CN102402788A - Method for segmenting three-dimensional ultrasonic image - Google Patents

Abstract

The invention discloses a method for segmenting a three-dimensional ultrasonic image, belonging to the technical field of the digital image processing. The method comprises the following steps: (1) preprocessing spots of the three-dimensional ultrasonic image by adopting a normalized anisotropic diffusion method of a three-dimensional wavelet according to the characteristics of the three-dimensional ultrasonic image to remove spot noise; (2) initializing the preprocessed three-dimensional ultrasonic image by adopting a Canny edge detection operator; and (3) segmenting the three-dimensional ultrasonic image three-dimensionally by using a B-Surface and GVF Snake based three-dimensional deformation model. The method disclosed by the invention can be used to rapidly and accurately segment the three-dimensional ultrasonic image and particularly has strong noise robustness. The method for automatically segmenting the three-dimensional ultrasonic image can be also used for segmenting other three-dimensional images such as CT (computed tomography) images, MRI (magnetic resonance images) and PET (position-emission tomography) images, thereby having high application value.

Description

A kind of dividing method of three-dimensional ultrasound pattern

Technical field

The invention belongs to the computer digital image process field, particularly a kind of three-dimensional ultrasound pattern cuts apart.

Background technology

Ultrasonic imaging is the ultrasound wave that utilizes specific wavelength; The technology that after surveying body, produces backscatter signals and form images, it has safe, advantages such as image taking speed is fast, easy to operate and low cost, and the application in clinical is very extensive; Especially abdominal organs detects; Like heart, liver, kidney etc., and three-dimensional imaging of fetus and realtime four-dimensional dynamic imaging etc., it is generaI investigation and one of most important means of clinical diagnosis.In addition, 3-D supersonic imaging technology also is widely used in the monitoring of animal reproduction, growth course, like the dynamic evaluation of early pregnancy detection, pathology examination, embryo's growth and development process etc.

In image quantization analysis, identification and computer-aided diagnosis and therapeutic process; Cutting apart of histoorgan, lesion region and tumour is very important link, and cutting apart of target or object is the basis of volume calculation, shape analysis, feature extraction, quantification diagnosis, motion analysis and tracking.

The dividing method of ultrasonoscopy, the space dimensionality angle from cutting apart can be divided into the dividing method based on the two-dimensional sequence image, and directly based on three-dimensional three-dimensional segmentation method.Wherein based on the disposal route of two-dimensional space, there is following shortcoming in they: (1) does not utilize three-dimensional spatial information; (2) two-dimensional silhouette of obtaining need be carried out topology at three dimensions and connected and interpolation, could rebuild degree of precision and complete 3-D view; (3) when volume calculated, two-dimensional silhouette at the interpolation arithmetic that three dimensions carries out, is often produced than mistake.And the three-dimensional segmentation method can overcome above deficiency, and this method can be divided three classes again, and (1) is based on the method for probability statistics model; (2) three-dimensional geometry deformation model; (3) method that combines with statistical model of geometric deformation model.Wherein,,, combined high-rise target priori again, therefore, when the complicated medical image of Processing Structure, demonstrated its unique advantages and adaptability widely owing to both utilized the image data information of bottom based on the dividing method of deformation model.But because the characteristic of ultrasonic imaging, the contrast and the resolution of ultrasonoscopy are lower, and organizational boundary and details are blured, especially wherein had a large amount of spots, make the very difficulty that becomes accurately cutting apart of three-dimensional ultrasound pattern.

Summary of the invention

The objective of the invention is to deficiency, a kind of dividing method of three-dimensional ultrasound pattern is provided, quickly and accurately ultrasonoscopy is implemented to cut apart to prior art.

The object of the invention is realized through following technical scheme:

A kind of dividing method of three-dimensional ultrasound pattern may further comprise the steps:

1) three-dimensional ultrasound pattern is carried out the spot pre-service, adopt normalization small echo anisotropy method of diffusion to remove speckle noise;

2) use the Canny edge detection operator image after pre-service is carried out initialization;

3) utilization is implemented three-dimensional segmentation based on the three-dimensional deformation model of B-Surface and GVF Snake.

