阅读说明：本技术 一种基于形状先验的肾脏图像分割方法 (Shape prior-based kidney image segmentation method ) 是由 吴梦麟 章世平 于 2021-07-28 设计创作，主要内容包括：本发明公开了一种基于形状先验的肾脏图像分割方法,包括如下步骤：S1、导入包含肾脏的人体CT序列图像,用户选择肾脏切片的起始位置、中间位置和结束位置的肾脏中心点；S2、对肾脏CT切片进行预处理,所述预处理包括去床、去噪、增强；S3、基于步骤S1结果,利用主动形状模型进行模板训练,最后进行初始分割,将分割的结果作为初始分割轮廓；S4、基于步骤S3的结果,将初始分割轮廓作为水平集方法的起始轮廓,进行进一步分割；S5、基于步骤S4的分割结果,利用CT序列图像的前后连续属性进行后处理,对分割结果进行修正。本发明通过少量的肾脏CT序列图像训练生成形状模板,从而对肾脏图像实现合理的分割,提高分割精度。(The invention discloses a kidney image segmentation method based on shape prior, which comprises the following steps: s1, importing a human body CT sequence image containing the kidney, and selecting the kidney center points of the initial position, the middle position and the end position of the kidney slice by a user; s2, preprocessing the kidney CT slice, wherein the preprocessing comprises bed removing, denoising and enhancing; s3, based on the result of the step S1, carrying out template training by using the active shape model, and finally carrying out initial segmentation, wherein the segmentation result is used as an initial segmentation contour; s4, based on the result of S3, further dividing the initial dividing contour as the initial contour of the level set method; s5, based on the segmentation result of the step S4, post-processing is carried out by utilizing the front and back continuous attributes of the CT sequence image, and the segmentation result is corrected. The shape template is generated through a small amount of kidney CT sequence image training, so that the kidney image is reasonably segmented, and the segmentation precision is improved.)

1. A kidney image segmentation method based on shape prior is characterized by comprising the following steps: the method comprises the following steps:

s1, importing a human body CT sequence image containing the kidney, and selecting the kidney center points of the initial position, the middle position and the end position of the kidney slice by a user;

s2, preprocessing the kidney CT slice, wherein the preprocessing comprises bed removing, denoising and enhancing;

s3, based on the result of the step S1, carrying out template training by using the active shape model, and finally carrying out initial segmentation, wherein the segmentation result is used as an initial segmentation contour;

s4, based on the result of S3, further dividing the initial dividing contour as the initial contour of the level set method;

s5, based on the segmentation result of the step S4, post-processing is carried out by utilizing the front and back continuous attributes of the CT sequence image, and the segmentation result is corrected.

2. The shape prior based kidney image segmentation method of claim 1, wherein: the step S1 includes the following sub-steps:

s1.1, according to the selection of a user on the initial position and the end position of a kidney slice, obtaining a CT sequence containing a kidney, and compressing the kidney sequence into 20 frames through interpolation calculation, wherein the interpolation calculation is a two-dimensional or three-dimensional algorithm;

s1.2, defining a 1 st frame as a starting frame, a 20 th frame as an ending frame and a 10 th frame as a sequence intermediate frame, and selecting a central point of a target kidney on images of the starting frame, the intermediate frame and the ending frame by a user;

s1.3, carrying out interpolation according to the central point coordinates of the initial frame and the intermediate frame to obtain the central point coordinates of the 2-9 frames, and similarly, obtaining the central point coordinates of the 11-19 frames according to the interpolation of the intermediate frame and the ending frame.

3. The shape prior based kidney image segmentation method of claim 1, wherein: in step S3, the template training includes the following sub-steps:

s3.1, extracting n groups of kidney contour annotation points to form 2 n-dimensional vector Z ═ Z (Z)₁,Z₂,...,Z_n) Calculating the mean value mu of the contour points, and carrying out data standardization on Z to obtain

Find outIs determined by the eigenvalue of the covariance matrix of (1) { lambda }₁,...,λ_nSorting the eigenvectors according to the magnitude of the eigenvalues, taking the first k rows to generate a matrix P, and expressing any kidney contour as

X＝μ+Pb (1)

