Lower limb fracture image high-precision segmentation method and system based on artificial intelligence

Abstract

The invention relates to the technical field of image processing, in particular to a lower limb fracture image high-precision segmentation method and system based on artificial intelligence, and the method comprises the steps: obtaining a plurality of initial extension bone lines in a leg connected domain; according to the distance between each initial extension bone line and the comparison edge, obtaining the texture degree of each initial extension bone line, screening out a plurality of initial extension bone lines, and obtaining a bone connected domain in the leg connected domain; according to the difference between the gradient difference of the central pixel point and the gradient difference of the to-be-grown pixel points, obtaining a growth coefficient between the central pixel point and each to-be-grown pixel point in the corresponding eight neighborhoods; growing to obtain a plurality of growth lines; according to the unevenness and the included angle of each growth line and the minimum distance between the growth line and the edge line of the bone connected domain, all the bone fracture lines are obtained, and observation and treatment of a doctor are assisted through all the bone fracture lines. The accuracy of high-precision segmentation of the lower limb fracture is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a high-precision segmentation method and system for lower limb fracture images based on artificial intelligence. Background Art

[0002] High-precision segmentation of lower limb fracture images is an important task in medical image analysis, especially significant for the diagnosis of fractures and the formulation of treatment plans. The goal of this task is to accurately segment the fracture area from the image data to help doctors judge the fracture type, severity, and treatment plan.

[0003] In order to perform high-precision segmentation on the lower limb fracture images of patients, the lower limb X-ray images of patients are collected for fracture recognition. However, due to the incomplete uniformity of tissues in different parts of the lower limb during the X-ray image collection process, the contrast in different parts of the image is different, resulting in unclear display of the bones of some patients, thus affecting the recognition of the fracture site and reducing the accuracy of high-precision segmentation of lower limb fractures. Summary of the Invention

[0004] The present invention provides a high-precision segmentation method and system for lower limb fracture images based on artificial intelligence to solve the existing problems.

[0005] The high-precision segmentation method and system for lower limb fracture images based on artificial intelligence of the present invention adopt the following technical solutions: An embodiment of the present invention provides a high-precision segmentation method for lower limb fracture images based on artificial intelligence, and the method includes the following steps: Obtain the lower limb fracture images of the patient; Segment the lower limb fracture images to obtain the leg connected domain; connect adjacent pixel points with similar gradient differences within the leg connected domain to obtain several initial bone lines; extend all the initial bone lines to obtain several initial extended bone lines, and initially screen out several initial extended bone lines according to the intersection quantity after extension; Obtain the texture degree of each initial extended bone line according to the distance between each initial extended bone line after initial screening and the corresponding comparison edge in the leg connected domain; re-screen out several initial extended bone lines through the average value of the texture degrees of each initial extended bone line and similar initial extended bone lines; Use the endpoints of each initial extended bone line after re-screening as the initial seed points, and grow according to the gradient differences between adjacent pixel points to obtain the bone connected domain in the leg connected domain; Obtain the in-bone pixels and out-of-bone pixels in the eight-neighborhood of the edge pixels of the bone connected region. Through the in-bone pixels and out-of-bone pixels in the eight-neighborhood of the edge pixels of the bone connected region, obtain the in-bone edge points and out-of-bone edge points; determine the break points based on the number of in-bone edge points and out-of-bone edge points in the eight-neighborhood of the edge pixels of the bone connected region, and use the break points as seed points; record the in-bone pixels in the eight-neighborhood of each central pixel as the to-be-grown pixels of each central pixel, and obtain the growth coefficient between the central pixel and each to-be-grown pixel in the corresponding eight-neighborhood according to the difference between the gradient difference of the central pixel and the gradient difference of the to-be-grown pixel; grow through the seed points and the growth coefficient to obtain several growth lines; Obtain the bone fracture coefficient of each growth line according to the unevenness, included angle and the minimum distance from the edge line of the bone connected region of each growth line, obtain all the bone fracture lines through the bone fracture coefficient, and assist doctors in observation and treatment through all the bone fracture lines.

[0006] Furthermore, the segmentation of the lower limb fracture image to obtain the leg connected region includes the following specific steps: Segment the lower limb fracture image through the semantic segmentation algorithm in the neural network to obtain the leg connected region.

[0007] Furthermore, the connection of adjacent pixel points with similar gradient differences in the leg connected region to obtain several initial bone lines includes the following specific steps: Obtain the gradient values of all pixel points inside the leg connected region of the patient through the canny edge detection algorithm, obtain the eight-neighborhood pixel points of each pixel point in the leg connected region, record the pixel point with the largest gradient value among the eight-neighborhood pixel points as the first pixel point of each pixel point, and record the pixel point in the eight-neighborhood with the gradient value closest to the gradient value between the eight-neighborhood pixel point and the central pixel point as the second pixel point of each pixel point; when the first pixel point and the second pixel point are the same pixel point, record the pixel point in the eight-neighborhood as the third pixel point; Connect all adjacent third pixel points to obtain several initial bone lines.

