CN103054563B - A kind of quantification of blood vessel wall image texture characteristic and extracting method - Google Patents

Abstract

The present invention proposes a kind of quantification and extracting method of blood vessel wall image texture characteristic, and the EARLY RECOGNITION for blood vessel wall pathological changes is diagnosed, and it comprises the following steps: utilize medical image system to gather multiframe continuous print blood vessel wall image; Blood vessel wall image described in each frame is chosen blood vessel wall inner membrance-middle rete as area-of-interest, and extract the textural characteristics value of described area-of-interest, then obtain its meansigma methods; Statistical analysis is carried out to described textural characteristics value, builds the textural characteristics population of parameters being used for the diagnosis of blood vessel wall pathological changes EARLY RECOGNITION.The quantification of described blood vessel wall image texture characteristic and extracting method can extract the textural characteristics value of any area-of-interest blood vessel wall, when contributing to blood vessel wall lesion detection, pathological changes is accurately located, and the suitability is stronger, without the need to carrying out HardwareUpgring to existing clinical imaging system, upgrade cost is low, easily accepted by hospital and doctor, facilitate clinical expansion.

Description

A kind of quantification of blood vessel wall image texture characteristic and extracting method

Technical field

The present invention relates to quantification and the extracting method of blood vessel wall image texture characteristic, particularly relate to a kind of quantification and extracting method of the blood vessel wall image texture characteristic based on image texture characteristic.

Background technology

Cardiovascular and cerebrovascular disease serious harm human health, has become day by day urgent great social common problem.Atherosclerosis is the arch-criminal of cardiovascular and cerebrovascular disease.Atherosclerosis is a long-term and complex process, if accurate identification and diagnosis can be realized to it in early days, just can intervening timely in early days and treating in disease, and effectively improve cure rate, reduction medical expense.Judge that atherosclerotic Main Basis is blood vessel structure clinically at present, function and Hemodynamic Factors, namely Ink vessel transfusing media thickness, with or without the elasticity of speckle, luminal stenosis degree, tremulous pulse and blood flow rate and flow etc., Main Diagnosis method has digital subtraction angiography, magnetic resonance angiography, CT angiography, pulse wave velocity and color doppler ultrasonography.These methods can detect the structure of blood vessel, function and hemodynamics.

In current noninvasive detection method, Ink vessel transfusing media thickness (IMT) measurement method is based on ultrasound image acquisition, ultra sonic imaging is simple to operation, but image resolution ratio is limited, and due to gathered image be video signal, the quantity of information that image can provide and certainty of measurement are all vulnerable to impact, even degree of accuracy also can only reach 0.1mm after amplifying, the IMT error recorded is larger.Pulse wave velocity (PWV) measurement method is that blood pulsations is injected aorta by heart, aorta wall produces pulse pressure wave, and conduct to peripheral blood vessel along blood vessel wall with certain speed, by measuring pulse wave translation time between two tremulous pulse record positions and distance, PWV can be calculated.But the sensitivity of PWV is poor, not easily find that slight arterial elasticity changes, and when PWV measures, body surface range measurement there is error, obviously can affect the accuracy of data.Arm index (ABI) measurement method measures the two artery of upper extremity of patient and bilateral posterior tibial artery and (or) dorsal artery of foot systolic pressure respectively by manual method or by automatic checkout equipment, calculates arm index (ABI) value.But it is narrow or inaccessible that the reduction of arm index (ABI) can only point out the above blood vessel of ankle to have, and accurately can not locate pathological changes.Therefore PWV and ABI is merely able to the average hardness of measurement one section of blood vessel.Flow-mediated vasodilation (FMD) detection technique is that one depends on and has high-resolution two dimensional image, M type or A type, is recorded by obtaining information and the method analyzed.This inspection can reflect and judge arterial vascular endothelial function because of it, therefore can make diagnosis to early stage arteriosclerosis.But because its operation is more loaded down with trivial details, higher to the requirement of operator, make it in clinical application, universal to be affected.Therefore, these detection methods blood vessel structure function not yet change early stage, usually all effectively cannot differentiate vascular lesion.

Summary of the invention

For the problems referred to above, the object of this invention is to provide a kind of quantification and extracting method of blood vessel wall image texture characteristic, the EARLY RECOGNITION for blood vessel wall pathological changes is diagnosed.

