Artificial intelligence-based ulcerative colitis image recognition method

Abstract

The invention discloses an ulcerative colitis image recognition method based on artificial intelligence, and the method comprises the steps: obtaining standard endoscope historical images of different intestinal examination positions, and enabling a plurality of standard endoscope historical images corresponding to each intestinal examination position to form a standard enteroscope group, calculating a curved surface smoothness standard value, an intestinal tract standard inner diameter and a blood vessel distribution standard state of an intestinal tract examination position corresponding to each standard enteroscope group; obtaining to-be-compared intestinal endoscopic examination images at different intestinal examination positions, sequentially grouping according to the intestinal examination positions to form a plurality of to-be-compared enteroscope groups, and calculating a to-be-detected curved surface smooth value, a to-be-detected intestinal inner diameter and a blood vessel distribution measurement state of the intestinal examination position corresponding to each to-be-compared enteroscope group; based on a comparison result of the smooth value of the to-be-measured curved surface, the inner diameter of the to-be-measured intestinal tract and the blood vessel distribution measurement state, performing constraint evaluation on the disease grade of the ulcerative colitis. According to the invention, the disease degree of ulcerative colitis can be subjected to constrained evaluation based on three dimensions.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to an ulcerative colitis image recognition method based on artificial intelligence. Background Art

[0002] Inflammatory bowel disease is a chronic, recurrent, inflammatory bowel disease state, mainly including ulcerative colitis and Crohn's disease. Among them, ulcerative colitis is a chronic nonspecific colon inflammation. The lesions mainly affect the colon mucosa, ranging from the distal colon to the proximal end, and even affecting the entire colon. The main clinical manifestations are diarrhea, abdominal pain, and mucus, pus and blood in the stool. It plays a vital role in the prediction of IBD complications and disease progression, and in the prevention and treatment of IBD complications.

[0003] Currently, the most common method for evaluating Crohn's disease activity index is the CD Best CDAI calculation method, in which the evaluation indicators include: number of loose stools, degree of abdominal pain, extraintestinal manifestations and complications, opioid antidiarrheal drugs, abdominal mass, decreased hematocrit value, etc. Crohn's disease is divided into active, mild, moderate and severe stages based on the Crohn's disease activity index score.

[0004] When evaluating the severity of ulcerative colitis, endoscopic images can be used to identify evaluation indicators of the intestinal lining, such as vascular conditions, erosion conditions, and bleeding conditions. For example, when the endoscope finds erythema, reduced vascular texture, and mild brittleness on the intestinal lining, it means that the intestine has mild lesions. When the endoscope finds obvious erythema, lack of vascular texture, brittleness, and erosion on the intestinal lining, it means that the intestine has moderate lesions. When the endoscope finds spontaneous bleeding and ulcer formation on the intestinal lining, it means that the intestine has severe lesions.

[0005] However, existing endoscopic images of ulcerative colitis are mostly manually identified by doctors to determine the grade of intestinal lesions in ulcerative colitis, and there is a lack of automated image recognition based on artificial intelligence. Summary of the Invention

[0006] The purpose of the present invention is to provide an artificial intelligence-based ulcerative colitis image recognition method to solve the technical problem in the prior art of using only a single lesion feature to evaluate symptomatic intestinal diseases, resulting in poor recognition accuracy.

[0007] In order to solve the above technical problems, the present invention specifically provides the following technical solutions: An artificial intelligence-based ulcerative colitis image recognition method, comprising: Step 100: Acquire standard endoscopic historical images at different intestinal examination positions, form a standard colonoscopy group with multiple standard endoscopic historical images corresponding to each intestinal examination position, split the standard endoscopic historical images in each standard colonoscopy group into multiple image fragments in order from the inside to the outside, perform independent image processing on each image fragment, and calculate the surface smoothing standard value and the standard inner diameter of the intestine for each intestinal examination position corresponding to the standard colonoscopy group; Step 200: identifying specific pixel points of each image fragment, and determining the standard state of blood vessel distribution at the intestinal examination position corresponding to each standard colonoscopy group; Step 300: Obtain intestinal endoscopic examination images to be compared at different intestinal examination locations, group them sequentially according to the intestinal examination locations to form a plurality of intestinal endoscopy groups to be compared, and calculate the surface smoothness value to be measured, the inner diameter of the intestinal tract to be measured, and the vascular distribution measurement status for each intestinal endoscopy group to be compared corresponding to the intestinal examination location; Step 400: Determine the disease grade of ulcerative colitis in a constrained evaluation manner based on the comparison results of the measured surface smoothness value, the measured intestinal inner diameter and the measured vascularity with the surface smoothness standard value, the intestinal standard inner diameter and the vascularity standard.

[0008] As a preferred embodiment of the present invention, in step 100, there is at least one standard endoscopic historical image in each standard colonoscopy group, and each standard endoscopic historical image in each standard colonoscopy group is split into multiple image fragments in order from the inside to the outside as follows: performing image noise reduction on each standard endoscopy historical image in each of the standard colonoscopy groups; Constructing the original image coordinate system of each standard endoscopic historical image, performing binarization image processing on each standard endoscopic historical image, and retaining the structural morphology of each standard endoscopic historical image; Identifying wrinkle rings contained in each standard endoscopic historical image after binary image processing, and marking the position of each wrinkle ring in the original image coordinate system; Each standard endoscope historical image is cut according to the position of the marked wrinkle ring to form a plurality of image fragments.

[0009] As a preferred embodiment of the present invention, In step 100, the method for determining the surface smoothness standard value of the intestinal surface of each standard colonoscopy group is as follows: Filling the intestinal surface between the two wrinkle rings in a mirror compensation manner so that a complete intestinal surface is formed between the two wrinkle rings, and using surface Gaussian curvature analysis to evaluate the smoothness of the intestinal inner wall of the intestinal surface between the two wrinkle rings; Determining a parameter distribution state of the smoothness of the intestinal inner wall of the intestinal curved surface formed between any two of the wrinkled rings in each standard endoscopic historical image; Based on the parameter distribution state of the intestinal wall smoothness of all standard endoscopic historical images in each of the standard colonoscopy groups, the extreme value of the standard intestinal smoothness of each standard endoscopic historical image and the average value of the standard intestinal smoothness of each standard endoscopic historical image are used as the surface smoothness standard values.