[0008] Furthermore, the extension of all the initial bone lines to obtain several initial extended bone lines and the initial screening of several initial extended bone lines according to the number of intersections after extension include the following specific steps: Obtain the tangent directions of the endpoints at both ends of each initial bone line. Along the tangent directions of the two endpoints, extend each initial bone line by half on each side to obtain the extended initial bone line, and denote the extended initial bone line as the initial extended bone line; obtain the number of initial extended bone lines that each initial extended bone line intersects with all other initial extended bone lines, and denote it as the intersection number of each initial extended bone line; perform linear normalization on the intersection numbers of all initial extended bone lines, and screen out the initial extended bone lines whose normalized intersection number is greater than or equal to the first preset threshold and screen out and retain the initial extended bone lines whose normalized intersection number is less than the first preset threshold to obtain several initial extended bone lines after the first screening.

[0009] Further, obtain the texture degree of each initial extended bone line according to the distance between each initial extended bone line after the first screening and the comparison edge corresponding to the connected region of the leg; screen out several initial extended bone lines again through the average value of the texture degrees of each initial extended bone line and the adjacent initial extended bone lines, and the specific steps are as follows: Obtain the points on the edge line of the leg connected region that are closest to the endpoints at both ends of each initial extended bone line, and denote them as reference points. Connect the two reference points to form a line, which is denoted as the comparison edge of each initial extended bone line; The calculation formula for the texture degree of each initial extended bone line is: ; In the formula, represents the information entropy of the minimum distance between all initial extended bone lines and the corresponding comparison edge, represents the th initial extended bone line and the corresponding comparison edge between the distance, represents the exponential function with the natural constant as the base, represents the th initial extended bone line's texture degree; Obtain the tangent directions of the two endpoints of each initial extended bone line. Along the tangent directions, obtain the two initial extended bone lines that are closest to the two endpoints, and denote them as the collinear bone lines of each initial extended bone line; calculate the average value of the texture degrees of each initial extended bone line and the collinear bone lines, and denote it as the texture average value of each initial extended bone line. Screen out the initial extended bone lines whose texture average value is greater than or equal to the second preset threshold and retain the initial extended bone lines whose texture average value is less than the second preset threshold to obtain several initial extended bone lines after the second screening.

[0010] Further, taking the endpoints of each initial extended bone line after re-screening as initial seed points, and growing according to the gradient difference between adjacent pixel points to obtain the bone connected region in the leg connected region, the specific steps are as follows: Obtain the eight-neighborhood of each endpoint on each initial extended bone line, record any pixel point in the eight-neighborhood of each endpoint as a reference pixel point, connect each endpoint with the reference pixel point to obtain a straight line between each endpoint and the reference point, record the direction perpendicular to the direction of the straight line as the perpendicular direction of the reference point, obtain the pixel points adjacent to the reference point on both sides in the corresponding perpendicular direction, record them as the two gradient points of the reference point, calculate the absolute value of the difference between the two gradient points, and record it as the gradient difference of the reference point; obtain the pixel points adjacent to the endpoint on both sides in the corresponding perpendicular direction, record them as the two gradient points of the endpoint, calculate the absolute value of the difference between the gray values of the two gradient points, and record it as the gradient difference of the endpoint; record the difference between the gradient difference of the reference point and the gradient difference of the endpoint as the gradient difference between each endpoint and the reference point; Taking all the endpoints on the initial extended bone line as initial seed points, grow according to the principle of minimizing the gradient difference between each endpoint and all pixel points in the eight-neighborhood, and record the grown region as the bone connected region in the leg connected region.

[0011] Further, obtain the in-bone pixel points and out-of-bone pixel points in the eight-neighborhood of the edge pixel points of the bone connected region, and obtain the in-bone edge points and out-of-bone edge points through the in-bone pixel points and out-of-bone pixel points in the eight-neighborhood of the edge pixel points of the bone connected region; determine the break points through the number of in-bone edge points and out-of-bone edge points in the eight-neighborhood of the edge pixel points of the bone connected region, the specific steps are as follows: Record the pixel points in the eight-neighborhood of the edge pixel points that are within the bone connected region as in-bone pixel points, and record the pixel points in the eight-neighborhood of the edge pixel points that are outside the bone connected region as out-of-bone pixel points; Calculate the average value of the gray values of all in-bone pixel points in the eight-neighborhood of each edge pixel point of the bone connected region, and record it as the in-bone gray average value of each edge pixel point of the bone connected region; calculate the average value of the gray values of all out-of-bone pixel points in the eight-neighborhood of each edge pixel point of the bone connected region, and record it as the out-of-bone gray average value of each edge pixel point of the bone connected region; Compare the difference in the gray value of each edge pixel point of the bone connected region with the in-bone gray average value and the out-of-bone gray average value. When the difference between the gray value of each edge pixel point of the bone connected region and the in-bone gray average value is less than or equal to the difference between the gray value of the edge pixel point and the out-of-bone gray average value, record the edge pixel point of the bone connected region as an in-bone edge point; otherwise, record the edge pixel point of the bone connected region as an out-of-bone edge point; Any edge pixel of a bone connected region that contains both an inner bone edge point and an outer bone edge point in the eight-neighborhood of an arbitrary pixel is denoted as a break point.

