Three-dimensional Image Reconstruction Accuracy Evaluation Method

Abstract

The present invention discloses a method for evaluating the reconstruction accuracy of three-dimensional images, comprising the steps of: collecting three-dimensional images containing a precision checker, wherein a plurality of checker points arranged according to a set rule are provided in the precision checker, including at least three coordinate reference points; extracting the actual coordinates of each candidate point used to represent each checker point in the image coordinate system, thereby obtaining; obtaining the coordinates of each coordinate reference point in the checker coordinate system according to the design parameters of the precision checker, and calculating the transformation matrix between the two by combining the actual coordinates of each coordinate reference point; transforming the actual coordinates of each candidate point to the checker coordinate system according to the transformation matrix, and accordingly making one-to-one correspondence with each corresponding checker point in the checker, so as to obtain the theoretical coordinates of each candidate point in the checker coordinate system; calculating the three-dimensional reconstruction accuracy according to the actual coordinates and the theoretical coordinates of each candidate point. The present invention can effectively evaluate the reconstruction accuracy of three-dimensional images and has higher accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular, to a method for evaluating the accuracy of three-dimensional image reconstruction. Background Art

[0002] The accuracy of three-dimensional image reconstruction is an important indicator for evaluating the quality of medical images. The principle of the navigation system is to associate the image coordinate system with the optical navigation coordinate system. Therefore, the accuracy of three-dimensional image reconstruction directly affects the system accuracy of navigation positioning. Thus, an evaluation method suitable for evaluating the accuracy of three-dimensional image reconstruction is required. Summary of the Invention

[0003] Object of the Invention: Aiming at the above deficiencies, the present invention proposes a method for evaluating the accuracy of three-dimensional image reconstruction, which can effectively evaluate the accuracy of three-dimensional image reconstruction and has higher accuracy.

[0004] Technical Solution:

[0005] A method for evaluating the accuracy of three-dimensional image reconstruction, comprising the steps of:

[0006] Collecting a three-dimensional image containing an accuracy checker, wherein the accuracy checker is provided with a plurality of check points arranged according to a set rule, including at least three coordinate reference points;

[0007] Extracting the actual coordinates of each candidate point used to represent each check point in the image coordinate system, and at the same time obtaining the actual coordinates of each coordinate reference point therein;

[0008] According to the design parameters of the accuracy checker, obtaining the coordinates of each coordinate reference point in the checker coordinate system, and calculating the transformation matrix between the two in combination with the actual coordinates of each coordinate reference point;

[0009] According to the transformation matrix, transforming the actual coordinates of each candidate point into the checker coordinate system, and accordingly corresponding each candidate point to the corresponding check point in the checker one by one, so as to obtain the theoretical coordinates of each candidate point in the checker coordinate system;

[0010] Calculating the three-dimensional reconstruction accuracy according to the actual coordinates and theoretical coordinates of each candidate point.

[0011] The specific method for extracting the actual coordinates of each candidate point and each coordinate reference point in the image coordinate system is: using adaptive threshold segmentation to obtain the three-dimensional coordinate data containing each candidate point, and calculating the centroid of each candidate point accordingly to obtain the coordinates of each candidate point in the image coordinate system.

[0012] The specific method for calculating the centroid of each candidate point to obtain the coordinates of each candidate point in the image coordinate system is:

[0013] Obtaining a voxel point in the three-dimensional coordinate data of each candidate point as a seed point;

[0014] Traverse and search for three-dimensional coordinate data, calculate the distances between all voxel points and all seed points. If the distance between a voxel point and a certain seed point is the closest, then cluster the voxel point as a nearby point of the seed point and include it as the set of voxel points contained in a candidate point;

[0015] Calculate the centroid of each candidate point based on the number of voxel points in the set of voxel points contained in a certain candidate point, the coordinate values of each voxel point, and the voxel gray value.

[0016] The specific method of obtaining a voxel point in the three-dimensional coordinate data of each candidate point as a seed point is as follows:

[0017] Traverse and search for the three-dimensional coordinate data obtained by segmentation that contains each candidate point, and take the first voxel point obtained as the first seed point and put it into the seed point set;

[0018] Traverse the remaining un-traversed voxel points in the three-dimensional coordinate data, search for the voxel point that is the closest to the first seed point and greater than the set threshold as a new seed point and update the seed point set;

[0019] Traverse the remaining un-traversed voxel points, search for the voxel point whose distance from the closest seed point in the seed point set is greater than the set threshold as a new seed point and update the seed point set, and repeat this step until all voxel points are traversed to obtain the final seed point set.

