CN104166976B - The dividing method of prospect in a kind of 3-D view - Google Patents

Abstract

The invention provides the dividing method of prospect in a kind of 3-D view, from 3-D view, namely obtain the method for foreground mask, described method comprises the steps: that (1) inputs described 3-D view and initial foreground mask M ⁽¹⁾; (2) the image constraint prospect probability of each pixel in described 3-D view is calculated; (3) based on current foreground mask M ^(k), calculate the shape constraining prospect probability of each pixel in described 3-D view , k is iterations; (4) based on described image constraint prospect probability and shape constraining prospect probability , obtain next foreground mask M ^(k+1); (5) if described current foreground mask M ^(k)with next foreground mask M ^(k+1)change be less than predetermined value or iterations k equals predetermined maximum iteration time, then described current foreground mask M ^(k)be the foreground mask that this method obtains; Otherwise make M ^(k)=M ^(k+1), k=k+1, and return step (3).The technical program not only realizes simply, fast, stably can also extract the prospect in 3 d medical images.

Description

The dividing method of prospect in a kind of 3-D view

Technical field

The present invention relates to image processing field, particularly relate to the dividing method of prospect in a kind of 3-D view.

Background technology

At present, for clinical medicine image, conventional automatic foreground extracting method comprises movable contour model and region increases.

Movable contour model is the method developed to initial mask boundary profile.The method utilizes interested feature in image, by the profile of initial mask as a whole curve carry out stretching and being out of shape, find the border of area-of-interest, therefore, to the insensitive for noise in image, and can overcome because the unintelligible segmentation caused in boundary local interested is revealed and segmentation by mistake.But the method is the energy function by optimizing by image itself and mask border curve combination, and develop to mask border, arithmetic speed is generally slower.

Region increases, and adopts the method for iteration, and from the boundary position of initial mask to external expansion, by image, the adjacent and pixel with feature of interest is progressively merged in mask, until search out the boundary position of area-of-interest with mask border.The algorithm realization of the method is simple, efficient, but, very responsive to the noise in image, and only considered the impact of mask adjacent area.

Summary of the invention

The problem that the present invention solves is to provide the dividing method of prospect in a kind of 3-D view, not only realizes simple, fast, stably can also extract the prospect in 3 d medical images.

In order to solve the problem, the invention provides the dividing method of prospect in a kind of 3-D view, comprising the steps:

(1) input described 3-D view, calculate the image constraint prospect probability of each pixel in described 3-D view; Remember that the foreground mask of described 3-D view is M, M ⁽¹⁾for initial foreground mask;

(2) based on current foreground mask M ^(k), calculate the shape constraining prospect probability of each pixel in described 3-D view wherein, k is iterations, k>=1;

(3) based on described image constraint prospect probability and shape constraining prospect probability obtain next foreground mask M ^(k+1);

(4) if described current foreground mask M ^(k)with next foreground mask M ^(k+1)change be less than predetermined value or iterations k equals predetermined maximum iteration time, then finishing iteration, described current foreground mask M ^(k)be the prospect of required segmentation; Otherwise then return step (2), k increases by 1.

The dividing method of prospect in a kind of 3-D view described above, wherein, calculates the image constraint prospect Probability p of each pixel in described 3-D view _{1 (X)}process comprise: 1) by described 3-D view binaryzation, obtain binary image; 2) described binary image is carried out spatial filtering and conversion, obtain described image constraint prospect Probability p _{1 (X)}.

The dividing method of prospect in a kind of 3-D view described above, wherein, obtains described image constraint prospect Probability p _{1 (X)}formula be:

$p_{1\left(X\right)}=g_1(G_{{\mathrm{\σ}}_1}\⊗I_b)$

Wherein, X represents the locus of arbitrary pixel in described 3-D view; g ₁for transforming function transformation function; for spatial filtering operator; represent convolution algorithm; I _bfor described binary image.