[0012] Furthermore, the growth coefficient between the central pixel and each to-be-grown pixel in the corresponding eight-neighborhood is obtained based on the difference between the gradient difference of the central pixel and the gradient difference of the to-be-grown pixel; growth is performed through the seed point and the growth coefficient to obtain a number of growth lines, and the specific steps are as follows: The growth coefficient between the central pixel and the th to-be-grown pixel in the corresponding eight-neighborhood is represented by the formula: ; In the formula, represents the gradient difference of the central pixel during the growth process, represents the gradient difference of the th to-be-grown pixel in the eight-neighborhood of the central point, is the absolute value symbol, represents the exponential function with the natural constant as the base, represents the growth coefficient between the central pixel and the th to-be-grown pixel in the corresponding eight-neighborhood; Growth is performed according to the seed point and the growth coefficient between two pixels, and pixels with a growth coefficient greater than or equal to the third preset threshold are grown until the growth ends, obtaining a number of growth lines.

[0013] Furthermore, the bone fracture coefficient of each growth line is obtained based on the unevenness, included angle, and minimum distance from the edge line of the bone connected region of each growth line. All bone fracture lines are obtained through the bone fracture coefficient, and all bone fracture lines assist doctors in observation and treatment. The specific steps are as follows: Obtain the edge chain code of the growth line, and record the average value of the differences between adjacent code numbers in the edge chain code of each growth line as the unevenness of each growth line; Obtain the midpoint on each growth line, obtain the included angle between the two endpoints and the midpoint straight line, and record it as the included angle of each growth line; obtain the minimum distance between the farthest point from the edge line of the bone connected region on each growth line and the edge line of the bone connected region, and record it as the minimum distance between each growth line and the edge line of the bone connected region; The bone fracture coefficient of each growth line is represented by the formula: ; In the formula, represents the th unevenness of the growth line, represents the The included angle of the growth lines denotes the minimum distance between the nth growth line and the edge line of the bone connection domain; denotes the linear normalization function, denotes the bone fracture coefficient of the nth growth line; Growth lines with a bone fracture coefficient greater than the fourth preset threshold are recorded as fracture lines. The fracture lines in the lower limb fracture image are marked, and the marked fracture line results assist doctors in observing and treating the patient's lower limb fracture condition.

[0014] The present invention also provides a high-precision segmentation system for lower limb fracture images based on artificial intelligence, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned high-precision segmentation methods for lower limb fracture images based on artificial intelligence are implemented.

[0015] The beneficial effects of the technical solution of the present invention are as follows: obtaining a number of initial extended bone lines within the leg connection domain; obtaining the texture degree of each initial extended bone line according to the distance between each initial extended bone line and the comparison edge, screening out a number of initial extended bone lines through the texture degree, and obtaining the bone connection domain in the leg connection domain; obtaining the growth coefficient between the central pixel point and each to-be-grown pixel point in the corresponding eight-neighborhood according to the difference between the gradient difference of the central pixel point and the gradient difference of the to-be-grown pixel point; and performing growth to obtain a number of growth lines, reducing the influence of the incomplete uniformity of tissues in different parts of the lower limb on the contrast in different parts of the image; obtaining all the fracture lines according to the unevenness, included angle of each growth line, and the minimum distance between the growth line and the edge line of the bone connection domain, and assisting doctors in observing and treating through all the fracture lines. The present invention improves the accuracy of high-precision segmentation of lower limb fractures. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 is the flowchart of the steps of the high-precision segmentation method for lower limb fracture images based on artificial intelligence of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following specifically describes, in conjunction with the accompanying drawings and preferred embodiments, the method and system for high-precision segmentation of lower limb fracture images based on artificial intelligence according to the present invention, including its specific implementation manners, structures, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0020] The following specifically describes the specific solutions of the method and system for high-precision segmentation of lower limb fracture images based on artificial intelligence provided by the present invention in conjunction with the accompanying drawings.

[0021] Please refer to Figure 1 , which shows the flowchart of the steps of the method for high-precision segmentation of lower limb fracture images based on artificial intelligence provided by one embodiment of the present invention. The method includes the following steps: Step S001: Collect lower limb fracture images of the patient.

[0022] It should be noted that since high-precision segmentation is performed on the lower limb fracture images of the patient, it is necessary to collect lower limb fracture images under X-ray for analysis.

[0023] Specifically, use X-ray to collect lower limb fracture images of the patient's leg. Among them, the lower limb fracture image is a grayscale image.

[0024] Thus, the lower limb fracture image is obtained.

[0025] Step S002: Segment the lower limb fracture image to obtain the leg connected domain; obtain several initial extended bone lines in the leg connected domain; obtain the texture degree of each initial extended bone line according to the distance between each initial extended bone line and the corresponding comparison edge; re-screen several initial extended bone lines through the average value of the texture degrees of each initial extended bone line and the adjacent initial extended bone lines; use the endpoints of each re-screened initial extended bone line as the initial seed points, and grow according to the gradient difference between adjacent pixel points to obtain the bone connected domain in the leg connected domain.