[0020] The selected range of the set threshold is set to be greater than the maximum diameter of the calibration points and less than half of the distance between two calibration points.

[0021] Obtain the coordinates of each coordinate reference point in the calibrator coordinate system according to the design parameters of the precision calibrator, and calculate the transformation matrix between the two by combining the actual coordinates of each coordinate reference point. Specifically:

[0022] Obtain the numbers in the image, and obtain the coordinates P of three coordinate reference points in the image coordinate system according to the design parameters of the precision calibrator i {p u ,p v ,p w};

[0023] Obtain the coordinates P of three coordinate reference points in the calibrator coordinate system according to the design parameters of the precision calibrator i '{p u ',p v ',p w '}, and compare it with the coordinates P of the three coordinate reference points in the image coordinate system obtained previously i {p u ,p v ,p wPerform least squares matching to obtain the least squares transformation matrix M between the two, specifically as follows:

[0024] P i '{p u ',p v ',p w [[ID=11']} = M * P i {p u ,p v ,p w}+δ

[0025] Obtain the transformation matrix M between the two by solving to minimize δ.

[0026] Specifically, the actual coordinates of each candidate point transformed to the checker coordinate system are corresponded one by one with the corresponding check points in the checker as follows:

[0027] x i = p i '.x / d

[0028] y i = p i '.y / d

[0029] z i = p i '.z / d

[0030] Among them, x i , y i , z i are the x-direction index, y-direction index, and z-direction index respectively, p i '.x, p i '.y, p i '.z are the x-coordinate, y-coordinate, and z-coordinate of the candidate point p i ' transformed to the checker coordinate system respectively, and d represents the Euclidean distance between adjacent check points.

[0031] Specifically, the three-dimensional reconstruction accuracy is calculated based on the actual coordinates and theoretical coordinates of each candidate point as follows:

[0032] The obtained theoretical coordinates of each candidate point in the checker coordinate system are P i ”{p1”, p2”, …, p n ”};

[0033] Perform least squares matching between them and the actual coordinates P i {p1, p2, …, p n} of each candidate point in the image coordinate system to obtain the least squares transformation matrix between the two, and transform the two to the checker coordinate system;

[0034] Calculate the actual coordinates P of each candidate point in the coordinate system of the calibrator i {p1, p2, …, p n} and the theoretical coordinates P i ”{p1”, p2”, …, p n ”}, and the Euclidean distance ε = |P k P k ”| between the corresponding points P k P k ”. Take the average value of the distances of all calculated points as the reconstruction error of the three-dimensional image.

[0035] Each of the coordinate reference points is asymmetrically arranged, with different sizes and all larger than other calibration points.

[0036] The precision calibrator is provided with several calibration points arranged according to a set rule, and the calibration point at the center of all calibration points is used as the first coordinate reference point, the calibration point with a first distance from the first coordinate reference point is used as the second coordinate reference point, and the calibration point with a second distance from the first coordinate reference point is used as the third coordinate reference point; the first distance and the second distance are different.

[0037] Beneficial effects: The present invention divides each calibration point through an adaptive threshold, and at the same time uses a weighted centroid extraction algorithm of calibration points to evaluate the reconstruction accuracy of the three-dimensional image. Compared with the prior art, the evaluation accuracy is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flowchart of the reconstruction accuracy evaluation of the present invention;

[0039] Figure 2 is a schematic diagram of the precision calibrator of the present invention;

[0040] Figure 3 is a schematic diagram of the threshold segmentation result of the present invention;

[0041] Figure 4 is a schematic diagram of the position deviation between the actual steel ball extraction coordinates and the actual steel ball theoretical reconstruction coordinates. DETAILED DESCRIPTION OF THE INVENTION

[0042] The following further clarifies the present invention in conjunction with the accompanying drawings and specific embodiments.

[0043] The specific flowchart of the three-dimensional image reconstruction accuracy evaluation method of the present invention is as Figure 1 shown, and includes the following steps:

[0044] (1) Collect a three-dimensional image containing a precision calibrator;

[0045] The precision calibrator adopted by the present invention is as Figure 2As shown, it is made of a light-transmitting material, with several calibration points arranged according to set rules inside, and at least three asymmetrically arranged coordinate reference points. In the present invention, the calibration point at the center of all calibration points is used as the first coordinate reference point, the calibration point with a first distance from the first coordinate reference point is used as the second coordinate reference point, and the calibration point with a second distance from the first coordinate reference point is used as the third coordinate reference point; the first distance and the second distance are different. By such a design, different coordinate reference points can be distinguished.