The quantification of blood vessel wall image texture characteristic and an extracting method, it comprises the following steps:

Medical image system is utilized to gather multiframe continuous print blood vessel wall image;

Blood vessel wall image described in each frame is chosen blood vessel wall inner membrance-middle rete as area-of-interest, and extract the textural characteristics value of described area-of-interest, then obtain its meansigma methods;

Statistical analysis is carried out to described textural characteristics value, builds the textural characteristics population of parameters being used for the diagnosis of blood vessel wall pathological changes EARLY RECOGNITION.

In the present invention one better embodiment, described medical image system is ultrasonic image-forming system, optical imaging system, CT imaging system or MRI imaging system.

In the present invention one better embodiment, described textural characteristics value comprises first-order statistical properties, fractal dimension texture analysis, gray level co-occurrence matrixes, grey scale difference statistics, local gray level difference matrix and statistical nature matrix.

In the present invention one better embodiment, described first-order statistical properties comprises gray average and the standard deviation of vascular wall area.

In the present invention one better embodiment, described fractal dimension texture analysis parameter is fractal dimension.

In the present invention one better embodiment, described gray level co-occurrence matrixes feature comprises contrast, dependency, energy, unfavourable balance distance, entropy.

In the present invention one better embodiment, described grey scale difference statistical nature comprises contrast, angle direction second moment, entropy, meansigma methods.

In the present invention one better embodiment, described local gray level difference matrix feature comprises roughness, contrast, frequency, complexity, texture strength.

In the present invention one better embodiment, described statistical nature matrix character comprises contrast, covariance, diversity.

In the present invention one better embodiment, the quantification of described blood vessel wall image texture characteristic and extracting method comprise employing grader further and classify to blood-vessel image that is healthy and pathological changes.

Compared to prior art, quantification and the extracting method of described blood vessel wall image texture characteristic provided by the invention have the following advantages: one, the suitability are comparatively strong, are applicable to the image of various medical image system and the fast imaging such as optics, optoacoustic system acquisition.The textural characteristics value of any area-of-interest blood vessel wall two, can be extracted, when contributing to blood vessel wall lesion detection, pathological changes is accurately located.Three, the quantification of described blood vessel wall image texture characteristic and extracting method are directly based on the blood-vessel image that medical image system gathers, clinician and the operator with certain medical imaging devices use experience can collect required image, are not easily affected by other factors.Four, the quantification of described blood vessel wall image texture characteristic and extracting method can be used as an image post processing software module integration in existing medical image system, to promote the function of medical image system, without the need to carrying out HardwareUpgring to existing clinical imaging system, upgrade cost is low, easily accepted by hospital and doctor, facilitate clinical expansion.

Above-mentioned explanation is only the general introduction of technical solution of the present invention, in order to technological means of the present invention can be better understood, and can be implemented according to the content of description, and can become apparent to allow above and other objects of the present invention, feature and advantage, below especially exemplified by embodiment, and coordinate accompanying drawing, be described in detail as follows.

Accompanying drawing explanation

The quantification of blood vessel wall image texture characteristic that Fig. 1 provides for one embodiment of the invention and the flow chart of extracting method.

Fig. 2 adopts high frequency ultrasound imaging system to the schematic diagram of the total arteriography of mouse carotid.

Fig. 3 selects blood vessel wall inner membrance-middle rete as the schematic diagram of area-of-interest.

Fig. 4 adopts the schematic diagram that high frequency ultrasound imaging system carries out imaging to the common carotid artery of anesthetized mice and the B ultrasonic image obtained.

Fig. 5 is the healthy mice common carotid artery and the Carotid B ultrasonic image of apoe knock out mice that adopt high frequency ultrasound imaging system acquisition.

Fig. 6 is using 14 effective textural characteristics populations of parameters as characteristic of division collection, adopts the schematic diagram that KNN grader is classified to healthy mice and the Carotid ultrasonoscopy of diseased mice.

Detailed description of the invention

Below in conjunction with drawings and the specific embodiments, the present invention is further detailed explanation.

Refer to Fig. 1, one embodiment of the invention provides a kind of quantification and extracting method of blood vessel wall image texture characteristic, and it comprises the following steps:

Step S101, medical image system is utilized to gather multiframe continuous print blood vessel wall image.