[0026] It should be noted that since there is a large gray level difference between the patient's leg area and the background area outside the leg, the patient's leg area can be preliminarily screened, and the specific information in the leg area can be analyzed.

[0027] Specifically, the lower limb fracture image is segmented by the semantic segmentation algorithm in the neural network to obtain the connected region of the patient's leg; among them, the loss function in the semantic segmentation algorithm is the cross-entropy loss function. Among them, the semantic segmentation algorithm is a well-known technology and will not be specifically described here.

[0028] Thus, the connected region of the patient's leg is obtained.

[0029] It should be noted that since the connected region of the patient's leg contains the patient's body tissues and bone parts, due to the penetrability of different parts of the body tissues, some bones are more obvious in the connected region, while the contrast between some bone parts and the body tissues is relatively low, resulting in the bones not being well distinguished; this may cause tissue pixels to be regarded as bone pixels when obtaining bone pixel points through the edge detection algorithm. Therefore, the lines detected by the edge are initially screened.

[0030] Furthermore, it should be noted that when segmenting bone pixel points, the gradient of their pixel points is relatively large. Therefore, when analyzing that the pixel point with the closest and largest gradient value to the corresponding pixel points in the eight-neighborhood of each pixel point is the same pixel point, the greater the possibility that the pixel point is a bone pixel point.

[0031] Specifically, the gradient values of all pixel points inside the connected region of the patient's leg are obtained through the canny edge detection algorithm, the eight-neighborhood pixel points of each pixel point in the connected region of the leg are obtained, the pixel point with the largest gradient value among the eight-neighborhood pixel points is recorded as the first pixel point of each pixel point, and the pixel point in the eight-neighborhood with the closest gradient value to the central pixel point is recorded as the second pixel point of each pixel point; when the first pixel point and the second pixel point are the same pixel point, the pixel points in the eight-neighborhood are recorded as the third pixel points. Among them, the canny edge detection algorithm is a well-known technology and will not be specifically described here.

[0032] All adjacent third pixel points are connected to obtain several initial bone lines.

[0033] It should be noted that since the lines of bones generally have regular texture changes, while other lines are relatively messy, the initial bone lines can be extended to check their intersection situations. When each bone line intersects with other bone lines more, it indicates that the bone line is relatively messy and may be a line caused by tissue interference. Therefore, it needs to be excluded.

[0034] Specifically, obtain the tangent directions of the endpoints at both ends of each initial bone line. Along the tangent directions of the two endpoints, extend each initial bone line by half on each side to obtain the extended initial bone line, and denote the extended initial bone line as the initial extended bone line. Obtain the number of initial extended bone lines that each initial extended bone line intersects with all other initial extended bone lines, and denote it as the intersection number of each initial extended bone line. Perform linear normalization on the intersection numbers of all initial extended bone lines, and screen out the initial extended bone lines whose normalized intersection numbers are greater than or equal to the first preset threshold and screen out and retain the initial extended bone lines whose normalized intersection numbers are less than the first preset threshold to obtain several initial extended bone lines after the first screening. Among them, in this embodiment, the first preset threshold is described as an example. In this embodiment is not specifically limited, and the implementer can determine it according to the specific situation.

[0035] It should be noted that the more consistent the minimum distances between all initial extended bone lines and the edge line of the leg connection region are, and the smaller the distances between each initial extended bone line and the edge line of the leg connection region are, the higher the texture degree of the initial extended bone line indicates.

[0036] Specifically, obtain the points on the leg connection region edge line that are closest to the endpoints at both ends of each initial extended bone line, and denote them as reference points. Connect the two reference points to form a line, which is denoted as the comparison edge of each initial extended bone line.

[0037] According to the distance between each initial extended bone line and the corresponding comparison edge, obtain the texture degree of each initial extended bone line, which is specifically expressed by the formula:[[]] ; In the formula,[[]] represents the information entropy of the minimum distances between all initial extended bone lines and the corresponding comparison edges,[[]] represents the th initial extended bone line and the corresponding comparison edge the distance between,[[]] represents the exponential function with the natural constant as the base,[[]] represents the th initial extended bone line's texture degree. Among them, the information entropy and the distance are both well-known technologies and will not be specifically elaborated here.

[0038] Among them, when the minimum distances between all the initial extended bone lines and the edge line of the leg connection region are more consistent, that is, the smaller the information entropy of the minimum distances between all the initial extended bone lines and the corresponding comparison edges, it indicates that the texture degree of the initial extended bone lines is greater; when the distance between each initial extended bone line and the edge line of the leg connection region is smaller, it indicates that the texture degree of this initial extended bone line is greater.

[0039] It should be noted that due to the layering of the patient's tissue in the lower limb fracture image of the patient, the leg of the patient will show gray-scale layering, so that an original initial extended bone line is divided into several initial extended bone lines. Therefore, the mean value of the texture degrees of adjacent initial extended bone lines of each initial extended bone line is obtained to determine whether this initial extended bone line needs to be screened out.