[0046] In the present invention, 7 layers of space are set inside the precision calibrator, with 49 calibration points evenly arranged in each layer, totaling 343 calibration points. The Euclidean distance between two adjacent calibration points is 20 mm; among them, the first distance is 40 mm and the second distance is 20 mm.

[0047] More specifically, in the present invention, the calibration points are steel balls, and all steel balls are divided into large steel balls with a diameter of 4 mm and small steel balls with a diameter of 2 mm. Among them, the small steel balls are used as calibration points and the large steel balls are used as coordinate reference points. As Figure 2 shown, one of the large steel balls is located at the center of all steel balls, denoted as the first large steel ball. The second large steel ball 2 is set at a distance of 40 mm from the first large steel ball, and the third large steel ball is set at a distance of 20 mm from the first large steel ball.

[0048] In the present invention, the precision calibrator is made of plexiglass material with good transparency.

[0049] The three-dimensional image of the present invention is obtained by CT acquisition.

[0050] (2) Segmentation to obtain the three-dimensional coordinate data including each candidate point;

[0051] Adaptive threshold segmentation is used to obtain the three-dimensional coordinate data including each candidate point to reduce the interference of impurities in the three-dimensional image on the target segmentation of the coordinate data in the three-dimensional image. As Figure 3 shown, the voxel points greater than the threshold are retained, and the voxel points less than the threshold are deleted;

[0052] (3) Calculate the centroid of each candidate point to obtain the coordinates of each candidate point in the image coordinate system;

[0053] (31) Select the voxel points corresponding to each candidate point as seed points;

[0054] a. Traverse and search the three-dimensional coordinate data obtained by segmentation in step (2), and take the first voxel point obtained as the first seed point and put it into the seed point set;

[0055] b. Traverse the remaining un-traversed voxel points in the three-dimensional coordinate data, and find the voxel point that is the closest to the first seed point and greater than the set threshold as the new seed point and update the seed point set;

[0056] c. Traverse the remaining un-traversed voxel points, and find the voxel point that is the closest to the seed point set and has a distance greater than the set threshold (that is, the distance between this voxel point and the closest seed point in the seed point set is greater than the set threshold) as the new seed point and update the seed point set. Repeat this step until all voxel points are traversed to obtain the final seed point set;

[0057] In the present invention, due to problems such as noise or acquisition angle, the number of seed points in the seed point set obtained in this step may not be consistent with the number of calibration points. Assume that n seed points are obtained here;

[0058] Among them, the selection range of the set threshold in the present invention is greater than the maximum diameter of the calibration points and less than half of the distance between two calibration points, so as to avoid the overlap of voxel points where the seed points envelope each other; in the present invention, it can be known that the set threshold range is [4mm, 10mm];

[0059] (32) According to the final seed point set obtained in step (31), cluster the voxel points according to each seed point, traverse and search the three-dimensional coordinate data obtained by segmentation in step (2), calculate the distances between all voxel points and each seed point in the final seed point set. If the distance between a voxel point and a certain seed point is the closest, then cluster this voxel point as the nearby point of this seed point and use it as the voxel point set included in a candidate point, so that the images of each candidate point can be separated from each other;

[0060] (33) Obtain the voxel point sets included in n candidate points according to the clustering result in step (32), and calculate the centroid of each candidate point accordingly;

[0061]

[0062] Among them, P i represents the centroid of the i-th candidate point, j represents the voxel point index included in the i-th candidate point, m represents the number of voxel points included in the i-th candidate point, ct_value j represents the voxel gray value of the j-th voxel point in the i-th candidate point, pt j represents the coordinate value of the j-th voxel point in the i-th candidate point in the image coordinate system, The sum of the voxel gray values of all voxel points included in the i-th candidate point;

[0063] By the above method of calculating the centroid of the calibration points, the greater the voxel gray value of a certain voxel point in the candidate point, the greater the weight of the calculation of the centroid of this candidate point;

[0064] Thus, the actual coordinate set P of each candidate point in the image coordinate system is obtained i{p1, p2, …, p n}, that is, the actual centroid coordinates of each candidate point in the image coordinate system;

[0065] (4) Extract the coordinates of the coordinate reference points;

[0066] According to the clustering in step (32), the checkpoints with the top three numbers of voxel points included in each candidate point can be obtained, so as to obtain three coordinate reference points. According to the centroid of the three coordinate reference points calculated in step (33), and according to the prior topological relationship of the three coordinate reference points in the precision checker designed in step (1), different coordinate reference points can be distinguished, so as to obtain the coordinates P i {p u , p v , p w} in the image coordinate system;