[0010] As a preferred solution of the present invention, in step 100, all standard endoscopy historical images in each standard enteroscopy group are divided into multiple intestinal surfaces according to their corresponding fold rings, and the average distance between each intestinal surface is calculated; The average value of the average distances of all intestinal surfaces of each standard endoscopic historical image is taken as the inner diameter of each intestinal tract of the standard endoscopic historical image; The inner diameters of the intestines of all standard endoscopic historical images within each standard colonoscopy group were calculated twice to obtain the standard inner diameter of the intestine corresponding to the intestinal examination position of each standard colonoscopy group.

[0011] As a preferred solution of the present invention, in step 200, the method for performing image processing on each image fragment to identify the specific pixel points of each image fragment is: Obtain the original HSV space f(x, y) of each image fragment, set the threshold range of H, S, and V in the original HSV space f(x, y), detect the intestinal wall blood vessels in each image fragment using a red object mask, and output an intestinal wall blood vessel detection image g(x, y), where the pixel value of red pixels in the intestinal wall blood vessel detection image g(x, y) is 255, and the pixel values ​​of pixels of other colors are 0; A two-dimensional coordinate system is constructed for each image fragment, and the pixel point with a pixel value of 255 is regarded as the specific pixel point.

[0012] Based on the distribution of the coordinate values ​​(x, y) of specific pixel points on the intestinal wall blood vessel detection image g(x, y) of each image fragment, the edge contour of the intestinal wall blood vessels formed by the specific pixel points is determined, and the distribution area set of the intestinal wall blood vessels of each image fragment is calculated.

[0013] The average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels on the image fragments corresponding to all standard endoscopic historical images in each of the standard colonoscopy groups are calculated, and the average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels are used as the standard state of the blood vessel distribution at the intestinal examination position corresponding to each of the standard colonoscopy groups.

[0014] As a preferred embodiment of the present invention, in step 300, intestinal endoscopic examination images of the subject are sequentially acquired according to the same intestinal examination position, wherein the intestinal endoscopic examination images acquired at the same intestinal examination position are divided into a colonoscopy group to be compared, and when determining the surface smoothness value to be measured of the intestinal surface of each colonoscopy group to be compared, the intestinal endoscopic examination image is first split into multiple surface images to be measured, and then the surface smoothness value to be measured of each of the intestinal endoscopic examination images is calculated, wherein the intestinal endoscopic examination image is first split into multiple surface images to be measured as follows: performing image noise reduction on the intestinal endoscopy images in each intestinal endoscopy group to be compared, and performing binarization image processing on the intestinal endoscopy images after noise reduction to preserve the inner surface structure morphology in the intestinal endoscopy images; Surface Gaussian curvature analysis is used to evaluate the smoothness of the intestinal wall of the binarized image of each intestinal endoscopy image to identify whether each intestinal endoscopy image after binarized image processing contains wrinkle rings; If not included, each intestinal endoscopy image is cut and split into multiple surface images to be tested in an equidistant order from inside to outside; If included, each intestinal endoscopy image is cut and split into multiple surface images to be tested according to the distribution position of the fold ring, and the image of each intestinal endoscopy image where the fold ring does not exist is split into multiple surface images to be tested according to the same spacing.

[0015] As a preferred solution of the present invention, the intestinal surface between the two cutting positions is filled in a mirror compensation manner, and the intestinal wall smoothness of each of the surface images to be tested is evaluated using surface Gaussian curvature analysis to construct an intestinal smoothness detection set for each of the intestinal endoscopy images; The extreme value of the intestinal smoothness detection to be tested is extracted from the intestinal inner wall smoothness set, and all the data of the intestinal smoothness detection set are averaged to obtain the average value of the intestinal smoothness detection to be tested of each intestinal endoscopic examination image to be compared, and the extreme value of the intestinal smoothness detection to be tested and the average value of the intestinal smoothness detection to be tested are used as the smoothness value of the surface to be tested.

[0016] As a preferred solution of the present invention, a two-dimensional coordinate system is reconstructed for each of the curved surface images to be measured, the average surface distance of each of the curved surface images to be measured is calculated, and the average of the average surface distances of all the curved surface images to be measured of the intestinal endoscopy image is used as the inner diameter of the intestinal tract to be measured of the intestinal endoscopy image; Construct a two-dimensional coordinate system for each surface image to be measured, obtain the original HSV space f'(x, y) of each image fragment, set the threshold range of H, S, and V in the original HSV space f'(x, y), detect the intestinal wall blood vessels in the surface image to be measured using a red object mask, and output the intestinal wall blood vessel binary image g'(x, y), where the pixel value of red pixels in the intestinal wall blood vessel binary image g'(x, y) is 255, and the pixel values ​​of pixels of other colors are 0; A two-dimensional coordinate system is constructed for each surface image to be measured, and the pixel points with a pixel value of 255 are regarded as specific pixel points.

[0017] Based on the distribution of the coordinate values ​​(x, y) of the specific pixel points on the binary image g'(x, y) of the intestinal wall blood vessels of each of the image fragments, the edge contours of the intestinal wall blood vessels formed by the specific pixel points are determined, and the distribution area set of the intestinal wall blood vessels of each of the curved surface images to be measured is calculated.

[0018] Calculate the average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels on the surface image to be measured corresponding to all the intestinal endoscopy inspection images in each of the said intestinal endoscopy groups to be compared, and use the average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels as the blood vessel distribution measurement status of the intestinal inspection position corresponding to each of the said intestinal endoscopy inspection images.

