CN105608707A - Image processing system and method for fluorescence confocal lung image - Google Patents

Abstract

The invention relates to an image processing system and method for a fluorescence confocal lung image. The method comprises the following steps: an image reading module reads a lung video or a lung image acquired by a fluorescence confocal lung image device; an image segmentation module extracts a contour of lung tissue in the lung video or the lung image so as to obtain a proportion of a high-light area; an image texture feature extraction module obtains correlation and entropy information of the lung video or the lung image through a grayscale symbiosis matrix, obtains variance information of the lung video or the lung image through Gabor wavelet transformation, and expresses texture information of the lung video or the lung image through the correlation, the entropy information and the variance information; and an output module outputs the lung video or the lung image after processing. By using such a structure, the image processing system and method for the fluorescence confocal lung image is simple in structure and convenient to operate, can accurately process the lung video or the lung image acquired by the fluorescence confocal lung image device and is wide in application scope.

Description

For image processing system and the method for fluorescence co-focusing lung image

Technical field

The present invention relates to technical field of image processing, relate in particular to medical image analysis technical field, specifically refer to a kind of for glimmeringImage processing system and the method for the burnt lung image of light copolymerization.

Background technology

Medical image analysis is subject to extensive concern at lung's illness detection field, especially the analysis of CT image. CT is external inspectionMeasurement equipment, and fluorescence co-focusing video imaging equipment is a kind of interior checkout equipment of body of novelty, meticulousr to the observation of lung.Because based endoscopic imaging is in vivo the process of a real time imagery, in the process checking, doctor probably ignores some containing focusLung areas, or the endoscope that need to move around determines the position containing focus, and these are all unfavorable for improving doctor's work effectThe accuracy of rate and inspection. The classification of fluorescence co-focusing lung image and recognition system can principium identification containing the lung image of focus,Thereby can send information to doctor, allow doctor can determine accurately and rapidly diseased region, thereby targetedly againCarry out careful observation.

Summary of the invention

The object of the invention is to overcome the shortcoming of above-mentioned prior art, provide a kind of and can realize based on image processing methodAuxiliary diagnosis platform, the lung's video being obtained by fluorescence co-focusing lung imaging device or lung images are carried out to the pin of Treatment AnalysisImage processing system to fluorescence co-focusing lung image and method.

To achieve these goals, the image processing system for fluorescence co-focusing lung image of the present invention and method have as followsForm:

Should be for the image processing system of fluorescence co-focusing lung image, its main feature is that described system comprises:

Image reading module, the lung's video or the lung images that gather in order to read fluorescence co-focusing lung image equipment;

Image is cut apart module, in order to the profile that extracts the lung tissue in described lung's video or described lung images to obtainThe ratio of highlight regions in described lung's video or described lung images;

Image texture characteristic extraction module, in order to obtain described lung's video or described lung images by gray level co-occurrence matrixesCorrelation and entropy information, and obtain the variance of described lung's video or described lung images by Gabor wavelet transformationInformation, and by described correlation, described entropy information and described variance information represent described lung's video or described inThe texture information of lung images;

Output module, in order to export lung after treatment video or lung images.

The invention still further relates to a kind of image processing method for fluorescence co-focusing lung image, it is characterized in that, described methodComprise the following steps:

(1) described image reading module reads lung's video or the lung images that fluorescence co-focusing lung image equipment gathers;

(2) described image cut apart profile that module extracts the lung tissue in described lung's video or described lung images withThe ratio of highlight regions in lung's video described in obtaining or described lung images;

(3) described image texture characteristic extraction module obtains described lung's video or described lung by gray level co-occurrence matrixesThe correlation of image and entropy information, and obtain described lung's video or described lung images by Gabor wavelet transformationVariance information, and by described correlation, described entropy information and described variance information represent described lung's video orThe texture information of described lung images;

(4) output module is exported lung after treatment video or lung images.

Further, described step (2) specifically comprises the following steps:

(2.1) described image is cut apart module and described lung's video or described lung images are carried out to K-means is cut apart, andBe initially two cluster centres;

(2.2) image described in is cut apart module and is traveled through successively the each pixel in lung's video or the lung images after cutting apart, andRecord the number of the pixel of highlight bar;

(2.4) described image is cut apart module and obtains the ratio of highlight regions in described lung's video or described lung images.

