CN103324934A - Blood vessel center line automatic extraction method based on parallel structure detection and clustering - Google Patents

Abstract

The invention provides a blood vessel center line automatic extraction method based on parallel structure detection and clustering. The blood vessel center line automatic extraction method solves the problems that in a contrastographic picture, the grey level distribution is not even, and due to a large amount of background noise, the vascular structure is hard to extract. The method comprises the steps of utilizing three-order differential operators for calculating the direction of an image border by arranging the position of the detection border of a gradient response operator; combining a curve model on the basis of straight lines, circular curves and Euler spiral lines to build an intrinsic curve bundle of image border points, adopting the intrinsic curve bundle to conduct description of the geometrical features on the image border, and extracting all contours of blood vessels according to the geometrical consistency of the image border; utilizing the parallel features of the borders of the blood vessels for completely detecting two parallel borders of the blood vessels, searching for the positions with the same distance as the two parallel borders, connecting the centers of all the blood vessels, and finally obtaining the center lines of the blood vessels. The method is high in precision, strong in adaptability and suitable for the fields of cardiovascular disease computer-assisted diagnosis and treatment.

Description

Based on the vessel centerline extraction method of parallel construction detection with cluster

Technical field

The present invention relates to a kind of vessel centerline extraction method based on parallel construction detection and cluster, belong to angiocardiopathy Computer aided decision field.

Background technology

Along with human living standard's raising, cardiovascular and cerebrovascular disease has become the No.1 cause of the death that affects human health.The average morbidity rate of China's coronary heart disease is 6.49%, and the incidence of disease and case fatality rate are zooming trend in recent years.Therefore, in early days cardiovascular and cerebrovascular disease is carried out quantitative Diagnosis and risk assessment and can prevent effectively that sb.'s illness took a turn for the worse ill, thereby prolong human expected life, improve the human life quality.

In preclinical therapy, the Coronary Artery Structure analysis all is based on the two-dimensional x-ray contrastographic picture.X ray angiography (X-RayAngiography, XRA) is considered diagnosis of coronary heart disease and definite " goldstandard " of therapeutic scheme.Because three-dimensional vessel foreshortening and overlapping, the disappearance of coronary artery structure three-dimensional information make the identification of coronary artery structure produce larger difficulty.The coronary artery three-dimensional reconstruction can be the space multistory information that the doctor provides coronary artery, can effectively assist the clinical medical diagnosis on disease of carrying out.For carrying out the coronary artery three-dimensional reconstruction, need accurately from two dimensional image, to be partitioned into blood vessel structure information, comprise center line, bifurcation and the diameter information of blood vessel.Wherein, the extraction of vessel centerline is part most crucial in the blood vessel segmentation.But, because blood vessel structure is complicated, and be subjected to the impact of contrast preparation larger during its imaging, cause gray scale easily to present unevenness, so that the central line pick-up of blood vessel difficulty increases.

Under desirable condition, the medical image medium vessels has many features with respect to other targets, has continuous center line, blood vessel such as blood vessel and has the width of two parallel boundary, blood vessel and brightness continuous and even variation etc.Vessel centerline shows as continuous curve between the parallel boundary.Therefore, can consider to utilize blood vessel parallel boundary feature to be used for extracting the center line of blood vessel, pass through again the parameter informations such as diameter, direction vector, bifurcation and vascular bending position of the vessel centerline information acquisition blood vessel of extraction.

A lot of vessel centerline extraction algorithms have been arranged at present, have obtained huge progress, but also had the problem of the following aspects:

1. traditional vessel centerline extraction algorithm needs a large amount of manually reciprocal processes, and the algorithm starting condition is set, and is not only consuming time, and is subject to interference from human factor, can't reappear.

2. in the vessel extraction process, because the coronary angiography image is high because of the background complexity, contrast is low, and the reason such as bulk information disappearance in the two dimensional image, blood vessel full-automatic dividing process is faced with great challenge.

3. the vessel borders that traditional vessel centerline extraction algorithm can be weak with topography's characteristic information can be disallowable, can not extract well the integral edge of blood vessel.

The structure of identification blood vessel is significant for research and the diagnosis and treatment of vascular diseases from medical image.In contrastographic picture, vascular tree is the human body three-dimensional blood vessel structure has lacked spatial information after x-ray bombardment the form of expression, can not clearly express the space structure of blood vessel.Therefore, accurate vessel extraction is particularly important for clinical demand.Traditional vessel extraction method not only needs a large amount of artificial interventional procedures, and it is large to relate to the algorithm calculated amount.