[0019] As a preferred embodiment of the present invention, in step 400, the method for implementing the constraint evaluation to identify symptomatic bowel disease is as follows: Calculate the difference between the extreme value of the standard intestinal smoothness corresponding to the standard endoscopic historical image and the extreme value of the intestinal smoothness to be tested in the intestinal endoscopic examination image, as well as the difference between the average value of the standard intestinal smoothness corresponding to the standard endoscopic historical image and the average value of the intestinal smoothness to be tested in the intestinal endoscopic examination image, and score the intestinal surface smoothness value of the intestinal endoscopic examination image based on the difference range; Based on the difference between the inner diameter of the intestine to be measured and the standard inner diameter of the intestine, scoring the inner diameter of the intestine in the intestinal endoscopy image based on the range of the difference; The difference between the average value of the distribution area of ​​the intestinal wall blood vessels corresponding to the standard endoscopic historical image and the average value of the distribution area of ​​the intestinal wall blood vessels corresponding to the intestinal endoscopic examination image at the same intestinal examination position, as well as the difference between the positive value of the distribution area of ​​the intestinal wall blood vessels corresponding to the standard endoscopic historical image and the positive value of the distribution area of ​​the intestinal wall blood vessels corresponding to the intestinal endoscopic examination image are calculated respectively, and the vascular bleeding status of the intestinal endoscopic examination image is scored based on the difference range.

[0020] As a preferred embodiment of the present invention, the smoothness value of the surface to be measured, the inner diameter of the intestine to be measured, and the IRS score of the blood vessel distribution measurement state of the intestinal endoscopy image are calculated respectively; Based on the numerical range of the IRS score, the ulcerative colitis lesion grade corresponding to the intestinal endoscopy image is automatically confirmed.

[0021] Compared with the prior art, the present invention has the following beneficial effects: The present invention processes standard endoscopic historical images to obtain evaluation indicators of multiple different dimensions, and calculates parameters such as surface smoothness value, intestinal inner diameter, and specific pixel distribution area set of intestinal endoscopic examination images of different intestinal examination positions, and then performs correlation evaluation based on a multi-parameter constrained evaluation method to evaluate the severity of ulcerative colitis. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0023] Figure 1 Schematic diagram of the flow of a method for recognizing symptomatic intestinal diseases according to an embodiment of the present invention; Figure 2 The standard endoscopic historical images of different intestinal examination positions according to an embodiment of the present invention; Figure 3 This is a colonoscopic endoscopic image of a patient suffering from severe intestinal diseases according to an embodiment of the present invention; Figure 4 This is a colonoscopic endoscopic image of a patient with moderate intestinal disease according to an embodiment of the present invention; DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] like Figure 1 As shown, an artificial intelligence-based ulcerative colitis image recognition method includes: Step 100: Obtain standard endoscopic historical images of different intestinal examination positions, form a standard colonoscopy group with multiple standard endoscopic historical images corresponding to each intestinal examination position, split the standard endoscopic historical images in each standard colonoscopy group into multiple image fragments in order from inside to outside, perform independent image processing on each image fragment, and calculate the surface smoothing standard value and the standard inner diameter of the intestine corresponding to the intestinal examination position of each standard colonoscopy group.

[0026] Step 200: Identify the specific pixel points of each image fragment and determine the standard state of blood vessel distribution at the intestinal examination position corresponding to each standard colonoscopy group.

[0027] Step 300: Obtain intestinal endoscopic examination images to be compared at different intestinal examination positions, and group them in sequence according to the intestinal examination positions to form multiple colonoscopy groups to be compared, and calculate the surface smoothness value to be measured, the inner diameter of the intestinal tract to be measured, and the vascular distribution measurement status of each intestinal endoscopic group to be compared corresponding to the intestinal examination position.

[0028] Step 400: Determine the disease grade of ulcerative colitis in a constrained evaluation manner based on the comparison results of the measured surface smoothness value, the measured intestinal inner diameter and the measured vascularity with the surface smoothness standard value, the intestinal standard inner diameter and the vascularity standard.

[0029] In this embodiment, artificial intelligence recognition technology is used to obtain three dimensions from intestinal images to evaluate the classification of symptomatic intestinal diseases. The specific implementation method is as follows: First, through image processing technology, the standard surface smoothness value, standard intestinal inner diameter and standard state of blood vessel distribution corresponding to each intestinal examination position are determined from the historical standard endoscopic images without symptoms.

[0030] Then, according to the intestinal examination positions corresponding to the standard endoscopic historical images, the subjects are sequentially subjected to colonoscopy examinations to obtain comparative intestinal endoscopic examination images of the same intestinal examination positions. Using image processing technology, the smoothness value of the surface to be measured, the inner diameter of the intestinal tract to be measured, and the vascular distribution measurement status corresponding to each intestinal examination position are determined from the comparative intestinal endoscopic examination images. Finally, the degree of symptomatic intestinal disease (specifically ulcerative colitis) is identified by constrained evaluation through comparison of the measured surface smoothness value, the measured intestinal inner diameter and the measured vascular distribution status with the surface smoothness standard value, the standard intestinal inner diameter and the standard vascular distribution status.

[0031] For example, through image processing technology, it is determined that the difference between the smoothness value of the surface to be measured and the standard value of the surface smoothness is within the maximum difference range, or the difference between the inner diameter of the intestine to be measured and the standard inner diameter of the intestine is within the maximum difference range, or the measured state of blood vessel distribution and the standard state of blood vessel distribution are within the maximum difference range. This means that the subject has a large ulcer or multiple ulcers in the intestine, and has symptoms of intestinal stenosis and massive bleeding, and therefore suffers from severe ulcerative colitis.

[0032] For example, through image processing technology, it is determined that the difference between the measured surface smoothness value and the surface smoothness standard value is in the medium difference range, or the difference between the measured intestinal diameter and the standard intestinal diameter is in the medium difference range, or the vascular distribution measurement status and the vascular distribution standard status are in the medium difference range, which means that the subject has superficial small ulcers in the intestine, small bleeding spots or mucosal thickening, and therefore has moderate ulcerative colitis.

[0033] For example, through image processing technology, it is determined that the difference between the smoothness value of the surface to be measured and the standard value of the surface smoothness is in a small difference range, or the difference between the inner diameter of the intestine to be measured and the standard inner diameter of the intestine is in a small difference range, or the measured blood vessel distribution state and the standard blood vessel distribution state are in a small difference range, which means that the subject has local erythema in the intestine or a small blood vessel distribution area, and therefore has mild ulcerative colitis.

