th region;

[0030] By using optical flow, each suspected adhesion region in the k-th region is matched with the suspected adhesion region in each matching region of the matching region sequence to obtain the matched suspected adhesion region for each suspected adhesion region in the k-th region.

[0031] For the a-th suspected adhesion zone in the k-th region, the matching suspected adhesion zones of the a-th suspected adhesion zone are arranged according to the order of the matching zones in the corresponding matching zone sequence to obtain the matching suspected adhesion zone sequence of the a-th suspected adhesion zone.

[0032] Based on the shape of the suspected adhesion region matched with the a-th suspected adhesion region, obtain the changed aspect ratio of the a-th suspected adhesion region;

[0033] Based on the positional differences between adjacent suspected adhesion regions in the sequence of suspected adhesion regions of the a-th suspected adhesion region, and the positional differences between the matching regions of the k-th region where the adjacent suspected adhesion regions are located, the degree of positional movement of the a-th suspected adhesion region is obtained.

[0034] The normalized result of the product of the changed aspect ratio and the degree of positional movement is taken as the probability of the a-th suspected adhesion zone.

[0035] When the probability is greater than the preset probability threshold, the a-th suspected adhesion region is identified as the adhesion region in the target image.

[0036] Furthermore, the method for obtaining the varying aspect ratio is as follows:

[0037] For any matching suspected adhesion region in the k-th region, the distance between the first and second specified points in the matching suspected adhesion region is taken as the first distance, the distance between the second and third specified points in the matching suspected adhesion region is taken as the second distance, the distance between the third and fourth specified points in the matching suspected adhesion region is taken as the third distance, and the distance between the first and fourth specified points in the matching suspected adhesion region is taken as the fourth distance.

[0038] The average of the first and third distances of all matching suspected adhesion regions in the a-th suspected adhesion region is obtained as the first feature value;

[0039] The average of the second and fourth distances of all matched suspected adhesion regions in the a-th suspected adhesion region is obtained as the second feature value;

[0040] The ratio of the second feature value to the first feature value is used as the aspect ratio of the a-th suspected adhesion zone.

[0041] Furthermore, the method for obtaining the degree of positional movement is as follows:

[0042] For any matching suspected adhesion region in the sequence of matching suspected adhesion regions for the a-th suspected adhesion region, obtain the distance between the centroid of the matching suspected adhesion region and the position of the centroid of the previous adjacent matching suspected adhesion region, and use it as the first length.

[0043] The distance between the centroid of the matched region containing the suspected adhesion zone and the centroid of the matched region containing the previous adjacent suspected adhesion zone is obtained and used as the second length.

[0044] The difference between the first length and the second length is used as a reference value for the positional change of the suspected adhesion area.

[0045] The average of the positional change reference values ​​of all matched suspected adhesion regions in the a-th suspected adhesion region is taken as the degree of positional movement of the a-th suspected adhesion region.

[0046] Furthermore, the method for obtaining the endometriosis lesion area is as follows:

[0047] Based on the grayscale distribution and shape of each region in the final image, the degree of anomaly in each region is obtained.

[0048] When the degree of abnormality exceeds the preset abnormality threshold, the corresponding area will be designated as the lesion area of ​​endometriosis.

[0049] Furthermore, the method for obtaining the degree of abnormality is as follows:

[0050] For any region in the final image, the difference between the average gray value of all pixels in that region and the average gray value of all pixels in the final image is obtained as the reference gray value difference for that region.

[0051] Obtain the average of the reference angles of all edge pixels in the region, and use it as the shape analysis value of the region;

[0052] The result of normalizing the ratio of the reference grayscale difference to the shape analysis value is used as the degree of anomaly in the region.

[0053] The present invention has the following beneficial effects:

[0054] This invention first obtains the matching region of each region in the target image in each reference image based on the overlap between each region in the target image and each region in each reference image, accurately determining the corresponding region of each region in the target image in each reference image, which helps improve the efficiency and accuracy of subsequent acquisition of adhesion regions. Then, based on the directional changes of the line segments connecting adjacent edge pixels on the edge lines of each region in the target image and its matching region, it obtains the suspected adhesion regions in each region of the target image and its matching region, initially determining the adhesion regions that meet the requirements. To more accurately acquire adhesion regions within the uterus and effectively avoid interference with subsequent acquisition of endometriosis lesion areas, this invention further enhances accuracy. Furthermore, based on the shape and positional changes of suspected adhesion areas in the matching regions of each region in the target image, adhesion areas in the target image are screened out to accurately determine the adhesion areas within the uterus. To obtain the endometriosis lesion area and enable accurate analysis of endometriosis, the image after removing the adhesion areas from the target image is used as the final image, avoiding the influence of adhesion areas on the identification of endometriosis lesion areas. Finally, based on the grayscale distribution and shape of each region in the final image, the endometriosis lesion area is accurately obtained, effectively improving the accuracy of endometriosis lesion area identification and facilitating accurate analysis of endometriosis. Attached Figure Description

[0055] 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.

