A C-arm X-ray machine image recognition method

Abstract

The present invention discloses a method for recognizing C-arm X-ray images. The method is characterized by detecting the dividing line between the thoracic and lumbar vertebrae through a U-Net model, automatically detecting the vertebrae in the spinal image using a YOLOV3 model, and identifying the thoracic and lumbar vertebrae. Finally, the C-arm X-ray image and the initial grayscale image are registered through a SIFT algorithm to identify a final C-arm X-ray recognition image. The method has the advantage of achieving good results in image recognition tasks by sorting based on region segmentation results after image detection. In order to solve the problem of lack of image field of view, an image registration method can be used since a full spinal X-ray film is provided before surgery. The feature-based image registration method is based on the matching relationship between salient points and landmarks in the image to perform registration. The method has high recognition efficiency and low error rate, and can be widely used in the field of medical image processing.

Description

Technical Field

[0001] The present invention relates to an X-ray machine image recognition method, in particular to a C-arm X-ray machine image recognition method. Background Art

[0002] C-arm X-ray images are widely used in orthopedic surgery as an important tool for intraoperative positioning because they can monitor the dynamic situation during surgery in real time. However, since C-arm X-ray images lack 3D spatial position information, doctors usually need to manually judge the position of the C-arm image based on the preoperative X-ray anteroposterior and lateral views. This has a certain possibility of error and is inefficient. Therefore, automatic C-arm X-ray image recognition can provide great convenience for doctors during surgery.

[0003] The main difficulties in image recognition using C-arm X-ray machines lie in vertebrae recognition and the limited field of view. Images typically display only three to five vertebrae, lacking information about the upper and lower spines. In vertebrae recognition, due to individual differences and the weak sensitivity of low-level image features to high-frequency information, both the statistical model-based and feature-learning-based methods used in traditional methods lack good judgment performance. Therefore, finding a vertebrae recognition method with good generalization and the ability to extract deep image features has become a major challenge in the current medical imaging field. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for recognizing C-arm X-ray machine images with high recognition efficiency and average recognition accuracy.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a method for recognizing images of a C-arm X-ray machine, comprising the following steps:

[0006] Step 1): A full spinal X-ray is obtained using a preoperative X-ray machine and converted into an initial grayscale image I1. Adaptive contrast enhancement is performed on I1 to obtain an enhanced image. The left and right boundaries S1 and S2 of the spinal image are obtained based on the enhanced image. The spinal image I2 is then cropped based on S1 and S2 on I1.

[0007] Step 2): After processing I1 through the U-Net network, a segmented binary classification image is generated, and the segmented binary classification image is cropped to obtain a cropped binary classification image. Then, the boundary line L1 between the cervical vertebrae and the thoracic vertebrae and the boundary line L2 between the thoracic vertebrae and the lumbar vertebrae are obtained based on the cropped binary classification image;

[0008] Step 3): Process I2 using the YOLOv3 model to obtain a detected spinal image. Then, based on L1 and L2, set the category label of the prediction box in the detected spinal image and amplify it to obtain an amplified detection image I3. I3 is overlaid on the I2 portion of I1 based on S1 and S2 to obtain the final full spine X-ray vertebra recognition image I4.

[0009] Step 4): Obtain a C-arm X-ray image I5 that matches the full spine X-ray film through a C-arm X-ray machine, obtain matching feature points based on I1 and I5 based on the SIFT algorithm, obtain the center coordinates (x0, y0) of the C-arm X-ray image to be identified, and finally crop I4 with (x0, y0) as the center and the radius R of I5 as the radius to obtain the final C-arm X-ray recognition image.