[0040] Furthermore, it should be noted that since the initial extended bone lines not affected by interference have relatively high texture degrees, it is necessary to exclude the initial extended bone lines with lower texture degrees caused by interference.

[0041] Specifically, obtain the tangent directions of the two endpoints of each initial extended bone line, and along the tangent directions, obtain the two initial extended bone lines closest to the two endpoints, and record them as the collinear bone lines of each initial extended bone line; calculate the mean value of the texture degrees of each initial extended bone line and its collinear bone lines, and record it as the texture mean value of each initial extended bone line. Screen out the initial extended bone lines with texture mean values greater than or equal to the second preset threshold , and retain the initial extended bone lines with texture mean values less than the second preset threshold to obtain several initial extended bone lines after re-screening. Among them, in this embodiment, the second preset threshold is described by way of example, and in this embodiment is not specifically limited, and the implementer can determine it according to the specific situation.

[0042] It should be noted that due to the different penetration degrees of different parts of the leg, the contrast of the image is uneven, resulting in incomplete initial extended bone lines. Therefore, it is necessary to grow complete bone edge lines and obtain the bone connection region.

[0043] Specifically, obtain the eight-neighborhood of each endpoint on each initial extended bone line. Denote any pixel point in the eight-neighborhood of each endpoint as a reference pixel point. Connect each endpoint to the reference pixel point to obtain a straight line between each endpoint and the reference point. Denote the direction perpendicular to the direction of the straight line as the perpendicular direction of the reference point. Obtain the pixel points adjacent to the reference point on both sides in the corresponding perpendicular direction, and denote them as the two gradient points of the reference point. Calculate the absolute value of the difference between the two gradient points, and denote it as the gradient difference of the reference point. Obtain the pixel points adjacent to the endpoint on both sides in the corresponding perpendicular direction, and denote them as the two gradient points of the endpoint. Calculate the absolute value of the difference between the two gradient points, and denote it as the gradient difference of the endpoint. Denote the difference between the gradient difference of the reference point and the gradient difference of the endpoint as the gradient difference between each endpoint and the reference point.

[0044] Through the above operations, obtain the gradient differences between each endpoint and all pixel points in all eight-neighborhoods. Take all endpoints on the initial extended bone line as initial seed points, and grow according to the principle of minimizing the gradient differences between each endpoint and all pixel points in all eight-neighborhoods. Denote the grown region as the bone connected region in the leg connected region.

[0045] So far, the bone connected region in the leg connected region is obtained.

[0046] Step S003: Obtain the growth coefficient between the central pixel point and each to-be-grown pixel point in the corresponding eight-neighborhood according to the difference between the gradient difference of the central pixel point and the gradient difference of the to-be-grown pixel point; and perform growth to obtain several growth lines. Obtain the bone fracture coefficient of each growth line according to the unevenness, included angle of each growth line, and the minimum distance from the edge line of the bone connected region.

[0047] It should be noted that since the fracture part will have a lower bone density due to cracks, resulting in less X-ray absorption in the fracture part, and thus the gray value of the fracture part will decrease. Therefore, it is necessary to analyze the decrease in gray value in the fracture edge.

[0048] Specifically, obtain the edge pixel points of the bone connected region. Denote the pixel points in the eight-neighborhood of the edge pixel points within the bone connected region as the inner-bone pixel points, and denote the pixel points in the eight-neighborhood of the edge pixel points outside the bone connected region as the outer-bone pixel points.

[0049] Calculate the mean gray value of all the in-bone pixel points in the eight-neighborhood of each edge pixel point of the bone connected region, and denote it as the in-bone gray mean of each edge pixel point of the bone connected region; calculate the mean gray value of all the out-of-bone pixel points in the eight-neighborhood of each edge pixel point of the bone connected region, and denote it as the out-of-bone gray mean of each edge pixel point of the bone connected region. Compare the difference between the gray value of each edge pixel point of the bone connected region and the in-bone gray mean and the out-of-bone gray mean. When the difference between the gray value of each edge pixel point of the bone connected region and the in-bone gray mean is less than or equal to the difference between the gray value of the edge pixel point and the out-of-bone gray mean, then mark the edge pixel point of the bone connected region as an in-bone edge point; otherwise, mark the edge pixel point of the bone connected region as an out-of-bone edge point.

[0050] Mark any edge pixel point of the bone connected region that contains an in-bone edge point and an out-of-bone edge point in the eight-neighborhood of any pixel point as a break point.

[0051] Mark the in-bone pixel points in the eight-neighborhood of each center pixel point as the to-be-grown pixel points of each center pixel point.

[0052] Take the break point as the seed point for growth, and perform growth through the gradient difference between adjacent pixel points; among them, the growth coefficient between the center pixel point and each to-be-grown pixel point in the corresponding eight-neighborhood is expressed by the formula: ; In the formula, represents the gradient difference of the center pixel point during the growth process, represents the gradient difference of the th to-be-grown pixel point in the eight-neighborhood of the center point, is the absolute value symbol, represents the exponential function with the natural constant as the base, represents the growth coefficient between the center pixel point and the th to-be-grown pixel point in the corresponding eight-neighborhood. Among them, the process of obtaining the gradient difference of the pixel point is in step S002.