Spot preprocess method in the step 1) improves picture quality in order to level and smooth spot, helps follow-up contour surface initialization and enforcement and cuts apart;

Specifically comprise the steps:

(1) ultrasonoscopy being carried out multi-scale wavelet and decompose, is high frequency and low frequency part with picture breakdown;

(2) according to the property taken advantage of spot model, to HFS, use normalization small echo anisotropy method of diffusion and handle, suppress spot;

(3) carry out wavelet inverse transformation, the image after reconstruct obtains denoising.

Said ultrasonoscopy is decomposed into 8 parts: HLL, HLH, HHL, HHH through 3 D wavelet; LHH, LHL, LLH, LLL, wherein LLL is a low frequency part, other is a HFS.

The property taken advantage of spot model in the step (2) is:

J(x，y，z)＝I(x，y，z)×n(x，y，z) (1)

Wherein (x, y z) represent original noise-free picture to I, and (x, y are to contain noise n (x, y, image z) z) to J.

The model of normalization small echo anisotropy diffusion is:

$\left\{\begin{array}{c}\frac{\∂J(x,y,z,t)}{\∂t}=\mathrm{div}[\tilde{c}(x,y,z,t)\▿J(x,y,z,t)]\\ J(x,y,z,0)=J_0(x,y,z)\end{array}\right.---\left(2\right)$

Wherein

is gradient of image and gray scale; Div representes divergence (divergence); is the normalization coefficient of diffusion, and it depends on gradient of image and gray scale, in the less even matter zone of gradient; Its value is bigger; With the stronger level and smooth inhibition noise of enforcement, otherwise, in the bigger zone of gradient; Organizational boundary or edge often; This moment, diffusion should have smaller value, thereby kept them as far as possible, and it is defined as:

$c\left(\right|\widetilde{\▿}g\left|\right)=1-\exp (-3.315/{\left(\right|\▿g|/\mathrm{kg}(x,y,z;t))}^4)---\left(3\right)$

Wherein k is a controlled variable, and its value is [0 1], and t representes iteration time or number of times.

Step 2) initialization in comprises the steps:

(1) to each tomography, utilizes the Canny edge detection operator, extract image outline through pretreated three-dimensional ultrasound pattern;

(2) near the center of object to be split, confirm an approximate central point by manual work;

(3) point from the center is being in the circumference in the center of circle with the central point, defines some equally distributed rays; Scan from inside to outside, ask the first intersection point of ray and profile, after all ray scannings finish; Obtain a series of intersection points; With these intersection points is the reference mark, can generate the B-Surface three-dimensional surface, as initial three-dimensional surface.

Three-dimensional segmentation in the step 3) is to adopt three-dimensional GVF Snake; From initial three-dimensional surface, drive above-mentioned B-surface reference mark through external force and move, thereby make the three-dimensional surface of object produce deformation; Approach real subject surface gradually, finally realize cutting apart of three dimensional object.

The external force of three-dimensional segmentation adopts the GVF model, and deformation model is B-Surface, and wherein the gradient vector flow model is:

$E_{\mathrm{gvf}}\left(v\right)=\&Integral;\&Integral;\mathrm{\&mu;}(s_x^2+{\mathrm{<figure-callout\; id="2"\; label="s\; y"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">s</figure-callout>}}_y^{}+t_x^2+t_y^2)+{|\&dtri;f|}^2{|v-\mathrm{\&Delta;f}|}^2\mathrm{dxdy}---\left(4\right)$

Wherein, v (x, y)=(s (x, y), t (x, y)), expression gradient vector, s _x, s _y, t _x, t _yBe respectively s (x, y) with t (x, y) at the local derviation of x and y direction,