[0056] Figure 1 A schematic flowchart illustrating an intelligent identification method for endometriosis lesions based on image processing, provided in one embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the distribution of adhesive bands in a region provided in an embodiment of the present invention;

[0058] Figure 3 This is a flowchart of a method for obtaining a suspected adhesion zone according to an embodiment of the present invention;

[0059] Figure 4 This is a flowchart illustrating a method for obtaining an adhesion zone according to an embodiment of the present invention.

[0060] Figure 5 This is a flowchart illustrating a method for obtaining lesion areas of endometriosis according to an embodiment of the present invention.

[0061] Figure 6 This is a structural diagram of an image processing-based intelligent identification system for endometriosis lesions provided in one embodiment of the present invention;

[0062] Figure 7 This is a schematic diagram of a computer device provided according to an embodiment of the present invention. Detailed Implementation

[0063] 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 the image processing-based intelligent identification method for endometriosis lesions proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0064] 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.

[0065] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent identification method for endometriosis lesions based on image processing provided by this invention.

[0066] Example 1:

[0067] This invention proposes an intelligent identification method for endometriosis lesions based on image processing. Please refer to [link / reference]. Figure 1 The diagram illustrates a schematic flowchart of an image processing-based intelligent identification method for endometriosis lesions provided by an embodiment of the present invention. The method includes the following steps:

[0068] Step S1: Acquire an image inside the uterus as the target image; acquire a preset number of consecutive reference images inside the uterus during the uterine traction process.

[0069] Specifically, to obtain images of endometriosis lesions, hysteroscopy is used to acquire images of the uterus as target images. Existing methods analyze pixel values ​​in various regions of the target image to identify endometriosis lesions, thus analyzing the endometriosis. However, in practice, the presence of adhesions in the endometrium can affect the accurate identification of endometriosis lesions; that is, adhesion areas are easily misidentified as endometriosis lesions. Therefore, to accurately identify endometriosis lesions in the target image and avoid interference from adhesion areas, this embodiment first needs to identify the adhesion areas in the target image. Considering that adhesions change significantly when the uterus is stretched, while tissue changes very little, this embodiment combines the different behaviors of adhesions and tissue under the influence of uterine stretching to identify adhesion areas. Therefore, this embodiment acquires a predetermined number of consecutive reference frames of images within the uterus during uterine stretching. In this embodiment, the preset quantity is set to 10. The implementer can set the size of the preset quantity according to the actual situation, and there is no limitation here.

[0070] It should be noted that the target image is an image of the uterus acquired without any external force, while the reference image is an image of the uterus acquired under traction. In this embodiment, the direction of the traction force is set horizontally to the right; however, the practitioner can set the direction of the traction force according to the actual situation, and it is not limited here. The traction force is within a reasonable range and is temporary, without causing any adverse effects on the uterus, and the adhesions will not rupture during the traction process. Furthermore, the hysteroscope's shooting position and angle remain consistent when acquiring the target image and each reference image. In this embodiment, the hysteroscope resolution is set to 1080p and the image acquisition frequency during the traction process is set to 30fps.

[0071] Step S2: Based on the overlap between each region in the target image and each region in each reference image, obtain the matching region of each region in the target image in each reference image.

[0072] Specifically, considering that adhesions typically exist between different tissues, to improve the efficiency of obtaining adhesion regions in the target image, this embodiment first uses the Canny edge detection algorithm to obtain each region in the target image and each reference image. Then, based on the overlap between each region in the target image and each region in each reference image, the corresponding region in each reference image for each region in the target image is obtained, i.e., the matching region. Subsequent analysis focuses only on each region in the target image and its matching region to efficiently determine the adhesion regions within each region of the target image. The Canny edge detection algorithm is a well-known technique and will not be elaborated upon further. It should be noted that when dividing the target image and reference images into regions, endometriosis lesions are also considered as a type of tissue.