Summary of the invention

The present invention is directed to the problems referred to above, proposed a kind of vessel centerline extraction method based on parallel construction detection and cluster, may further comprise the steps:

The first step, image border point detects: the position of gradient response operator detected image marginal point is set, adopts the direction of three rank differentiating operator computed image marginal points;

Second step, point feature in image border is described: the position, image border and the directional information that obtain according to detection, set up the intrinsic curve bundle of image border point in conjunction with the specific curves model, in order to the geometry consistance at Description Image edge, the intrinsic curve bundle has represented all curves that may become by the blood vessel profile border of this marginal point;

The 3rd step, image border curve cluster: comprise boundary curve cluster and the interior boundary curve cluster of image overall scope in the Image neighborhood scope;

In the 4th step, parallel construction detects: for the vessel borders that extracts, consider the parallel feature of vessel borders, make up parallel lines to model, and take this model as foundation, obtain two parallel boundary of blood vessel according to the parallel lines unit detection.

The 5th step, the blood vessel center line drawing: vessel centerline shows as continuous curve between the parallel boundary, based on two parallel boundary of detected blood vessel, seek and two positions that the parallel boundary distance is identical, obtain the blood vessel center point, this process that circulates connects all blood vessel center points, finally obtains the center line of blood vessel.

The neighborhood scope of the boundary curve cluster Given Graph picture in the described Image neighborhood scope of the 3rd step, marginal point in the territory forms one section discrete curve, intrinsic curve bundle for all image border points in the template, and according to its how much consistance, seek the common curve that it exists in the intrinsic curve bundle of all marginal points, connect the intrinsic curve bundle that all common curves form discrete curve.

Boundary curve cluster in the described image overall scope of the 3rd step is sought the geometry consistance of overall continuous curve by the intrinsic curve bundle between the discrete curve, reconstruct overall continuous curve, extract all boundary of blood vessel, rejecting can't be satisfied the conforming image border of intrinsic curve beam geometry point, thereby removes pseudo-blood vessel profile.

The vessel centerline extracting method advantage that the present invention proposes is:

1. whole vessel centerline leaching process full automation does not need manually alternately, has got rid of the artificial interference factor;

2. adopt three rank differentiating operator result of calculations to redefine image edge direction, utilize the edge transition form of presentation of intrinsic curve Shu Zuowei profile leaching process.The new direction at edge becomes the important parameter that curve connects in the edge connection procedure.According to how much consistance, effectively removed pseudo-blood vessel profile, and the complete vessel borders that kept.

3. take full advantage of the parallel characteristic information of vessel borders, algorithm can accurately be located the parallel lines structure of contrastographic picture medium vessels, and then two parallel boundary based on blood vessel extract vessel centerline quickly and accurately.

Description of drawings

Fig. 1 is based on the vessel centerline extraction method process flow diagram of parallel construction detection with cluster;

The geometry consistance computing method schematic diagram of Fig. 2 intrinsic curve interfascicular;

Fig. 3 blood vessel center point extraction algorithm schematic diagram.

Embodiment

Below in conjunction with accompanying drawing the present invention is done further introduction.Accompanying drawing 1 is based on the vessel centerline extraction method process flow diagram of parallel construction detection and cluster, specifically may further comprise the steps:

Step S101 inputs two-dimensional x-ray contrastographic picture f, detects position and the direction of all marginal points in the contrastographic picture.In concrete operations, the position of marginal point is by obtaining in the lower detection gradient operator of cartesian coordinate system peak response sub-pixel location:

$f_x^2{f}_{\mathrm{xx}}+{2f}_x{f}_y{f}_{\mathrm{xy}}+f_y^2{f}_{\mathrm{yy}}=0---\left(1\right)$

Wherein, $f_x=\frac{\∂f}{\∂x},$ $f_y=\frac{\∂f}{\∂y},$ $f_{\mathrm{xx}}=\frac{{\∂}^{2}f}{{\∂x}^2},$ $f_{\mathrm{xy}}=\frac{{\∂}^{2}f}{\∂x\∂y},$ $f_{\mathrm{yy}}=\frac{{\∂}^{2}f}{{\∂y}^2}$

And the direction at edge

Then be that one, two, three rank differentiating operators by image define:

$\stackrel{\→}{N}={\left[\begin{array}{c}{F}_x\\ F_y\end{array}\right]}_{F(x,y)=0}={\left[\begin{array}{c}{(f_x^2{f}_{\mathrm{xx}}+{2f}_x{f}_y{f}_{\mathrm{xy}}+f_y^2{f}_{\mathrm{yy}})}_x\\ {(f_x^2{f}_{\mathrm{xx}}+{2f}_x{f}_y{f}_{\mathrm{xy}}+f_y^2{f}_{\mathrm{yy}})}_y\end{array}\right]}_{F(x,y)=0}$

