Bone Bankart Injury Auxiliary Diagnosis Method and System Based on Image Recognition

Abstract

The present invention discloses an auxiliary diagnosis method and system for bony Bankart injury based on image recognition, including: through preprocessing and threshold segmentation of CT images, different tissue structures in CT images can be accurately distinguished, and by overlapping and analyzing the marked areas, the accuracy and intuitiveness of diagnosis are improved; randomly select edge pixel points and conduct detailed analysis, and judge whether they are normal by combining angle changes and gray values, reducing the interference of noise and artifacts, and making the image clearer. The judgment of all edge pixel points one by one ensures the overall improvement of image quality, and combined with feature matching, helps to construct a three-dimensional model, thereby helping to complete the recognition of Bankart injury of the patient's shoulder joint, providing strong support for the formulation of the patient's treatment plan.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and relates to a method and system for assisting in the diagnosis of osseous Bankart injury based on image recognition. Background Art

[0002] Osseous Bankart injury is a common type of shoulder joint injury, which involves avulsion injury of the anteroinferior glenoid labrum and its attached ligamentous structures, usually accompanied by fractures or bone defects in the anteroinferior glenoid. This kind of injury not only causes instability of the shoulder joint, but may also lead to a series of problems such as pain and dysfunction, seriously affecting the quality of life of patients. With the rapid development of computer technology and artificial intelligence, image recognition technology has gradually become an important auxiliary diagnostic means. When performing CT image examination on patients, due to possible slight tremors of patients during the CT image acquisition process due to pain, the acquired images are blurred, affecting the quality of CT images, thus resulting in abnormal recognition of osseous Bankart injury of the shoulder joint. Summary of the Invention

[0003] The purpose of the present invention is to solve the problem that the recognition of osseous Bankart injury of the shoulder joint is inaccurate due to blurred CT images in the prior art, and to provide a method and system for assisting in the diagnosis of osseous Bankart injury based on image recognition.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for assisting in the diagnosis of osseous Bankart injury based on image recognition, comprising:

[0006] Step 1, preprocessing the CT image of the patient's shoulder joint collected;

[0007] Step 2, dividing the region of the preprocessed shoulder joint CT image based on threshold segmentation technology;

[0008] Step 3, performing coincidence analysis on the image after threshold segmentation and the original CT image, and marking each region of the threshold segmentation on the original CT image;

[0009] Step 4, randomly selecting edge pixels of a region, and obtaining the possibility that the edge pixel is a normal edge pixel based on the angular change and gray value of the connection line between the edge pixel and adjacent edge pixels;

[0010] Step 5, determine whether the possibility that the edge pixel is a normal edge pixel is greater than a preset threshold. If it is greater, consider this edge pixel as a normal pixel; repeat Step 4 and Step 5 until the judgment of all edge pixels is completed to obtain a clear CT image of the patient's shoulder joint;

[0011] Step 6, perform adjacent image matching on the obtained several CT images of the patient's shoulder joint, and fuse the image information of the overlapping parts to construct a three-dimensional model of the patient's shoulder joint;

[0012] Step 7, input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

[0013] A further improvement of the present invention lies in:

[0014] Further, preprocess the CT image of the patient's shoulder joint collected, specifically: perform denoising processing and linear enhancement on the collected shoulder joint CT image to reduce the noise interference in the image and improve the clarity and contrast of the image.

[0015] Further, perform region division on the preprocessed shoulder joint CT image based on the threshold segmentation technology, specifically: according to the gray values of different tissue structures in the shoulder joint CT image, select the corresponding threshold, compare each pixel in the image with the threshold to generate a binary image, and then obtain the regions of different tissues.

[0016] Further, randomly select the edge pixels of the region, and based on the angular change and gray value between the connection line of this edge pixel and adjacent edge pixels, obtain the possibility that this edge pixel is a normal edge pixel, specifically:

[0017] Randomly select the edge pixels of the region to obtain the gray value of the th pixel in the th region on the th shoulder joint CT image ;

[0018] Based on the th pixel in the th region, select its adjacent edge pixel, connect the two pixels to obtain the included angle between the connection line and the horizontal direction ; repeat the operation to construct an angle data sequence between the outermost edge pixels of the th region, and then obtain the angular change trend of the connection line between adjacent edge pixels;

[0019] Based on the angular change trend between adjacent edge pixels and the The average gray value of the pixel points on the outermost edge of the th region on the th image. Judging from the pixel points themselves, the th region on the th image, the normality performance of the th pixel point on the outermost edge; further analyzing from the performance of the overall outer pixel points, the probability that the th region on the th pixel point on the outermost edge of the