[0073] Preferably, in one feasible embodiment, the method for obtaining the matching region is as follows: First, any region in the target image is taken as the target region. For any reference image, since the position of the hysteroscope is fixed, the position and size of the target image and the reference image are consistent. Therefore, in this embodiment, the target image is directly superimposed on the reference image to obtain the overlap area between each region in the reference image and the target region, which is taken as the reference area of ​​each region in the reference image. In order to accurately reflect the overlap between each region in the reference image and the target region, the ratio of the reference area of ​​any region in the reference image to the overall area of ​​the region is taken as the degree of overlap of the region. The greater the degree of overlap, the more likely that the region and the target region are the same uterine region. Therefore, in this embodiment, the region corresponding to the maximum degree of overlap in the reference image is taken as the matching region of the target region in the reference image.

[0074] At this point, the matching region in each reference image for each region in the target image is obtained.

[0075] Step S3: Based on the change in direction of the line segments connecting adjacent edge pixels on the edge lines of each region in the target image and its matching region, obtain the suspected adhesion zone region in each region of the target image and its matching region.

[0076] Adhesions are known to typically appear as cords between tissues, such as... Figure 2 The diagram shows the distribution of adhesion bands in a region, where... Figure 2 There are 3 adhesion zones corresponding to Figure 2 The area where adhesions occur. Through... Figure 2It can be seen that each adhesion zone region is composed of 4 edge pixels connected in sequence, and the 4 edge pixels are the turning points where the edge lines change significantly. Therefore, in this embodiment, the suspected adhesion zone region in each region of the target image and its matching region is obtained based on the change in the direction of the line segments connecting the adjacent edge pixels on the edge lines of each region in the target image and its matching region.

[0077] Preferably, in one feasible embodiment, the method for obtaining the suspected adhesion zone is described in [reference needed]. Figure 3 The document presents a flowchart of a method for obtaining a suspected adhesion zone, as provided in this embodiment. The method includes the following steps:

[0078] Step S301: Based on the change in direction of the line segments connecting adjacent edge pixels on the edge lines of each region in the target image and its matching region, obtain the target edge pixels of each region in the target image and its matching region.

[0079] It is known that an adhesion zone region consists of four inflection edge pixels whose edge line shapes change significantly. Therefore, this embodiment first obtains the target edge pixels of each region in the target image and its matching region based on the directional changes of the line segments connecting adjacent edge pixels on the edge lines of each region in the target image and its matching region. The edge pixels that construct the adhesion zone region must belong to the target edge pixels.

[0080] In one possible implementation of this embodiment, the method for obtaining the target edge pixel is as follows: For any edge pixel on the edge line of any region in the target image and any region in its matching region, the line segment connecting the edge pixel to its previous adjacent edge pixel is obtained as a first line segment, and the line segment connecting the edge pixel to its next adjacent edge pixel is obtained as a second line segment; the angle between the first line segment and the direction of the traction force on the uterus is taken as a first angle, and the angle between the second line segment and the direction of the traction force on the uterus is taken as a second angle; it should be noted that in this embodiment, the direction of the traction force on the uterus is horizontal to the right. To determine whether the edge pixel is a turning point where the shape of the region changes significantly, the absolute value of the difference between the first angle and the second angle is obtained as a reference angle for the edge pixel; the larger the reference angle, the more likely the edge pixel is to be the target edge pixel. Based on experience, this embodiment sets a preset angle range as follows: The implementer can set a preset angle range according to the actual situation, which is not limited here. When the reference angle is within the preset angle range, the edge pixel is taken as the target edge pixel of the region.

[0081] At this point, the target edge pixels of each region in the target image and its matching region are obtained.

[0082] Step S302: Based on the positional distribution of the target edge pixels, obtain the suspected adhesion zone region in each region of the target image and its matching region.

[0083] Depend on Figure 2 It can be seen that the adhesion zone is similar to a rectangle, and the direction of the length of the adhesion zone must be consistent with the direction of the tensile force. Therefore, in this embodiment, based on the position distribution of the target edge pixels, the suspected adhesion zone in each region of the target image and its matching region is obtained.