[0034] Therefore, this embodiment only needs to obtain the surface smoothness standard value, intestinal standard inner diameter and blood vessel distribution standard status of the standard endoscopic historical image corresponding to each intestinal inspection position through image processing, and obtain the surface smoothness value to be measured, intestinal inner diameter to be measured and blood vessel distribution measurement status in the intestinal endoscopic inspection image to be compared corresponding to each intestinal inspection position through image processing, which can constrain the evaluation of the severity of ulcerative colitis and improve the accuracy of the severity of symptomatic intestinal diseases.

[0035] In step 100, there is at least one standard endoscopic historical image in each standard colonoscopy group. Each standard endoscopic historical image in each standard colonoscopy group is split into multiple image fragments in order from the inside to the outside as follows: Image denoising was performed on each standard endoscopy historical image in each standard colonoscopy group.

[0036] The original image coordinate system of each standard endoscopic historical image is constructed, and each standard endoscopic historical image is subjected to binarization image processing to preserve the structural morphology of each standard endoscopic historical image.

[0037] The wrinkle rings contained in each standard endoscopic historical image after binary image processing are identified, and the position of each wrinkle ring in the original image coordinate system is marked.

[0038] Each standard endoscopic historical image is cut according to the position of the marked wrinkle ring to form multiple image fragments.

[0039] It should be noted that, in this embodiment, when shooting at a certain position of the intestine, an image of the intestinal segment in a certain longitudinal direction will be captured. Since the large intestine has circular muscles, the circular muscles regularly contract and relax to divide the intestinal contents into several segments, and then slowly move forward. This movement mode can enable the intestinal contents to fully contact with the intestinal mucosa, which is beneficial to the absorption of nutrients and the reabsorption of water. It is one of the more common forms of movement of the large intestine. Therefore, the captured image will have multiple wrinkled rings, which indicate the position of the circular muscles of the large intestine.

[0040] Therefore, in order to avoid the influence of the circular muscle of the large intestine on the calculation accuracy of the smoothness of the inner wall of the standard endoscopic historical image, this embodiment performs binarization image processing on each standard endoscopic historical image to retain the structural morphology of each standard endoscopic historical image, so as to identify the fold rings of each standard endoscopic historical image. The inner wall curved surface of the intestine is located between two fold rings, and the surface curvature analysis module is embedded in the image processing technology to determine the smoothness of the inner wall of the intestine curved surface between the two fold rings.

[0041] In step 100, the method for determining the surface smoothness standard value of the intestinal surface of each standard colonoscopy group is as follows: The intestinal surface between the two wrinkle rings is filled in a mirror compensation manner to form a complete intestinal surface between the two wrinkle rings. The surface Gaussian curvature analysis is used to evaluate the smoothness of the intestinal wall of the intestinal surface between the two wrinkle rings.

[0042] The parameter distribution state of the intestinal wall smoothness of the intestinal surface formed between two wrinkle rings in each standard endoscopic historical image is determined.

[0043] Based on the parameter distribution of the intestinal wall smoothness of all standard endoscopic historical images in each standard colonoscopy group, the extreme value of the standard intestinal smoothness of each standard endoscopic historical image and the average value of the standard intestinal smoothness of each standard endoscopic historical image are used as the surface smoothness standard values.

[0044] The mean is often used to describe the overall level or central tendency of a set of data. For example, in statistics, the mean is used to indicate the central tendency of a set of data, reflecting the overall average level of the data. Therefore, the mean of the standard intestinal smoothness of each standard endoscopic historical image, as one of the surface smoothness standard values, can reflect the overall condition of the intestinal wall at the depth that can be monitored at that colonoscopy location.

[0045] Extreme values ​​are used to describe the extreme conditions of data. In data analysis, maximum and minimum values ​​can help identify outliers or extreme conditions in a data set. This embodiment specifically uses the absolute value of the standard intestinal smoothness extreme value to represent the extreme conditions of the intestinal wall at the depth that can be monitored by the colonoscopy shooting position.

[0046] It should be noted that the calculation formula of Gaussian curvature is as follows: Gaussian curvature K = det(-dN / ds) / (det(dX / ds))^2 Among them, det represents the determinant of the matrix; dN / ds represents the derivative of the normal vector N with respect to the curve parameter s; dX / ds represents the derivative of the position vector X with respect to the curve parameter s.

[0047] Gaussian curvature represents the curvature of a surface. It is a scalar value that describes the degree of curvature of a surface. Generally speaking, the larger the Gaussian curvature, the more pronounced the surface's curvature, while the smaller the Gaussian curvature, the more pronounced the surface's flattening. A Gaussian curvature of zero indicates a flat surface. Gaussian curvature is related only to the curvature of the surface, not its area. For example, surfaces with different diameters at the same center have the same Gaussian curvature.

[0048] In this embodiment, although according to the depth of the colonoscopy shooting position, the center position of the colonoscopy image is specifically represented as the intestinal wall image taken at the distal end, and the enclosed area of ​​the intestinal curved surface between the two fold rings is small, and the periphery of the colonoscopy image is represented as the intestinal wall image taken at the proximal end, and the enclosed area of ​​the intestinal curved surface between the two fold rings is large, but using Gaussian curvature to represent the curvature of the intestinal curved surface between the two fold rings is specifically related to the curvature of the intestinal wall surface, and has nothing to do with the large enclosed area of ​​the intestinal curved surface between the two fold rings, thereby improving the accuracy of expressing the smoothness of the intestinal inner wall.

[0049] In this embodiment, the smoothness of the intestinal wall can indicate whether the intestinal mucosa has ulcers and erosions. When ulcers and erosions occur, the curvature of the intestinal wall increases. The more ulcers and erosions occur, the greater the curvature of the intestinal wall. In the standard endoscopic historical images of healthy people, since the intestinal mucosa is smooth, the Gaussian curvature is a concept of calculus, and the value of the Gaussian curvature is usually closely related to the shape of the surface. Specifically, for plane or spherical surfaces, their Gaussian curvatures are 0 and 1, respectively, and for non-regular surfaces such as saddle surfaces, their Gaussian curvatures are negative, that is, for the curve of the colonoscopy image of ulcerative colitis, its Gaussian curvature is negative.