[0010] The specific process is as follows:

[0011] Step 1)-1: Get the local mean of each point in I1, and record the local mean of the point with coordinates (i, j) in I1 as M(i, j). Where n is the preset horizontal local length parameter, m is the preset vertical length parameter, f(s, k) represents the grayscale value of the point with coordinates (s, k) in I1, 0≤i≤P, 0≤j≤Q, in≤s≤i+n, jm≤k≤j+m, P represents the horizontal total length of I1, and Q represents the vertical total length of I1;

[0012] Step 1)-2: Get the local standard deviation of each point in I1, and record the local standard deviation of the point with coordinates (i, j) in I1 as σ 2 (i, j),

[0013] Step 1)-3: Obtain the grayscale value of each point in the enhanced image and construct the enhanced image, wherein the grayscale value of the point with coordinates (i, j) in the enhanced image is defined as I(i, j), I(i, j) = M(i, j) + G(f(i, j) - M(i, j)), where α is the preset enhancement coefficient and 0<α<1, f(i, j) represents the gray value of the point with coordinates (i, j) in I1, M is the gray mean value of the entire I1,

[0014] Step 1)-4: Obtain the grayscale sum of all vertical components of the enhanced image, arrange the grayscale sum of all vertical components in ascending order according to the corresponding horizontal coordinates to form a vertical component grayscale sum sequence IH, and define the grayscale sum of the vertical components corresponding to all points with horizontal coordinate i in the vertical component grayscale sum sequence as IH(i). The average grayscale intensity of the enhanced grayscale image is defined as MI. The standard deviation of the grayscale values ​​of all points in the enhanced grayscale image is defined as SI. In IH, the grayscale sum of the vertical components greater than the threshold is screened out using MI+SI as the threshold, the grayscale sum of the vertical component with the smallest sequence number in IH among the grayscale sums of all the screened out vertical components is recorded as IH1, and the sequence number of IH1 is recorded as S1, the grayscale sum of the vertical component with the largest sequence number in IH among the grayscale sums of all the screened out vertical components is recorded as IH2, and the sequence number of IH2 is recorded as S2, S1 is used as the left boundary value in the horizontal direction to obtain the left boundary of the spine image, and S2 is used as the right boundary value in the horizontal direction to obtain the right boundary of the spine image, and then the spine image I2 is cropped from I1 according to S1 and S2;

[0015] Step 2)-1: Construct a U-Net network, set the rib area in I1, uniformly scale I1 to 672*384 pixels, and input it into the U-Net network to generate a segmented binary image. Set the grayscale of the rib area in the segmented binary image to G, 200≤G≤255, and set the grayscale of the non-rib area in the segmented binary image to 0;

[0016] Step 2)-2: Crop the segmented binary image according to S1 and S2 to obtain a cropped binary image, then obtain the mean vertical coordinate value MV1 of the first point with a grayscale value of G in all vertical components of the cropped binary image, and simultaneously obtain the mean vertical coordinate value MV2 of the last point with a grayscale value of G in each vertical component of the cropped binary image. The dividing line formed by connecting the points with vertical coordinates MV1 is used as the dividing line L1 between the cervical vertebrae and the thoracic vertebrae, and the dividing line formed by connecting the points with vertical coordinates MV2 is used as the dividing line L2 between the thoracic vertebrae and the lumbar vertebrae;

[0017] Step 3): Let the horizontal length of I2 be W1, set the vertebral area in I2, scale I2 to 256*608 pixels and input it into the YOLOv3 model, output the detected spine image, and then set the category label of the prediction box in the detected spine image based on MV1 and MV2 to obtain a detection image with category labels, wherein the category label of the prediction box whose center coordinates are between MV1 and MV2 is set to thoracic vertebrae, and the category label of the prediction box whose center coordinates are below MV2 is set to lumbar vertebrae, then enlarge the spine image with category labels to a pixel size of W1*Q, and finally generate an enlarged detection image I3, and cover the I2 part of I1 with I3 according to S1 and S2 to obtain the final full spine X-ray vertebra recognition image I4;

[0018] Step 4)-1: Obtain a C-arm X-ray image I5 that matches the full spine X-ray film using a C-arm X-ray machine, obtain matching feature points based on I1 and I5 using the SIFT algorithm, and record the feature point set in I1 as D1, D1 = D 11 , D 12 ...D 1v , record the feature point set in I5 as D2, D2 = D 21 , D 22 ...D 2v , where v is the number of groups of feature points obtained;

