CN104143190B - The dividing method and system organized in CT images - Google Patents
The dividing method and system organized in CT images Download PDFInfo
- Publication number
- CN104143190B CN104143190B CN201410357152.7A CN201410357152A CN104143190B CN 104143190 B CN104143190 B CN 104143190B CN 201410357152 A CN201410357152 A CN 201410357152A CN 104143190 B CN104143190 B CN 104143190B
- Authority
- CN
- China
- Prior art keywords
- profile
- images
- point
- split
- tissue
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present invention provides the dividing method and system organized in a kind of CT images, and method therein includes:The opacity of CT images is adjusted, the institutional framework in CT images is obtained;Tissue to be split is determined in acquired institutional framework, selects wherein one layer of the CT images of tissue to be split to interact formula segmentation, its process is:The energy cost function of the CT images under the opacity is constructed using gradient vector, organizing the position selected seed point that mutually closes on to be split, the profile of the current point generated according to energy cost function to a wherein tomographic image of the CT images that the optimal path between seed point is tissue to be split;The profile of the generation adjacent layer adjacent with the layer where profile;Then the profile of each tomographic image of the CT images of tissue to be split is generated using iterative method;The profile for combining each tomographic image of the CT images of tissue to be split obtains the tissue to be split of three-dimensional.Formula can be quickly and accurately interacted to three-dimensional CT image by the present invention to split.
Description
Technical field
The present invention relates to technical field of image segmentation, more specifically, it is related to the dividing method organized in a kind of CT images
And system.
Background technology
The three-dimensional segmentation of medical image can provide reliable foundation for clinical diagnosis and pathological research.But image is past
Toward there is complexity and ambiguity, all of segmentation problem still is can solve the problem that without a kind of method of standard at present, thus it is many
The work of segmentation cannot be automatically performed by computer, and split there is the big problem of workload by hand.Just because of this, people
Propose some man-machine interactivelies and computer is automatically positioned the method that is combined to realize the method for rapidly positioning of objective contour, this
Plant method for rapidly positioning and be roughly divided into two classes:
One class is Snake algorithms, which defines an energy curve, is together decided on by internal energy and external energy.
During deformation, external energy causes that curve closer to the edge in region, internal energy then maintain the flatness of curve.
The advantage of Snake algorithms is that view data, initial estimation, objective contour and the constraints based on rudimentary knowledge are all unified to be used
One characteristic extraction procedure is realized, it is known that rational initialization curve, by after successive ignition, curve energy automatic Fitting makes
Its energy function converges on the state of energy minima, but the shortcoming of Snake algorithms is that it is quicker to initial profile, noise
Sense, the inspection at edge need to depend on the regulation of weight, and the expense of time can not meet the requirement of real-time three-dimensional image compression.
Another kind of is staggered form contours segmentation, and it is intelligent scissors method that it represents method.Its basic thought is:Using La Pula
This zero-crossing method finds marginal point, determines the positional information at edge, recycles Robel operators, Canny operators or Sobel
Operator obtains the Grad of image to represent edge strength, and energy cost function is constructed according to edge strength, is constructing energy
On the basis of cost function the least energy between two seed points is tried to achieve using dynamic programming algorithm or dijkstra's algorithm
Spend path (i.e. optimal path).
Wherein, the computing formula of Laplce's zero-crossing method is as follows:Using Marr in
The calculation of the 1980 Gauss nuclear radius for being provided:W=3.35* σ+0.33;Kernel Size=2*W+1;Work as Kernel
During Size=5, σ=0.5;During Kernel Size=9, σ=1.1.Wherein, W represents Gauss nuclear radius, and Kernel Size are represented
Gaussian kernel size.
Assuming that 4 neighborhood points (vertical, horizontal direction) of the q for p, after trying to achieve LoG according to core size, the Laplce zero of p points
The value in crosspoint is:Wherein,
Because when image is processed, core is smaller more sensitive to details, and core is more big more can suppress noise, therefore, used here as multiple width
The method of core carrys out construction Laplce's zero crossing point value.By taking convolution kernel of the empirical value using 5*5 and 9*9 as an example, while specifying
The weighting effect of two cores is respectively 0.45 and 0.55, then the formula for obtaining Laplce's zero cross point of p points is as follows:fZ
(p)=0.45*B5Z+0.55*B9Z。
It is following to introduce respectively using Robel operators, Canny operators and Sobel operators after determining the positional information at edge
Carry out the example of rim detection.
The computing formula for carrying out rim detection using Robel operators is as follows:G (f (x, y))=| [f (x, y)-f (x+
1, y+1)]+[f (x+1, y)-f (x, y+1)], wherein f (x, y) represents the CT values of pixel (x, y) point, and G (f (x, y)) represents pixel
The Robel operator values of (x, y).
The general principle for carrying out rim detection using Canny operators is:First, image edge detection must is fulfilled for two bars
Part:Noise can effectively be suppressed and the position at edge must be accurately determined as far as possible;Then signal to noise ratio is surveyed with positioning product
Degree, obtains optimizing Approximation Operator.
The process for carrying out rim detection using Canny operators is as follows:
step1:Use Gaussian filter smoothing image;
step2:Amplitude and the direction of gradient are calculated with the finite difference of single order local derviation;
step3:Non-maxima suppression is carried out to gradient magnitude;
step4:Detected with dual threashold value-based algorithm and connection edge.