[0050] Furthermore, when ulcers and erosions occur in the intestinal mucosa, water accumulation and swelling will occur in the intestine, which will lead to a reduction in the inner diameter of the intestine. Based on the size of the inner diameter of the intestine, the water accumulation and swelling in the intestine can be judged, and then the severity of the ulcer and erosion can be determined. In other words, the size of the inner diameter of the intestine is positively correlated with the severity of the ulcer and erosion. Therefore, the severity of ulcerative colitis can also be determined by identifying the size of the inner diameter of the intestine in the intestinal endoscopic examination images to be compared corresponding to different intestinal examination positions.

[0051] In step 100, all standard endoscopy historical images in each standard colonoscopy group are divided into a plurality of intestinal surfaces according to their corresponding fold rings, and the average distance between each intestinal surface is calculated.

[0052] The average value of the average distances of all intestinal surfaces in each standard endoscopic historical image is taken as the inner diameter of each intestinal tract in the standard endoscopic historical image.

[0053] The inner diameters of the intestines of all standard endoscopic historical images within each standard colonoscopy group were calculated twice to obtain the standard inner diameter of the intestine corresponding to the intestinal examination position of each standard colonoscopy group.

[0054] Furthermore, when the intestinal mucosa is swollen or ulcerated and eroded, bleeding may occur on the mucosal surface. In order to compare the bleeding intestinal wall with the normal intestinal wall, it is necessary to identify the edge contours of the intestinal wall blood vessels of the normal intestinal wall. The average distribution area and the positive distribution area of ​​the blood vessels of the normal intestinal wall are used as the standard state of blood vessel distribution. In step 200, image processing is performed on each image fragment to identify the specific pixel points of each image fragment as follows: Get the original HSV space f(x,y) of each image fragment, set the threshold range of H, S, and V in the original HSV space f(x,y), use the red object mask to detect the intestinal wall blood vessels in each image fragment, and output the intestinal wall blood vessel detection image g(x,y), where the pixel value of the red pixel in the intestinal wall blood vessel detection image g(x,y) is 255, and the pixel values ​​of the pixels of other colors are 0; For each image fragment, a two-dimensional coordinate system is constructed, and the pixel point with a pixel value of 255 is used as the specific pixel point.

[0055] Based on the distribution of coordinate values ​​(x, y) of specific pixel points on the intestinal wall blood vessel detection image g(x, y) of each image fragment, the edge contour of the intestinal wall blood vessels formed by the specific pixel points is determined, and the intestinal wall blood vessel distribution area set of each image fragment is calculated.

[0056] The average distribution area and positive value of the intestinal wall blood vessels on the image fragments corresponding to all standard endoscopic historical images in each standard colonoscopy group were calculated, and the average distribution area and positive value of the intestinal wall blood vessels were used as the standard state of blood vessel distribution at the intestinal examination position corresponding to each standard colonoscopy group.

[0057] All standard endoscopy historical images in each standard colonoscopy group are converted into the HSV color space, the threshold ranges of H, S, and V are set, a red pixel mask is generated, and post-processing is performed to detect the red part in the standard endoscopy historical images, where the pixel value of the red part is 255 and the pixel value of the other parts is 0.

[0058] Specifically, the pixel point with a pixel value of 255 is used as the specific pixel point, the edge contour of the intestinal wall blood vessels composed of the specific pixel points is determined, and the average distribution area and the positive distribution area of ​​the intestinal wall blood vessels of each image fragment are calculated to represent the distribution morphology of the intestinal wall blood vessels and the vascular bleeding situation of the current captured image.

[0059] Therefore, based on step 100 and step 200, standard data in three dimensions, namely, the surface smoothness standard value, the standard inner diameter of the intestine, and the standard state of blood vessel distribution, corresponding to each intestinal examination position can be determined.

[0060] In step 300, intestinal endoscopic examination images of the subject are sequentially acquired at the same intestinal examination position, wherein the intestinal endoscopic examination images acquired at the same intestinal examination position are divided into a colonoscopy group to be compared. When determining the surface smoothness value to be measured of the intestinal surface of each colonoscopy group to be compared, the intestinal endoscopic examination image is first split into multiple surface images to be measured, and then the surface smoothness value to be measured of each intestinal endoscopic examination image is calculated. The implementation method of first splitting the intestinal endoscopic examination image into multiple surface images to be measured is as follows: performing image noise reduction on the intestinal endoscopy images in each colonoscopy group to be compared, and performing binarization image processing on the intestinal endoscopy images after noise reduction to preserve the internal surface structure morphology in the intestinal endoscopy images; Surface Gaussian curvature analysis is used to evaluate the smoothness of the intestinal wall of the binarized image of each intestinal endoscopy image to identify whether each intestinal endoscopy image after binarized image processing contains wrinkle rings; If not included, each intestinal endoscopy image is cut and split into multiple surface images to be tested in an equidistant order from inside to outside; If included, each intestinal endoscopy image is cut and split into multiple surface images to be tested according to the distribution position of the fold ring, and the image without fold ring in each intestinal endoscopy image is split into multiple surface images to be tested according to the same spacing.

[0061] For the intestinal endoscopy images of the subjects, if the patient does not have intestinal diseases and can identify the wrinkle rings of the intestinal endoscopy images, the intestinal endoscopy images are also divided into multiple surface images to be tested according to the wrinkle rings, and the surface Gaussian curvature analysis is used to evaluate the smoothness of the intestinal wall of each surface image to be tested.

[0062] For patients with intestinal diseases, such as Figure 3 and Figure 4 As shown, the inner wall of the intestine will swell due to ulcer erosion, and the wrinkle ring cannot be identified. At this time, each intestinal endoscopy image is cut and split into multiple surface images to be tested in an equidistant order from the inside to the outside.