Be shade of gray, || be the mould of vector, μ is that value is 0～1 controlled variable;

The B-Surface model is:

$p(u,v)=\underset{i=0}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{N}_{i,p}\left(u\right)N_{j,q}\left(v\right)B_{i,j}---\left(5\right)$

N wherein _{I, p}(u) and N _{J, q}(v) being respectively exponent number is the B-spline basis function of p and q, and u and v span are [0 1], B _{I, j}Be the reference mark, m, n are respectively row and column direction grid number;

Cubic B-spline can be described below:

$p(u,v)={\mathrm{UM}}_{R}Q{M}_R^T{V}^T---\left(6\right)$

Wherein

$M_R=\frac{1}{6}\left[\begin{array}{cccc}-1& 3& -3& 1\\ 3& -6& 3& 0\\ -3& 0& 3& 0\\ 1& 4& 1& 0\end{array}\right]---\left(8\right)$

Q is the reference mark matrix:

$Q=\left[\begin{array}{cccc}{Q}_{\mathrm{0,0}}& Q_{\mathrm{0,1}}& Q_{\mathrm{0,2}}& Q_{\mathrm{0,3}}\\ {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,0}}& {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,1}}& {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,2}}& {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,3}}\\ {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,0}}& {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,1}}& {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,2}}& {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,3}}\\ Q_{\mathrm{3,0}}& Q_{\mathrm{3,1}}& Q_{\mathrm{3,2}}& Q_{\mathrm{3,3}}\end{array}\right]---\left(9\right)$

Move and surface deformation at the reference mark in the said cutting procedure, accomplishes through following iteration:

Q ^t+1＝Q ^t+ΔQ (10)

$\mathrm{\&Delta;Q}=\mathrm{\&lambda;}{M}_R^{-1}{\left(U^{T}U\right)}^{-1}{F}_{Q}V{\left(V^{T}V\right)}^{-1}{\left(M_R^T\right)}^{-1}---\left(11\right)$

F wherein _QBe external force, depend on the GVF model;

(1) according to the B-Surface model in reference mark and the formula (6), the Calculation of Three Dimensional surface;

(2) according to three-dimensional surface and GVF Model Calculation external force F _Q

(3) generate new reference mark according to formula (10) and (11);

(4), calculate new three-dimensional surface according to formula (6) by new reference mark;

(5) if iterations reaches setting value, then stop, otherwise change step (2).

Compared with prior art, the present invention has following advantage:

1) three-dimensional ultrasound pattern carries out pre-service through the diffusion of 3 D wavelet anisotropy, and spot is by level and smooth effectively;

2) adopt based on the ray intersection algorithm under Canny edge detection operator and the polar coordinates, carry out the contour surface initialization, initial surface helps cutting apart fast and accurately near real subject surface;

3) pre-service is directly carried out at three dimensions with cutting apart all, can make full use of the space structure information of object on the one hand, and on the other hand, segmentation result is the three-dimensional surface of object, does not need the profile concatenation of two-dimentional dividing method.

4) B-Surface includes smoothness properties, does not need internal force, to noise tool strong robustness;

5) because B-Surface only needs a spot of reference mark, can generate the complex three-dimensional surface, therefore, it can save memory cost, can improve splitting speed and precision again.

Description of drawings

Fig. 1 is the process flow diagram of a kind of three-dimensional ultrasound pattern dividing method of the present invention;

Fig. 2 is a normalization small echo anisotropy method of diffusion process flow diagram of the present invention;

Fig. 3 is the 3 D wavelet exploded view, and 3-D view is broken down into 8 parts, HLL, HLH, HHL, HHH; LHH, LHL, LLH, LLL, wherein LLL is a low frequency part, other is for high frequency or contain HFS;

Fig. 4 is an embodiment of the invention heart three-dimensional ultrasound pattern and through the pretreated two-dimentional tangent plane of denoising; Fig. 4 (a) (b) is respectively original ultrasonic spot image and the result after small echo anisotropy DIFFUSION TREATMENT.