[0019] Step 4)-2: Let the radius of I5 be R, define the center coordinates of I5 as (R, R), obtain the relative X-direction coefficient and relative Y-direction coefficient of the feature point I5 to the center, and define the coordinates in I5 as (x 2i′ ,y 2i′ ) is the relative X-direction coefficient of the point x(2i′) , Where 0≤x 2i′ ≤2R,0≤y 2i′ ≤2R,, define the coordinates in I5 as (x 2i′ ,y 2i′ ) is the relative Y-direction coefficient of the point y(2i′) , Get the distance set r1 between each feature point in D1 and the next feature point, r1 = {r 11 , r 12 ,...,r 1(v-1)}, where r 11 Represents the distance between the first feature point and the second feature point, r 12 Represents the distance between the second feature point and the third feature point, r 1(v-1) Represents the distance between the second to last feature point and the last feature point, and obtains the distance set r2 between the feature points in D2, r2={r 21 , r 22 ,...,r 2(v-1)}, where r 21 Represents the distance between the first feature point and the second feature point, r 22 Represents the distance between the second feature point and the third feature point, r 2(v-1) Represents the distance between the second to last feature point and the last feature point, and defines the set of relative radius coefficients as r3, r3 = {r 31 , r 32 ,...,r 3(v-1)},

[0020] Step 4)-3: Define the center coordinates of the C-arm X-ray image to be identified as (x0, y0), where: (x 2k′ ,y 2k′ ) represents the coordinates of any feature point in I5, 0≤x 2k′ ≤2R,0≤y 2k′ ≤2R,p x(2k′) Indicates that the coordinates in I5 are (x 2k′ ,y 2k′ ) relative X-direction coefficient, r 3k′ Indicates the relative radius coefficient of the sequence number k' in r3, (x 2(k′+1) ,y 2(k′+1) ) means the coordinate in I5 is (x 2k′ ,y 2k′ )’s next feature point, p x(k′+1) Indicates that the coordinates in I5 are (x 2(k′+1) ,y 2(k′+1) ) relative X-direction coefficient, r 3(k′+1) represents the relative radius coefficient of the sequence number (k′+1) in r3, P y(2k′) Indicates that the coordinates in I5 are (x 2k′ ,y 2k′ ) point relative Y-direction coefficient, P y(k′+1) Indicates that the coordinates in I5 are (x 2(k′+1) ,y 2(k′+1) )’s relative Y-direction coefficient,

[0021] Step 4)-4: In I4, use (x0, y0) as the center and R as the radius to crop the image to obtain the final C-arm X-ray recognition image.

[0022] Compared with the existing technology, the advantages of the present invention are that it detects the boundary between the thoracic and lumbar vertebrae through the U-Net model, then uses the YOLOV3 model to automatically detect the vertebrae in the spinal image and identify the thoracic and lumbar vertebrae. Finally, the C-arm X-ray image and the initial grayscale image are registered using the SIFT algorithm to identify the final C-arm X-ray recognition image.

[0023] The above recognition method can achieve good results in image recognition tasks by performing image detection and then sorting based on the results of regional segmentation. In terms of the problem of lack of image field of view, since a full spinal X-ray is provided before surgery, the image registration method can be used to solve it. The feature-based image registration method is based on the salient points and landmarks in the image to establish a matching relationship between feature points for registration. It has high recognition efficiency and a low error rate, and can be widely used in the field of medical image processing. DETAILED DESCRIPTION

[0024] The present invention is described in further detail below.