Further, described image texture characteristic extraction module by gray level co-occurrence matrixes obtain described lung's video or described inCorrelation and the entropy information of lung images, specifically comprise the following steps:

(3.1) the entropy letter that in described image texture characteristic extraction module computed image, the gray level co-occurrence matrixes on different directions is correspondingBreath, energy, contrast, unfavourable balance distance and correlation;

(3.2) described image texture characteristic extraction module calculates respectively mean value, the mean value of energy, the contrast of entropy informationMean value, the mean value of unfavourable balance distance and the mean value of correlation;

(3.3) described image texture characteristic extraction module is according to the mean value of the mean value of described entropy information, energy, contrastMean value, the mean value of unfavourable balance distance and the mean value of correlation of degree obtains and represents described lung's video or the described figure of lungThe validity feature information of the texture information of picture.

Further, described step (3.1) is specially:

Described image texture characteristic extraction module is according to the gray level co-occurrence matrixes correspondence on different directions in following formula computed imageEntropy information, energy, contrast, unfavourable balance apart from and correlation:

$P(i,j|d,\mathrm{\θ})=\#\left\{\begin{array}{c}\[\left(k.l\right),(m.n)\]\∈(L_x\×L_y)\×(L_x\×L_y)\\ |d,\mathrm{\θ},f(k,l)=i,f(m,n)=j\end{array}\right\}$

Wherein, image is N with the pixel of vertical direction in the horizontal direction_c×N_r, the grey level quantization value that each pixel occurs is N_q，L_x＝{1,2,…,N_cWhat represent is horizontal direction, L_y＝{1,2,…,N_rWhat represent is vertical direction, and G={1,2 ..., N_qRepresentBe quantize after gray scale layer, at L_x×L_yIn scope distance be d angular separation be two pixels of θ in image, occur generalRate is P (i, j|d, θ), first prime number of what #{x} represented is set x, in presentation graphs picture all angles be θ be spaced apart d twoIn individual pixel, a gray level is that the phase adjacency pair that another gray level of i is j is counted, corresponding in gray level co-occurrence matrixes namely theValue on the capable j column position of i, (i, j) ∈ G × G in formula, θ is 0 °, 90 °, 45 °, 135 °.

Further, the described variance letter that passes through Gabor wavelet transformation and obtain described lung's video or described lung imagesBreath, specifically comprises the following steps:

(3.a) obtain frequency m=0, direction n=0,1 ..., eight Gabor kernel functions of 7 o'clock, to obtain described lungVideo or described lung images variance information;

(3.b) according to described lung's video or described lung images variance acquisition of information represent described lung's video orThe validity feature information of the texture information of described lung images.

Further, described step (4) specifically comprises following:

(4.1) described output module by ID3 algorithm obtain image about the ratio of highlight regions, entropy information, phase relation withAnd the decision tree of variance information;

(4.2) described output module is according to ratio, entropy information, the phase relation of described decision tree and described highlight regionsAnd variance information is exported lung after treatment video or lung images.

Adopt the image processing system for fluorescence co-focusing lung image and method in this invention, can realize based on imageThe auxiliary diagnosis platform of processing method, to the lung's video being obtained by fluorescence co-focusing lung imaging device or lung images placeReason is analyzed, thus improve doctor read sheet efficiency, it is simple in structure, has wide range of applications.

Brief description of the drawings

Fig. 1 is the flow chart of steps of the image processing method for fluorescence co-focusing lung image of the present invention.

Fig. 2 is of the present invention according to the flow chart of the treatment step of gray level co-occurrence matrixes.

Fig. 3 is the flow chart of the treatment step of output module of the present invention.

Detailed description of the invention

In order more clearly to describe technology contents of the present invention, conduct further description below in conjunction with specific embodiment.

Image processing system for fluorescence co-focusing lung image of the present invention comprises:

Image reading module, the lung's video or the lung images that gather in order to read fluorescence co-focusing lung image equipment;

Image is cut apart module, in order to the profile that extracts the lung tissue in described lung's video or described lung images to obtainThe ratio of highlight regions in described lung's video or described lung images;

Image texture characteristic extraction module, in order to obtain described lung's video or described lung images by gray level co-occurrence matrixesCorrelation and entropy information, and obtain the variance of described lung's video or described lung images by Gabor wavelet transformationInformation, and by described correlation, described entropy information and described variance information represent described lung's video or described inThe texture information of lung images;

Output module, in order to export lung after treatment video or lung images.