The dividing method of prospect in a kind of 3-D view described above, wherein, calculates described shape constraining prospect probability formula be:

$p_{2\left(X\right)}^{(k+1)}=g_2(G_{{\mathrm{\σ}}_2}\⊗M^{\left(k\right)})$

Wherein, g ₂for transforming function transformation function; for spatial filtering operator; represent convolution algorithm; K is iterations; X represents the locus of arbitrary pixel in described 3-D view.

The dividing method of prospect in a kind of 3-D view described above, wherein, obtains next foreground mask M described ^(k+1)process comprise:

1) by described image constraint prospect Probability p _{1 (X)}with described shape constraining prospect probability combination, obtaining each pixel in described 3-D view becomes the probability of prospect

2) to described prospect probability carry out binaryzation, obtain interim foreground mask;

3) described interim foreground mask is carried out space smoothing filtering and binaryzation, obtain next foreground mask M described ^(k+1).

The dividing method of prospect in a kind of 3-D view described above, wherein, the formula of described combination is wherein, w is described prospect probability middle p _{1 (X)}weight, 0 < w < 1.

The dividing method of prospect in a kind of 3-D view described above, wherein, the formula of described combination is $p_{\left(X\right)}^{(k+1)}=\frac{p_{1\left(X\right)}^w{\left(p_{2\left(X\right)}^{(k+1)}\right)}^{1-w}}{p_{1\left(X\right)}^w{\left(p_{2\left(X\right)}^{(k+1)}\right)}^{1-w}+{(1-p_{1\left(X\right)})}^w{(1-p_{2\left(X\right)}^{(k+1)})}^{1-w}}.$

The dividing method of prospect in a kind of 3-D view described above, wherein, obtains next foreground mask M described ^(k+1)formula be:

$M^{(k+1)}=I[(G_{{\mathrm{\&sigma;}}_3}\&CircleTimes;I[p^{(k+1)}>t_2\left]\right)>t_3]$

Wherein, for spatial filtering operator; represent convolution algorithm; I is target function, and when the inequality in its variable is set up, value is 1, otherwise value is 0; t ₂and t ₃for predetermined threshold.

The dividing method of prospect in a kind of 3-D view described above, wherein, the span of described predetermined value is 0.000001-0.001.

The dividing method of prospect in a kind of 3-D view described above, wherein, the span of described predetermined maximum iteration time is 10-20.

Compared with prior art, the whole process of the present invention can by after image binaryzation, then carries out spatial filtering and simple arithmetical operation completes, and compared with movable contour model, the complexity that the method realizes is low, and computing velocity is fast;

Further, Shape-based interpolation constraint and region increase, and use the shape facility of the characteristic sum mask of image self simultaneously, have higher stability.

Accompanying drawing explanation

Figure 1 shows that the schematic flow sheet of the dividing method of prospect in a kind of 3-D view of the embodiment of the present invention;

Figure 2 shows that the embodiment of the present invention obtains next foreground mask M described ^(k+1)schematic flow sheet;

Figure 3 shows that the effect schematic diagram of the segmentation prospect that embodiment of the present invention this method obtains and the segmentation prospect that traditional region growing methods obtains.

Embodiment

Set forth a lot of detail in the following description so that fully understand the present invention.But the present invention can be much different from alternate manner described here to implement, those skilled in the art can when without prejudice to doing similar popularization when intension of the present invention, therefore the present invention is by the restriction of following public concrete enforcement.

Secondly, the present invention utilizes schematic diagram to be described in detail, and when describing the embodiment of the present invention in detail, for ease of illustrating, described schematic diagram is example, and it should not limit the scope of protection of the invention at this.

Below in conjunction with drawings and Examples, the dividing method of prospect in a kind of 3-D view of the present invention is described in detail.In the 3-D view of the embodiment of the present invention, the dividing method of prospect as shown in Figure 1, first, performs step S1, input described 3-D view and initial foreground mask M ⁽¹⁾.Wherein, M is the foreground mask of described 3-D view.In the present embodiment, the 3-D view of input is CT abdomen images, and the prospect of required segmentation is the liver (or spleen) of CT image.