[0063] The surface smoothness value to be measured corresponding to each intestinal endoscopy examination image of the same intestinal examination position is compared with the surface smoothness standard value of the position to determine the difference between the surface smoothness value to be measured and the surface smoothness standard value, and the intestinal ulcer erosion condition corresponding to each intestinal examination position is determined based on the difference.

[0064] Similarly, the inner diameter of the intestine to be measured corresponding to each intestinal endoscopy examination image of the same intestinal examination position is compared with the standard inner diameter of the intestine at that intestinal examination position to determine the difference between the inner diameter of the intestine to be measured and the inner diameter of the intestine to be measured, and the intestinal swelling condition corresponding to each intestinal examination position is determined based on the difference.

[0065] The intestinal surface between the two cutting positions is filled in a mirror compensation manner, and the surface Gaussian curvature analysis is used to evaluate the smoothness of the intestinal wall of each surface image to be tested, and an intestinal smoothness detection set is constructed for each intestinal endoscopy image.

[0066] The extreme value of the intestinal smoothness detection to be tested is extracted from the intestinal inner wall smoothness set, and all the data of the intestinal smoothness detection set are averaged to obtain the average value of the intestinal smoothness detection to be tested of each intestinal endoscopic examination image to be compared, and the extreme value of the intestinal smoothness detection to be tested and the average value of the intestinal smoothness detection to be tested are used as the smoothness value of the surface to be tested.

[0067] A two-dimensional coordinate system is reconstructed for each surface image to be measured, the average surface distance of each surface image to be measured is calculated, and the average of the average surface distances of all surface images to be measured in the intestinal endoscopy image is used as the inner diameter of the intestinal tract to be measured in the intestinal endoscopy image.

[0068] A two-dimensional coordinate system was constructed for each surface image to be tested, and the original HSV space f'(x, y) of each image fragment was obtained. The threshold range of H, S, and V in the original HSV space f'(x, y) was set, and the intestinal wall blood vessels in the surface image to be tested were detected using the red object mask. The intestinal wall blood vessel binary image g'(x, y) was output, where the pixel value of the red pixel in the intestinal wall blood vessel binary image g'(x, y) was 255, and the pixel values ​​of the pixels of other colors were 0.

[0069] A two-dimensional coordinate system is constructed for each surface image to be measured, and the pixel points with a pixel value of 255 are regarded as specific pixel points.

[0070] Based on the distribution of the coordinate values ​​(x, y) of specific pixel points on the binary image g'(x, y) of the intestinal wall blood vessels of each image fragment, the edge contour of the intestinal wall blood vessels formed by the specific pixel points is determined, and the distribution area set of the intestinal wall blood vessels of each surface image to be measured is calculated.

[0071] Calculate the average distribution area and positive value of the intestinal wall blood vessels on the measured surface image corresponding to all intestinal endoscopy images in each intestinal endoscopy group to be compared, and use the average distribution area and positive value of the intestinal wall blood vessels as the blood vessel distribution measurement status of the intestinal examination position corresponding to each intestinal endoscopy image.

[0072] The above principles for determining the smoothness value of the surface to be measured and the measurement status of the vascular distribution at the intestinal examination position corresponding to each intestinal endoscopy image are the same as the implementation principles for determining the surface smoothness standard value and the vascular distribution standard status of the standard endoscopic historical image, and no further detailed description will be given.

[0073] After the above analysis, the smoothness value of the surface to be measured, the inner diameter of the intestine to be measured, and the measurement status of the blood vessel distribution corresponding to the intestinal endoscopic examination image of a certain intestinal examination position can be obtained. The smoothness value of the surface to be measured, the inner diameter of the intestine to be measured, and the measurement status of the blood vessel distribution corresponding to each intestinal examination position are compared with the surface smoothness standard value, the standard inner diameter of the intestine, and the standard status of the blood vessel distribution of the corresponding standard image to determine the comparison result of the intestinal examination position. If the smoothness value of the surface to be measured is much larger than the standard value of the surface smoothness, the inner diameter of the intestine to be measured is much smaller than the standard inner diameter of the intestine, or the measurement status of the blood vessel distribution is completely different from the standard status of the blood vessel distribution, then the ulcerative colitis at the intestinal examination position is determined to be severe.

[0074] If the measured surface smoothness value is greater than the surface smoothness standard value, the measured intestinal diameter is smaller than the intestinal standard diameter, or the measured vascularity status is approximately the same as the vascularity standard status, then the ulcerative colitis at the intestinal examination location is determined to be moderate.

[0075] If the measured surface smoothness value is slightly greater than the surface smoothness standard value, the measured intestinal diameter is slightly smaller than the intestinal standard diameter, or the measured vascularity status is approximately the same as the vascularity standard status, then the ulcerative colitis at the intestinal examination location is determined to be mild.

[0076] In step 400, the method for implementing the constraint evaluation to identify symptomatic bowel disease is as follows: The difference between the extreme value of the standard intestinal smoothness corresponding to the standard endoscopic historical image and the extreme value of the intestinal smoothness to be tested in the intestinal endoscopic examination image, as well as the difference between the average value of the standard intestinal smoothness corresponding to the standard endoscopic historical image and the average value of the intestinal smoothness to be tested in the intestinal endoscopic examination image, are calculated respectively. The intestinal surface smoothness value of the intestinal endoscopic examination image is scored based on the difference range. The intestinal surface smoothness value score is used to characterize the degree of deformation of the intestinal inner wall of the subject.

[0077] Based on the difference between the inner diameter of the intestine to be measured and the standard inner diameter of the intestine, the inner diameter of the intestine in the intestinal endoscopy image is scored based on the difference range. The intestinal diameter scoring result is used to characterize the degree of intestinal stenosis of the subject.

[0078] The difference between the average distribution area of ​​the intestinal wall blood vessels corresponding to the standard endoscopic historical images and the average distribution area of ​​the intestinal wall blood vessels corresponding to the intestinal endoscopic examination image at the same intestinal examination position, as well as the difference between the positive value of the distribution area of ​​the intestinal wall blood vessels corresponding to the standard endoscopic historical images and the positive value of the distribution area of ​​the intestinal wall blood vessels corresponding to the intestinal endoscopic examination image are calculated respectively. The vascular bleeding status of the intestinal endoscopic examination image is scored based on the difference range. The vascular bleeding status scoring result is used to characterize the degree of subcutaneous bleeding of the subject.