[0020] th image belongs to a normal edge pixel point. th region, the angle data sequence between the outermost edge pixel points ; obtaining the angle change trend of the connection lines between adjacent edge pixel points, specifically:

[0021] ;

[0022] The judgment of the normality performance of the th pixel point on the outermost edge of the th region on the th image, specifically:

[0023] ;

[0024] ;

[0025] Among them, represents the average gray value of the pixel points on the outermost edge of the th region on the th image; represents the number of selected outermost edge pixel points; when the gray difference of a certain pixel point is small and the adjacent angle change is small, it indicates that the probability that it belongs to a normal edge pixel point is large;

[0026] The analysis from the performance of the overall outer pixel points, the probability that the th pixel point on the outermost edge of the th region on the th image belongs to a normal edge pixel point, specifically:

[0027] ;

[0028] ;

[0029] Among them, represents the th region on the The average angular change magnitude of the pixel points on the outermost edge of a region denotes the number of the pixel points on the outermost edge of a region on the .

[0030] Further, determine whether the possibility that the edge pixel point is a normal edge pixel point is greater than a preset threshold. If it is greater, then consider this edge pixel point as a normal pixel point; until the judgment of all edge pixel points is completed to obtain a clear CT image of the patient's shoulder joint, specifically:

[0031] Set a threshold . When , it indicates that this pixel point belongs to a normal edge pixel point; when judging the outermost pixel points and there are abnormal edge pixel points, remove them from the region, select the pixel points adjacent to them and adjacent to their previous pixel points to replace these abnormal pixel points as the region edge; recalculate whether the selected pixel points are normal edge pixel points until all pixel points are normal edge pixel points.

[0032] Further, perform adjacent image matching based on the obtained several CT images of the patient's shoulder joint, and fuse the image information of the overlapping parts to construct a three-dimensional model of the patient's shoulder joint, specifically: Obtain the reliability magnitude of the matching results of the h-th group of matching points on the th CT image and the th CT image. If the reliability magnitude of the matching results is greater than the set threshold , it indicates that the matching results of the matching points on the CT image are correct; if the error rate of the matching results exceeds the preset value, then reshoot this angle and repeat the above steps; until the matching results are correct; and then fuse the image information of the overlapping parts, and use the registered and fused image information to construct a three-dimensional model of the patient's shoulder joint.

[0033] Further, obtain the reliability magnitude of the matching results of the h-th group of matching points on the th CT image and the th CT image, specifically:

[0034] ;

[0035] ;

[0036] Among them, denotes the The distance between the pixel points corresponding to the h-th group of matching points on the [[ID=]]th CT image and the pixel points on the edge of the bone joint in the horizontal direction. If there are two distances, select the shortest distance. Denotes the -th pixel value among the nine pixel points centered on the pixel point corresponding to the h-th group of matching points on the [[ID=]]th CT image. gray value of the pixel point; Denotes the ratio of successful matching among the 8-neighborhood pixel points around the pixel point corresponding to the h-th group of matching points on the [[ID=]]th CT image; the th CT image and the th CT image, and the h-th group of matching points are denoted as and ; Assume that among the pixel points within the 8-neighborhood of a certain pixel point, there are n pixel points for matching, and there are b groups of pixel points that match each other. .

[0037] Furthermore, input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury. Specifically: Use the three-dimensional models of known bony Bankart injuries and normal bones as training data to train the selected machine learning algorithm; Input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

[0038] An auxiliary diagnosis system for bony Bankart injury based on image recognition, including:

[0039] A preprocessing module that preprocesses the CT images of the patient's shoulder joint collected;

[0040] A region division module that divides the preprocessed shoulder joint CT image based on threshold segmentation technology;

[0041] A marking module that performs coincidence analysis on the threshold-segmented image and the original CT image, and marks each region of the threshold segmentation on the original CT image;

[0042] An acquisition module that randomly selects the edge pixel points of the region, and based on the angular change and gray value of the connection line between the edge pixel point and the adjacent edge pixel points, obtains the possibility that the edge pixel point is a normal edge pixel point;

[0043] A judgment module, which judges whether the possibility that the edge pixel point is a normal edge pixel point is greater than a preset threshold. If it is greater, it is considered that the edge pixel point is a normal pixel point until the judgment of all edge pixel points is completed, and a clear CT image of the patient's shoulder joint is obtained;

[0044] A matching module, which performs adjacent image matching based on the obtained several CT images of the patient's shoulder joint, and fuses the image information of the overlapping parts to construct a three-dimensional model of the patient's shoulder joint;