Then, perform step S2, calculate the image constraint prospect Probability p of each pixel in described 3-D view _{1 (X)}.Wherein, p _{1 (X)}=p _{1 (X)}(O=1|X), X is the locus of arbitrary pixel in described 3-D view, particularly, the image constraint prospect Probability p of arbitrary pixel p in described 3-D view is calculated _{1 (X)}process comprise: 1) by the setting method such as threshold value, K means Data Cluster Algorithm (K-means) by described 3-D view binaryzation, obtain binary image, extract at the pixel of interested gray value interval by gray-scale value in described 3-D view; 2) described binary image is carried out spatial filtering and conversion, obtain described image constraint prospect Probability p by formula (1) _{1 (X)}, described formula (1) is:

$p_{1\left(X\right)}=g_1(G_{{\mathrm{\&sigma;}}_1}\&CircleTimes;I_b)$

Wherein, X represents the locus of arbitrary pixel in described 3-D view; g ₁for transforming function transformation function; for spatial filtering operator; represent convolution algorithm; I _bfor described binary image.

In the present embodiment, first, adopt threshold method to carry out binaryzation to CT abdomen images, the pixel that gray-scale value in described CT abdomen images belongs to liver intensity value interval is extracted, obtains binary image.Then, by described formula (1), spatial filtering and conversion are carried out to binary image, obtain the image constraint prospect probability of each pixel in CT abdomen images.Wherein, for moving average filter, along the width cs of three-dimensional ₁=[2,2,2] mm, transforming function transformation function is g ₁(t)=t.

Then, step S3 is performed, based on current foreground mask M ^(k), calculate the shape constraining prospect probability of each pixel in described 3-D view wherein, k is iterations, k>=1, as k=1, and described foreground mask M ⁽¹⁾for the initial foreground mask inputted described in step S1.Particularly, by formula (2) to foreground mask M ^(k)do spatial filtering and conversion, obtain shape constraining prospect probability described formula (2) is:

$p_{2\left(X\right)}^{(k+1)}=g_2(G_{{\mathrm{\&sigma;}}_2}\&CircleTimes;M^{\left(k\right)})$

Wherein, g ₂for transforming function transformation function; for spatial filtering operator; represent convolution algorithm; K is iterations; X represents the locus of arbitrary pixel in described 3-D view.

In the present embodiment, for the liver in CT abdomen images, according to the liver initial foreground mask M of formula (2) to input ⁽¹⁾carry out spatial filtering and conversion, namely initial foreground mask carries out first time region growth (first time iteration), obtains the shape constraining prospect probability that arbitrary pixel in CT abdomen images carries out first time region growth the function of described conversion is g ₁(t)=t ².

Then, step S4 is performed, based on described image constraint prospect probability and shape constraining prospect probability obtain next foreground mask M ^(k+1).Particularly, next foreground mask M described is obtained ^(k+1)process as shown in Figure 2, first, perform step S201, by described image constraint prospect Probability p _{1 (X)}with described shape constraining prospect probability combination, obtaining each pixel in described 3-D view becomes the probability of prospect particularly, by linear suggestion pond, i.e. formula to image constraint prospect Probability p _{1 (X)}with shape constraining prospect probability combine, wherein, w is described prospect probability middle p _{1 (X)}weight, 0 < w < 1; Or by logarithm suggestion pond, i.e. formula $p_{\left(X\right)}^{(k+1)}=\frac{p_{1\left(X\right)}^w{\left(p_{2\left(X\right)}^{(k+1)}\right)}^{1-w}}{p_{1\left(X\right)}^w{\left(p_{2\left(X\right)}^{(k+1)}\right)}^{1-w}+{(1-p_{1\left(X\right)})}^w{(1-p_{2\left(X\right)}^{(k+1)})}^{1-w}},$ To image constraint prospect Probability p _{1 (X)}with shape constraining prospect probability combine.In the present embodiment, by logarithm suggestion pond, to the image constraint prospect Probability p obtained in step S2 _{1 (X)}with the shape constraining prospect probability obtained in step S3 combine, obtain arbitrary pixel in first time region growth CT abdomen images and belong to the probability of liver

Then, step S202 is performed, to described prospect probability carry out binaryzation, obtain interim foreground mask.Particularly, by all prospect probability compare with predetermined threshold, if be greater than described predetermined threshold, then assignment is 1, if be less than or equal to predetermined threshold, then assignment is 0.In the present embodiment, the probability of prospect will be become in step S201 the pixel assignment being greater than 0.5 is 1, and remaining pixel assignment is 0, namely obtains the interim foreground mask of first time region growth.