4. The shape prior based kidney image segmentation method of claim 1, wherein: in step S3, the segmentation includes the following sub-steps:

s3.2, enabling b to be 0, calculating a model contour X according to a formula 1, and moving the center of the contour to the center point of each frame of CT sequence kidney;

s3.3, searching contour points matched with the gray distribution along the normal distance of the points on the model contour X to form a target contour Y;

s3.4, according to the puroks transformation, aligning X to Y to obtain formula 2, i.e., Y ═ MX, where M is a transfer matrix that includes translation, rotation, and scaling;

s3.5, calculation formulaThereby projecting Y into the space of X by inverse transformation;

s3.6, inverse transformation of equation 1Updating the value of b while limiting the deformation, with a constraint | b_i|＜3λ_i；

And S3.7, repeating the steps 2.3-2.6 until convergence or the maximum execution times is reached, and obtaining an initial kidney contour preliminary segmentation result.

5. The shape prior based kidney image segmentation method of claim 1, wherein: the step S1 includes the following sub-steps:

s4.1, initializing a level set method by using the initial segmentation result of the kidney contour obtained in the step S3;

s4.2, iterating according to a formula 4, and calculating the evolution of a level set equation;

where phi denotes the level set function, delta denotes the laplacian,expressing a gradient operator, delta expressing a Dirichlet equation, v expressing a coefficient for controlling the expansion or contraction of the profile, and div expressing divergence calculation;representing the image edge weight, wherein I represents the current image, and constants gamma and eta represent the weights of a first term and a second term on the right side of a fourth equal sign of a formula respectively;

and S4.3, setting and limiting the evolution iteration times of the level set, wherein no more than 50 iterations can be set according to experience, and a zero level set equation phi which is 0 is the outline of the kidney segmentation.

6. The shape prior based kidney image segmentation method of claim 1, wherein: the step S5 includes the following sub-steps:

s5.1, on the basis of the segmentation result of the step S4, reserving the segmentation region with the largest area, deleting the rest small regions, and processing by using a connected domain;

s5.2, according to the assumption of continuity of the CT sequence, calculating the intersection ratio of the segmentation results of the two frames before and after, and if the ratio is too small, smoothing the segmentation results.

Technical Field

The invention belongs to the technical field of image processing, and particularly relates to a kidney image segmentation method based on shape prior.

Background

The kidney CT sequence image segmentation is an important premise for carrying out quantitative analysis on kidney diseases, has a supporting effect on disease diagnosis and treatment, and is also an important means for three-dimensional visualization, virtual surgery and biological system simulation of organ tissue images in computer-aided diagnosis. On the other hand, the kidney tissue structure is complex and the shape is diverse, and the shape of the kidney is greatly changed due to related diseases, so that the automatic segmentation of the kidney image becomes a difficulty in medical image segmentation.

Although the method based on deep learning is widely applied to the field of medical image segmentation in recent years, model training requires a large number of labeled samples, gray distribution differences among different devices, and interpretability problems, which are still key factors for restricting the deep learning method in clinical application. The traditional level set method based on the active contour drives the contour curve to gradually converge on the target boundary under the action of the energy function of the contour curve, can automatically process the topological change of the curve, and provides a good algorithm basis for segmenting the kidney CT sequence with complex topological shape and weak boundary characteristics. However, the segmentation accuracy of the level set method is affected by non-uniform gray levels of kidney tissues, close gray levels of organs, connectivity of adjacent organs and susceptibility to change of lesion morphology.

Disclosure of Invention

The present invention aims to solve the above problems and provide a novel kidney image segmentation method based on shape prior. The shape template is generated through a small amount of kidney CT sequence image training, so that reasonable segmentation is realized on the kidney image, and the segmentation precision is improved.

In order to achieve the purpose, the technical scheme of the invention is as follows:

a kidney image segmentation method based on shape prior comprises the following steps:

s1, importing a human body CT sequence image containing the kidney, and selecting the kidney center points of the initial position, the middle position and the end position of the kidney slice by a user;

s2, preprocessing the kidney CT slice, wherein the preprocessing comprises bed removing, denoising and enhancing;

s3, based on the result of the step S1, carrying out template training by using the active shape model, and finally carrying out initial segmentation, wherein the segmentation result is used as an initial segmentation contour;

s4, based on the result of S3, further dividing the initial dividing contour as the initial contour of the level set method;

s5, based on the segmentation result of the step S4, post-processing is carried out by utilizing the front and back continuous attributes of the CT sequence image, and the segmentation result is corrected.