[0045] A determination module, which inputs the three-dimensional model of the patient's shoulder joint into a trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] By preprocessing and threshold segmentation of the CT image, the present invention can accurately distinguish different tissue structures in the CT image. By overlapping and analyzing the marked areas, the accuracy and intuitiveness of the diagnosis are improved; randomly selecting edge pixel points and conducting detailed analysis, and combining the angle change and gray value to judge whether it is normal or not, reduces the interference of noise and artifacts, making the image clearer. The judgment of all edge pixel points one by one ensures the overall improvement of the image quality. Combining feature matching helps to construct a three-dimensional model, thus helping to complete the identification of Bankart injury of the patient's shoulder joint; providing strong support for the formulation of the patient's treatment plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0049] Figure 1 It is a schematic flow chart of the method for assisting in diagnosing bony Bankart injury based on image recognition of the present invention;

[0050] Figure 2 It is a schematic structural diagram of the system for assisting in diagnosing bony Bankart injury based on image recognition of the present invention;

[0051] Figure 3 It is a schematic diagram for replacing abnormal edge pixel points. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0053] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0054] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0055] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper", "lower", "horizontal", "inner", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, or the orientations or positional relationships in which the products of the invention are customarily placed during use, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the indicated devices or elements must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present invention. In addition, terms such as "first", "second", etc. are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0056] In addition, if the term "horizontal" appears, it does not mean that the component is required to be absolutely horizontal, but it can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and does not mean that the structure must be completely horizontal, but it can be slightly inclined.

[0057] In the description of the embodiments of the present invention, it should also be noted that unless otherwise clearly specified and limited, if terms such as "set", "installed", "connected", "connected" are understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0058] The following further describes the present invention in detail with reference to the accompanying drawings:

[0059] SeeFigure 1 , the present invention discloses an auxiliary diagnosis method for osseous Bankart injury based on image recognition, including:

[0060] S101, preprocess the CT image of the patient's shoulder joint collected;

[0061] Denoise and linearly enhance the collected shoulder joint CT image to reduce the noise interference in the image and improve the clarity and contrast of the image.

[0062] S102, divide the preprocessed shoulder joint CT image into regions based on threshold segmentation technology;

[0063] According to the gray values of different tissue structures in the shoulder joint CT image, select the corresponding threshold, compare each pixel in the image with the threshold to generate a binary image, and then obtain the regions of different tissues.

[0064] S103, perform coincidence analysis on the image after threshold segmentation and the original CT image, and mark each region of threshold segmentation on the original CT image;

[0065] S104, randomly select the edge pixel points of the region, and based on the angle change and gray value between the edge pixel point and the adjacent edge pixel point, obtain the possibility that the edge pixel point is a normal edge pixel point;

[0066] S104.1, randomly select the edge pixel points of the region, and obtain the gray value of the th pixel point in the th region on the th shoulder joint CT image ;

[0067] S104.2, based on the th pixel point in the th region, select its adjacent edge pixel point, connect the two pixel points, and obtain the included angle between the connection line and the horizontal direction; repeat the operation to construct the angle data sequence between the outermost edge pixel points of the th region, and then obtain the angle change trend between the adjacent edge pixel points;

[0068] The angle data sequence between the outermost edge pixel points of the th region ; Obtain the angle change trend between the adjacent edge pixel points, specifically:

[0069] ;

[0070] S104.3, based on the angle change trend of the connection lines between adjacent edge pixels and the average gray value of the pixels on the outermost edge of the th image in the th region, judge the normality performance of the th pixel on the outermost edge of the th region in the th image from the pixel itself; and then analyze from the performance of the overall outermost pixels, the probability that the th pixel on the outermost edge of the th region in the th image belongs to a normal edge pixel.

[0071] The judgment of the normality performance of the th pixel on the outermost edge of the th region in the th image is specifically as follows:

[0072] ;

[0073] ;

[0074] Among them, represents the average gray value of the pixels on the outermost edge of the th region in the th image; represents the number of the selected outermost edge pixels; when the gray difference of a certain pixel is small and the adjacent angle change is small, it indicates that the probability of it belonging to a normal edge pixel is large;

[0075] The analysis from the performance of the overall outermost pixels, the probability that the th pixel on the outermost edge of the th region in the th image belongs to a normal edge pixel is specifically as follows:

[0076] ;

[0077] ;

[0078] Among them, represents the average angle change size of the pixels on the outermost edge of the th region in the th image, represents the number of pixels with an angle difference greater than 10° between adjacent pixels on the outermost edge of the th region in the th image, The value is related to the state of the pixel points. If the pixel points are connected, then ; if these pixel points are separately distributed, then .