[0025] A method for recognizing images of a C-arm X-ray machine comprises the following steps:

[0026] Step 1): Obtain a full spinal X-ray using a preoperative X-ray machine and convert it into an initial grayscale image I1. Adaptively enhance the contrast of I1 to obtain an enhanced image. Based on the enhanced image, obtain the left and right boundaries S1 and S2 of the spinal image. Crop I1 based on S1 and S2 to obtain the spinal image I2. The specific process is as follows:

[0027] Step 1)-1: Get the local mean of each point in I1, and record the local mean of the point with coordinates (i, j) in I1 as M(i, j). Where n is a preset horizontal local length parameter, m is a preset vertical length parameter, f(s, k) represents the grayscale value of the point with coordinates (s, k) in I1, 0≤i≤P, 0≤j≤Q, in≤s≤i+n, jm≤k≤j+m, P represents the horizontal total length of I1, and Q represents the vertical total length of I1; n=10, m=15 can be taken;

[0028] Step 1)-2: Get the local standard deviation of each point in I1, and record the local standard deviation of the point with coordinates (i, j) in I1 as σ 2 (i, j),

[0029] Step 1)-3: Obtain the grayscale value of each point in the enhanced image and construct the enhanced image, wherein the grayscale value of the point with coordinates (i, j) in the enhanced image is defined as I(i, j), I(i, j) = M(i, j) + G(f(i, j) - M(i, j)), where α is the preset enhancement coefficient and 0<α<1, f(i, j) represents the gray value of the point with coordinates (i, j) in I1, M is the gray mean value of the entire I1,

[0030] Step 1)-4: Obtain the grayscale sum of all vertical components of the enhanced image, arrange the grayscale sum of all vertical components in ascending order according to the corresponding horizontal coordinates to form a vertical component grayscale sum sequence IH, and define the grayscale sum of the vertical components corresponding to all points with horizontal coordinate i in the vertical component grayscale sum sequence as IH(i). The average grayscale intensity of the enhanced grayscale image is defined as MI. The standard deviation of the grayscale values ​​of all points in the enhanced grayscale image is defined as SI. In IH, the grayscale sum of the vertical components greater than the threshold is screened out using MI+SI as the threshold, the grayscale sum of the vertical component with the smallest sequence number in IH among the grayscale sums of all the screened out vertical components is recorded as IH1, and the sequence number of IH1 is recorded as S1, the grayscale sum of the vertical component with the largest sequence number in IH among the grayscale sums of all the screened out vertical components is recorded as IH2, and the sequence number of IH2 is recorded as S2, S1 is used as the left boundary value in the horizontal direction to obtain the left boundary of the spine image, and S2 is used as the right boundary value in the horizontal direction to obtain the right boundary of the spine image, and then the spine image I2 is cropped from I1 according to S1 and S2;

[0031] Step 2)-1: Construct a U-Net network, set the rib area in I1, uniformly scale I1 to 672*384 pixels, and input it into the U-Net network to generate a segmented binary image. Set the grayscale of the rib area in the segmented binary image to G, 200≤G≤255, and set the grayscale of the non-rib area in the segmented binary image to 0;

[0032] Step 2)-2: Crop the segmented binary image according to S1 and S2 to obtain a cropped binary image, then obtain the mean vertical coordinate value MV1 of the first point with a grayscale value of G in all vertical components of the cropped binary image, and simultaneously obtain the mean vertical coordinate value MV2 of the last point with a grayscale value of G in each vertical component of the cropped binary image. The dividing line formed by connecting the points with vertical coordinates MV1 is used as the dividing line L1 between the cervical vertebrae and the thoracic vertebrae, and the dividing line formed by connecting the points with vertical coordinates MV2 is used as the dividing line L2 between the thoracic vertebrae and the lumbar vertebrae;

[0033] Step 3): Let the horizontal length of I2 be W1, set the vertebral area in I2, scale I2 to 256*608 pixels and input it into the YOLOv3 model, output the detected spine image, and then set the category label of the prediction box in the detected spine image based on MV1 and MV2 to obtain a detection image with category labels, wherein the category label of the prediction box whose center coordinates are between MV1 and MV2 is set to thoracic vertebrae, and the category label of the prediction box whose center coordinates are below MV2 is set to lumbar vertebrae, and then enlarge the spine image with category labels to a pixel size of W1*Q, and finally generate an enlarged detection image I3, and cover the I2 part of I1 with I3 according to S1 and S2 to obtain the final full spine X-ray vertebra recognition image I4; wherein, in the detected spine image, the prediction boxes are numbered in order from top to bottom for easy observation and comparison;