Fig. 5 among the present invention based on the result of the profile initial method of Canny edge detection operator and ray intersection algorithm.Fig. 5 (a) is certain two-dimentional tangent plane result after pre-service, and 5 (b) are the left ventricle zone, and 5 (c) are the initialization point, and they are as the initial control point of B-Surface deformation model.

Fig. 6 is the left ventricle three-dimensional surface of embodiment of the invention heart three-dimensional ultrasound pattern after the B-Surface deformation model is cut apart.

Embodiment

Below in conjunction with embodiment and accompanying drawing technical scheme of the present invention is done further to describe, but embodiment of the present invention is not limited thereto.

Like Fig. 1, shown in 2, the present invention includes following steps:

1) the pending three-dimensional ultrasound pattern of input;

2) image being carried out 3 D wavelet decomposes;

Three-dimensional ultrasound pattern is decomposed into HLL as shown in Figure 3, HLH, HHL, HHH; LHH, LHL, LLH, LLL 8 parts, wherein LLL is a low frequency part, other is a HFS.

3) use normalization anisotropy diffusion couple HFS and handle, suppress speckle noise;

Use normalization anisotropy diffusion couple HFS, i.e. subband HLL, HLH, HHL, HHH; LHH, LHL, LLH 7 parts are carried out smoothing processing.

The spot multiplicative noise model that the present invention considers is following:

J(x，y，z)＝I(x，y，z)×n(x，y，z) (1)

Wherein (x, y z) represent original (nothing is made an uproar) image to I, and (x, y are to contain noise n (x, y, image z) z) to J.

Consider the property the taken advantage of characteristic of ultrasonic spot, in even matter zone and nonuniformity zone, the calculating of gradient of image and gray scale is existed than big-difference by the influence of spot, thereby traditional anisotropy method of diffusion, can not remove spot effectively.In order to eliminate the influence of this otherness to shade of gray, the present invention adopts the normalization coefficient of diffusion to be:

$c\left(\right|\left|\widetilde{\&dtri;}g\right|\left|\right)=1-\exp (-3.315/{\left(\right|\left|\widetilde{\&dtri;}g\right||/\mathrm{kg}(x,y,z;t))}^4)---\left(2\right)$

Wherein expression normalization computing:

$\widetilde{\&dtri;}g=\mathrm{\&Delta;g}/g---\left(3\right)$

Then the normalization method of diffusion can be described below:

$\left\{\begin{array}{c}\frac{\&PartialD;J(x,y,z,t)}{\&PartialD;t}=\mathrm{div}\left[\tilde{c}(x,y,z,t)\right)\&dtri;J(x,y,z,t)]\\ J(x,y,z,0)=J_0(x,y,z)\end{array}\right.---\left(4\right)$

Utilize the inverse wavelet transform reconstructed image.

To the HFS after the normalization DIFFUSION TREATMENT shown in step 3) Chinese style (2) and (4),, use the inverse wavelet transform reconstructed image with low frequency part (like the LLL among Fig. 3).

Fig. 4 is through pretreated two-dimentional tangent plane, and visible most of spot is by level and smooth.

4) three-dimensional surface initialization

Initialization is carried out according to the following steps:

(1) to each tomography, utilizes the Canny edge detection operator, extract image outline through pretreated three-dimensional ultrasound pattern;

(2) near the center of object to be split, confirm an approximate central point by manual work;

(3) point from the center is being in the circumference in the center of circle with the central point, defines some equally distributed rays, scans from inside to outside, asks the first intersection point of ray and profile, after all ray scannings finish, obtains a series of intersection points; With these intersection points is the reference mark, can generate the B-Surface three-dimensional surface, as initial surface.

Fig. 5 is the initial profile and the CCP of certain two-dimentional tangent plane.