[0079] S105. Determine whether the possibility that the edge pixel point is a normal edge pixel point is greater than a preset threshold. If it is greater, then consider the edge pixel point as a normal pixel point; repeat S104 and S105 until the judgment of all edge pixel points is completed, and obtain a clear CT image of the patient's shoulder joint;

[0080] Set the threshold , when , it means that the pixel point belongs to a normal edge pixel point; when judging the outermost pixel points and there are abnormal edge pixel points, they will be excluded from the area, and the pixel points adjacent to them and adjacent to the previous pixel point will be selected to replace this abnormal pixel point as the area edge; recalculate whether the selected pixel points are normal edge pixel points until all pixel points are normal edge pixel points.

[0081] S106. Based on the obtained several CT images of the patient's shoulder joint, perform adjacent image matching, and fuse the image information of the overlapping parts to construct a three-dimensional model of the patient's shoulder joint;

[0082] Obtain the th CT image and the th CT image, and the reliability of the matching result of the hth group of matching points on them. If the reliability of the matching result is greater than the set threshold , it means that the matching result of the matching points on the CT image is correct; if the error rate of the matching result exceeds the preset value, then reshoot at this angle and repeat the above steps; until the matching result is correct; then fuse the image information of the overlapping parts, and use the registered and fused image information to construct a three-dimensional model of the patient's shoulder joint.

[0083] Obtain the th CT image and the th CT image, and the reliability of the matching result of the hth group of matching points on them. Specifically:

[0084] ;

[0085] ;

[0086] Among them, represents the distance between the pixel point corresponding to the hth group of matching points on the th CT image and the edge pixel point of the bone joint along the horizontal direction. If there are two distances, select the shortest distance; represents taking the The gray value of the th pixel among the nine pixels centered on the pixel corresponding to the h-th set of matching points on the CT image; the proportion of successful matching among the 8-neighborhood pixels around the pixel corresponding to the h-th set of matching points on the CT image; the h-th set of matching points on the CT image and the h-th set of matching points on the CT

[0087] image are denoted as

[0088] and

[0089] ; assuming that there are n pixels in the 8-neighborhood of a certain pixel for matching, and there are b groups of pixels that match each other, ;

[0087] S107, input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

[0088] Use the three-dimensional models of known bony Bankart injuries and normal bones as training data to train the selected machine learning algorithm; input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

[0089] See Figure 2 , the present invention discloses an auxiliary diagnosis system for bony Bankart injuries based on image recognition, including:

[0090] A preprocessing module that preprocesses the CT images of the patient's shoulder joint collected;

[0091] A region division module that divides the preprocessed shoulder joint CT image based on threshold segmentation technology;

[0092] A marking module that performs coincidence analysis on the image after threshold segmentation and the original CT image, and marks each region after threshold segmentation on the original CT image;

[0093] An acquisition module that randomly selects the edge pixels of the region and obtains the possibility that the edge pixel is a normal edge pixel based on the angular change and gray value between the edge pixel and the adjacent edge pixels;

[0094] A judgment module that judges whether the possibility that the edge pixel is a normal edge pixel is greater than a preset threshold. If it is greater, it is considered that the edge pixel is a normal pixel until the judgment of all edge pixels is completed to obtain a clear CT image of the patient's shoulder joint;

[0095] A matching module, which performs adjacent image matching based on a number of acquired CT images of the patient's shoulder joint, fuses the image information of the overlapping parts, and constructs a three-dimensional model of the patient's shoulder joint.

[0096] A determination module, which inputs the three-dimensional model of the patient's shoulder joint into a trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

[0097] Embodiment:

[0098] The present invention discloses an auxiliary diagnosis method for bony Bankart injury based on image recognition, including:

[0099] Step 1: Use a CT scanner to collect CT images of the patient's shoulder joint.

[0100] The present invention mainly performs recognition and analysis on whether there is a bony Bankart injury in the patient's shoulder joint. Through CT images, the glenoid labrum and the bone structure in the anteroinferior part of the glenoid cavity of the patient's shoulder joint can be clearly shown, which helps doctors make an early diagnosis of the patient's bone quality and helps complete the accurate recognition of bony Bankart injury and the formulation of treatment plans. The specific process is as follows:

[0101] First, ensure that the CT device is in good performance, and then use a high-resolution CT device to scan the patient's shoulder joint to obtain CT images of the patient's shoulder joint. Note: When the CT scanning machine is scanning, the patient needs to control their breathing state to minimize the influence of unstable breathing factors on the quality of the acquired CT images.

[0102] Perform denoising processing and linear enhancement on the acquired CT images of the shoulder joint to complete the preprocessing of the CT images and improve the image quality.

[0103] Step 2: Analyze based on the acquired image information to complete the three-dimensional reconstruction of the patient's shoulder joint.