As an improvement to the above technical solution, the step S1 includes the following sub-steps:

s1.1, according to the selection of a user on the initial position and the end position of the kidney slice, a CT sequence containing the kidney is obtained, and the kidney sequence is compressed into 20 frames through interpolation calculation, wherein the interpolation calculation is not limited to a two-dimensional or three-dimensional algorithm.

S1.2, defining a 1 st frame as a starting frame, a 20 th frame as an ending frame and a 10 th frame as a sequence intermediate frame, and selecting the central point of a target kidney on the images of the starting frame, the intermediate frame and the ending frame by a user.

S1.3, carrying out interpolation according to the central point coordinates of the initial frame and the intermediate frame to obtain the central point coordinates of the 2-9 frames, and similarly, obtaining the central point coordinates of the 11-19 frames according to the interpolation of the intermediate frame and the ending frame.

As an improvement to the above technical solution, in the step S3, the template training includes the following sub-steps:

s3.1, extracting n groups of kidney contour annotation points to form 2 n-dimensional vector Z ═ Z (Z)₁,Z₂,...,Z_n) Calculating the mean value mu of the contour points, and carrying out data standardization on Z to obtain

Find outIs determined by the eigenvalue of the covariance matrix of (1) { lambda }₁,...,λ_nSorting the eigenvectors according to the magnitude of the eigenvalues, taking the first k rows to generate a matrix P, and expressing any kidney contour as

X＝μ+Pb (1)

As an improvement to the above technical solution, in step S3, the dividing includes the following sub-steps:

s3.2, making b equal to 0, calculating a model contour X according to the formula 1, and moving the center of the contour to the central point of the kidney of each frame of CT sequence

And S3.3, searching contour points (shown in figure 2) matched with the gray distribution along the normal distance of the points on the model contour X to form a target contour Y.

S3.4, according to the puroks transformation, align X to Y to obtain formula 2, i.e., Y ═ MX, where M is a transfer matrix that includes translation, rotation, and scaling.

S3.5, calculation formula 3Thereby projecting Y into the space of X by an inverse transform.

S3.6, inverse transformation of equation 1Updating the value of b while limiting the deformation, with a constraint | b_i|＜3λ_i。

And S3.7, repeating the steps 2.3-2.6 until convergence or the maximum execution times is reached, and obtaining an initial kidney contour preliminary segmentation result.

As an improvement to the above technical solution, the step S1 includes the following sub-steps:

and S4.1, initializing a level set method by using the preliminary segmentation result of the kidney contour obtained in the step S3.

And S4.2, iterating according to a formula 4, and calculating the evolution of the level set equation.

Where phi denotes the level set function, delta denotes the laplacian,represents the gradient operator, δ represents the dirichlet equation, v represents the coefficient that controls the expansion or contraction of the profile, and div represents the divergence calculation.Representing the image edge weights, where I represents the current image. Constants γ and η represent the weights of the first term and the second term to the right of the equation four equal sign, respectively.

And S4.3, setting and limiting the evolution iteration number of the level set, wherein the iteration number is not more than 50 times according to experience, but is not limited to the empirical value. Wherein, the zero level set equation phi is 0, which is the contour of the kidney segmentation. The iteration number is limited, so that the segmentation result of the level set method does not deviate from the outline of the kidney represented by the active shape model too far, and the shape of the kidney is maintained.

As an improvement to the above technical solution, the step S5 includes the following sub-steps:

s5.1, on the basis of the result of the division in step S4, the divided region with the largest area is retained, and the remaining small regions are deleted.

S5.2, according to the assumption of continuity of the CT sequence, calculating the intersection ratio of the segmentation results of the two frames before and after, and if the ratio is too small, smoothing the segmentation results.

Compared with the prior art, the invention has the advantages and positive effects that:

the invention provides a novel kidney image segmentation method based on shape prior by generating a shape template through a small amount of kidney CT sequence image training, solves the problem of automatic segmentation of a kidney image and improves the segmentation precision.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.

FIG. 1 is a schematic flow chart of the present invention.

Fig. 2 is a schematic diagram of the target contour search of the present invention, which searches for target contour points matching with the gray distribution thereof in the normal vector direction through model contour points.

Fig. 3 is a diagram illustrating the result of kidney segmentation according to the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived from the embodiments of the present invention by a person skilled in the art without any creative effort, should be included in the protection scope of the present invention.