5) three-dimensional segmentation

From initial surface, utilize GVF Snake model based on B-Surface, drive above-mentioned B-surface reference mark through external force and move, thereby make the three-dimensional surface of object produce deformation, make three-dimensional surface converge to the real surface of target.

The external force of three-dimensional segmentation adopts the gradient vector flow model, and (gradient vector flow, GVF), deformation model is B-Surface, wherein

Gradient vector flow model (GVF) is:

$E_{\mathrm{gvf}}\left(v\right)=\&Integral;\&Integral;\mathrm{\&mu;}(s_x^2+s_{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">y</figure-callout>}}^2+t_x^2+t_y^2)+{|\&dtri;f|}^2{|v-\&dtri;f|}^2\mathrm{dxdy}---\left(5\right)$

Wherein, v (x, y)=(s (x, y), t (x, y)), expression gradient vector, s _x, s _y, t _x, t _yBe respectively s (x, y) with t (x, y) at the local derviation of x and y direction, Be shade of gray, || be the mould of vector, μ is that value is 0～1 controlled variable;

The B-Surface model is:

$p(u,v)=\underset{i=0}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{N}_{i,p}\left(u\right)N_{j,q}\left(v\right)B_{i,j}---\left(6\right)$

N wherein _{I, p}(u) and N _{J, q}(v) being respectively exponent number is the B-spline basis function of p and q, and u and v span are [0 1], B _{I, j}Be the reference mark, m, n are respectively row and column direction grid number;

Cubic B-spline can be described below:

$p(u,v)={\mathrm{UM}}_{R}Q{M}_R^T{V}^T---\left(7\right)$

Wherein

$M_R=\frac{1}{6}\left[\begin{array}{cccc}-1& 3& -3& 1\\ 3& -6& 3& 0\\ -3& 0& 3& 0\\ 1& 4& 1& 0\end{array}\right]---\left(9\right)$

Q is the reference mark matrix:

$Q=\left[\begin{array}{cccc}{Q}_{\mathrm{0,0}}& Q_{\mathrm{0,1}}& Q_{\mathrm{0,2}}& Q_{\mathrm{0,3}}\\ {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,0}}& {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,1}}& {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,2}}& {\mathrm{<figure-callout\; id="1"\; label="Q"\; filenames="HDA0000123702300000021.png,HDA0000123702300000022.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{1,3}}\\ {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,0}}& {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,1}}& {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,2}}& {\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HDA0000123702300000021.png"\; state="\{\{state\}\}">Q</figure-callout>}}_{\mathrm{2,3}}\\ Q_{\mathrm{3,0}}& Q_{\mathrm{3,1}}& Q_{\mathrm{3,2}}& Q_{\mathrm{3,3}}\end{array}\right]---\left(10\right)$

Move and surface deformation at the reference mark in the said cutting procedure, accomplishes through following iteration:

Q ^t+1＝Q ^t+ΔQ (11)

$\mathrm{\&Delta;Q}=\mathrm{\&lambda;}{M}_R^{-1}{\left(U^{T}U\right)}^{-1}{F}_{Q}V{\left(V^{T}V\right)}^{-1}{\left(M_R^T\right)}^{-1}---\left(12\right)$

F wherein _QBe external force, depend on the shade of gray of 3-D view;

(1) according to the B-Surface model in reference mark and the formula (7), the Calculation of Three Dimensional surface;

(2) calculate external force F according to three-dimensional surface and gradation of image _Q

(3) generate new reference mark according to formula (11) and (12);

(4), calculate new three-dimensional surface according to formula (7) by new reference mark;

(5) if iterations reaches setting value, then stop, otherwise change step (2).

The dividing method of a kind of three-dimensional ultrasound pattern that the present invention proposes, Fig. 1-Fig. 6 has explained the result of method flow and instance.Three-dimensional segmentation method described in the present invention relates to some controlled variable, and these parameters will comprehensively be adjusted and set to concrete data characteristics, so that result reaches best.The parameter that the set of present embodiment deal with data sets is following:

Step 2) 3 D wavelet in decomposes, and decomposition scale is 1;

Control constant k in the formula of step 3) (2), value are 0.75.