[0104] Process the collected CT images of the patient's shoulder joint. When a patient has a Bankart injury of the shoulder joint, there will be pain in the shoulder joint, resulting in slight body tremors of the patient (especially adolescent patients) during the CT image scanning due to the pain, causing the collected CT images to be blurred, thus affecting the doctor's identification of the Bankart injury of the patient's shoulder joint. Therefore, analyze the CT images of the patient's shoulder joint collected by the CT scanner to eliminate the blurring effect, and then further perform three-dimensional reconstruction to help identify the bony Bankart injury of the patient's shoulder joint. Therefore, the process of helping to complete the three-dimensional reconstruction of the patient's shoulder joint is as follows:

[0105] Step 2.1: Analyze the collected CT images of the patient to complete the refined processing of the shoulder joint edge.

[0106] Analyze the collected CT images of the patient's shoulder joint. On normal CT images of the shoulder joint, the shoulder joint has a relatively clear bony structure, that is, the bony edge is clear; when the collected CT images of the shoulder joint are blurred, the edge lines of the shoulder joint on the CT images are unclear, affecting the subsequent three-dimensional CT reconstruction. Therefore, analyze the CT images after threshold segmentation. Whether it is a normal shoulder joint or a damaged shoulder joint, the edge curve of the bone is relatively smooth without being affected by blurring (even if there are bone fragments, they are individual small lines protruding, and the overall still shows a smooth state). Based on this, analyze the CT images of the patient's shoulder joint collected to complete the refined processing of the bone edge.

[0107] The specific analysis is as follows:

[0108] (1) First, use the threshold segmentation technique to process the CT images of the shoulder joint.

[0109] The set threshold for threshold segmentation should be set small. This is because the CT images of the patient's shoulder joint are blurred, which will cause the gray values of some pixel points on the edge of the shoulder joint on the CT images to be relatively dark. Therefore, setting a small threshold makes these pixel points be included in the shoulder joint area (classified), and then distinguish and process them.

[0110] (2) Based on the CT images after threshold segmentation, select the pixel points with a gray value of 1 on the image, analyze whether there are pixel points with a gray value of 1 in its 8-neighborhood, and if so, connect them. Repeat the steps until the pixel points are connected to the initial pixel point, and the area surrounded by them is defined as area 1. Repeat the above steps to complete the area division of the CT images of the shoulder joint.

[0111] Note: Here, all adjacent pixels with a gray value of 1 are divided into one area. It should be noted that there are still pixels with a gray value of 0 inside the connected area of pixels with a gray value of 1, and these pixels are divided into the adjacent area (i.e., the area covering these pixels) of the above division.

[0112] (3) Since the shoulder joint CT image is blurred, the edge part of the shoulder joint area on the image is not precise enough, and specific analysis is required. Therefore, the present invention analyzes based on the areas divided on the threshold-segmented CT image, as follows:

[0113] (3.1) First, perform coincidence analysis on the threshold-segmented image and the original CT image, and mark each area segmented by the threshold on the original CT image. Obtain the gray value of the th pixel in the th area on the th shoulder joint CT image .

[0114] (3.2) Then, analyze along the th pixel (the pixel on the edge) in the th area, select its adjacent edge pixels, connect the two pixels, and obtain the angle between the connection line and the horizontal direction. Repeat the operation to complete the calculation of the angle connection line between the outermost edge pixels of the th area, and construct a data sequence:

[0115] ;

[0116] (3.3) Calculate the angle change of the connection line between adjacent edge pixels:

[0117] ;

[0118] (3.4) The edge of the shoulder joint of a normal human body is relatively smooth, and the gray values of the pixels on the edge are relatively uniform. Therefore, judge the normality performance of the th pixel on the outermost edge of the th area of the th image from the pixel itself:

[0119] ;

[0120] ;

[0121] In the formula, represents the average gray value of the pixels on the outermost edge of the th image in the th area ( indicating the outermost side). Indicates the number of outermost edge pixel points selected.

[0122] Logic: The above formula mainly analyzes based on the difference in pixel grayscale values and the difference in angles between adjacent pixels. For the shoulder joint of the human body, its edge is relatively smooth (the bone is relatively round). As the position of the pixel points changes, its angle will change to a certain extent, but the changed angle is generally small, and the grayscale performance is relatively similar. Therefore, when the grayscale difference of a certain pixel point is small and the adjacent angle change is small, it indicates that the possibility of it belonging to a normal edge pixel point is high.

[0123] (3.5) Further, combined with the performance of the overall outer pixels for analysis, to judge the possibility that the th pixel point on the outermost edge of the th region of the th image belongs to a normal edge pixel point:

[0124] ;

[0125] ;

[0126] In the formula, represents the average change in the angle of the pixel points on the outermost edge of the th region on the image. represents the number of pixel points on the outermost edge of the th region on the image where the angle difference between adjacent pixel points is greater than 10°. If these pixel points are connected, then ; if these pixel points are separately distributed, then .