The calculation for carrying out rim detection using Sobel operators is as follows:
The operator includes two group 3 × 3 of matrix, respectively transverse direction and longitudinal direction, divides by it and image are made into planar convolution
The brightness difference approximation of transverse direction and longitudinal direction is not drawn.If representing original image with A, Gx and Gy is represented through horizontal and vertical respectively
To the image of rim detection, its formula is as follows:
Formula under the transverse direction and longitudinal direction gradient approximation of each pixel of image may be used to combines to calculate gradient
Size:Then below equation is can use to calculate gradient direction:Wherein, if angle
Degree Θ is equal to zero, i.e. representative image and possesses longitudinal edge at this, and left is dark compared with right.
By above-mentioned example, it can be seen that the advantage of existing staggered form contours segmentation method is:Which raises algorithm
Flexibility and accuracy, its in different scale space, calculate Laplce's zero cross point and gradient intensity, reduce yardstick become
Change the influence to target detection, increase the robustness of algorithm.
Although this kind of staggered form contours segmentation method has above-mentioned advantage, it there is also deficiency.On the one hand, seed point it
Between short path energy value it is bigger than the advantage that the energy value in path long is accounted for, so the point on shortest path does not ensure that it is side
Edge point, so just cannot correctly track profile, and then cause not high to the contours segmentation degree of accuracy organized;On the other hand, in profit
When carrying out three-dimensional segmentation with this staggered form contours segmentation method, formula contours segmentation need to be interacted in two key stratums first,
Then using interpolation algorithm generate key stratum between intermediate layer profile, but traditional interpolation algorithm exist again noise-sensitive,
The deficiency such as discontinuity, time overhead be big, so further have impact on the effect of image segmentation.
The content of the invention
In view of the above problems, it is an object of the invention to provide the dividing method and system organized in a kind of CT images, with reality
Formula segmentation is now quickly and accurately interacted to three-dimensional CT image.
According to an aspect of the present invention, there is provided in a kind of CT images organize dividing method, including:
The opacity of CT images is adjusted, the institutional framework in CT images is obtained;
Tissue to be split is determined in acquired institutional framework, selects wherein one layer of the CT images of tissue to be split to enter
Row Interactive Segmentation, obtains the profile of a wherein tomographic image of the CT images of tissue to be split;Wherein, the process of Interactive Segmentation
For:Using the energy cost function of the CT images under gradient vector construction opacity, in the position mutually closed on tissue to be split
Selected seed point, the current point generated according to energy cost function is to the CT that the optimal path between seed point is tissue to be split
The profile of a wherein tomographic image of image.
Wherein, the dividing method organized in the CT images that the present invention is provided also includes:Generation is adjacent with the layer where profile
Adjacent layer profile;Wherein, the process of the profile of generation adjacent layer is:Using the layer where profile as key stratum, according to layer
Between correlation, the selected seed point on the adjacent layer adjacent with key stratum, according to energy cost function generate each pair seed point
Between optimal path for adjacent layer profile;Generate the profile of each tomographic image of the CT images of tissue to be split;Wherein, profit
The CT for generating tissue to be split using the profile identical process with generation adjacent layer between each pair adjacent layer with iterative method schemes
The profile of each tomographic image of picture;The profile for combining each tomographic image of the CT images of tissue to be split obtains the to be split of three-dimensional
Tissue.
Wherein, included using the process of gradient vector construction energy cost function:Using Gaussian function in the inclined of x, y direction
Derivative extracts the gradient vector of each pixel in CT images as convolution kernel;Using current point and the neighborhood adjacent with current point
Projection vector sum of the gradient vector of point on the field direction of current point 8 is used as energy cost function.
On the other hand, the present invention also provides the segmenting system organized in a kind of CT images, including:
Institutional framework acquiring unit, the opacity for adjusting CT images obtains the institutional framework in CT images;
Interactive Segmentation unit, for determining tissue to be split in acquired institutional framework, selects tissue to be split
Wherein one layer of CT images interact formula segmentation, obtain the profile of a wherein tomographic image of the CT images of tissue to be split;
Wherein, the process of Interactive Segmentation is:Using gradient vector construction opacity under CT images energy cost function, with treat
The segmentation position selected seed point that mutually closes on of tissue, the current point generated according to energy cost function is to optimal between seed point
Path is the profile of a wherein tomographic image of the CT images of tissue to be split.
The segmenting system organized in the CT images that the present invention is provided is further included:Adjacent layer outline generating unit, is used for
The profile of the generation adjacent layer adjacent with the layer where profile;Wherein, the process of the profile of generation adjacent layer is:By where profile
Layer as key stratum, according to layer dependencies, the selected seed point on the adjacent layer adjacent with key stratum, according to energy cost
Optimal path between each pair seed point of function generation is the profile of adjacent layer;
Outline generating unit, the profile of each tomographic image of the CT images for generating tissue to be split;Wherein, using repeatedly
Generate the CT images of tissue to be split using the profile identical process with generation adjacent layer between each pair adjacent layer for method
The profile of each tomographic image;
Assembled unit, the profile of each tomographic image of the CT images for combining tissue to be split obtains the to be split of three-dimensional
Tissue.
Using the dividing method and system organized in above-mentioned CT images of the invention, first by adjusting the resistance of image
Luminosity, the vital tissue structure of volume data in CT images is shown;Secondly, determine to treat in shown institutional framework point
Cut tissue wherein one layer interacts formula segmentation, and obtains the profile of a wherein tomographic image of tissue to be split.Wherein, entering
During row Interactive Segmentation, gradient intensity is replaced to construct energy cost function using gradient vector, from can overcome kind
The energy value of the short path shortcoming bigger than the advantage that the energy value in path long is accounted between son point, to ensure along profile heat input
Cost function is minimum, and then is solving preferably recognize weak edge while short path energy is dominant;Then according to layer
Between correlation, using iteration threshold method, fill out hole, quick morphology, transmitting light scheduling algorithm rapidly from the result of two-dimentional lasso trick
Three-dimensional is expanded to, so as to solve the problems, such as traditional interpolation algorithm noise-sensitive, discontinuous and time overhead is big, and then is reached fast
Speed, the purpose for interacting formula segmentation to CT images exactly.