As shown in fig. 1 to 3, the present embodiment discloses a kidney image segmentation method based on shape prior, which includes the following steps:

s1, importing a human body CT sequence image containing the kidney, and selecting the kidney center points of the initial position, the middle position and the end position of the kidney slice by a user;

s2, preprocessing the kidney CT slice, wherein the preprocessing comprises bed removing, denoising and enhancing;

s3, based on the result of the step S1, carrying out template training by using the active shape model, and finally carrying out initial segmentation, wherein the segmentation result is used as an initial segmentation contour;

s4, based on the result of S3, further dividing the initial dividing contour as the initial contour of the level set method;

s5, based on the segmentation result of the step S4, post-processing is carried out by utilizing the front and back continuous attributes of the CT sequence image, and the segmentation result is corrected.

The step S1 includes the following sub-steps:

s1.1, according to the selection of a user on the initial position and the end position of the kidney slice, a CT sequence containing the kidney is obtained, and the kidney sequence is compressed into 20 frames through interpolation calculation, wherein the interpolation calculation is not limited to a two-dimensional or three-dimensional algorithm.

S1.2, defining a 1 st frame as a starting frame, a 20 th frame as an ending frame and a 10 th frame as a sequence intermediate frame, and selecting the central point of a target kidney on the images of the starting frame, the intermediate frame and the ending frame by a user.

S1.3, carrying out interpolation according to the central point coordinates of the initial frame and the intermediate frame to obtain the central point coordinates of the 2-9 frames, and similarly, obtaining the central point coordinates of the 11-19 frames according to the interpolation of the intermediate frame and the ending frame.

In step S3, the template training includes the following sub-steps:

s3.1, extracting n groups of kidney contour annotation points to form 2 n-dimensional vector Z ═ Z (Z)₁,Z₂,...,Z_n) Calculating the mean value mu of the contour points, and carrying out data standardization on Z to obtain

Find outIs determined by the eigenvalue of the covariance matrix of (1) { lambda }₁,...,λ_nAnd (4) sorting the eigenvectors according to the magnitude of the eigenvalues, and taking the first k rows to generate a matrix P, so that any kidney contour can be expressed as:

X＝μ+Pb (1)

in step S3, the segmentation includes the following sub-steps:

s3.2, making b equal to 0, calculating a model contour X according to the formula 1, and moving the center of the contour to the central point of the kidney of each frame of CT sequence

And S3.3, searching contour points (shown in figure 2) matched with the gray distribution along the normal distance of the points on the model contour X to form a target contour Y.

S3.4, according to the puroks transformation, align X to Y to obtain formula 2, i.e., Y ═ MX, where M is a transfer matrix that includes translation, rotation, and scaling.

S3.5, calculation formula 3Thereby projecting Y into the space of X by an inverse transform.

S3.6, inverse transformation of equation 1Updating the value of b while limiting the deformation, with a constraint | b_i|＜3λ_i。

And S3.7, repeating the steps 2.3-2.6 until convergence or the maximum execution times is reached, and obtaining an initial kidney contour preliminary segmentation result.

The step S1 includes the following sub-steps:

and S4.1, initializing a level set method by using the preliminary segmentation result of the kidney contour obtained in the step S3.

And S4.2, iterating according to a formula 4, and calculating the evolution of the level set equation.

Where phi denotes the level set function, delta denotes the laplacian,represents the gradient operator, δ represents the dirichlet equation, v represents the coefficient that controls the expansion or contraction of the profile, and div represents the divergence calculation.Representing the image edge weights, where I represents the current image. Constants γ and η represent the weights of the first term and the second term to the right of the equation four equal sign, respectively.

And S4.3, setting and limiting the evolution iteration number of the level set, wherein the iteration number is not more than 50 times according to experience, but is not limited to the empirical value. Wherein, the zero level set equation phi is 0, which is the contour of the kidney segmentation. The iteration number is limited, so that the segmentation result of the level set method does not deviate from the outline of the kidney represented by the active shape model too far, and the shape of the kidney is maintained.

The step S5 includes the following sub-steps:

s5.1, on the basis of the result of the division in step S4, the divided region with the largest area is retained, and the remaining small regions are deleted.

S5.2, according to the assumption of continuity of the CT sequence, calculating the intersection ratio of the segmentation results of the two frames before and after, and if the ratio is too small, smoothing the segmentation results.

The invention provides a novel kidney image segmentation method based on shape prior by generating a shape template through a small amount of kidney CT sequence image training, and solves the automatic segmentation problem of the kidney image.