Then, perform step S203, described interim foreground mask is carried out space smoothing filtering and binaryzation, obtain next foreground mask M described ^(k+1), namely

$M^{(k+1)}=I[(G_{{\mathrm{\&sigma;}}_3}\&CircleTimes;I[p_{\left(X\right)}^{(k+1)}>t_2\left]\right)>t_3]$

Wherein, for spatial filtering operator; represent convolution algorithm; I is target function, and when the inequality in its variable is set up, value is 1, otherwise value is 0; t ₂and t ₃for predetermined threshold.

In the present embodiment, the interim foreground mask obtained in step S203 is carried out space smoothing filtering, foreground mask after level and smooth is carried out binaryzation, the assignment being greater than 0.5 by level and smooth interim foreground mask pixel value is 1, remaining pixel assignment is 0, and binary image is now foreground mask M ⁽²⁾.

Then, step S5, more described foreground mask M is performed ^(k)with foreground mask M ^(k+1)size or compare the size of iterations k and predetermined maximum iteration time, if current foreground mask M ^(k)with next foreground mask M ^(k+1)change be less than predetermined value or k equals predetermined maximum iteration time, then perform step S6, finishing iteration, described current foreground mask M ^(k)be the described prospect that need split; If current foreground mask M ^(k)with next foreground mask M ^(k+1)change be more than or equal to predetermined value or k is less than predetermined maximum iteration time, then return step S3, iterations k increases by 1.Wherein, the change between foreground mask can be the absolute value of both relative volume changes; The span of predetermined value is 0.000001-0.001; The span of predetermined maximum iteration time is 10-20.In the present embodiment, the foreground mask M will obtained in step S203 ⁽²⁾with the initial foreground mask M of the liver of input ⁽¹⁾volume compare, if be more than or equal to predetermined value, then re-execute step S3, and obtain foreground mask M ⁽²⁾method the same, try to achieve foreground mask M ⁽³⁾, then by foreground mask M ⁽²⁾with foreground mask M ⁽³⁾volume compare; If foreground mask M ⁽²⁾with the initial foreground mask M of the liver of input ⁽¹⁾the change of volume be less than predetermined value, then initial foreground mask M ⁽¹⁾be the prospect of required segmentation.With foreground mask M ⁽¹⁾with foreground mask M ⁽²⁾between relatively the same, foreground mask M ⁽²⁾with foreground mask M ⁽³⁾the change of volume be less than predetermined value, then foreground mask M ⁽²⁾be the prospect of required segmentation, otherwise, continue to calculate next foreground mask, until the change of the foreground mask of front and back twice is less than predetermined value.In addition, in the present embodiment, predetermined maximum iteration time is 12, if when iterating to the 12nd time, and foreground mask M ⁽¹²⁾with foreground mask M ⁽¹³⁾volume change be still more than or equal to predetermined value, now, also finishing iteration, foreground mask M ⁽¹²⁾be the prospect of required segmentation.

As shown in Figure 3, figure (a) and figure (b) is respectively the liver position and spleen position that this method obtains, the liver position that the region growing methods that figure (c) and figure (d) is respectively traditional obtains and spleen position.As can be seen from the figure, adopt identical gray threshold and iterations, there is a large amount of segmenting pixels point by mistake in traditional region growing methods, and said method is to the liver in CT abdomen images and spleen segmentation, well remains the shape of corresponding organ.

Although the present invention with preferred embodiment openly as above; but it is not for limiting the present invention; any those skilled in the art without departing from the spirit and scope of the present invention; the Method and Technology content of above-mentioned announcement can be utilized to make possible variation and amendment to technical solution of the present invention; therefore; every content not departing from technical solution of the present invention; the any simple modification done above embodiment according to technical spirit of the present invention, equivalent variations and modification, all belong to the protection domain of technical solution of the present invention.