In order to realize above-mentioned and related purpose, one or more aspects of the invention include will be explained in below and
The feature particularly pointed out in claim.Following explanation and accompanying drawing is described in detail some illustrative aspects of the invention.
However, what is indicated in terms of these is only some modes in the various modes for can be used principle of the invention.Additionally, of the invention
It is intended to include all these aspects and their equivalent.
Brief description of the drawings
By reference to the explanation below in conjunction with accompanying drawing and the content of claims, and with to it is of the invention more comprehensively
Understand, other purposes of the invention and result will be more apparent and should be readily appreciated that.In the accompanying drawings:
Fig. 1 is the dividing method schematic flow sheet organized in CT images according to the embodiment of the present invention;
Fig. 2 a and Fig. 2 b are respectively before being pre-processed to CT images and illustrate with the Contrast on effect after pre-processing
Figure;
Before Fig. 3 a and Fig. 3 b are respectively the direction projection using gradient vector and using gradient vector direction projection it
Contrast on effect schematic diagram afterwards;
Fig. 4 is the schematic diagram of the setting seed point according to the embodiment of the present invention;
Fig. 5 is the path schematic diagram between the calculating seed point according to the embodiment of the present invention;
Fig. 6 is the schematic diagram of the search adjacent pixel according to the embodiment of the present invention;
Fig. 7 is the segmentation result schematic diagram of the key stratum of the renal image according to the embodiment of the present invention;
Fig. 8 to Figure 13 is the segmentation result of the adjacent layer adjacent with key stratum of the renal image according to the embodiment of the present invention
Schematic diagram;
Figure 14 is that the three-dimensional kidney segmentation result obtained by the constitutional diagram 7 to Figure 13 according to the embodiment of the present invention is illustrated
Figure;
Figure 15 is the first logical construction block diagram of the segmenting system of tissue in the CT images for implement row according to the present invention;
Figure 16 is the second logical construction block diagram of the segmenting system of tissue in the CT images for implement row according to the present invention.
Identical label indicates similar or corresponding feature or function in all of the figs.
Specific embodiment
Specific embodiment of the invention is described in detail below with reference to accompanying drawing.
Because the present invention is that the tissue in CT images is split on the premise of based on opacity, therefore to this
Before invention is described in detail, first the concept to opacity carries out necessary explanation.
Opacity, also known as light resistance rate, light blocking coefficient, darkens coefficient.It is the number of the dielectric material light blocking ability for indicating light
Value.Opacity numerical value is bigger, illustrates that dielectric material transmitted light is fewer.For CT images, opacity affects CT images
Display resolution.It is therefore, high to the contours segmentation degree of accuracy organized for foregoing existing staggered form contours segmentation method,
And segmentation figure as when have the shortcomings that noise-sensitive, discontinuity and time overhead are big.The present invention is schemed by adjusting CT first
The opacity of picture, more clearly visible to show the institutional framework of the regulation of the volume data in CT images.
Then, determine that wherein one layer of tissue to be split interacts formula segmentation in shown institutional framework, and obtain
Obtain the profile of a wherein tomographic image of tissue to be split.Wherein, during the formula that interacts is split, taken using gradient vector
Energy cost function is constructed for gradient intensity, from energy of the energy value than path long that can overcome short path between seed point
The bigger shortcoming of advantage that value is accounted for, to ensure that along contour line energy cost function be minimum, and then is solving short path energy
Amount can preferably recognize weak edge while being dominant;Then according to layer dependencies, using iteration threshold method, hole, quick shape are filled out
The result of state, transmitting light scheduling algorithm rapidly from two-dimentional lasso trick expands to three-dimensional, so as to solve traditional interpolation algorithm noise
Sensitive, the discontinuous and big problem of time overhead, and then reach the mesh that formula segmentation is quickly and accurately interacted to CT images
's.
For the dividing method organized in the CT images for illustrating present invention offer, Fig. 1 shows according to embodiments of the present invention
CT images in organize dividing method flow.
As shown in figure 1, the dividing method organized in the CT images of present invention offer includes:
S110:The opacity of CT images is adjusted, the institutional framework in CT images is obtained.
Specifically, step S110 belongs to the pretreatment to CT images, and it passes through to adjust opacity, by volume data in CT images
Important institutional framework shown, so as to hide some medical inspections unwanted institutional frameworks.It should be noted that adjusting
Shown important institutional framework is to need to carry out some institutional frameworks of medical inspection after the opacity of whole CT images.Example
Such as, the structure in human stomach and kidney portion has been taken in a CT image, then after the opacity of adjustment this CT images, then can
The structure of stomach and kidney portion is shown, and the portion weave structure (such as tissue such as meat, blood vessel) around stomach and kidney portion
Can then be changed into transparent.
In order to more intuitively understand the purpose of adjustment CT image opacitys, Fig. 2 a and Fig. 2 b are respectively illustrated to CT images
Contrast on effect before being pre-processed and after being pre-processed.As shown in Figure 2 a and 2 b, by the light blocking to CT images
The change of degree, can remove the influence of adjacent tissue, so as to accelerate the arithmetic speed of algorithm.