[0067] In the present invention, the sizes of the three coordinate reference points can also be directly set to be different from each other and all greater than other checkpoints, so that different coordinate reference points can be directly distinguished by the number of voxel points included in each checkpoint;

[0068] (5) Coarse alignment;

[0069] According to the design parameters of the precision checker in step (1), obtain the coordinates P i '{p u ', p v ', p w '} of the three coordinate reference points in the checker coordinate system, and perform least-squares matching with the coordinates P i {p u , p v , p w} of the three coordinate reference points obtained above in the image coordinate system, and solve to obtain the least-squares transformation matrix M between the two, specifically:

[0070] P i '{p u ', p v ', p w '} = M * P i {p u , p v , p w} + δ

[0071] By solving, the transformation matrix M of the two is obtained to make δ the minimum value;

[0072] (6) Fine alignment;

[0073] According to the transformation matrix M obtained in step (5), the actual coordinate set P i{p1, p2, …, p n} is transformed into the coordinate system of the checker, and we get:

[0074] P i ' {p1', p2', …, p n '} = M * P i {p1, p2, …, p n}

[0075] where P i ' {p1', p2', …, p n '} is the actual coordinate set of each candidate point in the coordinate system of the checker;

[0076] Calculate the index of each candidate point and the same point in the coordinate system of the checker corresponding to it, and make a one-to-one correspondence accordingly. Specifically as follows:

[0077] x i = p i '.x / d

[0078] y i = p i '.y / d

[0079] z i = p i '.z / d

[0080] where x i 、y i 、z i are the x-direction index, y-direction index, and z-direction index respectively, p i '.x, p i '.y, p i '.z are the x-coordinate, y-coordinate, and z-coordinate of the candidate point p i ' transformed into the coordinate system of the checker, and d represents the Euclidean distance between adjacent check points;

[0081] Accordingly, calculate the theoretical coordinate set P i ” {p1”, p2”, …, p n ”} of each check point corresponding to each candidate point in the coordinate system of the checker;

[0082] Perform least squares matching on it and the actual coordinate set P i {p1, p2, …, p n} of each candidate point in the image coordinate system, and solve for the least squares transformation matrix K between the two. Specifically:

[0083] P i ” {p1”, p2”, …, p n ”} = K * P i{p1, p2, …, p n}+ δ

[0084] The transformation matrix K of the two is obtained by solving to minimize δ;

[0085] (7) Calculate the three-dimensional image reconstruction error;

[0086] According to the transformation matrix K calculated in step (6), the two are transformed into the checker coordinate system, and the actual coordinate set P i {p1, p2, …, p n} and the theoretical coordinate set P i ”{p1”, p2”, …, p n ”} The reconstruction error between them is specifically:

[0087] Calculate the actual coordinate set P i {p1, p2, …, p n} and the theoretical coordinate set P i ”{p1”, p2”, …, p n ”} The Euclidean distance ε = |P k P k ”| between the corresponding points P k P k ” is calculated, and the average value of the distances of all calculated points is used as the reconstruction error of the three-dimensional image, that is, the reconstruction accuracy of the three-dimensional image, with the unit of mm, as Figure 4 shown.

[0088] The present invention divides each check point through an adaptive threshold, and at the same time uses a weighted centroid extraction algorithm of the check point to evaluate the reconstruction accuracy of the three-dimensional image. Compared with the prior art, the evaluation accuracy is higher.

[0089] The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the specific details in the above embodiments. Within the technical concept scope of the present invention, various equivalent transformations (such as quantity, shape, position, etc.) can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. Three-dimensional image reconstruction accuracy evaluation method, characterized in that: Including the steps: Collect three-dimensional images including a precision checker, in which there are several calibration points arranged according to a set rule, including at least three coordinate reference points; Extract the actual coordinates of each candidate point used to represent each calibration point in the image coordinate system, and at the same time obtain the actual coordinates of each coordinate reference point among them, specifically including: Use adaptive threshold segmentation to obtain three-dimensional coordinate data including each candidate point, obtain the voxel points corresponding to each candidate point as seed points, traverse and search the three-dimensional coordinate data, calculate the distances between all voxel points and all seed points. If the distance between a voxel point and a certain seed point is the closest, then cluster the voxel point as the nearby point of the seed point and use it as the voxel point set included in a candidate point, and calculate the centroid of each candidate point according to the number of voxel points, the coordinate values of each voxel point, and the voxel gray value in the voxel point set included in a certain candidate point; Obtain the coordinates of each coordinate reference point in the checker coordinate system according to the design parameters of the precision checker, and calculate the transformation matrix between the two by combining the actual coordinates of each coordinate reference point; Transform the actual coordinates of each candidate point to the checker coordinate system according to the transformation matrix, and correspond them one by one with the corresponding calibration points in the checker, so as to obtain the theoretical coordinates of each candidate point in the checker coordinate system; Calculate the three-dimensional reconstruction accuracy according to the actual coordinates and theoretical coordinates of each candidate point.