[0079] The IRS scores of the measured surface smoothness value, the measured intestinal diameter, and the vascular distribution measurement status of the intestinal endoscopy images were calculated respectively.

[0080] Based on the numerical range of IRS score, the ulcerative colitis lesion grade corresponding to the intestinal endoscopy image is automatically confirmed.

[0081] That is, this embodiment combines the degree of intestinal stenosis, the degree of subcutaneous bleeding and the degree of intestinal wall deformation to restrictively evaluate the grade of symptomatic intestinal disease.

[0082] Specifically, the severity of ulcerative colitis is assessed based on the degree of intestinal stricture, differences in vascularity, and the degree of intestinal lining deformation.

[0083] Therefore, this embodiment processes standard endoscopic historical images to obtain evaluation indicators of multiple different dimensions, and calculates parameters such as surface smoothness value, intestinal inner diameter, and specific pixel distribution area set of intestinal endoscopic examination images of different intestinal examination positions, and then performs correlation evaluation based on a multi-parameter constrained evaluation method to evaluate the severity of symptomatic intestinal diseases and improve the accuracy of symptom severity evaluation.

[0084] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.

Claims

1. An artificial intelligence-based ulcerative colitis image recognition method, characterized in that: include: Step 100: Acquire standard endoscopic historical images at different intestinal examination positions, form a standard colonoscopy group with multiple standard endoscopic historical images corresponding to each intestinal examination position, split the standard endoscopic historical images in each standard colonoscopy group into multiple image fragments in order from the inside to the outside, perform independent image processing on each image fragment, and calculate the surface smoothing standard value and the standard inner diameter of the intestine for each intestinal examination position corresponding to the standard colonoscopy group; Step 200: identifying specific pixel points of each image fragment, and determining the standard state of blood vessel distribution at the intestinal examination position corresponding to each standard colonoscopy group; Step 300: Obtain intestinal endoscopic examination images to be compared at different intestinal examination locations, group them sequentially according to the intestinal examination locations to form a plurality of intestinal endoscopy groups to be compared, and calculate the surface smoothness value to be measured, the inner diameter of the intestinal tract to be measured, and the vascular distribution measurement status for each intestinal endoscopy group to be compared corresponding to the intestinal examination location; Step 400: Determine the disease grade of ulcerative colitis in a constrained evaluation manner based on the comparison results of the measured surface smoothness value, the measured intestinal inner diameter and the measured vascularity with the surface smoothness standard value, the intestinal standard inner diameter and the vascularity standard.

2. The artificial intelligence-based ulcerative colitis image recognition method according to claim 1, characterized in that: In step 100, there is at least one standard endoscopic historical image in each standard colonoscopy group, and each standard endoscopic historical image in each standard colonoscopy group is split into multiple image fragments in order from the inside to the outside as follows: performing image noise reduction on each standard endoscopy historical image in each of the standard colonoscopy groups; Constructing the original image coordinate system of each standard endoscopic historical image, performing binarization image processing on each standard endoscopic historical image, and retaining the structural morphology of each standard endoscopic historical image; Identifying wrinkle rings contained in each standard endoscopic historical image after binary image processing, and marking the position of each wrinkle ring in the original image coordinate system; Each standard endoscope historical image is cut according to the position of the marked wrinkle ring to form a plurality of image fragments.

3. The artificial intelligence-based ulcerative colitis image recognition method according to claim 2, characterized in that: In step 100, the method for determining the surface smoothness standard value of the intestinal surface of each standard colonoscopy group is as follows: Filling the intestinal surface between the two wrinkle rings in a mirror compensation manner so that a complete intestinal surface is formed between the two wrinkle rings, and using surface Gaussian curvature analysis to evaluate the smoothness of the intestinal inner wall of the intestinal surface between the two wrinkle rings; Determining a parameter distribution state of the smoothness of the intestinal inner wall of the intestinal curved surface formed between any two of the wrinkled rings in each standard endoscopic historical image; Based on the parameter distribution state of the intestinal wall smoothness of all standard endoscopic historical images in each of the standard colonoscopy groups, the extreme value of the standard intestinal smoothness of each standard endoscopic historical image and the average value of the standard intestinal smoothness of each standard endoscopic historical image are used as the surface smoothness standard values.

4. The artificial intelligence-based ulcerative colitis image recognition method according to claim 2, characterized in that: In step 100, all standard endoscopy historical images in each standard colonoscopy group are divided into a plurality of intestinal curved surfaces according to their corresponding fold rings, and the average distance between each intestinal curved surface is calculated; The average value of the average distances of all intestinal surfaces of each standard endoscopic historical image is taken as the inner diameter of each intestinal tract of the standard endoscopic historical image; The inner diameters of the intestines of all standard endoscopic historical images within each standard colonoscopy group were calculated twice to obtain the standard inner diameter of the intestine corresponding to the intestinal examination position of each standard colonoscopy group.

5. The artificial intelligence-based ulcerative colitis image recognition method according to claim 2, characterized in that: In step 200, the method for performing image processing on each image fragment to identify the specific pixel points of each image fragment is as follows: Obtain the original HSV space f(x, y) of each image fragment, set the threshold range of H, S, and V in the original HSV space f(x, y), detect the intestinal wall blood vessels in each image fragment using a red object mask, and output an intestinal wall blood vessel detection image g(x, y), where the pixel value of red pixels in the intestinal wall blood vessel detection image g(x, y) is 255, and the pixel values ​​of pixels of other colors are 0; A two-dimensional coordinate system is constructed for each image fragment, and the pixel point with a pixel value of 255 is regarded as the specific pixel point; Based on the distribution of coordinate values ​​(x, y) of specific pixel points on the intestinal wall blood vessel detection image g(x, y) of each image fragment, determine the edge contour of the intestinal wall blood vessels formed by the specific pixel points, and calculate the distribution area set of the intestinal wall blood vessels of each image fragment; The average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels on the image fragments corresponding to all standard endoscopic historical images in each of the standard colonoscopy groups are calculated, and the average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels are used as the standard state of the blood vessel distribution at the intestinal examination position corresponding to each of the standard colonoscopy groups.