[0053] Among them, the smaller the difference between the gradient difference of the center point and the gradient difference of the to-be-grown pixel points in the eight-neighborhood, the greater the possibility of growth between the two points; that is, the greater the growth coefficient between the corresponding two pixel points.

[0054] Perform growth according to the seed point and the growth coefficient between two pixel points, and grow the pixel points whose growth coefficient is greater than or equal to the third preset threshold until the growth ends. Among them, the third preset threshold is described by way of example. In this embodiment No specific limitations are imposed, and the implementer can decide according to the specific situation.

[0055] Thus, several growth lines are obtained.

[0056] It should be noted that since there are cracks generated by fractures and cartilage edges on the bone in the growth lines, and the fracture is a brittle fracture, the pixels on the cracks generated by the fracture are smooth and relatively straight, while the cartilage is rough and runs through the entire bone plane due to reasons such as life, and the fracture edge when the fracture runs through the entire bone is sharper.

[0057] Specifically, obtain the edge chain code of the growth line, and record the mean value of the differences between adjacent code numbers in the edge chain code of each growth line as the unevenness of each growth line; among them, the edge chain code is a well-known technology and will not be specifically described here.

[0058] Obtain the midpoint on each growth line, obtain the angle between the two endpoints and the midpoint line, and record it as the angle of each growth line; obtain the minimum distance between the farthest point from the bone connectivity domain edge line on each growth line and the bone connectivity domain edge line, and record it as the minimum distance between each growth line and the bone connectivity domain edge line.

[0059] According to the unevenness, angle, and minimum distance of each growth line, obtain the bone fracture coefficient of each growth line, which is specifically expressed by the formula: ; In the formula, represents the unevenness of the th growth line, represents the angle of the th growth line, represents the minimum distance between the th growth line and the bone connectivity domain edge line; represents the linear normalization function, represents the th growth line's bone fracture coefficient.

[0060] Among them, when the unevenness of each growth line is greater, it indicates that the possibility of this growth line being a bone fracture line is greater; when the angle of each growth line is larger and closer to 180 degrees, it indicates that the growth line is straighter, that is, the bone fracture coefficient corresponding to this growth line is larger. On the contrary, the smaller the angle, the smaller the bone fracture coefficient corresponding to this growth line; when the minimum distance between each growth line and the bone connectivity domain edge line is smaller, the closer the growth line is to the edge of the bone connectivity domain, that is, it more clearly indicates that the growth line is caused by a bone fracture.

[0061] Thus, the bone fracture coefficient of each growth line is obtained.

[0062] Step S004: Obtain all the fracture lines through the fracture coefficients of the growth lines to assist doctors in observation and treatment.

[0063] Mark the growth lines with fracture coefficients greater than the fourth preset threshold as fracture lines. Among them, the fourth preset threshold is described by way of example. In this embodiment is not specifically limited, and the implementer can determine it according to the specific situation.

[0064] Mark the fracture lines in the lower limb fracture image, and assist doctors to observe and treat the patient's lower limb fracture condition through the marked fracture line results.

[0065] It should be noted that the model used in this embodiment is only used to represent the negative correlation relationship and constrain the result of the model output to be within the interval. In specific implementation, it can be replaced with other models with the same purpose. This embodiment only takes the model as an example for description and does not specifically limit it. Among them refers to the input of the model.

[0066] This embodiment provides a high-precision segmentation system for lower limb fracture images based on artificial intelligence, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the high-precision segmentation method for lower limb fracture images based on artificial intelligence in steps S001 to S004.

[0067] So far, the present invention is completed.

[0068] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A high-precision segmentation method for lower limb fracture images based on artificial intelligence, characterized in that: The method comprises the following steps: Obtain images of the patient's lower extremity fractures; Segment the lower limb fracture image to obtain the leg connected domain; connect adjacent pixel points with similar gradient differences in the leg connected domain to obtain several initial bone lines; extend all initial bone lines to obtain several initial extended bone lines, and initially select several initial extended bone lines according to the number of intersections after extension; According to the distance between each initial extended bone line and the comparison edge corresponding to the leg connected domain after the initial screening, the texture degree of each initial extended bone line is obtained; through the average value of the texture degree of each initial extended bone line and the similar initial extended bone line, several initial extended bone lines are screened again; The endpoint of each initial extended bone line after rescreening is used as the initial seed point, and growth is performed according to the gradient difference between adjacent pixel points to obtain the bone connected domain in the leg connected domain; Obtain the pixel points inside the bone and the pixel points outside the bone in the eight neighborhoods of the edge pixel points of the bone connected domain, and obtain the bone inner edge point and the bone outer edge point through the pixel points inside the bone and the pixel points outside the bone in the eight neighborhoods of the edge pixel points of the bone connected domain; determine the breakpoint through the number of the bone inner edge point and the bone outer edge point in the eight neighborhoods of the edge pixel points of the bone connected domain, and use the breakpoint as the seed point; record the pixel points inside the bone in the eight neighborhoods of each central pixel point as the pixel points to be grown for each central pixel point, and obtain the growth coefficient between the central pixel point and each pixel point to be grown in the corresponding eight neighborhoods according to the difference between the gradient difference of the central pixel point and the gradient difference of the pixel point to be grown; grow through the seed point and the growth coefficient to obtain a plurality of growth lines; The bone fracture coefficient of each growth line is obtained based on the unevenness, angle and the minimum distance between each growth line and the edge line of the bone connected domain. All the bone fracture lines are obtained through the bone fracture coefficient, and all the bone fracture lines are used to assist doctors in observation and treatment.