Two-dimensional silhouette in the step 4) and three-dimensional surface initialization, it is 32 that control is counted;

Dividing method in the step 5), B-Surface grid are 8 * 8 * 8, and iterations is 10.

The foregoing description is a preferred implementation of the present invention; But embodiment of the present invention is not restricted to the described embodiments; Other any do not deviate from change, the modification done under spirit of the present invention and the principle, substitutes, combination, simplify; All should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (8)

1. the dividing method of a three-dimensional ultrasound pattern is characterized in that, may further comprise the steps:

1) three-dimensional ultrasound pattern is carried out the spot pre-service, adopt normalization small echo anisotropy method of diffusion to remove speckle noise;

2) use the Canny edge detection operator image after pre-service is carried out initialization;

3) utilization is implemented three-dimensional segmentation based on the three-dimensional deformation model of B-Surface and GVF Snake.

2. the dividing method of a kind of three-dimensional ultrasound pattern according to claim 1 is characterized in that, the spot pre-service in the step 1) comprises the steps:

(1) ultrasonoscopy being carried out multi-scale wavelet and decompose, is high frequency and low frequency part with picture breakdown;

(2) according to the property taken advantage of spot model, to HFS, use normalization small echo anisotropy method of diffusion and handle, suppress spot;

(3) carry out wavelet inverse transformation, the image after reconstruct obtains denoising.

3. normalization small echo anisotropy method of diffusion according to claim 2 is characterized in that said ultrasonoscopy is decomposed into 8 parts: HLL, HLH, HHL, HHH through 3 D wavelet; LHH, LHL, LLH, LLL, wherein LLL is a low frequency part, other is a HFS.

4. normalization small echo anisotropy method of diffusion according to claim 3 is characterized in that, the property the taken advantage of spot model in the step (2) is:

J(x，y，z)＝I(x，y，z)×n(x，y，z) (1)

Wherein (x, y z) represent original noise-free picture to I, and (x, y are to contain noise n (x, y, image z) z) to J.

5. normalization small echo anisotropy method of diffusion according to claim 4 is characterized in that, the model of normalization small echo anisotropy diffusion is:

$\left\{\begin{array}{c}\frac{\&PartialD;J(x,y,z,t)}{\&PartialD;t}=\mathrm{div}[\tilde{c}(x,y,z,t)\&dtri;J(x,y,z,t)]\\ J(x,y,z,0)=J_0(x,y,z)\end{array}\right.---\left(2\right)$

Wherein is gradient of image and gray scale; Div representes divergence (divergence); is the normalization coefficient of diffusion, and it depends on gradient of image and gray scale, in the less even matter zone of gradient; Its value is bigger; With the stronger level and smooth inhibition noise of enforcement, otherwise, in the bigger zone of gradient; Organizational boundary or edge often; This moment, diffusion should have smaller value, thereby kept them as far as possible, and it is defined as:

$c\left(\right|\widetilde{\&dtri;}g\left|\right)=1-\exp (-3.315/{\left(\right|\&dtri;g|/\mathrm{kg}(x,y,z;t))}^4)---\left(3\right)$

Wherein k is a controlled variable, and its value is [0 1], and t representes iteration time or number of times.

6. according to the dividing method of any described a kind of three-dimensional ultrasound pattern of claim 1～5, it is characterized in that step 2) in initialization comprise the steps:

(1) to each tomography, utilizes the Canny edge detection operator, extract image outline through pretreated three-dimensional ultrasound pattern;

(2) near the center of object to be split, confirm an approximate central point by manual work;

(3) point from the center is being in the circumference in the center of circle with the central point, defines some equally distributed rays; Scan from inside to outside, ask the first intersection point of ray and profile, after all ray scannings finish; Obtain a series of intersection points; With these intersection points is the reference mark, can generate the B-Surface three-dimensional surface, as initial three-dimensional surface.