[0084] In one possible implementation of this embodiment, the method for obtaining the suspected adhesion zone is as follows: For each region in the target image and any target edge pixel in any region of its matching region, the Euclidean distance between the target edge pixel and each target edge pixel in the region is obtained, and these distances are all taken as the first distance. The method for obtaining the Euclidean distance is a well-known technique and will not be elaborated further. The target edge pixel is taken as the first designated point, and the target edge pixel corresponding to the smallest first distance that is not the target edge pixel is taken as the second designated point. Straight lines are drawn through the first and second designated points, respectively, indicating the direction of the traction force on the uterus, and these lines are taken as target straight lines. Other target edge pixels in the region located on the target straight line passing through the first designated point are taken as fourth designated points. Other target edge pixels in the region located on the target straight line passing through the second designated point are taken as third designated points. It should be noted that if there are at least three target edge pixels on the target straight line, i.e., there may be multiple fourth and multiple third designated points, the Euclidean distance between each fourth designated point and each third designated point is obtained, and the two designated points corresponding to the smallest Euclidean distance are taken as the final third and fourth designated points, respectively. The region formed by connecting the first designated point, the second designated point, the third designated point, and the second designated point in sequence is considered as the suspected adhesion zone within that region.

[0085] At this point, each region in the target image and each suspected adhesion zone in its matching region have been obtained. It should be noted that once a suspected adhesion zone has been identified for a region, subsequent acquisitions of suspected adhesion zones for that region will no longer consider the target edge pixels corresponding to the already identified suspected adhesion zone.

[0086] Step S4: Based on the shape and positional changes of suspected adhesion regions in the matching regions of each region in the target image, filter out the adhesion regions in the target image.

[0087] It is known that when the uterus is stretched, both tissues and adhesions will move to some extent in the direction of the stretching force. Due to the poor elasticity of the tissues and the good extensibility of the adhesions, the position of the adhesion area in the reference image changes significantly compared to the corresponding tissue area. At the same time, the aspect ratio of the adhesion area increases with the acquisition order of the reference images. Therefore, in this embodiment, the adhesion area in the target image is accurately screened out based on the shape and position changes of the suspected adhesion area in the matching area of ​​each region in the target image.

[0088] Preferably, in one possible implementation of this embodiment, the method for obtaining the adhesion zone is described in [reference needed]. Figure 4 The document presents a flowchart of a method for obtaining an adhesion zone provided in this embodiment, which includes the following steps:

[0089] Step S401: Obtain the matching suspected adhesion region for each suspected adhesion region in each region of the target image.

[0090] It is known that adhesions will not break during traction on the uterus. In order to accurately analyze whether each suspected adhesion area in the target image is a real adhesion area, this embodiment first needs to determine the matching suspected adhesion area in each reference image for each suspected adhesion area in the target image.

[0091] To illustrate this clearly, this embodiment uses the k-th region in the target image as an example. The matching regions of the k-th region are arranged according to the acquisition order of the reference image, resulting in a matching region sequence for the k-th region. Then, using optical flow, each suspected adhesion region in the k-th region is matched with a suspected adhesion region in each matching region of the matching region sequence, resulting in a matching suspected adhesion region for each suspected adhesion region in the k-th region. The optical flow method is a well-known technique and will not be described in detail here.

[0092] At this point, the matching suspected adhesion region in each reference image is obtained for each suspected adhesion region in each region of the target image.

[0093] Step S402: Obtain the aspect ratio of each suspected adhesion zone in each region of the target image.

[0094] The larger the aspect ratio of the matching suspected adhesion region within a region of the target image, the more significant the change in the suspected adhesion region during traction, and the more likely the suspected adhesion region in the target image is a real adhesion region. Therefore, this embodiment obtains the changing aspect ratio of each suspected adhesion region in each region of the target image based on the shape of the matching suspected adhesion region. The larger the changing aspect ratio, the more likely the corresponding suspected adhesion region in the target image is a real adhesion region.

[0095] To clarify the explanation, this embodiment uses the a-th suspected adhesion region in the k-th region as an example for analysis. For any matching suspected adhesion region in the a-th suspected adhesion region of the k-th region, the distance between the first and second specified points in the matching suspected adhesion region is taken as the first distance, the distance between the second and third specified points in the matching suspected adhesion region is taken as the second distance, and the distance between the third and fourth specified points in the matching suspected adhesion region is taken as the third distance. The distance between points is taken as the fourth distance; at this point, the first and third distances can be assumed to be the width of the suspected adhesion region, and the second and fourth distances can be assumed to be the length of the suspected adhesion region; in order to obtain the aspect ratio of the suspected adhesion region, the average of the first and third distances of all the suspected adhesion regions of the a-th suspected adhesion region is further obtained as the first feature value; the average of the second and fourth distances of all the suspected adhesion regions of the a-th suspected adhesion region is obtained as the second feature value; the ratio of the second feature value to the first feature value is taken as the changing aspect ratio of the a-th suspected adhesion region.

[0096] A method for obtaining the aspect ratio of the a-th suspected adhesion region in the target image, and for obtaining the aspect ratio of each suspected adhesion region in the target image.