[0127] Logic: The above formula mainly uses the pixel points calculated in step (3.4) to judge its normality performance from the pixel points themselves and combines the overall pixel points for analysis and calculation. When the angle differences between the overall edge pixel points are all small, combined with the individual pixel points, it indicates that the shoulder joint edge is relatively smooth. If there is a slightly large and continuous angle difference between adjacent pixel points, it may be a problem caused by a joint crack, and it is recognized as a potential pixel point if it is considered normal; but when these pixel points are not continuously distributed and are scattered, they are determined to be abnormal edge pixel points.

[0128] (3.6) Set a threshold , when the calculated If so, it indicates that the pixel belongs to a normal edge pixel. Repeat the above steps to complete the judgment of the outermost pixels. When there are abnormal edge pixels, they will be excluded from the region, and the pixel adjacent to it and adjacent to its previous pixel will be selected to replace this abnormal pixel as the region edge. As Figure 3 shown.

[0129] (3.7) Recalculate whether the newly selected pixels are normal edge pixels until all pixels are normal edge pixels. Here, 5 layers of pixels are selected to complete the calculation.

[0130] So far, the processing of the bone edge on the CT image of the patient's shoulder joint has been completed.

[0131] Step 2.2, based on the above-precised CT image of the patient, further construct a three-dimensional model of the patient's shoulder joint.

[0132] The present invention has completed the processing of the bone edge on the CT image of the shoulder joint, making the edge of the shoulder joint on the CT image clearer and more reliable. Then, multiple two-dimensional shoulder joint CT images are collected for three-dimensional reconstruction. There needs to be a certain overlapping area between these multiple CT images. Based on the specific information shown in the above-processed CT image, adjacent images are subjected to region matching, and then a smooth and continuous three-dimensional model is generated to help analyze whether there is a bony Bankart injury in the shoulder joint in the follow-up. The specific analysis is as follows:

[0133] First, based on Step 2.1, the CT images of the patient collected at different angles are processed to eliminate the influence of ambiguity. Then, the feature matching algorithm is used to perform feature matching on the collected adjacent images. Here, the th CT image and the th CT image, and the rd group of matching points are denoted as and .

[0134] Then analyze whether the matching result of the th CT image and the th CT image on the rd group of matching points is reliable. Calculate the matching reliability of the th CT image and the th CT image on the rd group of matching points:

[0135] ;

[0136] ;

[0137] In the formula, represents the The distance between the pixel points corresponding to the th group of matching points on the th CT image and the pixel points on the edge of the bone joint along the horizontal direction (if there are two distances, select the shortest distance). The grayscale value of the th pixel point among the nine pixel points centered on the pixel points corresponding to the th group of matching points on the th CT image.

[0138] Note: Assume that among the pixel points in the 8-neighborhood of a certain pixel point, n pixel points are matched, and there are b groups of pixel points that are mutually matched, .

[0139] Logic: Since on the shoulder joint CT image, the pixel points in the bone area are all relatively similar, there may be cases of incorrect matching. Therefore, based on the invariance of the position relationship of the matching points on the two CT images and combined with the invariance of their gray colors for analysis. When a group of matching points show the same distance from the bone edge on the two images, and the surrounding pixel points have the same color and the matching groups are the same, it indicates that the matching result of the pixel point is correct.

[0140] Set the threshold , when the calculated , it indicates that the matching result of the matching points on the CT image is correct. When the number of incorrect matching groups exceeds 5%, the image is reshot at this angle, and the above steps are repeated.

[0141] Based on the results of the above analysis, using the existing technology, the image information of the overlapping part is fused, and using the registered and fused image information, a complete three-dimensional model can be constructed.

[0142] Thus, the three-dimensional reconstruction of the patient's shoulder joint CT image is completed.

[0143] Step 3, based on the above three-dimensional reconstruction of the patient's shoulder joint CT image, help complete the identification of the bony Bankart injury of the patient's shoulder joint.

[0144] In the above process, the present invention has completed the three-dimensional reconstruction of the patient's shoulder joint CT image and generated a three-dimensional model of the patient's shoulder joint. Further, based on the three-dimensional model of the patient's shoulder joint, existing machine learning models can be used for identification and analysis. The specific process is as follows:

[0145] A three-dimensional model of a known osseous Bankart lesion and normal bone is used as training data to train a selected machine learning algorithm. The generated three-dimensional model of the patient's shoulder joint is input into the trained machine learning model for analysis and diagnosis to determine the presence of an osseous Bankart lesion, as well as the degree and location of the lesion.