2. The three-dimensional image reconstruction accuracy evaluation method according to claim 1, characterized in that: The specific operation of obtaining the voxel points corresponding to each candidate point as seed points is: Traverse and search the segmented three-dimensional coordinate data including each candidate point, and take the first obtained voxel point as the first seed point and put it into the seed point set; Traverse the remaining un-traversed voxel points in the three-dimensional coordinate data, and find the voxel point that is the closest to the first seed point and greater than the set threshold as the new seed point and update the seed point set; Traverse the remaining un-traversed voxel points, and find the voxel point that is greater than the set threshold from the distance to the closest seed point in the seed point set as the new seed point and update the seed point set. Repeat this step until all voxel points are traversed to obtain the final seed point set.

3. The three-dimensional image reconstruction accuracy evaluation method according to claim 2, wherein: The selection range of the set threshold is set to be greater than the maximum diameter of the calibration points and less than half of the distance between two calibration points.

4. The three-dimensional image reconstruction accuracy evaluation method according to claim 1, characterized in that: Obtain the coordinates of each coordinate reference point in the checker coordinate system according to the design parameters of the precision checker, and calculate the transformation matrix between the two by combining the actual coordinates of each coordinate reference point, specifically: Obtain the numbers in the image, and obtain the coordinates P of the three coordinate reference points in the image coordinate system according to the design parameters of the precision checker i {p u , p v , p w}; Obtain the coordinates P of the three coordinate reference points in the coordinate system of the calibrator according to the design parameters of the precision calibrator i '{p u ',p v ',p w '}, and perform least squares matching with the coordinates P of the three coordinate reference points obtained above in the image coordinate system i {p u ,p v ,p w}, and solve for the least squares transformation matrix M between the two, specifically: P i '{p u ',p v ',p w '} = M * P i {p u ,p v ,p w}+δ Solve to obtain the transformation matrix M between the two so that δ is the minimum value.

5. The three-dimensional image reconstruction accuracy evaluation method according to claim 1, characterized in that: The specific operation of transforming the actual coordinates of each candidate point to the checker coordinate system according to the transformation matrix and corresponding them one by one with the corresponding calibration points in the checker is: x i = p i '.x / d y i = p i '.y / d z i = p i '.z / d where x i , y i , z i are the x-direction index, y-direction index, and z-direction index respectively, and p i '.x, p i '.y, p i '.z are the x-coordinate, y-coordinate, and z-coordinate of the candidate point p i ' transformed to the checker coordinate system, and d represents the Euclidean distance between adjacent check points.

6. The three-dimensional image reconstruction accuracy evaluation method according to claim 1, characterized in that: The specific operation of calculating the three-dimensional reconstruction accuracy according to the actual coordinates and theoretical coordinates of each candidate point is: The theoretical coordinates of each candidate point obtained in the calibrator coordinate system are P i ”{p1”, p2”, …, p n ”}; Compare it with the actual coordinates P of each candidate point in the image coordinate system i {p1, p2, …, p n} to perform least squares matching, solve to obtain the least squares transformation matrix between the two, and transform the two to the checker coordinate system; Calculate the actual coordinates P of each candidate point in the verifier coordinate system i {p1,p2,…,p n } and theoretical coordinates P i ”{p1”,p2”,…,p n ”} corresponding point P k and P k "The Euclidean distance ε=|P k P k ”|, the average distance of all points calculated is used as the reconstruction error of the 3D image.

7. The three-dimensional image reconstruction accuracy evaluation method according to claim 1, characterized in that: Each of the coordinate reference points is arranged asymmetrically, with different sizes and all larger than other calibration points.

8. The three-dimensional image reconstruction accuracy evaluation method according to claim 1, wherein: The precision checker is provided with a number of check points arranged according to a set rule, and the check point at the center of all check points is used as the first coordinate reference point, the check point with a first distance from the first coordinate reference point is used as the second coordinate reference point, and the check point with a second distance from the first coordinate reference point is used as the third coordinate reference point; the first distance and the second distance are different.

Images