[0097] Step S403: Obtain the degree of positional movement of each suspected adhesion zone in each region of the target image.

[0098] During the traction process on the uterus, the changes in adhesions are significantly different from those in other tissues. Changes in the adhesion areas are more pronounced. To accurately analyze the changes in each suspected adhesion area in the target image when the uterus is stretched, the matching suspected adhesion areas for each suspected adhesion area in the target image are arranged according to their order in the corresponding matching region sequence, thus obtaining a matching suspected adhesion area sequence for each suspected adhesion area in the target image. For example, for the a-th suspected adhesion area in step S402, the matching suspected adhesion areas for the a-th suspected adhesion area are arranged according to their order in the corresponding matching region sequence, thus obtaining a matching suspected adhesion area sequence for the a-th suspected adhesion area. This helps to more accurately reflect the changes in the a-th suspected adhesion area during the traction process on the uterus. Furthermore, in this embodiment, based on the positional differences between adjacent matched suspected adhesion regions in the sequence of matched suspected adhesion regions for each suspected adhesion region in the target image, and the positional differences between the matched regions in which adjacent matched suspected adhesion regions are located, the degree of positional movement of each suspected adhesion region in the target image is obtained. The greater the degree of positional movement, the more likely the corresponding suspected adhesion region in the target image is a real adhesion region.

[0099] To clearly illustrate the process of filtering real adhesion regions in the target image, this embodiment takes the a-th suspected adhesion region in step S402 as an example for analysis. For any matching suspected adhesion region in the matching suspected adhesion region sequence of the a-th suspected adhesion region, the Euclidean distance between the centroid of the matching suspected adhesion region and the centroid of the previous adjacent matching suspected adhesion region is obtained as the first length; then, the Euclidean distance between the centroid of the matching region where the matching suspected adhesion region is located and the centroid of the matching region where the previous adjacent matching suspected adhesion region is located is obtained as the second length; when the difference between the first length and the second length is greater... The larger the value of the first length and the second length, the greater the impact on the suspected adhesion area during uterine traction, and the more likely the a-th suspected adhesion area is to be a true adhesion area. The absolute value of the difference between the first and second lengths is then used as a reference value for the positional change of the suspected adhesion area. The larger the reference value, the more likely the a-th suspected adhesion area is to be a true adhesion area. To more accurately analyze the changes of the a-th suspected adhesion area during traction, the average of the reference values ​​for the positional changes of all suspected adhesion areas is used as the degree of positional movement of the a-th suspected adhesion area. It should be noted that the first suspected adhesion area in the sequence of suspected adhesion areas for the a-th suspected adhesion area does not have a preceding adjacent suspected adhesion area. In this embodiment, the a-th suspected adhesion area is considered as the preceding adjacent suspected adhesion area of ​​the first suspected adhesion area in the sequence of suspected adhesion areas for the a-th suspected adhesion area. The method for obtaining the centroid is a well-known technique and will not be described further.

[0100] At this point, the degree of positional movement of each suspected adhesion zone in the target image is obtained.

[0101] Step S404: Obtain the probability of each suspected adhesion zone in each region of the target image.

[0102] It is known that a larger change in aspect ratio and a greater degree of positional movement indicate that the corresponding suspected adhesion region in the target image is more likely to be a real adhesion region. Therefore, this embodiment normalizes the product of the change in aspect ratio and the degree of positional movement for each suspected adhesion region in the target image, and uses this normalized result as the probability of each suspected adhesion region in the target image. Specifically, this embodiment uses a norm normalization function to normalize the product of the change in aspect ratio and the degree of positional movement.

[0103] At this point, the probability of each suspected adhesion zone in each region of the target image is obtained.

[0104] Step S405: Obtain the adhesion zone region in the target image.

[0105] The greater the probability, the more likely the corresponding suspected adhesion region in the target image is to be the actual adhesion region. Therefore, this embodiment sets a preset probability threshold of 0.6. Implementers can set the size of the preset probability threshold according to the actual situation, which is not limited here. When the probability is greater than the preset probability threshold, the corresponding suspected adhesion region is identified as the adhesion region in the target image.

[0106] At this point, the adhesion zone region in the target image has been obtained.

[0107] Step S5: The image after removing the adhesion area in the target image is taken as the final image; based on the grayscale distribution and shape of each region in the final image, the endometriosis lesion area is obtained.