S120:Tissue to be split is determined in acquired institutional framework, the CT images of tissue to be split is selected wherein
One layer interacts formula segmentation, obtains the profile of a wherein tomographic image of the CT images of tissue to be split;Wherein, Interactive Segmentation
Process be:The energy cost function of the CT images under the opacity is constructed using gradient vector, adjacent with tissue to be split
Near position selected seed point, is to be split to the optimal path between seed point according to the current point that energy cost function is generated
The profile of a wherein tomographic image of the CT images of tissue.
Wherein, included using the process of gradient vector construction energy cost function:Using Gaussian function in the inclined of x, y direction
Derivative extracts the gradient vector of each pixel in CT images as convolution kernel;Using current point and the neighborhood adjacent with current point
The projection vector sum of the gradient vector on the field direction of current point 8 of point as current point energy cost function.
Specifically, in order to improve the robustness to noise and discrete point, the present invention is using Gaussian function in the inclined of x, y direction
Derivative extracts the gradient vector of each pixel as convolution kernel.Convolution kernel formula is as follows:
Accounted for than the energy value in path long due to the energy value of short path between seed point in traditional intelligent scissors method
Advantage is bigger, it is impossible to the profile of correct tracing figure picture.The present invention is using current point and the ladder of the neighborhood point adjacent with current point
Energy cost function of projection vector sum of the vector on the neighborhood direction of current point 8 as CT images is spent, specific method is as follows:
Gradient vector generally represents that the gradient vector D put on contour line represents gray scale between adjacent pixel with unit vector
The maximum direction of saltus step, the unit vector D ' perpendicular to vectorial D represents the tangent vector of the point on contour line.It is so more known
Gradient vector D and Di (i=1-8) of P and its 8 neighborhood point, by the vertical vector D ' and Di ' of D and Di in corresponding neighborhood side
To being projected.When gradient vector D is bigger in neighborhood direction projection, its vertical vector D ' is just smaller in neighborhood direction projection,
Therefore using the minimum neighborhood point of projection vector as next profile point.Use gradient vector direction projection cause contour line with
The difference of the characteristic value on non-contour line is more obvious.This method overcomes the energy value of short path between seed point than road long
The bigger shortcoming of advantage that the energy value in footpath is accounted for, it is ensured that along contour line energy cost function be minimum.
Specifically, the gradient vector such as point p is expressed as D (p)=[Ix(p),Iy(p)], the gradient of 8 neighborhood point q of p points to
Amount is expressed as D (q)=[Ix(q),Iy(q)].If direction L (p, q) of the direction at edge and p points to q points is more close, then D (p)
It is bigger with projections of the D (q) on vectorial L (p, q).The vertical vector D'(p of D (p))=[Iy(p),-Ix(p)] and D (q) hang down
Straight vector D'(q)=[Iy(q),-Ix(q)] projection on vectorial L (p, q) is just smaller.Gradient sides of the point p in L (p, q) direction
To characteristic formula:
fD(p, q)=D'(p) L (p, q)+D'(q) L (p, q)
As described above, the present invention is adjacent in current point 8 using the gradient vector of current point and the neighborhood point adjacent with current point
Projection vector sum on the direction of domain as CT images energy cost function, can be while solving short path energy and being dominant
Preferably recognize weak edge.By taking the CT images for splitting cephalophyma as an example, in order to more intuitively understand, Fig. 3 a and Fig. 3 b show respectively
Before having gone out the direction projection using gradient vector and using the Contrast on effect after the direction projection of gradient vector, such as Fig. 3 a and
Shown in Fig. 3 b, using after the present invention split cephalophyma profile closer to hemotoncus edge.Further, the present invention is provided
The dividing method organized in CT images is further comprising the steps of:
S130:The profile of the generation adjacent layer adjacent with the layer where profile;Wherein, the process of the profile of generation adjacent layer
For:Using the layer where profile as key stratum, according to layer dependencies, the selected seed on the adjacent layer adjacent with key stratum
Point, is the profile of adjacent layer according to the optimal path between each pair seed point that energy cost function is generated.
S140:Generate the profile of each tomographic image of the CT images of tissue to be split;Wherein, using iterative method in each pair phase
Each tomographic image of the CT images of tissue to be split is generated between adjacent bed using the profile identical process with generation adjacent layer
Profile.
Wherein, the process for generating the profile of the adjacent layer adjacent with the layer where the profile that step S120 is generated includes:
Input of the profile seed point as adjacent layer is equidistantly chosen on the profile of key stratum, is used on the layer adjacent with key stratum
Iteration method carries out coarse segmentation;(the tiny of organization internal to be split is filled using the result for filling out hole algorithm filling coarse segmentation
Cavity), the influence of pseudo-edge around institutional framework to be split is then removed using opening operation;By profile seed point and key stratum
Center of mass point connects the subpoint of profile seed point and center of mass point in adjacent layer respectively as the input of adjacent layer, is formed with barycenter
Transmitting ray plot centered on point, sets the seed point of adjacent layer on the transmitting radiation direction of transmitting ray plot;Wherein, it is both
Point in coarse segmentation result is again simultaneously the seed point that the maximum point of gradient is just set to adjacent layer;According to energy cost function
The optimal path between each pair seed point is calculated, and new profile boundary point is marked on the profile of key stratum;Searched using path
Be linked in sequence for profile boundary point and generate the profile of adjacent layer by rope algorithm.
Because there is noise-sensitive, discontinuity and the deficiency such as time overhead is big in traditional interpolation algorithm.In order to solve annual reporting law
To the sensitivity of noise, the present invention is filled after coarse segmentation is carried out using iteration threshold method calculating partition threshold using hole algorithm is filled out
The influence of pseudo-edge is eliminated while filling organization internal minuscule hole after the result of coarse segmentation using opening operation.