6. The method for ulcerative colitis image recognition based on artificial intelligence according to claim 1, characterized in that: In step 300, intestinal endoscopic examination images of the subject are sequentially acquired according to the same intestinal examination position, wherein the intestinal endoscopic examination images acquired at the same intestinal examination position are divided into a colonoscopy group to be compared, and when determining the surface smoothness value to be measured of the intestinal surface of each colonoscopy group to be compared, the intestinal endoscopic examination image is first split into multiple surface images to be measured, and then the surface smoothness value to be measured of each of the intestinal endoscopic examination images is calculated, wherein the intestinal endoscopic examination image is first split into multiple surface images to be measured as follows: performing image noise reduction on the intestinal endoscopy images in each intestinal endoscopy group to be compared, and performing binarization image processing on the intestinal endoscopy images after noise reduction to preserve the inner surface structure morphology in the intestinal endoscopy images; Surface Gaussian curvature analysis is used to evaluate the smoothness of the intestinal wall of the binarized image of each intestinal endoscopy image to identify whether each intestinal endoscopy image after binarized image processing contains wrinkle rings; If not included, each intestinal endoscopy image is cut and split into multiple surface images to be tested in an equidistant order from inside to outside; If included, each intestinal endoscopy image is cut and split into multiple surface images to be tested according to the distribution position of the fold ring, and the image of each intestinal endoscopy image where the fold ring does not exist is split into multiple surface images to be tested according to the same spacing.

7. The artificial intelligence-based ulcerative colitis image recognition method according to claim 6, characterized in that: Filling the intestinal surface between the two cutting positions in a mirror compensation manner, using surface Gaussian curvature analysis to evaluate the smoothness of the intestinal wall of each of the surface images to be tested, and constructing an intestinal smoothness detection set for each of the intestinal endoscopy images; The extreme value of the intestinal smoothness detection to be tested is extracted from the intestinal inner wall smoothness set, and all the data of the intestinal smoothness detection set are averaged to obtain the average value of the intestinal smoothness detection to be tested of each intestinal endoscopic examination image to be compared, and the extreme value of the intestinal smoothness detection to be tested and the average value of the intestinal smoothness detection to be tested are used as the smoothness value of the surface to be tested.

8. The artificial intelligence-based ulcerative colitis image recognition method according to claim 7, characterized in that: reconstructing a two-dimensional coordinate system for each of the curved surface images to be measured, calculating the average surface distance of each of the curved surface images to be measured, and taking the average of the average surface distances of all the curved surface images to be measured of the intestinal endoscopy image as the inner diameter of the intestinal tract to be measured of the intestinal endoscopy image; Construct a two-dimensional coordinate system for each surface image to be measured, obtain the original HSV space f'(x, y) of each image fragment, set the threshold range of H, S, and V in the original HSV space f'(x, y), detect the intestinal wall blood vessels in the surface image to be measured using a red object mask, and output the intestinal wall blood vessel binary image g'(x, y), where the pixel value of red pixels in the intestinal wall blood vessel binary image g'(x, y) is 255, and the pixel values ​​of pixels of other colors are 0; A two-dimensional coordinate system is constructed for each surface image to be measured, and the pixel point with a pixel value of 255 is regarded as the specific pixel point; Based on the distribution of coordinate values ​​(x, y) of specific pixel points on the intestinal wall blood vessel binary image g'(x, y) of each image fragment, the edge contour of the intestinal wall blood vessel formed by the specific pixel points is determined, and the distribution area set of the intestinal wall blood vessels of each of the curved surface images to be measured is calculated; Calculate the average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels on the surface image to be measured corresponding to all the intestinal endoscopy inspection images in each of the said intestinal endoscopy groups to be compared, and use the average distribution area and the positive value of the distribution area of ​​the intestinal wall blood vessels as the blood vessel distribution measurement status of the intestinal inspection position corresponding to each of the said intestinal endoscopy inspection images.

9. The method for ulcerative colitis image recognition based on artificial intelligence according to claim 1, characterized in that: In step 400, the method for implementing the constraint evaluation to identify symptomatic bowel disease is as follows: Calculate the difference between the extreme value of the standard intestinal smoothness corresponding to the standard endoscopic historical image and the extreme value of the intestinal smoothness to be tested in the intestinal endoscopic examination image, as well as the difference between the average value of the standard intestinal smoothness corresponding to the standard endoscopic historical image and the average value of the intestinal smoothness to be tested in the intestinal endoscopic examination image, and score the intestinal surface smoothness value of the intestinal endoscopic examination image based on the difference range; Based on the difference between the inner diameter of the intestine to be measured and the standard inner diameter of the intestine, scoring the inner diameter of the intestine in the intestinal endoscopy image based on the range of the difference; The difference between the average value of the distribution area of ​​the intestinal wall blood vessels corresponding to the standard endoscopic historical image and the average value of the distribution area of ​​the intestinal wall blood vessels corresponding to the intestinal endoscopic examination image at the same intestinal examination position, as well as the difference between the positive value of the distribution area of ​​the intestinal wall blood vessels corresponding to the standard endoscopic historical image and the positive value of the distribution area of ​​the intestinal wall blood vessels corresponding to the intestinal endoscopic examination image are calculated respectively, and the vascular bleeding status of the intestinal endoscopic examination image is scored based on the difference range.

10. The artificial intelligence-based ulcerative colitis image recognition method according to claim 9, characterized in that: Calculating the smoothness value of the surface to be measured, the inner diameter of the intestine to be measured, and the IRS score of the vascular distribution measurement status of the intestinal endoscopy image respectively; Based on the numerical range of the IRS score, the ulcerative colitis lesion grade corresponding to the intestinal endoscopy image is automatically confirmed.

Images