[0108] Specifically, to avoid interference from adhesion areas in the identification of endometriosis lesion areas, this embodiment uses the image after removing adhesion areas from the target image as the final image. Then, each region in the final image is analyzed. Because there are obvious differences in color and shape between endometriosis lesions and normal uterine tissue, and the shape of endometriosis lesions is more regular, that is, more circular, this embodiment obtains the endometriosis lesion area based on the grayscale distribution and shape of each region in the final image.

[0109] Preferably, in one feasible embodiment, the method for obtaining the endometriosis lesion area is described in [reference needed]. Figure 5 The document presents a flowchart of a method for obtaining lesions in endometriosis, as provided in this embodiment. The method includes the following steps:

[0110] Step S501: Based on the grayscale distribution and shape of each region in the final image, obtain the degree of anomaly of each region in the final image.

[0111] It is known that endometriosis lesions and normal uterine tissue differ significantly in color and shape. Therefore, this embodiment first uses the Canny edge detection algorithm to obtain each region in the final image. Then, based on the grayscale distribution and shape of each region in the final image, the degree of abnormality of each region is obtained. The greater the degree of abnormality, the more likely the corresponding region in the final image is an endometriosis lesion.

[0112] Preferably, in one feasible embodiment of this method, the method for obtaining the degree of anomaly is as follows: For any region in the final image, the absolute value of the difference between the average gray value of all pixels in that region and the average gray value of all pixels in the final image is obtained as the reference gray value difference for that region; the larger the reference gray value difference, the more likely the region is to be an endometriosis lesion region; in order to more accurately analyze whether the region is an endometriosis lesion region, it is necessary to further analyze whether the shape of the region is regular. Therefore, the average value of the reference included angle of all edge pixels in the region is obtained as the shape analysis value of the region; the smaller the shape analysis value, the closer the shape of the region is to a regular circle, indirectly indicating that the region is more likely to be an endometriosis lesion region; it should be noted that since the region is a closed region, the shape analysis value of the region must be greater than 0. Furthermore, in this embodiment, the result of normalizing the ratio of the reference gray value difference to the shape analysis value is used as the degree of anomaly of the region. In this embodiment, the ratio of reference grayscale difference to shape analysis value is normalized using the norm normalization function.

[0113] At this point, the degree of anomaly in each region of the final image is obtained.

[0114] Step S502: When the degree of abnormality is greater than the preset abnormality threshold, the corresponding area is designated as the lesion area of ​​endometriosis.

[0115] The greater the degree of abnormality, the more likely the corresponding area is to be an endometriosis lesion. Therefore, this embodiment sets a preset abnormality threshold of 0.7. Implementers can set the size of the preset abnormality threshold according to the actual situation, which is not limited here. When the abnormality is greater than the preset abnormality threshold, the corresponding area is regarded as an endometriosis lesion area.

[0116] This accurately identifies the lesion area of ​​endometriosis, effectively avoiding the influence of adhesion areas and facilitating accurate analysis of endometriosis.

[0117] In summary, this embodiment acquires a target image of the uterus and a reference image of the uterus during uterine traction; based on the overlap between the target image and the reference image, it acquires a matching region for each region of the target image; based on the directional changes of the line segments connecting adjacent edge pixels, it acquires suspected adhesion regions; based on the shape and position of the suspected adhesion regions in the matching regions, it acquires the adhesion regions in the target image; and based on the grayscale distribution and shape of the regions in the final image after removing the adhesion regions from the target image, it acquires the endometriosis lesion region. This invention, by acquiring adhesion regions, effectively avoids interference from adhesion regions during the acquisition of endometriosis lesion regions, thus improving the accuracy of acquiring endometriosis lesion regions.

[0118] Example 2:

[0119] This invention also proposes an intelligent identification system for endometriosis lesions based on image processing. Please refer to [link to relevant documentation]. Figure 6 The diagram illustrates a structural diagram of an image processing-based intelligent identification system for endometriosis lesions provided by an embodiment of the present invention. The system includes: an image acquisition module 10, a matching region acquisition module 20, a suspected adhesion region acquisition module 30, an adhesion region acquisition module 40, and an endometriosis lesion region acquisition module 50.

[0120] The image acquisition module 10 is used to acquire images inside the uterus as target images; and to acquire a preset number of consecutive reference images inside the uterus during uterine traction.

[0121] The matching region acquisition module 20 is used to acquire the matching region of each region in the target image in each reference image based on the overlap between each region in the target image and each region in each reference image.

[0122] The suspected adhesion zone acquisition module 30 is used to acquire the suspected adhesion zone in each region of the target image and its matching region based on the change in the direction of the line segment connecting the adjacent edge pixels on the edge line of each region in the target image and its matching region.