$={\left[\begin{array}{c}{2f}_x{f^2}_{\mathrm{xx}}+{2f}_x{f^2}_{\mathrm{xy}}+{2f}_y{f}_{\mathrm{xx}}{f}_{\mathrm{xy}}+{2f}_y{f}_{\mathrm{yy}}{f}_{\mathrm{xy}}+{2f}_x{f}_y{f}_{\mathrm{xxy}}+{2f}_y^2{f}_{\mathrm{xyy}}+f_x^2{f}_{\mathrm{xxx}}\\ {2f}_y{f^2}_{\mathrm{yy}}+{2f}_y{f^2}_{\mathrm{xy}}+{2f}_x{f}_{\mathrm{xx}}{f}_{\mathrm{xy}}+{2f}_x{f}_{\mathrm{yy}}{f}_{\mathrm{xy}}+{2f}_x{f}_y{f}_{\mathrm{xyy}}+{2f}_x^2{f}_{\mathrm{xxy}}+f_y^2{f}_{\mathrm{yyy}}\end{array}\right]}_{F(x,y)=0}---\left(2\right)$

Wherein, $F(x,y)=f_x^2{f}_{\mathrm{xx}}+{2f}_x{f}_y{f}_{\mathrm{xy}}+f_y^2{f}_{\mathrm{yy}}.$

Step S102 based on image border point position and the directional information that detection obtains, makes up the intrinsic curve bundle of image border in conjunction with multistage curve model.Any continuous curve available parameter form is expressed as:

$k\left(s\right)=k_0+\frac{\mathrm{dk}}{\mathrm{ds}}\left(0\right)s+\frac{1}{2}\frac{d^2}{{\mathrm{ds}}^2}\left(0\right)s^2+o\left(s^3\right)---\left(3\right)$

K (s) is curvature, and s is arc length.

The height of how many signature song line model exponent numbers of number of parameters in the formula (3), parameter is fewer, and signature song line model exponent number is lower.When parameter satisfies

The time, this zeroth order curve definitions straight line L ₁When parameter satisfies

The time, this first order curve has defined circular arc L ₂When parameter satisfies The time, this curve of order 2 represents Euler's helix L ₃(EulerSpirals); When parameter satisfies k ₀≠ 0, $\frac{\mathrm{dk}}{\mathrm{ds}}\left(0\right)\≠0,$ $\frac{d^{n}k}{{\mathrm{ds}}^n}\left(0\right)\≠0,$ … $\frac{d^{n+1}k}{{\mathrm{ds}}^{n+1}}\left(0\right)=0$ The time, this curve definitions N rank curve Taylor polynomials.Exponent number is higher, and curve model is more accurate, but computation process is more unstable.Therefore, the present invention does not consider the curve that three rank are above, sets up the low order curve model that comprises straight line, circular arc, Euler's helix, and expressing possibility with this becomes the profile of vessel borders fragment.

A given image border point by its position and directional information, can define cluster by the low order pencil of curves of this marginal point in conjunction with above-mentioned curve model.Curve model is unique to be determined in order to make, and can determine straight line and circular arc by its position and the directional information of marginal point; But Euler's helix also needs the rate of change γ of curvature κ and curvature to determine except the position and direction parameter at edge.Therefore simultaneously, the disturbance on marginal existence position and the direction (△ x, △ θ) makes up the pencil of curves of four-dimensional parameter space (△ x, △ θ, κ, γ) Description Image marginal point.In concrete operations, discrete (△ x, △ θ) obtains (△ x, 0, △ x) and (△ θ, 0, △ θ), finally obtains the intrinsic curve bundle of image border point.The intrinsic curve bundle has represented the curve that might become blood vessel profile border by this edge.

Step S103 makes up the intrinsic curve bundle of each marginal point of contrastographic picture according to step S102, in certain neighborhood scope (templates such as 7 * 7), seek the common curve that exists in the intrinsic curve bundle of all marginal points wherein.

In four-dimensional parameter space (△ x, △ θ, κ, γ), the intrinsic curve bundle of an image border point can represent with a closed region in the four-dimentional space.In specific operation process, discrete parameter (△ x, △ θ) is fixed as (△ x, 0, △ x) and (△ θ, 0, △ θ), after △ x and △ θ determine, the intrinsic curve bundle of this marginal point is at parameter (κ, γ) space can be with the For Polygons Representation in the two dimensional surface, as shown in Figure 2, and polygonal region β ₁And β ₂Represent respectively the intrinsic curve bundle of two marginal points, the plane that exists together by calculating two polygonal crossing areas, can obtain the common pencil of curves between two image border points, the geometry consistance that Here it is between the point of image border.Keep how much consistent boundary chains in the neighborhood, produced a series of orderly marginal point combinations, form discrete curve.These marginal points can couple together by the low order curve model and form profile fragment in the neighborhood scope.