In the present embodiment, medical image system is adopted to gather N frame continuous print blood vessel wall image, N contains the image at least one cardiac cycle, and such as, the image acquisition frame frequency FR of medical image system is 100 frames/second, human heartbeat's frequency f is 60Hz/ minute, cardiac cycle Tc=60/f=1 second; Then N=m × FR × Tc=100m frame (m=1,2,3 ...).Namely N should be the integral multiple of 100, as shown in Figure 2.

Be understandable that, described medical image system can be ultrasonic image-forming system, optical imaging system (as X-ray machine), CT imaging system or MRI imaging system etc.In the present embodiment, be described for the ultrasonoscopy of ultrasonic image-forming system collection.

Step S103, on blood vessel wall image described in each frame, choose blood vessel wall inner membrance-middle rete as area-of-interest, and extract the textural characteristics value of described area-of-interest, then obtain the meansigma methods of described textural characteristics value.

Be understandable that, blood vessel wall inner membrance is made up of subcutaneous connective tissue in simple squamous epithelium cell and skim, the equal echo light belt on the ultrasound image in a fine rule shape continuous and derivable; In blood vessel wall, rete is a blanking bar, and form primarily of smooth muscle cell and elastic [connective, blood vessel wall theca externa is made up of blood vessel wall outermost layer loose connective tissue, in the euphotic zone more become clear compared with inner membrance.Therefore, blood vessel wall shows as typically " two-wire bar " on the ultrasound image, namely two parallel strong echoes by a low echo area or no echo area separate.In the present embodiment, " two-wire bar " is selected to be area-of-interest on the ultrasound image, as shown in Figure 3.

In the present embodiment, the textural characteristics value of described area-of-interest comprises first-order statistical properties, fractal dimension texture analysis, gray level co-occurrence matrixes, grey scale difference statistics, local gray level difference matrix and statistical nature matrix, below will be introduced these textural characteristics values.

1, described first-order statistical properties is also called grey level histogram, and the general evaluation system characteristic of its reflection piece image intensity profile, it mainly comprises gray average and the standard deviation of vascular wall area.

In the present embodiment, if the gray value of any point (x, y) is I (x, y) in described region of interest area image (M × N), then the meansigma methods of described area-of-interest gray scale is:

$\mathrm{ROIavg}=\frac{1}{M\×N}\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}I(x,y)---\left(1\right)$

Wherein, ROIavg characterizes the average gray value of area-of-interest.

The standard deviation of area-of-interest gray scale is:

$\mathrm{ROIstd}=\sqrt{\frac{1}{M\×N-1}\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}{(I(x,y)-\mathrm{ROIavg})}^2}---\left(2\right)$

Wherein, ROIstd characterizes the uniformity coefficient of intensity profile in described area-of-interest, and ROIstd is less, then intensity profile is more even.

2, fractal dimension is as the tolerance of imaging surface degree of irregularity, and it is consistent with the perception of human vision to imaging surface coarse texture degree, and namely fractal dimension is larger, and corresponding imaging surface is more coarse; Otherwise the imaging surface of the less correspondence of fractal dimension is more smooth.In the present embodiment, described fractal dimension texture analysis parameter is fractal dimension.

Be understandable that, the fractal dimension D of the interesting image regions will estimated in theory _fdetermined by following formula:

E(ΔI ²)＝c(Δr) ^(6-2Df)(3)

In formula, E (.) is expected value; Δ I is the change of gray value, Δ I=I (x ₂, y ₂)-I (x ₁, y ₁); Δ r is the change of space length, Δ r=I (x ₂, y ₂)-I (x ₁, y ₁); C is constant.Formula (3) can be reduced to:

E(|ΔI|)＝k(Δr) ^H(4)

In formula, k=E (| Δ I|) _{Δ r=1}, then D _f=3-H, takes the logarithm to formula (4) both sides simultaneously, then has:

logE(|ΔI|)＝logk+Hlog(Δr)(5)

Under two-dimensional coordinate system, with logE (| Δ I|), for vertical coordinate, log (Δ r) is abscissa, draws some discrete points, discrete point is fitted to straight line, finally utilizes method of least square to obtain the slope of straight line, is H.Therefore fractal dimension D can be calculated _f.