[0123] The adhesion zone acquisition module 40 is used to filter out adhesion zones in the target image based on the shape and positional changes of suspected adhesion zones in the matching region of each region in the target image.

[0124] The endometriosis lesion area acquisition module 50 is used to obtain the image after removing the adhesion area in the target image as the final image; and to obtain the endometriosis lesion area based on the gray-scale distribution and shape of each area in the final image.

[0125] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the intelligent identification system for endometriosis lesions based on image processing and the intelligent identification method for endometriosis lesions based on image processing provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0126] Example 3:

[0127] This invention also proposes an intelligent identification device for endometriosis lesions based on image processing. The device includes a memory and a processor. The memory stores executable program code, and the processor calls and executes the executable program code to perform the intelligent identification method for endometriosis lesions based on image processing provided in the embodiments of this application. Specifically, the device may be a chip, component, or module. The chip may include a connected processor and memory; the memory stores instructions, and when the processor calls and executes the instructions, the chip can perform the intelligent identification method for endometriosis lesions based on image processing provided in the above embodiments.

[0128] Furthermore, this application also protects a computer device; please refer to [link to relevant documentation]. Figure 7 The computer device includes a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402. When the processor 402 executes the computer program 403, the computer device can execute any of the aforementioned image processing-based intelligent identification methods for endometriosis lesions.

[0129] Example 4:

[0130] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to achieve the intelligent identification method for endometriosis lesions based on image processing provided in the above embodiment.

[0131] Example 5:

[0132] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to realize the intelligent identification method for endometriosis lesions based on image processing provided in the above embodiment.

[0133] In this embodiment, the device, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0134] 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.

[0135] 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. An intelligent identification method for endometriosis lesions based on image processing, characterized in that, The method comprises the following steps: acquiring an image in the uterus as a target image; acquiring a preset number of continuous frames of reference images in the uterus during uterine traction; according to the coincidence of each region in the target image and each region in each reference image, acquiring the matching region of each region in the target image in each reference image; according to the direction change of the line segment connecting adjacent edge pixels on the edge line of each region in the target image and its matching region, acquiring the suspected adhesion zone in the matching region of each region in the target image; according to the shape and position change of the suspected adhesion zone in the matching region of each region in the target image, screening the adhesion zone in the target image; taking the image after removing the adhesion zone in the target image as a final image; and acquiring the endometriosis lesion region according to the gray scale distribution in each region in the final image and the shape of each region. 2.The image processing based endometriosis lesion intelligent identification method of claim 1, wherein, The acquisition method of the matching region is as follows: taking any region in the target image as a target region, and for any reference image, superimposing the target image and the reference image to acquire the overlapping area of each region in the reference image and the target region as the reference area of each region in the reference image; for any region in the reference image, taking the ratio of the reference area of the region to the overall area of the region as the overlap degree of the region; taking the region corresponding to the maximum overlap degree in the reference image as the matching region of the target region in the reference image. 3.The image processing based endometriosis lesion intelligent identification method of claim 1, wherein, The acquisition method of the suspected adhesion zone is as follows: according to the direction change of the line segment connecting adjacent edge pixels on the edge line of each region in the target image and its matching region, acquiring the target edge pixel of each region in the target image and its matching region; according to the position distribution of the target edge pixel, acquiring the suspected adhesion zone in the matching region of each region in the target image and its matching region.

4. The intelligent endometriosis lesion identification method based on image processing according to claim 3, characterized in that, The acquisition method of the target edge pixel is as follows: for any edge pixel on the edge line of any region in each region in the target image and its matching region, acquiring the line segment connecting the edge pixel and its previous adjacent edge pixel as a first line segment, and acquiring the line segment connecting the edge pixel and its next adjacent edge pixel as a second line segment; taking the included angle between the first line segment and the direction of the traction force on the uterus as a first included angle, and taking the included angle between the second line segment and the direction of the traction force on the uterus as a second included angle; acquiring the difference between the first included angle and the second included angle as a reference included angle of the edge pixel; when the reference included angle is within a preset included angle range, taking the edge pixel as the target edge pixel of the region.