Step S104 seeks the geometry consistance of overall contour curve by the intrinsic curve bundle between the discrete curve, detect the blood vessel profile in the contrastographic picture.Obtained a large amount of discrete curves among the step S103, its intrinsic curve bundle is at parameter (κ, γ) space can be expressed as the polygon closed region in the two dimensional surface, S103 is similar with step, by calculating the area that intersects between a plurality of polygons, obtain the common pencil of curves between many discrete curves, can obtain overall continuous curve, finally extract all boundary of blood vessel.

Step S105 for the vessel borders that extracts, has obvious parallel lines feature according to blood vessel in the picture structure form of expression, adopts parallel lines to model (" line " comprises straight line and curve), designs a kind of image parallel lines feature extraction algorithm.

Article two, smooth curve L and S is if L and S can be divided into corresponding line segment to (the line segment here comprises straight-line segment and the two-layer meaning of segment of curve), l ₁With s ₁, l ₂With s ₂..., l _nWith s _n(n 〉=1, n ∈ N), and each line segment is to satisfying condition:

1）l ₁∪l ₂∪l ₃…∪l _n=L， s ₁∪s ₂∪s ₃…∪s _n=S；

2) for any a pair of l wherein _kWith s _k, k=1,2 ..., n, n ∈ N can find a direction Make l _kWith s _kAll internal point of surrounding make progress to l from the party _kWith s _kDistance and be definite value, think that then these two lines are one group of parallel lines pair.

Step S106 has accurately located the main parallel construction of former figure medium vessels in step S105, for per two parallel boundary, adopt the central point of the method searching blood vessel of progressive track-while-scan.Arbitrary region is chosen initial point P at random between acquired two parallel boundary _k, get a P _kContiguous borderline image border point direction is reference direction u _k, at reference direction u _kUpper carrying out forward respectively and backward progressive tracking.

As shown in Figure 3, to be tracked as forward example, according to reference direction u _k, with a P _kDistance is that the position of d obtains and direction u _kVertical direction g _k, direction g _kIntersect at 2 A with parallel boundary _K+dAnd B _K+d, line taking section A _K+dB _K+dMid point get P _K+d, this is a central point of blood vessel, this process that circulates obtains initial point P _kAll blood vessel center points forward.Initial point P _kProgressive following principle is identical backward, obtains initial point P _kBehind the blood vessel center point of following the tracks of backward, connect all central points between the parallel boundary, finally obtain the center line of blood vessel.

Although present invention is described with reference to preferred embodiment; but the above example does not consist of the restriction of protection domain of the present invention; any in spirit of the present invention and principle modification, be equal to and replace and improvement etc., all should be included in the claim protection domain of the present invention.

Claims (3)

1. based on the vessel centerline extraction method of parallel construction detection with cluster, it is characterized in that, may further comprise the steps:

The first step, image border point detects: the position of gradient response operator detected image marginal point is set, adopts the direction of three rank differentiating operator computed image marginal points;

Second step, point feature in image border is described: the position, image border and the directional information that obtain according to detection, set up the intrinsic curve bundle of image border point in conjunction with the specific curves model, in order to the geometry consistance at Description Image edge, the intrinsic curve bundle has represented all curves that may become by the blood vessel profile border of this marginal point;

The 3rd step, image border curve cluster: comprise boundary curve cluster and the interior boundary curve cluster of image overall scope in the Image neighborhood scope;

In the 4th step, parallel construction detects: for the vessel borders that extracts, consider the parallel feature of vessel borders, make up parallel lines to model, and take this model as foundation, obtain two parallel boundary of blood vessel according to the parallel lines unit detection;

The 5th step, the blood vessel center line drawing: vessel centerline shows as continuous curve between the parallel boundary, based on two parallel boundary of detected blood vessel, seek and two positions that the parallel boundary distance is identical, obtain the blood vessel center point, this process that circulates connects all blood vessel center points, finally obtains the center line of blood vessel.

2. as claimed in claim 1 based on the vessel centerline extraction method of parallel construction detection with cluster, it is characterized in that, in the boundary curve clustering method in the described Image neighborhood scope of the 3rd step, answer the neighborhood scope of Given Graph picture, marginal point in the scope forms one section discrete curve, intrinsic curve bundle for all image border points in the template, and according to its how much consistance, seek the common curve that it exists in the intrinsic curve bundle of all marginal points, connect the intrinsic curve bundle that all common curves form discrete curve.

3. as claimed in claim 1 based on the vessel centerline extraction method of parallel construction detection with cluster, it is characterized in that, in the boundary curve clustering method in the described image overall scope of the 3rd step, seek the geometry consistance of overall continuous curve by the intrinsic curve bundle between the discrete curve, reconstruct overall continuous curve, extract all boundary of blood vessel, rejecting can't be satisfied the conforming image border of intrinsic curve beam geometry point, thereby removes pseudo-blood vessel profile.

Images