7. the dividing method of a kind of three-dimensional ultrasound pattern according to claim 6; It is characterized in that the three-dimensional segmentation in the step 3) is to adopt three-dimensional GVF Snake, from initial three-dimensional surface; Driving above-mentioned B-surface reference mark through external force moves; Thereby make the three-dimensional surface of object produce deformation, approach real subject surface gradually, finally realize cutting apart of three dimensional object.

8. the dividing method of a kind of three-dimensional ultrasound pattern according to claim 7 is characterized in that, the external force of three-dimensional segmentation adopts the GVF model, and deformation model is B-Surface, and wherein the gradient vector flow model is:

$E_{\mathrm{gvf}}\left(v\right)=\&Integral;\&Integral;\mathrm{\&mu;}(s_x^2+s_y^2+t_x^2+t_y^2)+{|\&dtri;f|}^2{|v-\mathrm{\&Delta;f}|}^2\mathrm{dxdy}---\left(4\right)$

Wherein, v (x, y)=(s (x, y), t (x, y)), expression gradient vector, s _x, s _y, t _x, t _yBe respectively s (x, y) with t (x, y) at the local derviation of x and y direction,

Be shade of gray, || be the mould of vector, μ is that value is 0～1 controlled variable;

The B-Surface model is:

$p(u,v)=\underset{i=0}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{N}_{i,p}\left(u\right)N_{j,q}\left(v\right)B_{i,j}---\left(5\right)$

N wherein _{I, p}(u) and N _{J, q}(v) being respectively exponent number is the B-spline basis function of p and q, and u and v span are [01], B _{I, j}Be the reference mark, m, n are respectively row and column direction grid number;

Cubic B-spline can be described below:

$p(u,v)={\mathrm{UM}}_{R}Q{M}_R^T{V}^T---\left(6\right)$

Wherein

$M_R=\frac{1}{6}\left[\begin{array}{cccc}-1& 3& -3& 1\\ 3& -6& 3& 0\\ -3& 0& 3& 0\\ 1& 4& 1& 0\end{array}\right]---\left(8\right)$

Q is the reference mark matrix:

$Q=\left[\begin{array}{cccc}{Q}_{\mathrm{0,0}}& Q_{\mathrm{0,1}}& Q_{\mathrm{0,2}}& Q_{\mathrm{0,3}}\\ Q_{\mathrm{1,0}}& Q_{\mathrm{1,1}}& Q_{\mathrm{1,2}}& Q_{\mathrm{1,3}}\\ Q_{\mathrm{2,0}}& Q_{\mathrm{2,1}}& Q_{\mathrm{2,2}}& Q_{\mathrm{2,3}}\\ Q_{\mathrm{3,0}}& Q_{\mathrm{3,1}}& Q_{\mathrm{3,2}}& Q_{\mathrm{3,3}}\end{array}\right]---\left(9\right)$

Move and surface deformation at the reference mark in the said cutting procedure, accomplishes through following iteration:

Q ^t+1＝Q ^t+ΔQ (10)

$\mathrm{\&Delta;Q}=\mathrm{\&lambda;}{M}_R^{-1}{\left(U^{T}U\right)}^{-1}{F}_{Q}V{\left(V^{T}V\right)}^{-1}{\left(M_R^T\right)}^{-1}---\left(11\right)$

F wherein _QBe external force, depend on the GVF model;

Partitioning algorithm of the present invention carries out according to the following steps:

(1) according to the B-Surface model in reference mark and the formula (6), the Calculation of Three Dimensional surface;

(2) according to three-dimensional surface and GVF Model Calculation external force F _Q

(3) generate new reference mark according to formula (10) and (11);

(4), calculate new three-dimensional surface according to formula (6) by new reference mark;

(5) if iterations reaches setting value, then stop, otherwise change step (2).

Images