In actual applications, the image processing method for fluorescence co-focusing lung image of the present invention specifically comprises the following steps:

(1) utilize QT development platform and C Plus Plus to realize classification and the recognition system of fluorescence co-focusing lung image. Comprise figureCut apart module, image texture extraction module and output module as read module, image.

(2) image reading module obtains lung's video and the image that fluorescence co-focusing image documentation equipment gathers.

(3) image is cut apart module and the lung image of processing is carried out to the second grade segmentation of K-means and calculate highlight regions in imageRatio. Specific implementation process is as follows:

(3-1) lung image of input is carried out to K-means and cut apart, be initially two cluster centres;

(3-2) for the lung image after cutting apart be the image of a binaryzation, successively each the pixel note on traversing graph pictureThe pixel number of record highlight regions;

(3-3) obtain the ratio that just can draw image highlight regions after the number of highlight regions point in image;

(4) extract the textural characteristics of image with gray level co-occurrence matrixes, draw can distinguish pulmonary tuberculosis image, healthy lung image,Effective textural characteristics of canceration lung image. As shown in Figure 2, specific implementation process is as follows:

(4-1) hypothesis image is N with the pixel of vertical direction in the horizontal direction_c×N_r, the grey level quantization value that each pixel occursFor N_q，L_x＝{1,2,…,N_cWhat represent is horizontal direction, L_y＝{1,2,…,N_rWhat represent is vertical direction,G＝{1,2,…,N_qWhat represent is the gray scale layer after quantizing, at L_x×L_yIn scope, distance is that d angular separation is two pictures of θThe probability that unit occurs in image:

$P(i,j|d,\mathrm{\θ})=\#\left\{\begin{array}{c}\[\left(k.l\right),(m.n)\]\∈(L_x\×L_y)\×(L_x\×L_y)\\ |d,\mathrm{\θ},f(k,l)=i,f(m,n)=j\end{array}\right\}$

In formula, what #{x} represented is first prime number of set x, specific to namely owning in presentation graphs picture in the computational process of above formulaAngle be θ to be spaced apart a gray level in two pixels of d be that the phase adjacency pair that another gray level of i is j is counted, corresponding to ashThe value on the capable j column position of i namely in degree co-occurrence matrix. (i, j) ∈ G × G in formula, θ general quantity turn to level (0 °),Vertically (90 °), diagonal (45 °), this four direction of back-diagonal (135 °). Thereby show that input lung image is in levelGray level co-occurrence matrixes on (0 °), vertical (90 °), diagonal (45 °), back-diagonal (135 °) four direction;

(4-2) calculate corresponding entropy, energy, contrast, the unfavourable balance distance, relevant of gray level co-occurrence matrixes on these four different directionsThese five characteristic quantities of property;

(4-3) four groups of entropys that obtain, energy, contrast, unfavourable balance distance, correlative character are measured to mean value, average with this groupThe textural characteristics of value tag token image;

(4-4) on pulmonary tuberculosis image, healthy lung image and canceration lung image, verify entropy, energy, contrast, unfavourable balance apart from,The validity of these five features of correlation, draws and numerically really can distinguish pulmonary tuberculosis image, healthy lung image and cancerationThe validity feature of lung image;

(4-5) result of the test shows that in entropy, energy, contrast, unfavourable balance distance, correlation, effective textural characteristics is entropy and relevantProperty, and correlation numerically can distinguish pulmonary tuberculosis image, healthy lung image and canceration lung image, and entropy is numericallyCan distinguish healthy lung image and canceration lung image; Therefore in this image processing process, carry out image according to above-mentioned featureAnalyze.

(5) extract the textural characteristics of image with Gabor wavelet transformation, draw can distinguish pulmonary tuberculosis image, healthy lung image,Effective textural characteristics of canceration lung image. Specific implementation process is as follows:

(5-1) two-dimensional Gabor function g (x, y) can be expressed as:

$g(x,y)=\left(\frac{1}{2{\mathrm{\π\σ}}_x{\mathrm{\σ}}_y}\right)\exp \{-\frac{1}{2}\[\frac{x^2}{{\mathrm{\σ}}_x^2}+\frac{y^2}{{\mathrm{\σ}}_y^2}\]+2\mathrm{\π}jwx\}$

The Fourier transform of g (x, y) is G (μ, v):

$G(\mathrm{\μ},v)=\exp \{-\frac{1}{2}\[\frac{{(\mathrm{\μ}-W)}^2}{{\mathrm{\σ}}_x^2}+\frac{v^2}{{\mathrm{\σ}}_y^2}\]\}$