5. The intelligent endometriotic lesion identification method based on image processing according to claim 3, characterized in that, The acquisition method of the suspected adhesion zone is as follows: for any target edge pixel of any region in each region in the target image and its matching region, acquiring the distance between the target edge pixel and each target edge pixel in the region as a first distance; taking the target edge pixel as a first specified point, and taking the target edge pixel corresponding to the smallest first distance as a second specified point; Respectively, the straight line passing through the first designated point and the second designated point as the direction of the pulling force of the uterus is taken as the target straight line, and other target edge pixel points in the region located on the target straight line passing through the first designated point are taken as the fourth designated point; other target edge pixel points in the region located on the target straight line passing through the second designated point are taken as the third designated point; The region formed by sequentially connecting the first designated point, the second designated point, the third designated point and the second designated point is taken as the suspected adhesion zone in the region.

6. The intelligent endometriosis lesion identification method based on image processing according to claim 5, characterized in that, The adhesion zone acquisition method is: For the kth region in the target image, the matching regions of the kth region are arranged according to the acquisition order of the reference images, and a matching region sequence of the kth region is obtained; Each suspected adhesion zone in the kth region is matched with each suspected adhesion zone in each matching region in the matching region sequence by using the optical flow method, and a matching suspected adhesion zone of each suspected adhesion zone in the kth region is obtained; For the ath suspected adhesion zone in the kth region, the matching suspected adhesion zones of the ath suspected adhesion zone are arranged according to the arrangement order of the matching regions in the corresponding matching region sequence, and a matching suspected adhesion zone sequence of the ath suspected adhesion zone is obtained; According to the shape of the matching suspected adhesion zone of the ath suspected adhesion zone, the change aspect ratio of the ath suspected adhesion zone is obtained; According to the position difference between adjacent matching suspected adhesion zones in the matching suspected adhesion zone sequence of the ath suspected adhesion zone and the position difference between the matching regions of the kth region where the adjacent matching suspected adhesion zones are located, the position movement degree of the ath suspected adhesion zone is obtained; The product of the change aspect ratio and the position movement degree is normalized, and the result is taken as the possibility of the ath suspected adhesion zone; When the possibility is greater than a preset possibility threshold, the ath suspected adhesion zone is taken as the adhesion zone in the target image.

7. The image processing-based intelligent endometriosis lesion identification method of claim 6, wherein, The change aspect ratio acquisition method is: For any matching suspected adhesion zone of the ath suspected adhesion zone in the kth region, the distance between the first designated point and the second designated point in the matching suspected adhesion zone is taken as the first distance, the distance between the second designated point and the third designated point in the matching suspected adhesion zone is taken as the second distance, the distance between the third designated point and the fourth designated point in the matching suspected adhesion zone is taken as the third distance, and the distance between the first designated point and the fourth designated point in the matching suspected adhesion zone is taken as the fourth distance; The average of the first distance and the third distance of all matching suspected adhesion zones of the ath suspected adhesion zone is taken as the first characteristic value; The average of the second distance and the fourth distance of all matching suspected adhesion zones of the ath suspected adhesion zone is taken as the second characteristic value; The ratio of the second characteristic value to the first characteristic value is taken as the change aspect ratio of the ath suspected adhesion zone.

8. The intelligent identification method for endometriosis lesions based on image processing as described in claim 6, characterized in that, The position movement degree acquisition method is: For any one matched suspected adhesion zone in the sequence of the a-th suspected adhesion zone, a distance between the matched suspected adhesion zone and a position where a centroid of a previous adjacent matched suspected adhesion zone is located is obtained as a first length; a distance between a position where the matched suspected adhesion zone is located and a position where a centroid of a matching region of the previous adjacent matched suspected adhesion zone is located is obtained as a second length; a difference between the first length and the second length is taken as a position change reference value of the matched suspected adhesion zone; a mean value of the position change reference values of all the matched suspected adhesion zones of the a-th suspected adhesion zone is taken as a position moving degree of the a-th suspected adhesion zone. 9.The image processing based endometriosis lesion intelligent identification method of claim 4, wherein, The method for obtaining the endometriosis lesion region is: According to a gray distribution in each region in the final image and a shape of each region, an abnormality degree of each region in the final image is obtained; When the abnormality degree is greater than a preset abnormality degree threshold, the corresponding region is taken as the endometriosis lesion region.

10. The intelligent identification method for endometriosis lesions based on image processing as described in claim 9, characterized in that, The method for obtaining the abnormality degree is: For any one region in the final image, a difference between a mean value of gray values of all pixel points in the region and a mean value of gray values of all pixel points in the final image is taken as a reference gray difference of the region; A mean value of reference angles of all edge pixel points of the region is taken as a shape analysis value of the region; A result of normalizing a ratio of the reference gray difference and the shape analysis value is taken as the abnormality degree of the region.

Images