[0146] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An auxiliary diagnosis method for bony Bankart injury based on image recognition, characterized in that, Including: Step 1: Preprocess the CT images of the patient's shoulder joint collected; Step 2: Based on the threshold segmentation technology, divide the regions of the preprocessed shoulder joint CT images; Step 3: Perform coincidence analysis on the images after threshold segmentation and the original CT images, and mark each region of the threshold segmentation on the original CT images; Step 4: Randomly select the edge pixel points of the region, and based on the angular change and gray value between the edge pixel point and the adjacent edge pixel points, obtain the possibility that the edge pixel point is a normal edge pixel point; Randomly select the edge pixels of the region to obtain the gray value of the th pixel in the th region on the th shoulder joint CT image; Based on the th pixel in the th region, select its adjacent edge pixels, connect the two pixels, and obtain the angle between the connection line and the horizontal direction; Repeat the operation to construct the angle data sequence between the outermost edge pixels of the th region, and then obtain the angle change trend of the connection lines between adjacent edge pixels; Based on the angle change trend of the connection lines between adjacent edge pixels and the average gray value of the pixels on the outermost edge of the th image in the th region, judge the normality performance of the th pixel on the outermost edge of the th region of the th image from the pixel itself; Further analyze from the performance of the overall outer pixels, the probability that the th pixel on the outermost edge of the th region of the th image belongs to a normal edge pixel; The angle data sequence between the outermost edge pixel points of the th region; obtaining the angle change trend of the line connecting adjacent edge pixel points, specifically: ； The determination of the normalcy performance of the outermost edge of the pixel points on the outermost edge of the nth region of the nth image, specifically: ； ； Among them, represents the average gray value of the pixel points on the outermost edge of the th region on the th image; represents the number of selected outermost edge pixel points; when the gray value difference of a certain pixel point is small and the adjacent angle change is small, it indicates that the possibility of it belonging to a normal edge pixel point is high; Analyzing from the performance of the overall outer pixel points, for the th image, the probability that the th pixel point on the outermost edge of the th region belongs to a normal edge pixel point is specifically as follows: ； ； Among them, represents the average angular change magnitude of the pixel points on the outermost edge of the th region in the th image. represents the number of pixel points with an angular difference greater than 10° between adjacent pixel points on the outermost edge of the th region in the th image. The value of is related to the state of the pixel points. If the pixel points are connected, then ; if these pixel points are separately distributed, then Step 5: Judge whether the possibility that the edge pixel point is a normal edge pixel point is greater than the preset threshold. If it is greater, then consider the edge pixel point as a normal pixel point; repeat Step 4 and Step 5 until the judgment of all edge pixel points is completed to obtain clear CT images of the patient's shoulder joint; Step 6: Based on the obtained several CT images of the patient's shoulder joint, perform adjacent image matching, and fuse the image information of the overlapping parts to construct a three-dimensional model of the patient's shoulder joint; Step 7: Input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

2. The method for assisting in the diagnosis of bony Bankart injury based on image recognition according to claim 1, wherein The preprocessing of the CT images of the patient's shoulder joint collected specifically is: perform denoising processing and linear enhancement on the collected shoulder joint CT images to reduce the noise interference in the images and improve the clarity and contrast of the images.

3. The method for auxiliary diagnosis of bony Bankart injury based on image recognition according to claim 2, characterized in that, The region division of the preprocessed shoulder joint CT images based on the threshold segmentation technology specifically is: according to the gray values of different tissue structures in the shoulder joint CT images, select the corresponding thresholds, compare each pixel in the image with the thresholds to generate a binary image, and then obtain the regions of different tissues.

4. The method for auxiliary diagnosis of bony Bankart injury based on image recognition according to claim 3, wherein The judgment of whether the possibility that the edge pixel point is a normal edge pixel point is greater than the preset threshold. If it is greater, then consider the edge pixel point as a normal pixel point; until the judgment of all edge pixel points is completed to obtain clear CT images of the patient's shoulder joint specifically is: Set a threshold , when , it means that the pixel belongs to a normal edge pixel; when judging the outermost pixel, if there is an abnormal edge pixel, it will be excluded from the region, and the pixel adjacent to it and adjacent to its previous pixel will be selected to replace this abnormal pixel as the region edge; recalculate whether the selected pixel is a normal edge pixel until all pixels are normal edge pixels.