3, gray level co-occurrence matrixes feature describes one of modal method of textural characteristics, and its main purpose is pixel and the distribution situation of gray scale in space in statistical picture.

Its computational methods are as follows: set the intensity value ranges of image as [0, L], by calculating gray level co-occurrence matrixes, obtain the feature texture matrix that a size is (L+1) (L+1).In this method, two very important parameter: distance d and angle θ must be considered, if select different θ and d, different feature texture matrixes can be obtained.About angle θ, it is definition like this, if I is (x, y) be (x at coordinate, y) the gray value intensity of position, and I (x, y) be 1 with the distance of surrounding 8 neighbors, under this definition, I (x, y), I (x+1, y) with I (x-1, y) angle of three pixels is 0 degree, I (x, y), I (x+1, and I (x-1 y+1), y-1) angle of three pixels is 45 degree, the rest may be inferred I (x, y), I (x, and I (x y+1), y-1) angle is 90 degree, I (x, y), I (x-1, and I (x+1 y+1), y-1) angle is 135 degree.

Get any point (x, y) in image (M × N) and depart from its another point (x+a, y+b), these 2 corresponding gray values are (i, j).Order point moves on whole picture, then can obtain different (i, j) value, if the combination that the progression of gray value is k1 then (i, j) has square kind of k1.For whole picture, count the number of times that each (i, j) value occurs, then be arranged in a square formation, then they are normalized to the probability P (i of appearance by the total degree using (i, j) to occur, j), namely this square formation becomes gray level co-occurrence matrixes.

Following normalization is carried out to gray level co-occurrence matrixes:

According to normalized gray level co-occurrence matrixes, following textural characteristics parameter can be obtained: contrast (Contrast), dependency (Correlation), energy (Energy), unfavourable balance are apart from (Homogeneity) and entropy (Entropy).Wherein:

Contrast

The contrast reflection definition of image and the degree of the texture rill depth.Texture rill is darker, and its contrast is larger, and visual effect is more clear; Otherwise contrast is little, then rill is shallow, and effect is fuzzy.

Dependency $\mathrm{Correlation}=\frac{\underset{i=0}{\overset{L}{\mathrm{\Σ}}}\underset{j=0}{\overset{L}{\mathrm{\Σ}}}p(i,j)-u_x{u}_y}{{\mathrm{\σ}}_x{\mathrm{\σ}}_y}---\left(8\right)$

Wherein, u _xand σ _xp _xmeansigma methods and standard deviation, u _yand σ _yp _ymeansigma methods and standard deviation, and p _xwith p _ybe defined as follows:

$p_x\left(i\right)=\underset{j=0}{\overset{L}{\mathrm{\Σ}}}p(i,j)---\left(9\right)$

$p_y\left(j\right)=\underset{i=0}{\overset{L}{\mathrm{\Σ}}}p(i,j)---\left(10\right)$

Relativity measurement spatial gray level co-occurrence matrix element is expert at or similarity degree on column direction, and therefore, dependency size reflects local gray level dependency in image.When matrix element value even equal time, dependency is just large; On the contrary, if matrix pixel value differs greatly, dependency is little.

Energy $\mathrm{Energy}=\underset{i=0}{\overset{L}{\mathrm{\Σ}}}\underset{j=0}{\overset{L}{\mathrm{\Σ}}}{\left(p(i,j)\right)}^2---\left(11\right)$

Energy is that gradation of image is evenly distributed a tolerance of degree and texture thickness, and when image texture strand is careful, intensity profile is even, energy value is comparatively large, otherwise, less.

Unfavourable balance distance $\mathrm{Homogeneity}=\underset{i=0}{\overset{L}{\mathrm{\Σ}}}\underset{j=0}{\overset{L}{\mathrm{\Σ}}}\frac{p(i,j)}{1+{(i-j)}^2}---\left(12\right)$

Unfavourable balance apart from the readability Sum fanction degree of reflection texture, clean mark, regular comparatively strong, be easy to describe, inverse gap value is larger; Rambling, be difficult to describe, inverse gap value is less.