2. The high-precision segmentation method for lower limb fracture images based on artificial intelligence according to claim 1 is characterized in that: The step of segmenting the lower limb fracture image to obtain the leg connected domain includes the following specific steps: The lower limb fracture images are segmented using the semantic segmentation algorithm in the neural network to obtain the leg connected domain.

3. The high-precision segmentation method for lower limb fracture images based on artificial intelligence according to claim 1 is characterized in that: The step of connecting adjacent pixel points with similar gradient differences in the leg connected domain to obtain a plurality of initial bone lines includes the following specific steps: Obtain the gradient values ​​of all pixels in the connected domain of the patient's leg by using the Canny edge detection algorithm, obtain the eight-neighborhood pixel points of each pixel in the connected domain of the leg, record the pixel point with the largest gradient value in the eight-neighborhood as the first pixel point of each pixel point, and record the pixel points in the eight-neighborhood with the closest gradient value to the central pixel point as the second pixel point of each pixel point; when the first pixel point and the second pixel point are the same pixel point, record the pixel point in the eight-neighborhood as the third pixel point; All adjacent third pixel points are connected to obtain several initial bone lines.

4. According to claim 1, the method for high-precision segmentation of lower limb fracture images based on artificial intelligence is characterized in that: The method of extending all initial bone lines to obtain a plurality of initial extended bone lines and initially selecting a plurality of initial extended bone lines according to the number of intersections after extension includes the following specific steps: Obtain the tangent direction of the two end points of each initial bone line, extend the corresponding initial bone line by half on each side along the tangent direction of the two end points, obtain the extended initial bone line, and record the extended initial bone line as the initial extended bone line; obtain the number of initial extended bone lines that each initial extended bone line intersects with all other initial extended bone lines, and record it as the number of intersections of each initial extended bone line; perform linear normalization on the number of intersections of all initial extended bone lines, and make the normalized number of intersections greater than or equal to the first preset threshold The initial extended bone lines are screened out, and the normalized intersection number is less than the first preset threshold The initial extended bone lines are screened out and retained to obtain several initial extended bone lines after the initial screening.

5. According to claim 1, the method for high-precision segmentation of lower limb fracture images based on artificial intelligence is characterized in that: The method comprises the following specific steps: obtaining the texture of each initial extended bone line according to the distance between each initial extended bone line and the comparison edge corresponding to the leg connected domain after the initial screening; and screening out a plurality of initial extended bone lines again according to the average of the textures of each initial extended bone line and the similar initial extended bone lines. Obtain the points where the two end points of each initial extended bone line are closest to the edge line of the leg connected domain, record them as reference points, and record the line connecting the two reference points as the comparison edge of each initial extended bone line; The calculation formula for the texture degree of each initial extended bone line is: ； In the formula, The information entropy representing the minimum distance between all initial extended bone lines and the corresponding comparison edges, Indicates Initial extension bone line Compare the corresponding edges Between distance, represents an exponential function with a natural constant as base, Indicates The texture of the initial extended bone line; Obtain the tangent directions of the two end points of each initial extended bone line, and obtain the two initial extended bone lines closest to the two end points along the tangent directions, and record them as the collinear bone lines of each initial extended bone line; Calculate the average value of the texture of each initial extended bone line and the collinear bone line, record it as the texture average of each initial extended bone line, and set the texture average greater than or equal to the second preset threshold The initial extended bone lines are screened out, and the texture mean is less than the second preset threshold The initial extended bone lines are retained, and several initial extended bone lines are obtained after re-screening.

6. The method for high-precision segmentation of lower limb fracture images based on artificial intelligence according to claim 1, characterized in that: The end point of each initial extended bone line after rescreening is used as the initial seed point, and growth is performed according to the gradient difference between adjacent pixel points to obtain the bone connected domain in the leg connected domain, including the following specific steps: Obtain the eight neighborhoods of each endpoint on each initial extended bone line, record any pixel point in the eight neighborhoods of each endpoint as a reference pixel point, connect each endpoint with the reference pixel point, obtain a straight line between each endpoint and the reference point, record the direction perpendicular to the straight line direction as the vertical direction of the reference point, obtain the pixel points adjacent to the reference point on both sides of the corresponding vertical direction, record them as the two gradient points of the reference point, calculate the absolute value of the difference between the two gradient points, and record it as the gradient difference of the reference point; obtain the pixel points adjacent to the endpoint on both sides of the corresponding vertical direction, record them as the two gradient points of the endpoint, calculate the absolute value of the difference between the grayscale values ​​of the two gradient points, and record it as the gradient difference of the endpoint; record the difference between the gradient difference of the reference point and the gradient difference of the endpoint as the gradient difference between each endpoint and the reference point; All endpoints on the initial extended bone line are used as initial seed points, and growth is performed according to the principle of minimizing the gradient difference between each endpoint and all pixels in all eight neighborhoods. The grown area is recorded as the bone connected domain in the leg connected domain.