[0034] Step 4)-1: Obtain a C-arm X-ray image I5 that matches the full spine X-ray film using a C-arm X-ray machine, obtain matching feature points based on I1 and I5 using the SIFT algorithm, and record the feature point set in I1 as D1, D1 = D 11 , D 12 ...D 1v , record the feature point set in I5 as D2, D2 = D 21 , D 22 ...D 2v , where v is the number of groups of feature points obtained;

[0035] Step 4)-2: Let the radius of I5 be R, define the center coordinates of I5 as (R, R), obtain the relative X-direction coefficient and relative Y-direction coefficient of the feature point I5 to the center, and define the coordinates in I5 as (x 2i′ ,y 2i′ ) is the relative X-direction coefficient of the point x(2i′) , Where 0≤x 2i′ ≤2R,0≤y 2i′ ≤2R,, define the coordinates in I5 as (x 2i ′,y 2i′ The relative Y-direction coefficient of the point is p y(2i′) , Get the distance set r1 between each feature point in D1 and the next feature point, r1 = {r 11 , r 12 ,...,r 1(v-1)}, where r 11 Represents the distance between the first feature point and the second feature point, r 12 Represents the distance between the second feature point and the third feature point, r 1(v-1) Represents the distance between the second to last feature point and the last feature point, and obtains the distance set r2 between the feature points in D2, r2={r 21， r 22 ,...,r 2(v-1)}, where r 21 Represents the distance between the first feature point and the second feature point, r 22 Represents the distance between the second feature point and the third feature point, r 2(v-1) Represents the distance between the second to last feature point and the last feature point, and defines the set of relative radius coefficients as r3, r3 = {r 31 , r 32 ,...,r 3(v-1)},

[0036] Step 4)-3: Define the center coordinates of the C-arm X-ray image to be identified as (x0, y0), where: (x 2k′ ,y 2k′ ) represents the coordinates of any feature point in I5, 0≤x 2k′ ≤2R,0≤y 2k′ ≤2R,p x(2k′) Indicates that the coordinates in I5 are (x 2k′ ,y 2k′ ) relative X-direction coefficient, r 3k′ Indicates the relative radius coefficient of the sequence number k' in r3, (x 2(k′+1) ,y 2(k′+1) ) means the coordinate in I5 is (x 2k′ ,y 2k′ )’s next feature point, p x(k′+1) Indicates that the coordinates in I5 are (x 2(k′+1) ,y 2(k′+1) ) relative X-direction coefficient, r 3(k′+1) represents the relative radius coefficient of the sequence number (k′+1) in r3, p y(2k′) Indicates that the coordinates in I5 are (x 2k′ ,y 2k′ ) point relative Y-direction coefficient, p y(k′+1) Indicates that the coordinates in I5 are (x 2(k′+1) ,y 2(k′+1) )’s relative Y-direction coefficient,

[0037] Step 4)-4: In I4, use (x0, y0) as the center and R as the radius to crop the image to obtain the final C-arm X-ray recognition image.

[0038] The method of the above embodiment was used to conduct actual experimental tests. The images used in the experiment were 400 X-ray images from the hospital. Without repeatedly calculating the offset error, the C-arm X-ray recognition images identified by this method had an average recognition accuracy of 82.885% compared with the results of manual annotation.