Entropy $\mathrm{Entropy}=\underset{i=0}{\overset{L}{\mathrm{\Σ}}}\underset{j=0}{\overset{L}{\mathrm{\Σ}}}p(i,j)\log \left(p(i,j)\right)---\left(13\right)$

Entropy is the tolerance of the quantity of information that Description Image has, and shows the complicated process of image, and when complicated process height, entropy is comparatively large, otherwise then less.

4, the ultimate principle of grey scale difference statistic law describes the grey scale change situation between each pixel of texture image and adjacent image point thereof.

Image f's (x, y) is a bit (x, y), this point and it have the gray scale difference value Δ I of the point (x+ Δ x, y+ Δ y) of slight distance σ=(Δ x, Δ y) to be grey scale difference.If the institute of gray scale difference score value likely value has m level, then form the vector of a m dimension, calculating the rectangular histogram of Δ I, is the probability density p of i by the known gray scale difference value of rectangular histogram _σ(i).When σ gets smaller value, and gray scale difference value is the probability density p of i _σwhen () is larger i, illustrate that image texture is more coarse, otherwise illustrate that texture is thinner.Described grey scale difference statistics comprises contrast (Contrast), angle direction second moment (AngularSecondMoment), entropy (Entropy), meansigma methods (Mean).Wherein:

Contrast $\mathrm{con}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{i}^2{p}_{\mathrm{\δ}}\left(i\right)---\left(14\right)$

Contrast is grey scale difference probability density p _σsecond moment, such as, p _σabout the moment of inertia of initial point.

Angle direction second moment $\mathrm{ASM}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\left[p_{\mathrm{\δ}}\left(i\right)\right]}^2---\left(15\right)$

Work as p _σvalue homogeneous phase simultaneously, ASM is minimum; And work as p _σvalue when differing greatly, ASM is larger.

Entropy $\mathrm{ENT}=-\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{p}_{\mathrm{\δ}}\left(i\right)\log p_{\mathrm{\δ}}\left(i\right)---\left(16\right)$

Work as p _σvalue homogeneous phase simultaneously, ENT is maximum; And work as p _σvalue when differing greatly, ENT is less.

Meansigma methods $\mathrm{MEAN}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{p}_{\mathrm{\δ}}\left(i\right)---\left(17\right)$

Work as p _σvalue close to zero time, MEAN is less; Otherwise MEAN is larger.

5, local gray level difference matrix feature, if the window of gray level image f (k, l) and W × W (generally getting 3 or 5) W=2d+1, matrix of averaging

$\stackrel{\‾}{A}(k,l)=\frac{1}{W^2-1}\left[\underset{m=-d}{\overset{d}{\mathrm{\Σ}}}\underset{n=-d}{\overset{d}{\mathrm{\Σ}}}f(k+m,l+n)\right],(m,n)\≠\left(\mathrm{0,0}\right)---\left(18\right)$

Ask S (i) again, S (i) represents that gray scale is all pixels of i and corresponding Mean Matrix the absolute value sum of difference:

Described local gray level difference matrix feature comprises roughness (Coarseness), contrast (Contrast), frequency (Busyness), complexity (Complexity), texture strength (TextureStrength).Wherein:

Roughness

G in formula _hfor gray value maximum in gray level image; ε is less constant, stops f _cosfor infinitary value; P (i) represents that gray scale is the probability of the pixel appearance of i.F _cosbe worth larger interval scale gray difference less, be image ratio more coarse.

Contrast $f_{\mathrm{con}}=\left[\frac{1}{N_g(N_g-1)}\underset{i=0}{\overset{\mathrm{Gh}}{\mathrm{\Σ}}}\underset{j=0}{\overset{\mathrm{Gh}}{\mathrm{\Σ}}}{P}_i{P}_j{(i-j)}^2\right]\left[\frac{1}{n^2}\underset{i=0}{\overset{\mathrm{Gh}}{\mathrm{\Σ}}}S\left(i\right)\right]---\left(21\right)$

N in formula _gfor the sum of gray difference, be expressed as:

$N_g=\underset{i=0}{\overset{\mathrm{Gh}}{\mathrm{\Σ}}}{Q}_i,Q_i=\left\{\begin{array}{cc}1& P\left(i\right)\≠0\\ 0& P\left(i\right)=0\end{array}\right.---\left(22\right)$

F _conthe larger adjacent area gray-scale intensity that represents of value differs greatly.

