CN102346911A - Method for segmenting blood vessel in digital subtraction angiography (DSA) image sequence - Google Patents

Abstract

The invention discloses a method for segmenting a blood vessel in a digital subtraction angiography (DSA) image sequence. The method comprises the following steps of: removing noise from a background, performing background separation on an angiography sequence image by a fuzzy C-mean clustering algorithm, and extracting the blood vessel of the angiography sequence image in the angiography image sequence by a background difference method. The method can be used for effectively segmenting the angiography sequence, and is simple in operation and high in segmenting efficiency.

Description

The method of in the digital subtraction angiography image sequence, cutting apart blood vessel

Technical field

The invention belongs to the medical imaging technology field, in particular to a kind of method of in the digital angiographic image sequence, cutting apart blood vessel.

Technical background

(Digital Subtraction Angiography is a kind of new x-ray imaging system DSA) to the digital subtraction angiography technology, is the product that routine angiography and robot calculator image processing techniques combine.The imaging ultimate principle of DSA is the angiogram X line fluoroscopic image that will be examined after the position is not injected contrast preparation and injected contrast preparation; Respectively after the image amplifier gain; Again with high-resolution television camera tube scanning; Image segmentation is become many lattices; Make matrixing; The video image that formation is made up of the pixel in the lattice; Through logarithm amplification and analog to digital conversion is the numeral of different numerical value; Forming digital picture also stores respectively; Import then that robot calculator is handled and the numerical information of two width of cloth images is subtracted each other; The difference signal of the different numerical value that obtain; Become common simulating signal with digital-to-analog conversion through the contrast enhancing again; Obtained the removal bone; Muscle and other soft tissue only stay the subtraction image of simple blood vessel image, show through display.

Image through DSA handles makes the image of blood vessel more clear, and is safer when carrying out intervene operation.X ray coronarogram picture is widely adopted in clinical diagnosis, yet owing to cardiac cycle is beated, the decay of X ray and the unequal unpredictable factor of exposure make extraction coronary artery blood vessel and overall topology very difficult.Nowadays, the vessel extraction algorithm mainly is divided into following six big types: (1) based on pattern-recognition, (2) based on model, (3) based on tracking, (4) based on artificial intelligence, based on neural network, (6) are detected based on tube (5).Yet most of coronary artery vessel extraction algorithms all need certain man-machine interaction, and this is not only consuming time, and have subjective judgement, easy error.

Summary of the invention:

The present invention proposes a kind of method of automatic extraction coronary artery, and this method can effectively be cut apart the vessel extraction in the digital subtraction angiography image sequence.

The method of automatic extraction coronary artery provided by the invention the steps include:

1) background denoising: reduce contrastographic picture because the erroneous segmentation that brightness irregularities produces.

2) background separation of radiography sequence image:, in the contrastographic picture sequence, extract background template through the fuzzy C-means clustering algorithm.

3) vessel extraction of radiography sequence image: the difference method through the front and back frame extracts the blood vessel that needs frame.

Description of drawings

Below in conjunction with accompanying drawing and embodiment the present invention is further specified.

Fig. 1 a is the original contrastographic picture before this method denoising;

Fig. 1 b is the contrastographic picture after this method denoising;

Fig. 2 a and Fig. 2 b are the blood vessel of front and back two frame the same areas;

Fig. 3 a is first frame of one section radiography sequence;

Fig. 3 b is the tenth frame of one section radiography sequence;

Fig. 3 c is the 20 frame of one section radiography sequence;

Fig. 3 d is the 30 frame of one section radiography sequence;

Fig. 3 e is the 40 frame of one section radiography sequence;

Fig. 3 f is one section background image that the radiography sequence extracts;

Fig. 4 a is the first frame blood vessel segmentation image after this method blood vessel segmentation;

Fig. 4 b is the tenth a frame blood vessel segmentation image after this method blood vessel segmentation;

Fig. 4 c is the 20 a frame blood vessel segmentation image after this method blood vessel segmentation;

Fig. 4 d is the 40 a frame blood vessel segmentation image after this method blood vessel segmentation;

Embodiment

1) background denoising

Because the X ray unequal factor of making public, make to exist large stretch of speck on the contrastographic picture, and these specks move along with the change of image in image sequence, these specks have great influence to the background separation algorithm.The purpose of background denoising is that the speck on these contrastographic pictures is removed.Here we utilize nonlinear filtering to remove speck.Key step is following: at first be to the average gray of each width of cloth image calculation entire image of contrastographic picture sequence, be designated as In_average.Secondly, all pixels being divided into two types according to the average gray size, be type one less than the pixel of In_average, is class two greater than the pixel of In_average.The average gray of the pixel in the compute classes two is designated as In_average2.For the pixel of gray scale, its gray scale is made as In_average2 above In_average2.Like this, just obtain the result shown in Fig. 1 b, removed the speck in the image.

2) background separation of radiography sequence image

The contrastographic picture sequence has following characteristic: 1, can find that through observing the contrastographic picture sequence at former frames of sequence, contrast preparation does not also enter into the coronary artery blood vessel, this moment, the coronary artery blood vessel can not develop under radiography, did not observe the coronary artery blood vessel in the image; Along with the injection of contrast preparation, the coronary artery blood vessel begins to appear, and the gray scale that shows as vessel segment increases in image gradually; In the latter stage of contrastographic picture sequence, contrast preparation flows out gradually, and the coronary artery blood vessel fades away in image.Therefore, in the contrastographic picture sequence, there is not the absolute threshold value that can distinguish blood vessel and background.2, the position of coronary artery blood vessel is different in the image of different frame with shape, and shown in Fig. 2 a and Fig. 2 b, two frame contrastographic pictures are more or less the same before and after can seeing.

2.1) background separation threshold matrix and difference threshold matrix

Because in the contrastographic picture sequence; Can not exist an absolute threshold value background and blood vessel can be separated; Therefore at first define a threshold matrix, the entry of a matrix element is the threshold value that each pixel can be separated background and blood vessel in the image, and the size of matrix is exactly the size of image pixel matrix.

Definition 1: in one section contrastographic picture sequence

; Wherein n is a picture number in the sequence; M1*m2 is the resolution ratio of contrastographic picture, deposits

Background separation threshold matrix T=t (i, j) with corresponding background separation threshold matrix difference matrix R=r (i, j), wherein (i, j) ∈ R ², every image (i, j) the background separation threshold value of individual pixel in the expression sequence.

2.2) fuzzy C-means clustering algorithm (FCM)

Fuzzy C-means clustering is to confirm that with degree of membership each data point belongs to a kind of clustering algorithm of a certain type.It is divided into m ambiguity group to n vector (ai), and asks every group cluster centre, makes the cost function of non-similarity index reach minimum.What fuzzy C-means clustering adopted is fuzzy the division, makes each given data point confirm that 0,1 degree of membership it belongs to the degree of each group with value.Simultaneously, being subordinate to matrix U allows to exist value at 0,1 element.Usually, the degree of membership of a data set with always equal 1:

$\underset{i=1}{\overset{c}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}=1,\∀j=1,...,n---\left(1\right)$

The objective function of FCM is:

$J(U,c_1,...,c_c)=\underset{i=1}{\overset{c}{\mathrm{\Σ}}}{J}_i=\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\underset{j}{\overset{n}{\mathrm{\Σ}}}{u}_{\mathrm{ij}}^m{d}_{\mathrm{ij}}^2---\left(2\right)$

0≤u wherein _Ij≤1, c _iBe the cluster centre of ambiguity group I, d _Ij=|| c _i-x _j|| be defined as the Euclid distance between i cluster centre and j data points, and m ∈ [1, ∞) be a weighted index.

2.3) threshold matrix element t (i, calculating j)

We are that (i, gray-scale value j) takes out r, as data set H with coronarogram as all pixels of sequence.Do following analysis for data set H: utilize the fuzzy C-means clustering algorithm that data set H is divided into two types of H _MaxWith H _Min, obtain the fuzzy clustering center C of these two data sets _HmaxAnd C _Hmin, fuzzy membership U _HmaxAnd U _Hmin, standard variance D _HmaxAnd D _HminFor threshold matrix, if t (i, j)=C _Hmin, can cause the contrastographic picture medium vessels manyly must be classified as background template, if t (i, j)=C _HminThe pixel that can feasiblely originally be background is not classified as background template and becomes noise.For difference matrix, if (i j) approaches zero explanation fuzzy clustering center C to r more _HmaxAnd C _HminApproaching more, just in sequence, (i j) nearly all is background pixel to pixel; On the contrary, if (i, j) (i's big more remarked pixel j) beats in sequence clearly r, and (i is j) more near the blood vessel gray-scale value for pixel.At this moment, C _HmaxJust more near the average gray of background pixel, C _HminJust more near the average gray of blood vessel pixel.

The definition 2: background separation threshold matrix T=t (i, j) in element t (i, j) definition as follows:

$t(i,j)=U_{H_{\max }}(C_{H_{\max }}-D_{H_{\max }})+U_{H_{\min }}(C_{H_{\min }}-D_{H_{\min }})---\left(3\right)$

Wherein, U _Hmax+ U _Hmin=1.

The definition 3: background separation threshold matrix T=t (i, j) in element t (i, j) definition as follows:

$r(i,j)=C_{H_{\max }}-C_{H_{\min }}---\left(4\right)$

2.4) calculating vessel extraction threshold value

By the background separation difference threshold matrix of above-mentioned definition, definition vessel extraction threshold value T _VesselBe the minimum value of background separation difference threshold value, just the critical gray-scale value of pixel from the blood vessel pixel jump to background pixel.Here we use the fuzzy C-means clustering algorithm equally, and (i, j) value of all elements in the matrix utilizes FCM that data set I is divided into two types of I as data set I with R=r _MaxWith I _Min, obtain the fuzzy clustering center C of these two data sets _ImaxAnd C _Imin, standard variance D _ImaxAnd D _Imin, then:

$T_{\mathrm{vessel}}=I_{\min }+D_{I_{\min }}---\left(5\right)$

2.5) the background template extraction

Below our two matrixes that will go out through previous calculations, with background template from coronarogram as extracting the sequence.Background template be defined as with sequence in image the gray matrix B of same resolution is arranged _M1*m2=b (i, j).Like this, open the gray matrix of contrastographic picture in the sequence each

With background separation threshold matrix T _M1*m2(i, j), at first we define each (Im) to=t _zA background template B is arranged _k=b _k(i, j), wherein

$B_k=b_k(i,j)=\left\{\begin{array}{c}t(i,j)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;I_k(i,j)<t(i,j)\\ I_k(i,j)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;I_k(i,j)>t(i,j)\end{array}\right.---\left(6\right)$

Like this, the background template of contrastographic picture sequence is:

$B_{m1*m2}={\left(b(i,j)\right)}_{m1*m2}={\left(\frac{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{b}_k(i,j)}{n}\right)}_{m1*m2}---\left(7\right)$

3) vessel extraction of radiography sequence image

Characteristic 2 by the radiography sequence can know that for k frame radiography figure, best background separation template must be the k-1 two field picture, yet the k-1 two field picture possibly comprise blood vessel, so our sequence background template B of needing said extracted to go out simultaneously _M1*m2Suppose k-1, k frame contrastographic picture I _K-1, I _kThe vessel extraction image be respectively V _K-1, V _kAt V _K-1In, if the gray-scale value of certain pixel is 255, represent that this pixel is at I _K-1In be background pixel, be I at the background separation stencil value of this pixel this moment _K-1Pixel value at this point; Equally, at V _K-1In, if certain pixel value is not 255, just this pixel is the blood vessel pixel, at this time, is the pixel value of sequence background template at this point at the background separation stencil value of this pixel.Algorithm obtains the blood vessel background template matrix V E of every later radiography figure since the 2nd two field picture _k, k=2,3 ..., with this matrix with

Relatively, utilize the 2.2 vessel extraction threshold values that calculate simultaneously, obtain vessel extraction image V _k, with V _kBe used for above-mentioned algorithm, obtain the vessel extraction image of all radiography figure.

$V_k=\mathrm{abs}({\mathrm{VE}}_k-{\mathrm{Im}}_k)=\left\{\begin{array}{c}0\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\mathrm{abs}{({\mathrm{VE}}_k-{\mathrm{Im}}_k)}_{(i,j)}<=T_{\mathrm{vessel}}\\ 255\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\mathrm{abs}{({\mathrm{VE}}_k-{\mathrm{Im}}_k)}_{(i,j)}>T_{\mathrm{vessel}}\end{array}\right.---\left(8\right)$

Claims (4)

1. the invention discloses a kind of method of in the digital angiographic image sequence, cutting apart blood vessel.

2. one kind is adopted the described method of in the digital angiographic image sequence, cutting apart blood vessel of claim 1, it is characterized in that may further comprise the steps:

1) background denoising.

2) background separation of radiography sequence image.

3) vessel extraction of radiography sequence image.

3. by the method for cutting apart blood vessel in the described angiographic image sequences of claim 2, it is characterized in that what the method for described background denoising adopted is non-linear filtering method.

4. by the method for cutting apart blood vessel in the described angiographic image sequences of claim 2, it is characterized in that the fuzzy C-means clustering algorithm that the background separating method of described radiography sequence image adopts.

Images