Claims

1. A method for recognizing images of a C-arm X-ray machine, characterized in that The following steps are involved: Step 1): A full spinal X-ray is obtained using a preoperative X-ray machine and converted into an initial grayscale image I1. Adaptive contrast enhancement is performed on I1 to obtain an enhanced image. The left and right boundaries S1 and S2 of the spinal image are obtained based on the enhanced image. The spinal image I2 is then cropped based on S1 and S2 on I1. Step 2): After processing I1 through the U-Net network, a segmented binary classification image is generated. The segmented binary classification image is cropped according to S1 and S2 to obtain a cropped binary classification image. Then, the boundary line L1 between the cervical vertebrae and the thoracic vertebrae and the boundary line L2 between the thoracic vertebrae and the lumbar vertebrae are obtained based on the cropped binary classification image; Step 3): Process I2 using the YOLOv3 model to obtain a detected spinal image. Then, based on L1 and L2, set the category label of the prediction box in the detected spinal image and amplify it to obtain an amplified detection image I3. I3 is overlaid on the I2 portion of I1 based on S1 and S2 to obtain the final full spine X-ray vertebra recognition image I4. Step 4): Use a C-arm X-ray machine to obtain a C-arm X-ray image I5 that matches the full spine X-ray film. Based on I1 and I5, obtain the matching feature points based on the SIFT algorithm, obtain the center coordinates (x0, y0) of the C-arm X-ray image to be identified, and finally crop I4 with (x0, y0) as the center and the radius R of I5 as the radius to obtain the final C-arm X-ray recognition image.