5. The method for assisting in the diagnosis of bony Bankart injury based on image recognition according to claim 4, wherein Based on the obtained several CT images of the patient's shoulder joint, adjacent image matching is performed, and the image information of the overlapping part is fused to construct a three-dimensional model of the patient's shoulder joint. Specifically: Obtain the th CT image and the th CT image, and the reliability of the matching results of the hth group of matching points on the CT images. If the reliability of the matching results is greater than the set threshold , it indicates that the matching results of the matching points on the CT images are correct; if the error rate of the matching results exceeds the preset value, then this angle is photographed again, and the above steps are repeated; until the matching results are correct; then the image information of the overlapping part is fused, and a three-dimensional model of the patient's shoulder joint is constructed using the registered and fused image information.

6. The method for auxiliary diagnosis of bony Bankart injury based on image recognition according to claim 5, wherein The reliability of the matching result of the h-th group of matching points on the th CT image and the th CT image is as follows: ； ； Among them, represents the distance between the pixel point corresponding to the h-th group of matching points on the -th CT image and the pixel point on the edge of the bone joint along the horizontal direction. If there are two distances, the shortest distance is selected; represents the gray value of the -th pixel point among the nine pixel points centered on the pixel point corresponding to the h-th group of matching points on the -th CT image; represents the proportion of successful matching among the 8-neighborhood pixel points around the pixel point corresponding to the h-th group of matching points on the -th CT image; The h-th group of matching points on the -th CT image and the -th CT image are denoted as and ; Assume that there are n pixel points in the 8-neighborhood of a certain pixel point for matching, and there are b groups of pixel points that are mutually matched, .

7. The method for auxiliary diagnosis of bony Bankart injury based on image recognition according to claim 6, wherein The input of the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury specifically is: use the three-dimensional models of known bony Bankart injuries and normal bones as training data to train the selected machine learning algorithm; input the three-dimensional model of the patient's shoulder joint into the trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

8. An auxiliary diagnosis system for osseous Bankart injury based on image recognition, characterized in that, Including: A preprocessing module, which preprocesses the CT images of the patient's shoulder joint collected; A region division module, which divides the regions of the preprocessed shoulder joint CT images based on the threshold segmentation technology; A marking module, which performs coincidence analysis on the images after threshold segmentation and the original CT images, and marks each region of the threshold segmentation on the original CT images; An acquisition module that randomly selects edge pixels of a region and obtains the possibility that the edge pixel is a normal edge pixel based on the angular change and gray value of the connection line between the edge pixel and adjacent edge pixels; Randomly select the edge pixels of the region to obtain the gray value of the th pixel in the th region on the th shoulder joint CT image; Based on the th pixel in the th region, select its adjacent edge pixels, connect the two pixels, and obtain the angle between the connecting line and the horizontal direction; Repeat the operation to construct the angle data sequence between the outermost edge pixels of the th region, and then obtain the angle change trend of the connecting lines between adjacent edge pixels; Based on the angle change trend of the connecting lines between adjacent edge pixels and the average value of the gray values of the pixels on the outermost edge of the th image in the th region, judge the normality of the th pixel on the outermost edge of the th region of the th image from the pixel itself; Then analyze from the performance of the overall outer pixels, the probability that the th pixel on the outermost edge of the th region of the th image belongs to a normal edge pixel; The angle data sequence between the outermost edge pixels of the th region; obtaining the angle change trend of the connection line between adjacent edge pixels, specifically: ； The determination of the normalcy performance of the outermost pixel points on the outermost edge of the th area of the th image, specifically: ； ； Among them, represents the average gray value of the pixel points on the outermost edge of the th region on the th image; represents the number of selected outermost edge pixel points; when the gray value difference of a certain pixel point is small and the adjacent angle change is small, it indicates that the possibility of it belonging to a normal edge pixel point is high; Analyzing from the performance of the overall pixel points on the outside, for the th image, the th pixel point on the outermost edge of the th area has a certain probability of being a normal edge pixel point, specifically: ； ； Among them, represents the average angular change magnitude of the pixel points on the outermost edge of the th region in the th image. represents the number of pixel points with an angular difference greater than 10° between adjacent pixel points on the outermost edge of the th region in the th image. The value of is related to the state of the pixel points. If the pixel points are connected, then ; if these pixel points are separately distributed, then A judgment module that judges whether the possibility that the edge pixel is a normal edge pixel is greater than a preset threshold. If it is greater, the edge pixel is considered a normal pixel until the judgment of all edge pixels is completed to obtain a clear CT image of the patient's shoulder joint; A matching module that performs adjacent image matching based on the obtained several CT images of the patient's shoulder joint and fuses the image information of the overlapping parts to construct a three-dimensional model of the patient's shoulder joint; A determination module that inputs the three-dimensional model of the patient's shoulder joint into a trained machine learning model for analysis and diagnosis to determine whether there is a bony Bankart injury and the degree and location of the injury.

Images