Additionally, the present invention equidistantly selects the center of mass point of point on key stratum initial profile and key stratum as adjacent layer
Input, connects the subpoint of profile seed point and center of mass point in adjacent layer respectively, forms a launching light centered on barycenter
Line chart, transmitting light direction on find be both coarse segmentation result in point again gradient maximum point as adjacent layer kind
Sub- point, it is ensured that the continuity of interlayer profile.Between the energy cost function, the calculating each pair seed point that are constructed according to gradient vector
Optimal path, mark new profile boundary point, using path search algorithm, make these points to be correctly linked in sequence, and know
Not single profile.
Specifically, the present invention eliminates influence of the pseudo-edge to detection object edge using a series of operation:
First, ask (Niblck adaptive thresholds method 15*15) adjacent with key stratum seed point adjacent using iteration method
Segmentation threshold in domain, because the organization internal split always includes uniform CT values, therefore can always find one suitably
Partition threshold will be organized and separated with background;Then, being projected in neighborhood in adjacent layer using the partition threshold tried to achieve carries out region life
Length obtains coarse segmentation result;Finally, opening operation is carried out using hole algorithm and quick morphological operation is filled out to coarse segmentation result, is filled
The influence of pseudo-edge is eliminated while organization internal minuscule hole, its area is not substantially changed while smoothing its edge.
Wherein, iteration threshold method specific implementation method is as follows:
1st, gradation of image scope iGrayMin and iGrayMax, iGrayRange=m_iGrayMax-m_ are obtained
iGrayMin+1;
2nd, traversing graph picture, obtains grey level histogram;
3rd, the average gray iThreshold in the range of 0 to iGrayRange-1 is tried to achieve;
4th, try to achieve the average gray in the range of 0 to iThreshold and be assigned to iThresholdLow;
5th, try to achieve the average gray in the range of iThreshold to iGrayRange-1 and be assigned to iThresholdHigh;
6th, iThresholdMid=(iThresholdLow+iThresholdHigh)/2;
If the 7, iThresholdMid is equal to iThreshold termination iteration, optimum threshold is equal to, otherwise iThreshold
=iThresholdMid, goes to 4.
After the influence of removal adjacent layer pseudo-edge, the present invention finds the seed point of adjacent layer using following methods:Equidistantly
Ground selection key stratum initial profile on point and key stratum initial profile center of mass point as adjacent layer input.Connection respectively is closed
The barycenter and profile seed point of key layer form a transmitting ray plot centered on barycenter, such as Fig. 4 in the subpoint of adjacent layer
It is shown.Since subpoint of the key stratum profile seed point in adjacent layer, sought using the method for continuity point along the direction of barycenter
Look for new seed point, point of first for running into coarse segmentation result as candidate point, if not having in set continuity point
Find the point bigger than its Grad, then using it as new seed point, be otherwise substituted for the bigger point of Grad.Finding
To after the seed point of adjacent layer, each pair seed point is calculated using the energy cost function and Dijkstra of gradient vector construction
Method calculates optimal cost path between seed.In order to reduce the error that erroneous judgement seed point is caused to segmentation, application method such as Fig. 5 institutes
Show.The path a little between 1 and point 3 is tried to achieve for the first time, the path a little between 2 and point 4 is tried to achieve for the second time, mark by that analogy new
The boundary point of profile, the last present invention using a path search algorithm by these points to be correctly linked in sequence, and know
Do not go out the profile border of maximum.It is described in detail below:
Since first profile point, adjacent pixel is searched for counterclockwise.As shown in fig. 6, Dark grey is to have visited
The point asked, light gray is the point not accessed, it is known that current point p, and 0 to 3 counter clockwise direction accesses neighborhood point as illustrated.If
The point for having accessed is run into, is judged to a profile, if this profile is largest contours being preserved, otherwise abandoned.Should
Algorithm is repeated, and until all of boundary point is all assigned to a contour path, and is set backtracking algorithm and is located reason list picture
Plain line connects the situation of profile.
The following is an exemplary implementation:
S150:The profile for combining each tomographic image of the CT images of tissue to be split obtains the tissue to be split of three-dimensional.
Wherein, after the opacity of the CT images is readjusted as needed so that a part of structure is visible, another part knot
Structure becomes transparent.Algorithm re-starts the optimal path between the seed point and seed point for calculating each layer.In calculating process
The influence of transparent pixels is not considered, and each layer of profile of CT images understands automatic Fitting to the profile in new opacity undertissue,
Each layer of CT image outline of combination obtains the image segmentation result of three-dimensional.
Specifically, by taking the CT images for splitting kidney as an example, Fig. 7 is the two-dimentional Interactive Segmentation result of key stratum, Fig. 8 to figure
13 is the result that adjacent layer is split automatically.Constitutional diagram 7 obtains use inventive algorithm as shown in figure 14 and splits kidney to Figure 13
Three-dimensional visualization result.
Corresponding with the above method, the present invention also provides the segmenting system organized in a kind of CT images, and Figure 15 shows root
First logical construction of the segmenting system organized in the CT images for implementing row according to the present invention.
As shown in figure 15, the segmenting system 1500 organized in the CT images that the present invention is provided includes institutional framework acquiring unit
1510 and Interactive Segmentation unit 1520.
Wherein, institutional framework acquiring unit 1510 is used to adjust the opacity of CT images, obtains the knot of tissue in CT images
Structure.