2. A C-arm X-ray machine image recognition method according to claim 1, characterized in that The specific process is as follows: Step 1)-1: Get the local mean of each point in I1, and record the local mean of the point with coordinates (i, j) in I1 as M(i, j). Where n is the preset horizontal local length parameter, m is the preset vertical length parameter, f(s,k) represents the grayscale value of the point with coordinates (s,k) in I1, 0≤i≤P, 0≤j≤Q, in≤s≤i+n, jm≤k≤j+m, P represents the horizontal total length of I1, and Q represents the vertical total length of I1; Step 1)-2: Get the local standard deviation of each point in I1, and record the local standard deviation of the point with coordinates (i, j) in I1 as σ 2 (i,j), Step 1)-3: Obtain the grayscale value of each point in the enhanced image and construct the enhanced image, wherein the grayscale value of the point with coordinates (i, j) in the enhanced image is defined as I(i, j), I(i, j) = M(i, j) + G(f(i, j) - M(i, j)), where α is the preset enhancement coefficient and 0<α<1, f(i,j) represents the gray value of the point with coordinates (i,j) in I1, M is the gray mean value of the entire I1, Step 1)-4: Obtain the grayscale sum of all vertical components of the enhanced image, arrange the grayscale sum of all vertical components in ascending order according to the corresponding horizontal coordinates to form a vertical component grayscale sum sequence IH, and define the grayscale sum of the vertical components corresponding to all points with horizontal coordinate i in the vertical component grayscale sum sequence as IH(i). The average grayscale intensity of the enhanced grayscale image is defined as MI. The standard deviation of the grayscale values ​​of all points in the enhanced grayscale image is defined as SI. In IH, the grayscale sum of the vertical components greater than the threshold is screened out using MI+SI as the threshold, the grayscale sum of the vertical component with the smallest sequence number in IH among the grayscale sums of all the screened out vertical components is recorded as IH1, and the sequence number of IH1 is recorded as S1, the grayscale sum of the vertical component with the largest sequence number in IH among the grayscale sums of all the screened out vertical components is recorded as IH2, and the sequence number of IH2 is recorded as S2, S1 is used as the left boundary value in the horizontal direction to obtain the left boundary of the spine image, and S2 is used as the right boundary value in the horizontal direction to obtain the right boundary of the spine image, and then the spine image I2 is cropped from I1 according to S1 and S2; Step 2)-1: Construct a U-Net network, set the rib area in I1, uniformly scale I1 to 672*384 pixels, and input it into the U-Net network to generate a segmented binary image. Set the grayscale of the rib area in the segmented binary image to G, 200≤G≤255, and set the grayscale of the non-rib area in the segmented binary image to 0; Step 2)-2: Crop the segmented binary image according to S1 and S2 to obtain a cropped binary image, then obtain the mean vertical coordinate value MV1 of the first point with a grayscale value of G in all vertical components of the cropped binary image, and simultaneously obtain the mean vertical coordinate value MV2 of the last point with a grayscale value of G in each vertical component of the cropped binary image. The dividing line formed by connecting the points with vertical coordinates MV1 is used as the dividing line L1 between the cervical vertebrae and the thoracic vertebrae, and the dividing line formed by connecting the points with vertical coordinates MV2 is used as the dividing line L2 between the thoracic vertebrae and the lumbar vertebrae; Step 3): Let the horizontal length of I2 be W1, set the vertebral area in I2, scale I2 to 256*608 pixels and input it into the YOLOv3 model, output the detected spine image, and then set the category label of the prediction box in the detected spine image based on MV1 and MV2 to obtain a detection image with category labels, wherein the category label of the prediction box whose center coordinates are between MV1 and MV2 is set to thoracic vertebrae, and the category label of the prediction box whose center coordinates are below MV2 is set to lumbar vertebrae, then enlarge the spine image with category labels to a pixel size of W1*Q, and finally generate an enlarged detection image I3, and cover the I2 part of I1 with I3 according to S1 and S2 to obtain the final full spine X-ray vertebra recognition image I4; Step 4)-1: Obtain a C-arm X-ray image I5 that matches the full spine X-ray film using a C-arm X-ray machine, obtain matching feature points based on I1 and I5 using the SIFT algorithm, and record the feature point set in I1 as D1, D1 = D 11 ,D 12 …D 1v , record the feature point set in I5 as D2, D2 = D 21 ,D 22 …D 2v , where v is the number of groups of feature points obtained; Step 4)-2: Let the radius of I5 be R, define the center coordinates of I5 as (R, R), obtain the relative X-direction coefficient and relative Y-direction coefficient of the feature point I5 to the center, and define the coordinates in I5 as (x 2i' ,y 2i' ) is the relative X-direction coefficient of the point x(2i′) , Where 0≤x 2i′ ≤2R,0≤y 2i′ ≤2R, define the coordinates in I5 as (x 2i' ,y 2i' The relative Y-direction coefficient of the point is p y(2i′) , Get the distance set r1 between each feature point in D1 and the next feature point, r1 = {r 11 ,r 12 ,…,r 1(v-1) }, where r 11 Represents the distance between the first feature point and the second feature point, r 12 Represents the distance between the second feature point and the third feature point, r 1(v-1) Represents the distance between the second to last feature point and the last feature point, and obtains the distance set r2 between the feature points in D2, r2={r 21 ,r 22 ,…,r 2(v-1) }, where r 21 Represents the distance between the first feature point and the second feature point, r 22 Represents the distance between the second feature point and the third feature point, r 2(v-1) Represents the distance between the second to last feature point and the last feature point, and defines the set of relative radius coefficients as r3, r3 = {r 31 ,r 32 ,…,r 3(v-1) }, Step 4)-3: Define the center coordinates of the C-arm X-ray image to be identified as (x0, y0), where: (x 2k′ ,y 2k′ ) represents the coordinates of any feature point in I5, 0≤x 2k′ ≤2R,0≤y 2k′ ≤2R,p x(2k′) Indicates that the coordinates in I5 are (x 2k′ ,y 2k′ ) relative X-direction coefficient, r 3k′ Indicates the relative radius coefficient of the sequence number k' in r3, (x 2(k′+1) ,y 2(k′+1) ) means the coordinate in I5 is (x 2k′ ,y 2k′ )’s next feature point, p x(k′+1) Indicates that the coordinates in I5 are (x 2(k′+1) ,y 2(k′+1) ) relative X-direction coefficient, r 3(k′+1) represents the relative radius coefficient of the sequence number (k′+1) in r3, p y(2k′) Indicates that the coordinates in I5 are (x 2k′ ,y 2k′ ) point relative Y-direction coefficient, p y(k′+1) Indicates that the coordinates in I5 are (x 2(k′+1 ),y 2(k′+1) )’s relative Y-direction coefficient, Step 4)-4: In I4, use (x0, y0) as the center and R as the radius to crop the image to obtain the final C-arm X-ray recognition image.

Images