Claims (7)

1. the dividing method of prospect in 3-D view, is characterized in that, comprise the steps:

(1) input described 3-D view, calculate the image constraint prospect probability of each pixel in described 3-D view; Remember that the foreground mask of described 3-D view is M, initial foreground mask is M ⁽¹⁾;

(2) based on current foreground mask M ^(k), calculate the shape constraining prospect probability of each pixel in described 3-D view wherein, k is iterations, k>=1;

(3) based on described image constraint prospect probability and shape constraining prospect probability obtain next foreground mask M ^(k+1);

(4) if described current foreground mask M ^(k)with next foreground mask M ^(k+1)change be less than predetermined value or iterations k equals predetermined maximum iteration time, then finishing iteration, described current foreground mask M ^(k)be the prospect of required segmentation; Otherwise then return step (2), k increases by 1;

Wherein, the image constraint prospect Probability p of each pixel in described 3-D view is calculated _{1 (Χ)}process comprise: 1) by described 3-D view binaryzation, obtain binary image; 2) described binary image is carried out spatial filtering and conversion, obtain described image constraint prospect Probability p _{1 (Χ)};

Based on current foreground mask M ^(k), calculate the shape constraining prospect probability of each pixel in described 3-D view formula be: g ₂for transforming function transformation function; for spatial filtering operator; represent convolution algorithm; K is iterations; X represents the locus of arbitrary pixel in described 3-D view;

Based on described image constraint prospect probability and shape constraining prospect probability obtain next foreground mask M ^(k+1)process comprise: 1) by described image constraint prospect Probability p _{1 (Χ)}with described shape constraining prospect probability combination, obtaining each pixel in described 3-D view becomes the probability of prospect 2) to described prospect probability carry out binaryzation, obtain interim foreground mask; 3) described interim foreground mask is carried out space smoothing filtering and binaryzation, obtain next foreground mask M described ^(k+1).

2. the dividing method of prospect in 3-D view as claimed in claim 1, is characterized in that, obtain described image constraint prospect Probability p _{1 (Χ)}formula be:

$p_{1\left(X\right)}=g_1(G_{{\mathrm{\&sigma;}}_1}\&CircleTimes;I_b)$

Wherein, X represents the locus of arbitrary pixel in described 3-D view; g ₁for transforming function transformation function; for spatial filtering operator; represent convolution algorithm; I _bfor described binary image.

3. the dividing method of prospect in a kind of 3-D view as claimed in claim 1, it is characterized in that, the formula of described combination is wherein, w is described prospect probability middle p _{1 (Χ)}weight, 0 < w < 1.

4. the dividing method of prospect in a kind of 3-D view as claimed in claim 1, it is characterized in that, the formula of described combination is

$p_{\left(X\right)}^{(k+1)}=\frac{p_{1\left(X\right)}^{w}g{\left(p_{2\left(X\right)}^{(k+1)}\right)}^{1-w}}{p_{1\left(X\right)}^{w}g{\left(p_{2\left(X\right)}^{(k+1)}\right)}^{1-w}+{(1-p_{1\left(X\right)})}^{w}g{(1-p_{2\left(x\right)}^{(k+1)})}^{1-w}}.$

5. the dividing method of prospect in a kind of 3-D view as claimed in claim 1, is characterized in that, obtain next foreground mask M described ^(k+1)formula be:

$M^{(k+1)}=I[(G_{{\mathrm{\&sigma;}}_3}\&CircleTimes;I[p_{\left(X\right)}^{(k+1)}>t_2\left]\right)>t_3]$

Wherein, for spatial filtering operator; represent convolution algorithm; I is target function, and when the inequality in its variable is set up, value is 1, otherwise value is 0; t ₂and t ₃for predetermined threshold.

6. the dividing method of prospect in a kind of 3-D view as claimed in claim 1, it is characterized in that, the span of described predetermined value is 0.000001-0.001.

7. the dividing method of prospect in a kind of 3-D view as claimed in claim 1, it is characterized in that, the span of described predetermined maximum iteration time is 10-20.