Interactive Segmentation unit 1520 selects to be split group for determining tissue to be split in acquired institutional framework
Wherein one layer of the CT images knitted interacts formula segmentation, obtains the wheel of a wherein tomographic image of the CT images of tissue to be split
It is wide;Wherein, the process of Interactive Segmentation is:The energy cost function of the CT images under opacity is constructed using gradient vector,
With the position selected seed point organized and mutually close on to be split, the current point according to energy cost function generation is between seed point
Optimal path is the profile of a wherein tomographic image of the CT images of tissue to be split.
Wherein, Interactive Segmentation unit 1520 is further included:Gradient vector extraction unit and energy cost function are generated
Unit (not shown in figure).Wherein, gradient vector extraction unit, makees for the partial derivative using Gaussian function in x, y direction
It is convolution kernel, extracts the gradient vector of each pixel in CT images;Energy cost function generation unit be used for using current point and
Projection vector sum of the gradient vector of the neighborhood point adjacent with current point on the field direction of current point 8 as current point energy
Amount cost function.
Further, Figure 16 shows the second logic of the segmenting system organized in CT images according to embodiments of the present invention
Structure.As shown in figure 16, the segmenting system organized in the CT images that the present invention is provided also includes:Adjacent layer outline generating unit
1530th, outline generating unit 1540 and result obtaining unit 1550.
Wherein, adjacent layer outline generating unit 1530 is used for the profile of the adjacent layer for generating adjacent with the layer where profile;
Wherein, the process of the profile of generation adjacent layer is:Using the layer where profile as key stratum, according to layer dependencies, with pass
Selected seed point on the adjacent adjacent layer of key layer, be according to the optimal path between each pair seed point that energy cost function is generated
The profile of adjacent layer.
Wherein, adjacent layer outline generating unit is further included:Coarse segmentation unit, fills unit, seed point setup unit,
Profile boundary point indexing unit and profile boundary point connection unit (not shown in figure).
Wherein, coarse segmentation unit is used to equidistantly choose profile seed point on the profile of key stratum as the defeated of adjacent layer
Enter, coarse segmentation is carried out using iteration method on the layer adjacent with key stratum.
Fills unit is used to, using the minuscule hole for filling out hole algorithm filling organization internal to be split, then be gone using opening operation
Except the influence of pseudo-edge around tissue to be split.
Seed point setup unit is used for the center of mass point of profile seed point and key stratum as the input of adjacent layer, connects respectively
Cock wheel exterior feature seed point and center of mass point form the transmitting ray plot centered on center of mass point, in launching light in the subpoint of adjacent layer
The seed point of adjacent layer is set on the transmitting radiation direction of line chart;Wherein, be both point in coarse segmentation result simultaneously and gradient
Maximum point is just set to the seed point of adjacent layer.
Profile boundary point indexing unit is used for according to the optimal path between energy cost function calculating each pair seed point, and
New profile boundary point is marked on the profile of key stratum.
Profile boundary point connection unit is used to that profile boundary point be linked in sequence generation adjacent layer using path search algorithm
Profile.
Outline generating unit 1540 is used for the profile of each tomographic image of the CT images for generating tissue to be split;Wherein, profit
The CT for generating tissue to be split using the profile identical process with generation adjacent layer between each pair adjacent layer with iterative method schemes
The profile of each tomographic image of picture.
The profile that assembled unit 1550 is used for each tomographic image of the CT images for combining tissue to be split obtains treating for three-dimensional
Segmentation tissue.
Describe the dividing method of tissue in CT images of the invention in an illustrative manner above with reference to accompanying drawing and be
System.It will be understood by those skilled in the art, however, that in the CT images proposed for the invention described above organize dividing method and
System, can also make various improvement on the basis of present invention is not departed from.Therefore, protection scope of the present invention should be by
The content of appending claims determines.
Claims (6)
1. the dividing method organized in a kind of CT images, including:
The opacity of CT images is adjusted, the institutional framework in the CT images is obtained;
Tissue to be split is determined in acquired institutional framework, selects wherein one layer of the CT images of the tissue to be split to enter
Row Interactive Segmentation, obtains the profile of a wherein tomographic image of the CT images of the tissue to be split;Wherein,
The process of the Interactive Segmentation is:The energy cost letter of the CT images under the opacity is constructed using gradient vector
Number, in the position selected seed point mutually closed on the tissue to be split, according to the current point that the energy cost function is generated
To the profile of a wherein tomographic image of the CT images that the optimal path between the seed point is the tissue to be split;Wherein,
Process using gradient vector construction energy cost function includes:
Using Gaussian function x, y direction partial derivative as convolution kernel, extract the gradient of each pixel in the CT images to
Amount;
Using projection of the gradient vector of current point and the neighborhood point adjacent with the current point on the neighborhood direction of current point 8 to
Amount sum is used as energy cost function.
2. the dividing method organized in CT images as claimed in claim 1, also includes:
The profile of the generation adjacent layer adjacent with the layer where the profile;Wherein, the process of the profile of the adjacent layer is generated
For:Using the layer where the profile as key stratum, according to layer dependencies, selected on the adjacent layer adjacent with the key stratum
Seed point is taken, according to the wheel that the optimal path between each pair seed point that the energy cost function is generated is the adjacent layer
It is wide;
Generate the profile of each tomographic image of the CT images of the tissue to be split;Wherein, using iterative method in each pair adjacent layer
Between using the CT images for generating with the profile identical process for generating the adjacent layer tissue to be split each layer of figure
The profile of picture;
The profile for combining each tomographic image of the CT images of the tissue to be split obtains the tissue to be split of three-dimensional.