In formula, σ_xσ_yBe the mean square deviation of Gaussian function in X-axis and Y-axis, W is the frequency bandwidth of Gabor small echo. ConventionallyIn situation, in the time getting W=0.5, Gabor small echo can reach the effect of mating with human visual system preferably;

(5-2) carry out appropriate rotation expansion and change of scale using g (x, y) as generating function and to it, can obtain one group from phaseLike wave filter, be Gabor small echo:

g_m,n(x,y)＝a^-mg(x',y')

In formula, a > 1 be the change of scale factor. X ', y ' are expressed as:

x'＝a^-m(xcosθ+ysinθ)

y'＝a^-m(-xsinθ+ycosθ)

θ＝nπ/k

In formula, m represents corresponding yardstick, and n ∈ [0, k] represents corresponding direction, and k is total direction number.

(5-3) hypothesis with I (x, y) represent piece image its Gabor wavelet transformation can be defined as:

$G_{m,n}(x,y)=\underset{x_1{y}_1}{\Σ}I(x,y){g_{mn}}^{*}(x-x_{1,}y-y_1)$

In formula, what " * " represented is conjugate complex number.

(5-3) can calculating energy information E (m, n) according to the result of 2 dimension discrete Gabor wavelet transformations, what its represented is yardstickFor m, direction are the upper energy information of n, can be represented by the formula:

$E(m,n)=\underset{x}{\Σ}\underset{y}{\Σ}\left|G_{m,n}(x,y)\right|$

(5-4) the energy information textural characteristics of token image effectively to a certain extent, but adopt energy information to characterizeProcess in easily cause error in classification. In Practical Calculation, normally characterize textural characteristics by standard variance, σ_mnCanBe expressed as:

${\mathrm{\σ}}_{mn}={\[\frac{\underset{x}{\Σ}\underset{y}{\Σ}{\left(\right|G_{mn}(x,y)|-{\mathrm{\μ}}_{mn})}^2}{M\×N}\]}^{1/2}$

In formula ${\mathrm{\μ}}_{mn}=\frac{E(m,n)}{MN}.$

(5-5) get frequency m=0, direction n=0,1 ..., 7 totally 8 Gabor kernel functions ask the variance yields of lung image.

(5-6) ask the average of 8 variances, and verify whether be to distinguish healthy lung image and cancer by the average drawingBecome effective textural characteristics of lung image.

(6), according to the effective characteristic of division drawing, adopt ID3 Algorithm for Training to go out corresponding decision tree, thereby realize lung's shadowPicture output. As shown in Figure 3, specific implementation process is as follows:

(6-1), to the part lung image in database, adopt ID3 Algorithm for Training to go out about correlation, entropy, highlight regions ratioThe decision tree of example, these four features of variance.

(6-2) the final output of removing to realize image with the decision tree of training out.

Adopt the image processing system for fluorescence co-focusing lung image and method in this invention, can realize based on imageThe auxiliary diagnosis platform of processing method, to the lung's video being obtained by fluorescence co-focusing lung imaging device or lung images placeReason is analyzed, thus improve doctor read sheet efficiency, it is simple in structure, has wide range of applications.

In this description, the present invention is described with reference to its specific embodiment. But, still can make obviously variousAmendment and conversion and do not deviate from the spirit and scope of the present invention. Therefore, description and accompanying drawing should be considered to illustrative but not limitProperty processed.

Claims (7)

1. for an image processing system for fluorescence co-focusing lung image, it is characterized in that, described system comprises:

Image reading module, the lung's video or the lung images that gather in order to read fluorescence co-focusing lung image equipment;

Image is cut apart module, in order to the profile that extracts the lung tissue in described lung's video or described lung images to obtainThe ratio of highlight regions in described lung's video or described lung images;

Image texture characteristic extraction module, in order to obtain described lung's video or described lung images by gray level co-occurrence matrixesCorrelation and entropy information, and obtain the variance of described lung's video or described lung images by Gabor wavelet transformationInformation, and by described correlation, described entropy information and described variance information represent described lung's video or described inThe texture information of lung images;

Output module, in order to export lung after treatment video or lung images.