7. The method for high-precision segmentation of lower limb fracture images based on artificial intelligence according to claim 6, characterized in that: The method of obtaining the pixel points inside the bone and the pixel points outside the bone in the eight neighborhoods of the edge pixel points of the bone connected domain, obtaining the bone inner edge point and the bone outer edge point through the pixel points inside the bone and the pixel points outside the bone in the eight neighborhoods of the edge pixel points of the bone connected domain; and determining the breakpoint through the number of the bone inner edge point and the bone outer edge point in the eight neighborhoods of the edge pixel points of the bone connected domain includes the following specific steps: The pixel points of the eight neighboring pixel points of the edge pixel point within the bone connected domain are recorded as the pixel points inside the bone, and the pixel points of the eight neighboring pixel points of the edge pixel point outside the bone connected domain are recorded as the pixel points outside the bone; Calculate the mean grayscale value of all pixels in the bone in the eight neighborhoods of each edge pixel in the bone connected domain, and record it as the mean grayscale value of each edge pixel in the bone connected domain; Calculate the average grayscale value of all pixels outside the bone in the eight neighborhoods of each edge pixel in the bone connected domain, and record it as the average grayscale value outside the bone of each edge pixel in the bone connected domain; Compare the difference between the grayscale value of each edge pixel point of the bone connected domain and the grayscale mean inside the bone and the grayscale mean outside the bone. When the difference between the grayscale value of each edge pixel point of the bone connected domain and the grayscale mean inside the bone is less than or equal to the difference between the grayscale value of the edge pixel point and the grayscale mean outside the bone, the edge pixel point of the bone connected domain is recorded as an edge point inside the bone; otherwise, the edge pixel point of the bone connected domain is recorded as an edge point outside the bone; Any edge pixel point in the bone connected domain that contains an inner edge point of a bone and an outer edge point of a bone in the eight-neighborhood of any pixel point is recorded as a breakpoint.

8. The method for high-precision segmentation of lower limb fracture images based on artificial intelligence according to claim 7, characterized in that: The method of obtaining a growth coefficient between the central pixel and each pixel to be grown in the corresponding eight neighborhoods according to the difference between the gradient difference of the central pixel and the gradient difference of the pixel to be grown; and obtaining a plurality of growth lines by growing through the seed point and the growth coefficient includes the following specific steps: The center pixel and the corresponding eight neighborhoods The formula for the growth coefficient between the pixels to be grown is expressed as: ； In the formula, Indicates the gradient difference of the central pixel during the growth process, The eight neighborhoods of the center point The gradient difference of the pixels to be grown is is the absolute value symbol, represents an exponential function with a natural constant as base, Represents the center pixel and the corresponding eight neighborhoods The growth coefficient between the pixels to be grown; Grow according to the growth coefficient between the seed point and the two pixel points, and make the growth coefficient greater than or equal to the third preset threshold The pixel points are grown until the growth is completed, and several growth lines are obtained.

9. The method for high-precision segmentation of lower limb fracture images based on artificial intelligence according to claim 1, characterized in that: According to the unevenness, angle and minimum distance between each growth line and the edge line of the bone connection domain, the bone fracture coefficient of each growth line is obtained, all the bone fracture lines are obtained through the bone fracture coefficient, and all the bone fracture lines are used to assist doctors in observation and treatment, including the following specific steps: Obtain the edge chain code of the growth line, and record the average of the differences between the numbers of adjacent codes in the edge chain code of each growth line as the roughness of each growth line; Get the midpoint of each growth line, get the angle between the two endpoints and the midpoint straight line, and record it as the angle of each growth line; get the minimum distance between the endpoint farthest from the edge line of the bone connected domain on each growth line and the edge line of the bone connected domain, and record it as the minimum distance between each growth line and the edge line of the bone connected domain; The bone fracture coefficient of each growth line is expressed by the formula: ； In the formula, Indicates The unevenness of the growth lines, Indicates The angle between the growth lines, Indicates The minimum distance between a growth line and the edge line of the bone connected domain; represents the linear normalization function, Indicates The bone fracture coefficient of the growth line; The bone fracture coefficient is greater than the fourth preset threshold The growth line is recorded as the bone fracture line, and the bone fracture line in the lower limb fracture image is marked. The marked bone fracture line results assist doctors in observing and treating patients' lower limb fractures.

10. A high-precision segmentation system for lower limb fracture images based on artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the high-precision segmentation method for lower limb fracture images based on artificial intelligence as described in any one of claims 1-9 are implemented.

Images