3. the dividing method organized in CT images as claimed in claim 2, wherein, generate adjacent with the layer where the profile
The process of profile of adjacent layer include:
Input of the profile seed point as adjacent layer is equidistantly chosen on the profile of the key stratum, with the key stratum phase
On adjacent layer coarse segmentation is carried out using iteration method;
Using the minuscule hole for filling out the hole algorithm filling organization internal to be split, then tissue to be split is removed using opening operation
The influence of surrounding pseudo-edge;
Using the center of mass point of the profile seed point and the key stratum as the input of adjacent layer, the profile seed is connected respectively
Point and the center of mass point form the transmitting ray plot centered on the center of mass point, in the transmitting in the subpoint of adjacent layer
The seed point of adjacent layer is set on the transmitting radiation direction of ray plot;Wherein, it is both simultaneously and terraced point in coarse segmentation result
The maximum point of degree is just set to the seed point of adjacent layer;
Optimal path between each pair seed point is calculated according to the energy cost function, and in the profile subscript of the key stratum
The new profile boundary point of note;
The profile boundary point is linked in sequence using path search algorithm generates the profile of the adjacent layer.
4. the segmenting system organized in a kind of CT images, including:
Institutional framework acquiring unit, the opacity for adjusting CT images obtains the institutional framework in the CT images;
Interactive Segmentation unit, for determining tissue to be split in acquired institutional framework, selects the tissue to be split
Wherein one layer of CT images interact formula segmentation, obtain the wheel of a wherein tomographic image of the CT images of the tissue to be split
It is wide;Wherein,
The process of the Interactive Segmentation is:The energy cost letter of the CT images under the opacity is constructed using gradient vector
Number, in the position selected seed point mutually closed on the tissue to be split, according to the current point that the energy cost function is generated
To the profile of a wherein tomographic image of the CT images that the optimal path between the seed point is the tissue to be split;Wherein,
The Interactive Segmentation unit is further included:
Gradient vector extraction unit, for using Gaussian function in the partial derivative in x, y direction as convolution kernel, extracts the CT figures
The gradient vector of each pixel as in;
Energy cost function generation unit, exists for the gradient vector using current point and the neighborhood point adjacent with the current point
Projection vector sum on the neighborhood direction of current point 8 is used as energy cost function.
5. the segmenting system organized in CT images as claimed in claim 4, also includes:
Adjacent layer outline generating unit, the profile for generating the adjacent layer adjacent with the layer where the profile;Wherein, generate
The process of the profile of the adjacent layer is:Using the layer where the profile as key stratum, according to layer dependencies, with it is described
Selected seed point on the adjacent adjacent layer of key stratum, according to optimal between each pair seed point that the energy cost function is generated
Path is the profile of the adjacent layer;
Outline generating unit, the profile of each tomographic image of the CT images for generating the tissue to be split;Wherein, using repeatedly
For method the tissue to be split is generated between each pair adjacent layer using with the profile identical process for generating the adjacent layer
The profile of each tomographic image of CT images;
Assembled unit, the profile of each tomographic image of the CT images for combining the tissue to be split obtains the to be split of three-dimensional
Tissue.
6. the segmenting system organized in CT images as claimed in claim 5, wherein, the adjacent layer outline generating unit enters one
Step includes:
Coarse segmentation unit, for equidistantly choosing input of the profile seed point as adjacent layer on the profile of the key stratum,
On the layer adjacent with the key stratum coarse segmentation is carried out using iteration method;
Fills unit, for using the minuscule hole for filling out the hole algorithm filling organization internal to be split, then using opening operation
Remove the influence of pseudo-edge around tissue to be split;
Seed point setup unit, for using the center of mass point of the profile seed point and the key stratum as adjacent layer input,
The subpoint of the profile seed point and the center of mass point in adjacent layer is connected respectively, forms the hair centered on the center of mass point
Ray plot is penetrated, the seed point of adjacent layer is set on the transmitting radiation direction of the transmitting ray plot;Wherein, it is both coarse segmentation knot
Maximum point is again simultaneously the seed point that the maximum point of gradient is just set to adjacent layer in fruit;
Profile boundary point indexing unit, for calculating the optimal path between each pair seed point according to the energy cost function,
And new profile boundary point is marked on the profile of the key stratum;
Profile boundary point connection unit, the phase is generated for the profile boundary point to be linked in sequence using path search algorithm
The profile of adjacent bed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410357152.7A CN104143190B (en) | 2014-07-24 | 2014-07-24 | The dividing method and system organized in CT images |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410357152.7A CN104143190B (en) | 2014-07-24 | 2014-07-24 | The dividing method and system organized in CT images |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104143190A CN104143190A (en) | 2014-11-12 |