2. realize the image processing method for fluorescence co-focusing lung image, its spy based on system claimed in claim 1Levy and be, described method comprises the following steps:

(1) described image reading module reads lung's video or the lung images that fluorescence co-focusing lung image equipment gathers;

(2) described image cut apart profile that module extracts the lung tissue in described lung's video or described lung images withThe ratio of highlight regions in lung's video described in obtaining or described lung images;

(3) described image texture characteristic extraction module obtains described lung's video or described lung by gray level co-occurrence matrixesThe correlation of image and entropy information, and obtain described lung's video or described lung images by Gabor wavelet transformationVariance information, and by described correlation, described entropy information and described variance information represent described lung's video orThe texture information of described lung images;

(4) output module is exported lung after treatment video or lung images.

3. the image processing method for fluorescence co-focusing lung image according to claim 2, is characterized in that, described inStep (2) specifically comprise the following steps:

(2.1) described image is cut apart module and described lung's video or described lung images are carried out to K-means is cut apart, andBe initially two cluster centres;

(2.2) image described in is cut apart module and is traveled through successively the each pixel in lung's video or the lung images after cutting apart, andRecord the number of the pixel of highlight bar;

(2.4) described image is cut apart module and obtains the ratio of highlight regions in described lung's video or described lung images.

4. the image processing method for fluorescence co-focusing lung image according to claim 2, is characterized in that, described inImage texture characteristic extraction module obtain the correlation of described lung's video or described lung images by gray level co-occurrence matrixesAnd entropy information, specifically comprise the following steps:

(3.1) the entropy letter that in described image texture characteristic extraction module computed image, the gray level co-occurrence matrixes on different directions is correspondingBreath, energy, contrast, unfavourable balance distance and correlation;

(3.2) described image texture characteristic extraction module calculates respectively mean value, the mean value of energy, the contrast of entropy informationMean value, the mean value of unfavourable balance distance and the mean value of correlation;

(3.3) described image texture characteristic extraction module is according to the mean value of the mean value of described entropy information, energy, contrastMean value, the mean value of unfavourable balance distance and the mean value of correlation of degree obtains and represents described lung's video or the described figure of lungThe validity feature information of the texture information of picture.

5. the image processing method for fluorescence co-focusing lung image according to claim 4, is characterized in that, described inStep (3.1) be specially:

Described image texture characteristic extraction module is according to the gray level co-occurrence matrixes correspondence on different directions in following formula computed imageEntropy information, energy, contrast, unfavourable balance apart from and correlation:

$P\left(i,j|d,\mathrm{\θ}\right)=\#\left\{\begin{array}{c}\[\left(k.l\right),\left(m.n\right)\]\∈\left(L_x\×L_y\right)\×\left(L_x\×L_y\right)\\ |d,\mathrm{\θ},f\left(k,l\right)=i,f\left(m,n\right)=j\end{array}\right\}$

Wherein, image is N with the pixel of vertical direction in the horizontal direction_c×N_r, the grey level quantization value that each pixel occurs is N_q，L_x＝{1,2,...,N_cWhat represent is horizontal direction, L_y＝{1,2,...,N_rWhat represent is vertical direction, and G={1,2 ..., N_qRepresentBe quantize after gray scale layer, at L_x×L_yIn scope distance be d angular separation be two pixels of θ in image, occur generalRate is P (i, j|d, θ), first prime number of what #{x} represented is set x, in presentation graphs picture all angles be θ be spaced apart d twoIn individual pixel, a gray level is that the phase adjacency pair that another gray level of i is j is counted, corresponding in gray level co-occurrence matrixes namely theValue on the capable j column position of i, (i, j) ∈ G × G in formula, θ is 0 °, 90 °, 45 °, 135 °.

6. the image processing method for fluorescence co-focusing lung image according to claim 2, is characterized in that, described inThe variance information of passing through Gabor wavelet transformation and obtain described lung's video or described lung images, specifically comprise following stepRapid:

(3.a) obtain frequency m=0, direction n=0,1 ..., eight Gabor kernel functions of 7 o'clock, to obtain described lungVideo or described lung images variance information;

(3.b) according to described lung's video or described lung images variance acquisition of information represent described lung's video orThe validity feature information of the texture information of described lung images.

7. the image processing method for fluorescence co-focusing lung image according to claim 2, is characterized in that, described inStep (4) specifically comprise following:

(4.1) described output module by ID3 algorithm obtain image about the ratio of highlight regions, entropy information, phase relation withAnd the decision tree of variance information;

(4.2) described output module is according to ratio, entropy information, the phase relation of described decision tree and described highlight regionsAnd variance information is exported lung after treatment video or lung images.