Frequency

Frequency reflects the difference degree of a certain pixel and the grey scale pixel value around it, when the spatial variations of gray scale is larger, and f _busybe worth higher.

Complexity $f_{\mathrm{com}}=\underset{i=0}{\overset{\mathrm{Gh}}{\mathrm{\Σ}}}\underset{j=0}{\overset{\mathrm{Gh}}{\mathrm{\Σ}}}(\left(\right|i-j\left|\right)/\left(n^2(p\left(i\right)+p\left(j\right))\right))(P\left(i\right)S\left(i\right)+P\left(j\right)S\left(j\right))---\left(24\right)$

The visual information amount of assorted degree reflection image texture, when quantity of information is larger, f _combe worth larger.

Texture strength

The pixel comprised when image is easily definition, apparent, and fstr value is larger.

6, statistical nature matrix character, statistical nature matrix to can be used in computed image pixel to the statistical property of some distances.Make (x, y) for a bit in image, I (x, y) is the gray value of this point, and δ=(Δ x, Δ y) is space length vector.Described statistical nature matrix character comprises contrast (Contrast), covariance (Covariance), diversity (Dissimilarity).Wherein:

Contrast C ON (δ)=E{ [I (x, y)-I (x+ Δ x, y+ Δ y ²)] (26)

In formula, E{} is expected value.

Covariance COV (δ)=E{ [[I (x, y)-η] [I (x+ Δ x, y+ Δ y)-η] } (27)

In formula, η is gradation of image meansigma methods.

Diversity factor

Step S105, statistical analysis is carried out to described textural characteristics value, build the textural characteristics population of parameters being used for the diagnosis of blood vessel wall pathological changes EARLY RECOGNITION.

In the present embodiment, adopt Minitab statistical software to carry out statistical analysis to described textural characteristics value, draw the textural characteristics population of parameters that there were significant differences, and create described textural characteristics population of parameters, the EARLY RECOGNITION for blood vessel wall pathological changes is diagnosed.

Further, the quantification of described blood vessel wall image texture characteristic and extracting method comprise and adopt the blood-vessel image of grader to healthy body and pathological changes body to classify.Namely from described textural characteristics population of parameters, best characteristic parameter is chosen according to clinical knowledge and experience, and adopt grader, as blood vessel (as the common carotid artery) ultrasonoscopy of KNN (k-NearestNeighbor) grader to healthy body and pathological changes body is classified.Thus, when utilizing the quantification of described blood vessel wall image texture characteristic and extracting method to carry out blood vessel wall lesion detection, classification results can be used to guide the differentiation of vascular lesion, the foundation that the early diagnosis and therapy for cardiovascular disease provides.

For verifying the described quantification of blood vessel wall image texture characteristic provided by the invention and the feasibility of extracting method and effectiveness, below adopt the quantification of described blood vessel wall image texture characteristic and extracting method to VisualSonicsvevo2100 (VisualsonicsInc., Toronto, Canada) the total arterial images of mouse carotid of high frequency ultrasound imaging system acquisition carried out (comprising normal mouse and apoe knock out mice) quantification and the extraction of textural characteristics.Wherein, Vevo2100 is equipped with MS-550D linear array probe, and mid frequency is 40MHz.First isoflurane gas (flow rate 1L/min) anesthetized mice of 1% pure oxygen is used, then the chaeta at imaging position is removed with depilatory cream, be fixed on examining table, the body temperature of mice is made to remain on 36 ~ 38 ° by hot plate, adopt high frequency ultrasound imaging system Vevo2100 to the common carotid artery imaging of anesthetized mice, as shown in Figure 4.Carry out experimentation to 14 normal mouses and 16 apoe knock out mice (high lipid food is fed 36 weeks), two kinds of Carotid ultrasonoscopys of mice as shown in Figure 5.

Adopt the quantification of blood vessel wall image texture characteristic of the present invention and the textural characteristics value of extracting method quantification and normal mouse and apoe knock out mice, then with Minitab statistics software it is analyzed and draw 14 textural characteristics populations of parameters that there were significant differences, as shown in table 1, thus effectively distinguish normal mice and clpp gene deratization.

Table 1. normal mouse and apoe knock out mice different texture eigenvalue statistic analysis result

Textural characteristics value	Normal mouse (n=14)	Apoe knock out mice (n=16)	p value
				Gray average *	201.49±8.84	141.4±23.4	0.0001
Standard deviation	23.45±2.62	23.76±4.69	0.825
				Fractal dimension *	2.046±0.043	2.014±0.037	0.041
Contrast *	9.14±3.19	6.36±2.38	0.013
				Dependency	0.889±0.033	0.909±0.039	0.129
Energy *	0.034±0.007	0.022±0.005	0.0001
				Unfavourable balance distance	0.518±0.03	0.517±0.038	0.94
Entropy *	1.317±0.048	1.408±0.057	0.0001
				Contrast-1*	17.39±7.05	26.3±11.4	0.016
Angle direction second moment *	0.184±0.026	0.135±0.025	0.0001
				Entropy-1*	0.838±0.075	0.952±0.09	0.001
Meansigma methods-1*	0.0128±0.002	0.0162±0.003	0.001
				Roughness *	6.1±1.63	11.53±3.49	0.0001
Contrast-2	0.55±0.186	0.443±0.151	0.098
				Frequency *	2.73±1.42E-5	1.55±0.97E-5	0.016
Complexity *	4600±1837	7076±3446	0.02
				Texture strength *	9397±4332	31187±17728	0.0001
Contrast-3	1.27±0.225	1.22±0.107	0.436
				Covariance	0.797±0.123	0.87±0.083	0.073
Diversity *	1.266±0.159	0.892±0.22	0.01

Choose wherein 14 effective textural characteristics values (the textural characteristics value that in table, " * " identifies) and build population of parameters as feature set, and adopt the Carotid ultrasonoscopy of KNN grader to healthy mice and diseased mice to classify.Ultrasonoscopy in conjunction with 30 known classification results estimates 29 samples as training, and nicety of grading can reach 93.1%, as shown in Figure 6.

From experimental result, the quantification of described blood vessel wall image texture characteristic and extracting method, based on image texture characteristic, can be used for the differentiation instructing vascular lesion, for the early diagnosis treatment of cardiovascular disease provides foundation.

Compared to prior art, quantification and the extracting method of described blood vessel wall image texture characteristic provided by the invention have the following advantages: one, the suitability are comparatively strong, are applicable to the image of various medical image system and the generation of the fast imaging techniques such as optics, optoacoustic.The textural characteristics value of any area-of-interest blood vessel wall two, can be extracted, when contributing to blood vessel wall lesion detection, pathological changes is accurately located.Three, the quantification of described blood vessel wall image texture characteristic and extracting method are directly based on the blood-vessel image that various medical image system gathers, clinician and the operator with certain medical imaging devices use experience can collect required image, are not easily affected by other factors.Four, the quantification of described blood vessel wall image texture characteristic and extracting method can be used as an image post processing software module integration in existing medical image system, to promote the function of medical image system.Therefore without the need to carrying out HardwareUpgring to existing clinical imaging system, upgrade cost is low, is easily accepted by hospital and doctor, facilitates clinical expansion.

The above, only embodiments of the invention, not any pro forma restriction is done to the present invention, although the present invention discloses as above with embodiment, but and be not used to limit the present invention, any those skilled in the art, do not departing within the scope of technical solution of the present invention, make a little change when the technology contents of above-mentioned announcement can be utilized or be modified to the Equivalent embodiments of equivalent variations, in every case be do not depart from technical solution of the present invention content, according to any simple modification that technical spirit of the present invention is done above embodiment, equivalent variations and modification, all still belong in the scope of technical solution of the present invention.

Claims (1)

1. the quantification of blood vessel wall image texture characteristic and an extracting method, it is characterized in that, quantification and the extracting method of described blood vessel wall image texture characteristic comprise the following steps:

Medical image system is utilized to gather multiframe continuous print blood vessel wall image;

Blood vessel wall image described in each frame is chosen blood vessel wall inner membrance-middle rete as area-of-interest, and extract the textural characteristics value of described area-of-interest, then obtain its meansigma methods;

Carry out statistical analysis to described textural characteristics value, build the textural characteristics population of parameters being used for the diagnosis of blood vessel wall pathological changes EARLY RECOGNITION, described textural characteristics value comprises first-order statistical properties, gray level co-occurrence matrixes and statistical nature matrix,

Wherein, described first-order statistical properties comprises gray average and the standard deviation of vascular wall area, and the gray average of described vascular wall area is:

$ROIavg=\frac{1}{M\×N}{\mathrm{\Σ}}_{x=1}^M{\mathrm{\Σ}}_{y=1}^{N}I(x,y)$

The standard deviation of described vascular wall area is:

$ROIstd=\sqrt{\frac{1}{M\×N-1}{\mathrm{\Σ}}_{x=1}^M{\mathrm{\Σ}}_{y=1}^N{\left(I\right(x,y)-ROIavg)}^2}$

Any point (x in the image (M × N) of described vascular wall area, y) gray value is I (x, y), ROIavg characterizes the average gray value of area-of-interest, and ROIstd characterizes the uniformity coefficient of intensity profile in described area-of-interest;

Wherein, described gray level co-occurrence matrixes feature comprises contrast, and dependency, energy, unfavourable balance distance, entropy, described contrast is $Contrast={\mathrm{\Σ}}_{n=0}^L{n}^2\left\{{\mathrm{\Σ}}_{i=0}^L{\mathrm{\Σ}}_{j=0}^{L}p(i,j)\right\},|i-j|=n,$ The described contrast reflection definition of image and the degree of the texture rill depth, texture rill is darker, and its contrast is larger, and visual effect is more clear; Otherwise contrast is little, then rill is shallow, and effect is fuzzy;

Described dependency is $Correlation\frac{{\mathrm{\Σ}}_{i=0}^L{\mathrm{\Σ}}_{j=0}^{L}p(i,j)-u_x{u}_y}{{\mathrm{\σ}}_x{\mathrm{\σ}}_y};$

Described energy is $Energy={\mathrm{\Σ}}_{i=0}^L{\mathrm{\Σ}}_{i=0}^{L}p{(i,j)}^2;$

Described unfavourable balance is apart from being $Homogeneity={\mathrm{\Σ}}_{i=0}^L{\mathrm{\Σ}}_{i=0}^L\frac{p(i,j)}{1+{(i,j)}^2};$

Described entropy is $Entropy={\mathrm{\Σ}}_{i=0}^L{\mathrm{\Σ}}_{j=0}^{L}p(i,j)log\left(p\right(i,j\left)\right),$ Wherein, the intensity value ranges of image is [0, L], and (i, j) is for any point (x, y) in image (M × N) and depart from gray value corresponding to its another point (x+a, y+a), get any point (x in image (M × N), and depart from its another point (x+a y), y+b), these 2 corresponding gray values are (i, j), about angle θ, it is definition like this, if I is (x, y) be (x at coordinate, y) the gray value intensity of position, and I (x, y) be 1 with the distance of surrounding 8 neighbors, under this definition, I (x, y), I (x+1, y) with I (x-1, y) angle of three pixels is 0 degree, I (x, y), I (x+1, and I (x-1 y+1), y-1) angle of three pixels is 45 degree, the rest may be inferred I (x, y), I (x, and I (x y+1), y-1) angle is 90 degree, I (x, y), I (x-1, and I (x+1 y+1), y-1) angle is 135 degree, u _xand σ _xp _xmeansigma methods and standard deviation, u _yand σ _yp _ymeansigma methods and standard deviation, and p _xwith p _ybe defined as follows:

$p_x\left(i\right)={\mathrm{\Σ}}_{j=0}^{L}p(i,j);$

$p_x\left(j\right)={\mathrm{\Σ}}_{i=0}^{L}p(i,j);$

Described statistical nature matrix character comprises contrast, covariance, diversity factor, wherein:

Described contrast is CON (δ)=E{ [I (x, y)-I (x+ Δ x, y+ Δ y)] ²; Described contrast can be used to the degree of pixel resolution and the texture rill depth in computed image;

Described covariance be COV (δ)=E{ [[I (x, y)-η] [I (x+ Δ x, y+ Δ y)-η] }

Described diversity factor is DSS (δ)=E{|I (x, y)-I (x+ Δ x, y+ Δ y) | };

In formula, E{} is expected value, η is gradation of image meansigma methods, (x, y) is a bit in image, and I (x, y) is the gray value of this point, and δ=(Δ x, Δ y) is space length vector.