CN104143190B true CN104143190B (en) | 2017-06-09 |
Family
ID=51852358
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410357152.7A Active CN104143190B (en) | 2014-07-24 | 2014-07-24 | The dividing method and system organized in CT images |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104143190B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105513055B (en) * | 2015-11-26 | 2018-09-14 | 东软集团股份有限公司 | Divide the method and apparatus of tissue in CTA images |
CN105844617A (en) * | 2016-03-17 | 2016-08-10 | 电子科技大学 | Brain parenchyma segmentation realization based on improved threshold segmentation algorithm |
CN106780470A (en) * | 2016-12-23 | 2017-05-31 | 浙江大学 | CT image nipple automated detection methods |
CN108932716B (en) * | 2017-05-26 | 2020-09-22 | 无锡时代天使医疗器械科技有限公司 | Image segmentation method for dental images |
CN108447063B (en) * | 2017-12-15 | 2020-06-19 | 浙江中医药大学 | Multi-modal nuclear magnetic resonance image segmentation method for glioblastoma |
CN108122240B (en) * | 2017-12-19 | 2020-06-12 | 东软集团股份有限公司 | Method, device and equipment for determining target area in three-dimensional image |
CN108682001A (en) * | 2018-03-30 | 2018-10-19 | 深圳市深图医学影像设备有限公司 | The measurement method and device of bone density |
CN109118499B (en) * | 2018-07-11 | 2020-06-16 | 武汉科技大学 | Layer segmentation method of coherent light tomography image based on backtracking shortest path algorithm |
CN109272476B (en) * | 2018-08-03 | 2021-12-31 | 东软医疗***股份有限公司 | PET/CT image fusion drawing method, device, equipment and storage medium |
CN110766693B (en) * | 2018-09-06 | 2022-06-21 | 北京连心医疗科技有限公司 | Method for jointly predicting radiotherapy structure position based on multi-model neural network |
KR20220012407A (en) * | 2020-08-17 | 2022-02-03 | 상하이 센스타임 인텔리전트 테크놀로지 컴퍼니 리미티드 | Image segmentation method and apparatus, electronic device and storage medium |
CN111899268B (en) * | 2020-08-17 | 2022-02-18 | 上海商汤智能科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
CN114266789B (en) * | 2021-12-28 | 2023-05-23 | 沈阳东软智能医疗科技研究院有限公司 | Infarct region segmentation method, device, equipment and medium based on brain image |
CN116152274B (en) * | 2023-04-23 | 2023-08-04 | 之江实验室 | Image segmentation method, device, equipment and storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1584936A (en) * | 2004-06-07 | 2005-02-23 | 东软飞利浦医疗设备***有限责任公司 | Three-dimensional dividing method for medical images |
-
2014
- 2014-07-24 CN CN201410357152.7A patent/CN104143190B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1584936A (en) * | 2004-06-07 | 2005-02-23 | 东软飞利浦医疗设备***有限责任公司 | Three-dimensional dividing method for medical images |
Non-Patent Citations (6)
Title |
---|
3D live-wire-based semi-automatic segmentation of medical images;Ghassan Hamarneh 等;《Medical Imaging 2005: Image Processing》;20051231;第5747卷;第1597-1603页 * |
Segmentation of the central-chest lymph nodes in 3D MDCT images;Kongkuo Lu 等;《Computers in Biology and Medicine》;20110930;第41卷(第9期);第780-789 * |
一种医学图像序列的交互式分割方法;王卫平 等;《电子质量》;20091130(第11期);第1-4页 * |
医学图像序列的交互式分割方法研究;吴炳荣;《中国优秀硕士学位论文全文数据库》;20110315(第3期);第I138-1298页 * |
基于特征点集搜索的三维序列livewire分割方法;金勇 等;《合肥工业大学学报(自然科学版)》;20110531;第34卷(第5期);第691-694页 * |
序列医学图像三维分割的一种方法;杨金柱 等;《***仿真学报》;20051231;第17卷(第12期);引言,第1.1节,第1.3节,1.4节,第2节,第3节,第5节,图3,图7,图8 * |
Also Published As
Publication number | Publication date |
---|---|
CN104143190A (en) | 2014-11-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104143190B (en) | The dividing method and system organized in CT images | |
US9978150B2 (en) | Method and system for spatial segmentation of anatomical structures | |
US8170304B2 (en) | Modeling cerebral aneurysms in medical images | |
EP2827301B1 (en) | Image generation device, method, and program | |
WO2006024974A1 (en) | Feature weighted medical object contouring using distance coordinates | |
CN107633514B (en) | Pulmonary nodule peripheral blood vessel quantitative evaluation system and method | |
US8224057B2 (en) | Method and system for nodule feature extraction using background contextual information in chest x-ray images | |
CN107622503A (en) | A kind of layering dividing method for recovering image Ouluding boundary | |
Yang | Local smoothness enforced cost volume regularization for fast stereo correspondence | |
CN106537451A (en) | Method and device for extracting vascular ridge point based on image gradient vector flow field | |
CN115631194B (en) | Method, device, equipment and medium for identifying and detecting intracranial aneurysm | |
CN107492105A (en) | A kind of variation dividing method based on more statistical informations | |
WO2020240455A1 (en) | A computer based method for classifying a mass of an organ as a cyst | |
CN113989178A (en) | CTA image blood vessel segmentation and flat scan image prediction based on self-supervision learning | |
Chen et al. | An automated approach for fibrin network segmentation and structure identification in 3D confocal microscopy images | |
Bukenya et al. | A review of blood vessel segmentation techniques | |
Zhao et al. | Automated 4D Segmentation of Aortic Magnetic Resonance Images. | |
KR101054043B1 (en) | Mothod of generating 3d sterioscopic image from 2d medical image | |
Pécot et al. | Non parametric cell nuclei segmentation based on a tracking over depth from 3d fluorescence confocal images | |
Oliveira et al. | Automatic Couinaud Liver and Veins Segmentation from CT Images. | |
Nikravanshalmani et al. | Three-dimensional semi-automatic segmentation of intracranial aneurysms in CTA | |
WO2014030262A1 (en) | Shape data-generating program, shape data-generating method and shape data-generating device | |
Li et al. | Improved live-wire algorithm for kidney image segmentation | |
KR102102369B1 (en) | Method and apparatus for estimating matching performance | |
Yan et al. | Segmentation of liver metastasis on CT images using the marker-controlled watershed and fuzzy connectedness algorithms |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |