CN103578099A - Method for extracting tumor elasticity characteristics based on ultrasonic elastography - Google Patents

Abstract

The invention provides a method for extracting tumor elasticity characteristics based on ultrasonic elastography. The method comprises the following steps that medical images and elasticity images are collected; the medical images are preprocessed; the tumor marginal information of the preprocessed medical images is automatically extracted; the tumor marginal information is fused into the elasticity images; an elasticity characteristic parameter group, fused with the tumor marginal information, of the elasticity images is extracted. According to the method for extracting the tumor elasticity characteristics based on the ultrasonic elastography, the tumor marginal information is extracted through the medical images, meanwhile, the tumor marginal information is fused into the elasticity images, the elasticity characteristic parameter group based on the elasticity images is built and quantized and can describe the attributes of tumors from different angles, and therefore the subjective difference of clinicians is eliminated, and diagnosis is more accurate and more scientific.

Description

The extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging

Technical field

The present invention relates to image processing techniques, particularly a kind of extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging.

Background technology

Elastogram is that tissue is applied to an external drive, and under the physics law effects such as pressure-deformation, tissue will produce a response, according to this response, obtain elastogram figure.After own elasticity imaging concept proposes, Ultrasonic Elasticity Imaging has obtained development rapidly in the more than ten years recently.

The elasticity average of the elastic characteristic parameter area-of-interest based on ultrasonic Transient elastography technology that the medical journal > > of < < PLA the 36th volume o. 11th 1131-1133 page in 2011 and < < Chinese medicine image technology > > the 28th volume the 3rd phase 529-533 page in 2012 proposes, but the instantaneous elasticity detection technique adopting is a kind of one-dimensional image technology, although can quantitative organize average elasticity modulus value, but cannot expand to two-dimension elastic imaging and obtain tissue elasticity distributed intelligence, generally be only applicable to detect dispersivity pathology, < < Chinese medicine image technology > > the 26th volume the 9th phase 1682-1684 page in 2010, the elastic characteristic elastic image distribution characteristics based on ultrasonic real-time elastogram technology that < < modern biomedical progress > > the 26th volume the 9th phase 492-494 page in 2010 and < < West China medical science > > the 25th volume the 2nd phase 294-297 page in 2010 proposes, although provided clinical diagnostic applications elastic characteristic the most widely, but elastogram technology is a kind of quasistatic elasticity of compression imaging technique in real time, can only provide relative displacement/strain figure, cannot provide the concrete numerical value of tissue local hardness, lack objectivity and science, < < Chinese medicine image technology > > the 18th volume the 7th phase 589-591 page in 2009 and < < world Chinese digest the elastic characteristic rate of strain ratio based on ultrasonic real-time elastogram technology that magazine > > the 18th volume the 30th phase 3254-3258 page in 2010 proposes, it is the new elastic characteristic proposing in the recent period, but from extracting two-dimension displacement/strain figure qualitatively, although more reflect the firmness change degree of focus than simple elasticity distribution feature objective quantitative, but this feature cannot be carried out the lateral comparison between different focuses equally, the rate of strain ratio interval that for example the pernicious occupying lesion of liver on cirrhosis basis and benign liver tumours venereal disease become exists certain overlapping, only rely on the identification of rate of strain ratio easily to cause mistaken diagnosis.

Therefore, conventional ultrasound elastogram technology there is many limitation, make the qualitative or sxemiquantitative of the elastic characteristic that obtains, lack objectivity and repeatability, cannot analyze quantitatively, affected by diagnosis person's subjectivity larger, cannot set up diagnostic criteria generally acknowledged, standard, limited Ultrasonic Elasticity Imaging widespread use clinically.

Summary of the invention

Based on this, be necessary to propose a kind of extracting method of the ultrasonic tumour elastic characteristic based on ultrasonic elastograph imaging, for instructing the good pernicious differentiation of tumour, for the pathological research of tumour and early diagnosis treatment provide new foundation.

An extracting method for the tumour elastic characteristic of ultrasonic elastograph imaging, comprises the steps: to gather medical image and elastic image; Pre-service to described medical image; Automatically extract the borderline tumor information of medical image after pretreatment; By the borderline tumor information fusion of extracting in corresponding elastic image; Extract the elastic characteristic population of parameters of the elastic image that merges described borderline tumor information.

The above-mentioned tumour elastic characteristic extracting method based on elastogram, utilize medical image to extract borderline tumor information, simultaneously by this borderline tumor information fusion in elastic image, and create and the elastic characteristic population of parameters of quantification based on elastic image, this elastic characteristic population of parameters can be described the attribute of tumour from different perspectives, thereby eliminate clinician's subjective differences, make to diagnose more accurate and the science of becoming.

Accompanying drawing explanation

The process flow diagram of the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging that Fig. 1 provides for the embodiment of the present invention.

The pretreated method flow diagram of the B ultrasonic image that Fig. 2 provides for the embodiment of the present invention.

The borderline tumor information of the B ultrasonic image of the tumor of breast that Fig. 3 provides for the embodiment of the present invention.

The borderline tumor information fusion of extracting in the B ultrasonic image that Fig. 4 provides for the embodiment of the present invention is to the schematic diagram of elastic image.

Fig. 5 chooses schematic diagram for the elastic image area-of-interest that inventive embodiments provides.

Fig. 6 is for choosing schematic diagram in the gray level co-occurrence matrixes zoning that the embodiment of the present invention provides.

The quantization method process flow diagram of the quantization parameter value that Fig. 7 provides for the embodiment of the present invention.

Fig. 8 is for choosing schematic diagram in the tumour peripheral organization region that the embodiment of the present invention provides.

Fig. 9 is for choosing schematic diagram in the tumor center region that the embodiment of the present invention provides.

The B ultrasonic image of the fibroadenoma of breast that Figure 10 (a) provides for the embodiment of the present invention.

The lesion segmentation result figure of the fibroadenoma of breast that Figure 10 (b) provides for the embodiment of the present invention.

The springform spirogram of the fibroadenoma of breast that Figure 10 (c) provides for the embodiment of the present invention.

The B ultrasonic image of the infiltration ductal carcinomas of breast that Figure 11 (a) provides for the embodiment of the present invention.

The lesion segmentation result figure of the infiltration ductal carcinomas of breast that Figure 11 (b) provides for the embodiment of the present invention.

The springform spirogram of the infiltration ductal carcinomas of breast that Figure 11 (c) provides for the embodiment of the present invention.

Embodiment

Refer to Fig. 1, a kind of extracting method of tumour elastic characteristic of elastogram, concrete steps are as follows:

Step S10: input medical image and elastic image.

Elastogram technology is a kind of comparatively ripe imaging technique, according to this technology, can obtain elastic image, medical image can be B ultrasonic image or CT image or MRI (magnetic resonance) image or X-ray (X ray) image, in embodiment provided by the invention, medical image is preferably B ultrasonic image.

Step S20: the pre-service to described medical image.

Refer to Fig. 2, for the embodiment of the present invention provide to the pretreated process flow diagram of B ultrasonic image, step S20 is specially:

Step S21: medical image is carried out to speckle noise filtering processing.

The coherence of ultrasonic imaging causes the intrinsic speckle noise of B ultrasonic image, and speckle noise has reduced picture quality, has especially covered some detailed information of image, brings difficulty to the subsequent treatment such as rim detection, feature extraction of image.Nonlinear properties processing that clinical sonography system is built-in is (as log-compressed, low-pass filtering etc.), the dynamic range of compression and back wave envelope signal is to adapt to the small dynamic range of display device, and the speckle noise model of explicit log-compressed B ultrasonic image can be expressed as:

${\mathrm{<figure-callout\; id="0"\; label="I"\; filenames="HSA00000761302200031.png"\; state="\{\{state\}\}">I</figure-callout>}}_0=I+\sqrt{I}n---\left(1\right)$

In formula (1), I is original signal, I ₀for observation signal, n is zero-mean, and standard variance is σ _ngaussian noise.

Speckle noise model based on (1) formula, adopt a kind of anisotropic diffusion filtering device (Speckle Reducing Anisotropic Diffusion, SRAD), can be in noise reduction, retain and even strengthen the marginal information in image, Anisotropic Diffusion Model can be expressed as:

$\frac{\&PartialD;I}{\&PartialD;t}=\mathrm{div}[c\left(q\right)\&dtri;I]$

I(t＝0)＝I ₀ (2)

In formula (2), q is instantaneous variation coefficient operator (Instantaneous Coefficient of Variation, ICOV), is the edge detector in SARD, is expressed as:

$q=\sqrt{\frac{(1/2){\left(\right|\&dtri;I|/I)}^2-(1/16){({\&dtri;}^{2}I/I)}^2}{{[1+(1/4)({\&dtri;}^{2}I/I)]}^2}}---\left(3\right)$

In formula (3), q has comprised gradient operator

and Laplace operator

second derivative character can be used for distinguishing the grey scale change caused by noise and the grey scale change being caused by edge, so combination

with make the rim detection in speckle noise environment more accurate.Coefficient of diffusion c (q) is expressed as:

$c\left(q\right)=\frac{1}{1+(q^2-q_0{\left(t\right)}^2)/\left[q_0{\left(t\right)}^2(1+q_0{\left(t\right)}^2)\right]}---\left(4\right)$

In formula (4), t represents diffusion time, q ₀(t) be diffusion thresholding, ideally reflect that speckle noise in image is uniformly distributed the statistical property in region:

$q_0\left(t\right)=\frac{\sqrt{\mathrm{var}\left[z\left(t\right)\right]}}{\stackrel{\&OverBar;}{z\left(t\right)}}---\left(5\right)$

In formula (5), z (t) represents that speckle noise is uniformly distributed region, var[z (t)] and

the variance and the average that represent respectively this region.In specific implementation process, be generally approximately: q ₀(t) ≈ q ₀exp (ρ t), parameter q ₀get 1 with ρ, iterations is 100.

Step S22: the medical image of processing through speckle noise filtering is carried out to smoothing processing.

Through the speckle noise lower gray-scale value of filtered B ultrasonic image region (as inside tumor), exist less bright details, higher gray-scale value region (as soft tissue around) exists less dark-coloured details, this is mainly comprised of inside tumor microtexture (as blood vessel, calcification etc.), can be level and smooth by morphologic filtering device.Here adopt alternating sequence filtering, by the structural elements of a series of continuous increases, usually carry out switching filtering, structural element is got collar plate shape, radius 2～5.

Step S30: the borderline tumor information of automatically extracting medical image after pretreatment.

Refer to Fig. 3, the borderline tumor information of the B ultrasonic image of the tumor of breast providing for the embodiment of the present invention.The present invention adopts the borderline tumor extracting method based on Chan-Vese model: be specially definition image I ₀coordinate set be Ω, ω is the subgraph being defined in Ω, the border that curve C is ω, curve C is divided into inside (C) and two regions of outside (C) by image.If c ₁and c ₂represent respectively the average gray in these two regions, and represent inside (C) and outside (C) with level set function φ, when following energy function hour, the zero level collection of level set function φ is the object boundary curve C of expectation:

$F(c_1,{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HSA00000761302200031.png"\; state="\{\{state\}\}">c</figure-callout>}}_2,\mathrm{\&phi;})=\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}|\&dtri;H\left(\mathrm{\&phi;}\right)|\mathrm{dxdy}+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+$

${\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}{|{\mathrm{\&mu;}}_0-c_1|}^{2}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}{|{\mathrm{\&mu;}}_0-{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HSA00000761302200031.png"\; state="\{\{state\}\}">c</figure-callout>}}_2|}^2(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}---\left(6\right)$

In formula (6), μ, ν, λ ₁, λ ₂be all constant coefficient, H (φ) is Heaviside function.The φ developing for each step, c ₁and c ₂can be calculated by two formulas below:

$c_1\left(\mathrm{\&phi;}\right)=\frac{{\&Integral;}_{\mathrm{\&Omega;}}{u}_0(x,y)H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}}{{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}}---\left(7\right)$

$c_2\left(\mathrm{\&phi;}\right)=\frac{{\&Integral;}_{\mathrm{\&Omega;}}{u}_0(x,y)(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}}{{\&Integral;}_{\mathrm{\&Omega;}}(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}}---\left(8\right)$

Defined function φ in time t develops, and by formula (6), can derive the Euler-Lagrange equation about φ (t, x, y):

$\frac{\&PartialD;\mathrm{\&phi;}}{\&PartialD;t}={\mathrm{\&delta;}}_{\mathrm{\&epsiv;}}\left(\mathrm{\&phi;}\right)[\mathrm{\&mu;}\mathrm{div}\left(\frac{\&dtri;\mathrm{\&phi;}}{|\&dtri;\mathrm{\&phi;}|}\right)-v-{\mathrm{\&lambda;}}_1{(u_0-c_1)}^2+{\mathrm{\&lambda;}}_2{(u_0-c_2)}^2]=0,\mathrm{in}(0,\&infin;)\&times;\mathrm{\&Omega;}$

φ(0，x，y)＝φ ₀(x，y)，inΩ

$\frac{{\mathrm{\&delta;}}_{\mathrm{\&epsiv;}}\left(\mathrm{\&phi;}\right)}{|\&dtri;\mathrm{\&phi;}|}\frac{\&PartialD;\mathrm{\&phi;}}{\&PartialD;\stackrel{\&OverBar;}{n}}=0,\mathrm{on}\&PartialD;\mathrm{\&Omega;}---\left(9\right)$

In formula (9), δ _ε(φ) be the approximate of Dirac function,

represent the outer normal vector on border, δ ₀(x, y) is the symbolic distance function being obtained by initial boundary.Solving equation (9), the zero level collection of its steady state solution is the object boundary curve C of expectation.

The edge extracting that is appreciated that tumour also can adopt other conventional image partition method.

Step S40: by described borderline tumor information fusion in described elastic image.

Refer to Fig. 4, the borderline tumor information fusion of extracting in the B ultrasonic image providing for the embodiment of the present invention is to the schematic diagram of elastic image.The clinical sonography system with elastogram function generally shows B ultrasonic image and the elastic image of human body same area simultaneously, therefore, can be directly by the borderline tumor information fusion of extracting in B ultrasonic image in elastic image, be partitioned into lesion region, be designated as TUMORarea.Be appreciated that the borderline tumor information that other medical images for example extract in CT image or MRI image or X-ray image also can be fused in elastic image.

Step S50: the elastic characteristic population of parameters of the elastic image of borderline tumor information described in Quantitative fusion.

Elastic characteristic population of parameters comprises elasticity number parameter.Elasticity number parameter comprises the elastic modulus mean value of area-of-interest, the mean value in the elastic modulus standard deviation of area-of-interest, the mean value of lesion region, the standard deviation of lesion region, surrounding tissue region and elasticity ratio.

Referring again to Fig. 4, binding of pathological is gained knowledge and doctor's clinical experience, it is generally acknowledged: optimum focus is softer, focus inner elastomeric distribution uniform is less with the flexible contrast of surrounding tissue; Pernicious focus shows as harder, and the elasticity distribution inhomogeneous (as blood vessel, calcification etc.) of focus inside is larger with the flexible contrast of surrounding tissue.Pseudocolour picture is recovered to after elastic mould value gray-scale map, in conjunction with doctor's clinical experience and image, processes knowledge.

Refer to Fig. 5, for the elastic image area-of-interest that inventive embodiments provides is chosen schematic diagram.Wherein, solid white line rectangular area is area-of-interest.

In embodiment provided by the invention, area-of-interest comprises lesion region and surrounding tissue region.Area-of-interest is designated as ROIarea, and lesion region is designated as TUMORarea, and surrounding tissue region is designated as SUDarea.Wherein, the choosing method of area-of-interest is:

Step 1: go out horizontal boundary rectangle according to the tumor's profiles curve calculation that merges the elastic image of borderline tumor information.

Step 2: to four direction continuation, forming one, to comprise tumour and size be that the rectangular area of 2～3 times, tumour is as area-of-interest by boundary rectangle.Area-of-interest is designated as ROIarea.

What be appreciated that surrounding tissue region extracts the region of lesion region TUMORarea remainder for region of interest ROI area.

In embodiment provided by the invention, the quantization method of the elastic modulus mean value of ROIarea is:

$\mathrm{ROIavg}=\frac{1}{N_{\mathrm{ROI}}}\underset{i=1}{\overset{N_{\mathrm{ROI}}}{\mathrm{\&Sigma;}}}{e}_i---\left(10\right)$

In formula (10), N _rOIthe number of all pixels in ROIarea, e _iit is the elastic mould value that pixel is corresponding.ROIavg has characterized the average hardness of ROIarea.

In embodiment provided by the invention, the quantization method of the standard deviation of ROIarea elastic modulus definition is:

$\mathrm{ROIstd}=\sqrt{\frac{1}{N_{\mathrm{ROI}}-1}\underset{i=1}{\overset{N_{\mathrm{ROI}}}{\mathrm{\&Sigma;}}}{(e_i-\mathrm{ROIavg})}^2}---\left(11\right)$

In formula (11), N _rOIthe number of all pixels in ROIarea, e _iit is the elastic mould value that pixel is corresponding.ROIstd has characterized the degree of uniformity of elasticity distribution in ROIarea, and ROIstd is less, and elasticity distribution is more even.

In embodiment provided by the invention, the quantization method of TUMORarea elastic modulus mean value is:

$\mathrm{TUMORavg}=\frac{1}{N_{\mathrm{TUMOR}}}\underset{i=1}{\overset{N_{\mathrm{TUMOR}}}{\mathrm{\&Sigma;}}}{e}_i---\left(12\right)$

In formula (12), N _tUMORit is the number of all pixels in TUMORarea; e _iit is the elastic mould value that pixel is corresponding.TUMORavg has characterized the average hardness of tumour.

In embodiment provided by the invention, the quantization method of the standard deviation of TUMORarea elastic modulus is:

$\mathrm{TUMORstd}=\sqrt{\frac{1}{N_{\mathrm{TUMOR}}-1}\underset{i=1}{\overset{N_{\mathrm{TUMOR}}}{\mathrm{\&Sigma;}}}{(e_i-\mathrm{ROIavg})}^2}---\left(13\right)$

In formula (13), N _tUMORit is the number of all pixels in TUMORarea; e _iit is the elastic mould value that pixel is corresponding.TUMORstd has characterized the degree of uniformity of tumour elasticity distribution, and TUMORstd is less, and elasticity distribution is more even.

In embodiment provided by the invention, the elastic modulus mean value quantization method of SUDarea is:

$\mathrm{SUDavg}=\frac{1}{N_{\mathrm{SUD}}}\underset{i=1}{\overset{N_{\mathrm{SUD}}}{\mathrm{\&Sigma;}}}{e}_i---\left(14\right)$

In formula (14), N _sUDit is the number of all pixels in SUDarea; e _iit is the elastic mould value that pixel is corresponding.SUDavg has characterized the average hardness of normal surrounding tissue.

In embodiment provided by the invention, elasticity ratio quantization method is:

$\mathrm{Eratio}=\frac{\mathrm{TUMORavg}}{\mathrm{SUDavg}}---\left(15\right)$

In formula (15), TUMORavg is the average hardness of TUMORarea; SUDavg is the average hardness of SUDarea.Eratio has characterized the relative resilient intensity of variation in tumour and normal surrounding tissue region.

Elastic characteristic population of parameters also comprises elastic image parametric texture.Elastic image parametric texture comprises: the elastic image parametric texture of lesion region and the elastic image parametric texture of gray level co-occurrence matrixes image-region.The elastic image parametric texture of lesion region comprises: histogram normalization variance, histogram degree of bias descriptor, histogram kurtosis descriptor, histogram consistance descriptor and entropy of histogram.The elastic image parametric texture of gray level co-occurrence matrixes image-region comprises co-occurrence matrix energy descriptor, co-occurrence matrix contrast descriptor, co-occurrence matrix unfavourable balance square and co-occurrence matrix entropy.

In embodiment provided by the invention, lesion region TUMORarea histogram normalization variance quantization method is:

$\mathrm{H\; var}=\frac{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(z_i-m)}^{2}p\left(z_i\right)}{{(L-1)}^2}---\left(16\right)$

In formula (16), z _irepresent that elastic mould value is mapped to a stochastic variable after [0,255]; M represents the average after elastic modulus corresponding to the interior all pixels of TUMORarea shines upon; p(z _i) be the grey level histogram of TUMORarea; L is possible number of greyscale levels, generally gets 256.Hvar normalizes to scope [0,1], characterizes the discrete distribution situation of elastic mould value in tumour.

In embodiment provided by the invention, the quantization method of histogram degree of bias descriptor is:

$\mathrm{Hskew}=\frac{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(z_i-m)}^{3}p\left(z_i\right)}{{(L-1)}^2}---\left(17\right)$

Z in formula (17) _irepresent that elastic mould value is mapped to a stochastic variable after [0,255], m represents the average after elastic modulus corresponding to the interior all pixels of lesion region shines upon, p (z _i) be the grey level histogram of lesion region, L is possible number of greyscale levels, the asymmetric degree of Hskew token image histogram distribution.Hskew is larger, represents that histogram distribution is more asymmetric, otherwise more symmetrical.

In embodiment provided by the invention, the quantization method of histogram kurtosis descriptor is:

$\mathrm{Hkurt}=\frac{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(z_i-m)}^{4}p\left(z_i\right)}{{(L-1)}^2}---\left(18\right)$

Z in formula (18) _irepresent that elastic mould value is mapped to a stochastic variable after [0,255], m represents the average after elastic modulus corresponding to the interior all pixels of lesion region shines upon, p (z _i) be the grey level histogram of lesion region, L is possible number of greyscale levels, Hkurt characterizes elastic image and is distributed in the approximate state while approaching average, in order to judge whether the elasticity distribution of image concentrates near average elasticity value.Hkurt is less, represents more concentrated; Otherwise overstepping the bounds of propriety loose.

In embodiment provided by the invention, TUMORarea histogram consistance descriptor is defined as:

$\mathrm{Henergy}=\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{p}^2\left(z_i\right)---\left(19\right)$

In formula (19), z _irepresent that elastic mould value is mapped to a stochastic variable after [0,255]; p(z _i) be the grey level histogram of TUMORarea; L is possible number of greyscale levels.Henergy characterizes the degree of uniformity of tumour elasticity distribution, and distribution uniform duration is larger, otherwise less.

In embodiment provided by the invention, TUMORarea entropy of histogram is defined as:

$\mathrm{Hentropy}=-\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}p\left(z_i\right){\log }_{2}p\left(z_i\right)---\left(20\right)$

In formula (20), z _irepresent that elastic mould value is mapped to a stochastic variable after [0,255]; p(z _i) be the grey level histogram of TUMORarea; L is possible number of greyscale levels.Hentropy also characterizes the homogeneity of tumour elasticity distribution.

Refer to Fig. 6, for choosing schematic diagram in the gray level co-occurrence matrixes zoning that the embodiment of the present invention provides.In Fig. 6, solid white line rectangular area is the minimum circumscribed rectangular region of tumour, is used for calculating the characteristic parameter based on gray level co-occurrence matrixes.Be designated as MBBarea, calculating distance is 2, from top to bottom the gray level co-occurrence matrixes G of direction.

In embodiment provided by the invention, the choosing method of gray level co-occurrence matrixes image-region is: according to the tumor's profiles opisometer that merges the elastic image of borderline tumor information, calculate minimum circumscribed rectangular region as gray level co-occurrence matrixes image-region.

In embodiment provided by the invention, the co-occurrence matrix energy descriptor quantization method of MBBarea is:

$\mathrm{ASM}=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{\left\{G(i,j)\right\}}^2---\left(21\right)$

In formula (21), G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K.ASM token image gray scale (elasticity) be evenly distributed degree and texture fineness degree.Elastic image distributes, and more evenly value is larger, otherwise less.

In embodiment provided by the invention, the co-occurrence matrix contrast descriptor quantization method of MBBarea is:

$\mathrm{CON}=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{(i-j)}^{2}G(i,j)---\left(22\right)$

In formula (22), G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K.The degree of the sharpness of CON token image and the texture rill depth.Texture rill is darker, and CON is larger, and visual effect is more clear; Otherwise less.

In embodiment provided by the invention, the co-occurrence matrix unfavourable balance square quantization method of MBBarea is:

$\mathrm{IDM}=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}\frac{G(i,j)}{1+{(i-j)}^2}---\left(23\right)$

In formula (23), G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K.The homogeney of IDM token image, the intensity of variation of tolerance image texture part.Between the zones of different of the larger explanation image texture of IDM, lack variation, part is very even.

In embodiment provided by the invention, the co-occurrence matrix entropy quantization method of MBBarea is:

$\mathrm{ENT}=-\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}G(i,j){\log }_{2}G(i,j)---\left(24\right)$

In formula (24), G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K.Non-uniform degree or the complexity of ENT token image texture.In co-occurrence matrix, all elements has in maximum randomness, space co-occurrence matrix when all values is almost equal, and when in co-occurrence matrix, element disperses to distribute, ENT is larger.

Elastic characteristic population of parameters also comprises quantization parameter.

Refer to Fig. 7, the quantization method process flow diagram of the quantization parameter value providing for the embodiment of the present invention, comprises

Step S51: lesion region is done dilation operation and obtained tumour peripheral organization region.

Refer to Fig. 8, for choosing schematic diagram in the tumour peripheral organization region that the embodiment of the present invention provides, wherein, the region between solid white line and grey solid line is tumour peripheral organization region.Adopt morphological images disposal route, the collar plate shape structural element that setting radius is 20, to lesion region, TUMORarea does dilation operation, obtains tumour peripheral organization region, is expressed as TUMORSUDarea.

Step S52: described lesion region is done erosion operation and obtained pathology central area.

Refer to Fig. 9, for choosing schematic diagram in the tumor center region that the embodiment of the present invention provides, wherein, the region between solid white line and grey solid line is tumor center region.Set radius and be 1/5 collar plate shape structural element of TUMORarea boundary rectangle minor axis length, TUMORarea is done to erosion operation, obtain pathology central area, be expressed as CENTERarea.

Step S53: be defined as follows parameter:

$\mathrm{TUMORsoft}=\frac{N_{\mathrm{soft}}}{N},$ $\mathrm{TUMORhard}=\frac{N_{\mathrm{hard}}}{N}---\left(25\right)$

In formula (25), N represents the number of all pixels in TUMORarea; N _softrepresent that the interior elastic mould value of TUMORarea is less than the pixel number of ROIavg; Nhard represents that TUMORarea elastic mould value is greater than the pixel number of ROIavg.

$\mathrm{TUMORSUDsoft}=\frac{M_{\mathrm{soft}}}{M},$ $\mathrm{TUMORSUDhard}=\frac{M_{\mathrm{hard}}}{M}---\left(26\right)$

In formula (26), M represents the number of all pixels in TUMORSUDarea; M _softrepresent that the interior elastic mould value of TUMORSUDarea is less than the pixel number of ROIavg; M _hardin representing, TUMORSUDarea elastic mould value is greater than the pixel number of ROIavg.

$\mathrm{CENTERhard}=\frac{L_{\mathrm{hard}}}{L}---\left(27\right)$

In formula (27), L represents the number of all pixels in CENTERarea; L _hardrepresent that CENTERarea Elastic value is greater than the pixel number of ROIavg.

Step S54: elasticity scoring is as follows:

When TUMORsoft >=90%, quantization parameter value Escore is 1;

When TUMORsoft-TUMORhard >=10%, quantization parameter value Escore is 2;

When TUMORhard-TUMORsoft >=10%, quantization parameter value Escore is 3;

When TUMORhard >=80%, TUMORSUFsoft >=70%, quantization parameter value Escore is 4;

When TUMORhard >=90%, TUMORSUFhard >=50%, quantization parameter value Escore is 5.

The relative distribution situation of lesion region elasticity that above-mentioned quantization parameter employing elasticity point system has carried out quantization signifying, but need to combine with elasticity number parameter, could make diagnosis accurately to tumour.

The above embodiment of the present invention, the quantization method of 16 elastic characteristic parameters more than having provided, these characteristic parameters group has described the attribute of tumour from different perspectives.In actual applications, by strong, objective to these recognition capabilities, repeatable strong characteristic parameter, carry out decision making package, assist clinicians is diagnosed, and draws result more accurately.

By specific embodiment, to of the present invention, be further described below.

The ultrasonic tumor of breast image (the conventional B ultrasonic image and the elastic modulus image that comprise focus) that adopts method proposed by the invention to gather by Supersonic Aixplorer Ultrasound Instrument attached San hospital of Zhongshan University Ultrasonography has carried out quantification and the extraction of elastic characteristic.This apparatus preparation has shearing wave elastogram module, adopts SC15-4 linear array probe, and frequency range is 4-15MHz.During image data, first under two-dimensional ultrasound pattern, carry out the scanning of mammary gland horizontal stroke, vertical each tangent plane, find, after tumor of breast, to observe the Infiltrating of size, form, border, internal echo and the surrounding tissue of focus, record the position of focus.Then be switched to ultrasonic elastograph imaging checking mode, sampling frame is approximately 2-3 times of size of tumor, if focus is larger, the part of only getting focus is positioned at sampling frame.Double width shows two-dimentional B ultrasonic image and elastic image simultaneously, keeps the stable of probe during operation as far as possible, maintains 2 seconds, the static B ultrasonic image of Cryopreservation and elastic image.

See also Fig. 9, Figure 10, table 1 and table 2.The lesion region that has provided respectively adenofibroma and infitrating ductal carcinoma is cut apart and characteristic quantification extraction result, can find out, optimum focus hardness is less, and focus inner elastomeric distribution uniform is less with the flexible contrast of surrounding tissue; Pernicious focus hardness is larger, and the elasticity distribution of focus inside is inhomogeneous, larger with the flexible contrast of surrounding tissue.From clinical trial result, the extracting method of the tumour elastic characteristic of the elastogram that the present invention proposes, can be used for instructing good, the pernicious differentiation of tumour, for pathological research and the early diagnosis treatment of tumour provides new foundation.

Table 1 elastic characteristic parameter (1)

Table 2 elastic characteristic parameter (2)

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

Claims (12)

1. an extracting method for the tumour elastic characteristic based on ultrasonic elastograph imaging, is characterized in that, comprises the steps:

Gather medical image and elastic image;

Described medical image is carried out to pre-service;

Automatically extract the borderline tumor information of medical image after pretreatment;

By the borderline tumor information fusion of extracting in corresponding elastic image;

Extract the elastic characteristic population of parameters of the elastic image that merges described borderline tumor information.

2. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 1, is characterized in that, described pre-service comprises the steps:

Medical image is carried out to speckle noise filtering processing;

The medical image of processing through speckle noise filtering is carried out to smoothing processing.

3. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 1, is characterized in that, described automatic extraction adopts the borderline tumor extracting method based on Chan-Vese model.

4. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 3, it is characterized in that, described elastic characteristic population of parameters comprises elasticity number parameter, and described elasticity number parameter comprises the elastic modulus mean value of area-of-interest, the mean value in the elastic modulus standard deviation of area-of-interest, the mean value of lesion region, the standard deviation of lesion region, surrounding tissue region and elasticity ratio.

5. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 4, is characterized in that, described area-of-interest comprises described lesion region and described surrounding tissue region, and the choosing method of described area-of-interest is:

According to the tumor's profiles curve calculation that merges the elastic image of described borderline tumor information, go out horizontal boundary rectangle;

By described boundary rectangle, to four direction continuation, forming one, to comprise tumour and size be that the rectangular area of 2～3 times, tumour is as described area-of-interest.

6. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 4, is characterized in that, the quantization method of the elastic modulus mean value of described area-of-interest is:

$\mathrm{ROIavg}=\frac{1}{N_{\mathrm{ROI}}}\underset{i=1}{\overset{N_{\mathrm{ROI}}}{\mathrm{\&Sigma;}}}{e}_i$

N in formula _rOIfor the number of all pixels in described area-of-interest, e _ibe the elastic mould value that pixel is corresponding, ROIavg has characterized the average hardness of area-of-interest;

The quantization method of the elastic modulus standard deviation of described area-of-interest is:

$\mathrm{ROIstd}=\sqrt{\frac{1}{N_{\mathrm{ROI}}-1}\underset{i=1}{\overset{N_{\mathrm{ROI}}}{\mathrm{\&Sigma;}}}{(e_i-\mathrm{ROIavg})}^2}$

N in formula _rOIfor the number of all pixels in described area-of-interest, e _ibe the elastic mould value that pixel is corresponding, ROIstd has characterized the degree of uniformity of elasticity distribution in area-of-interest;

The quantization method of the mean value of described lesion region is:

$\mathrm{TUMORavg}=\frac{1}{N_{\mathrm{TUMOR}}}\underset{i=1}{\overset{N_{\mathrm{TUMOR}}}{\mathrm{\&Sigma;}}}{e}_i$

N in formula _tUMORnumber for all pixels in described lesion region; e _ifor elastic mould value corresponding to pixel, TUMORavg has characterized the average hardness of tumour;

The quantization method of the standard deviation of described lesion region is:

$\mathrm{TUMORstd}=\sqrt{\frac{1}{N_{\mathrm{TUMOR}}-1}\underset{i=1}{\overset{N_{\mathrm{TUMOR}}}{\mathrm{\&Sigma;}}}{(e_i-\mathrm{ROIavg})}^2}$

N in formula _tUMORnumber for all pixels in described lesion region; e _ifor elastic mould value corresponding to pixel, TUMORstd has characterized the degree of uniformity of tumour elasticity distribution;

The quantization method of the mean value in described surrounding tissue region is:

$\mathrm{SUDavg}=\frac{1}{N_{\mathrm{SUD}}}\underset{i=1}{\overset{N_{\mathrm{SUD}}}{\mathrm{\&Sigma;}}}{e}_i$

N in formula _sUDnumber for all pixels in surrounding tissue region; e _ibe the elastic mould value that pixel is corresponding, SUDavg has characterized the average hardness of normal surrounding tissue;

The quantization method of described elasticity ratio is:

$\mathrm{Eratio}=\frac{\mathrm{TUMORavg}}{\mathrm{SUDavg}}$

In formula, TUMORavg is the average hardness of lesion region; SUDavg is the average hardness in surrounding tissue region, and Eratio has characterized the relative resilient intensity of variation in tumour and normal surrounding tissue region.

7. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 1 or 5, it is characterized in that, described elastic characteristic population of parameters also comprises elastic image parametric texture, and described elastic image parametric texture comprises: the elastic image parametric texture of lesion region and the elastic image parametric texture of gray level co-occurrence matrixes image-region.

8. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 7, it is characterized in that, the elastic image parametric texture of described lesion region comprises: histogram normalization variance, histogram degree of bias descriptor, histogram kurtosis descriptor, histogram consistance descriptor and entropy of histogram; The elastic image parametric texture of described gray level co-occurrence matrixes image-region comprises co-occurrence matrix energy descriptor, co-occurrence matrix contrast descriptor, co-occurrence matrix unfavourable balance square and co-occurrence matrix entropy.

9. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 8, is characterized in that, the choosing method of described gray level co-occurrence matrixes image-region is:

According to the tumor's profiles opisometer that merges the elastic image of described borderline tumor information, calculate minimum circumscribed rectangular region as described gray level co-occurrence matrixes image-region.

10. the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging according to claim 7, is characterized in that, the quantization method of described histogram normalization variance is:

$\mathrm{H\; var}=\frac{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(z_i-m)}^{2}p\left(z_i\right)}{{(L-1)}^2}$

Z in formula _irepresent that elastic mould value is mapped to a stochastic variable after [0,255]; M represents the average after elastic modulus corresponding to the interior all pixels of lesion region shines upon; p(z _i) be the grey level histogram of lesion region; L is possible number of greyscale levels, generally gets 256, H var and normalizes to scope [0,1], characterizes the discrete distribution situation of elastic mould value in tumour;

The quantization method of described histogram degree of bias descriptor is:

$\mathrm{Hskew}=\frac{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(z_i-m)}^{3}p\left(z_i\right)}{{(L-1)}^2}$

Z in formula _irepresent that elastic mould value is mapped to a stochastic variable after [0,255], m represents the average after elastic modulus corresponding to the interior all pixels of lesion region shines upon, p (z _i) be the grey level histogram of lesion region, L is possible number of greyscale levels, the asymmetric degree of Hskew token image histogram distribution;

The quantization method of described histogram kurtosis descriptor is:

$\mathrm{Hkurt}=\frac{\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(z_i-m)}^{4}p\left(z_i\right)}{{(L-1)}^2}$

Z in formula _irepresent that elastic mould value is mapped to a stochastic variable after [0,255], m represents the average after elastic modulus corresponding to the interior all pixels of lesion region shines upon, p (z _i) be the grey level histogram of lesion region, L is possible number of greyscale levels, Hkurt characterizes elastic image and is distributed in the approximate state while approaching average;

The quantization method of described histogram consistance descriptor is:

$\mathrm{Henergy}=\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{p}^2\left(z_i\right)$

Z in formula _irepresent that elastic mould value is mapped to a stochastic variable after [0,255], p (z _i) be the grey level histogram of lesion region, L is possible number of greyscale levels, Henergy characterizes the degree of uniformity of tumour elasticity distribution;

The quantization method of described entropy of histogram is:

$\mathrm{Hentropy}=-\underset{i=0}{\overset{L-1}{\mathrm{\&Sigma;}}}p\left(z_i\right){\log }_{2}p\left(z_i\right)$

Z in formula _irepresent that elastic mould value is mapped to a stochastic variable after [0,255], p (z _i) be the grey level histogram of lesion region, L is possible number of greyscale levels, Hentropy characterizes the homogeneity of tumour elasticity distribution;

The quantization method of described co-occurrence matrix energy descriptor is:

$\mathrm{ASM}=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{\left\{G(i,j)\right\}}^2$

In formula, G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K, ASM token image intensity profile degree of uniformity and texture fineness degree;

The quantization method of described co-occurrence matrix contrast descriptor is:

$\mathrm{CON}=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}{(i-j)}^{2}G(i,j)$

In formula, G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K, the degree of the sharpness of CON token image and the texture rill depth;

The quantization method of described co-occurrence matrix unfavourable balance square is:

$\mathrm{IDM}=\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}\frac{G(i,j)}{1+{(i-j)}^2}$

In formula, G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K, the homogeney of IDM token image, the intensity of variation of tolerance image texture part;

The quantization method of described co-occurrence matrix entropy is:

$\mathrm{ENT}=-\underset{i=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{K}{\mathrm{\&Sigma;}}}G(i,j){\log }_{2}G(i,j)$

In formula, G (i, j) represents that size is each element value of the gray level co-occurrence matrixes G of K * K, non-uniform degree or the complexity of ENT token image texture.

11. according to the extracting method of the tumour elastic characteristic based on ultrasonic elastograph imaging described in claim 1 or 5 or 6, it is characterized in that, described elastic characteristic population of parameters also comprises quantization parameter, and the quantization method of described quantization parameter value is:

Described lesion region is done dilation operation and obtained tumour peripheral organization region;

Described lesion region is done erosion operation and obtained pathology central area;

Be defined as follows parameter:

$\mathrm{TUMORsoft}=\frac{N_{\mathrm{soft}}}{N},$ $\mathrm{TUMORhard}=\frac{N_{\mathrm{hard}}}{N}$

In formula, N represents the number of all pixels in described lesion region, N _softrepresent that the interior elastic mould value of described lesion region is less than the pixel number of described ROIavg, Nhard represents that described lesion region elastic mould value is greater than the pixel number of described ROIavg,

$\mathrm{TUMORSUDsoft}=\frac{M_{\mathrm{soft}}}{M},$ $\mathrm{TUMORSUDhard}=\frac{M_{\mathrm{hard}}}{M}$

In formula, M represents the number of all pixels in described tumour peripheral organization region, M _softin expression tumour peripheral organization region, elastic mould value is less than the pixel number of described ROIavg, M _hardin representing, tumour peripheral organization region elastic mould value is greater than the pixel number of described ROIavg,

$\mathrm{CENTERhard}=\frac{L_{\mathrm{hard}}}{L}$

In formula, L represents the number of all pixels in described pathology central area, L _hardrepresent that described pathology central area Elastic value is greater than the pixel number of described ROIavg;

Elasticity scoring: when TUMORsoft >=90%, quantization parameter value Escore is 1;

When TUMORsoft-TUMORhard >=10%, quantization parameter value Escore is 2;

When TUMORhard-TUMORsoft >=10%, quantization parameter value Escore is 3;

When TUMORhard >=80%, TUMORSUFsoft >=70%, quantization parameter value Escore is 4;

When TUMORhard >=90%, TUMORSUFhard >=50%, quantization parameter value Escore is 5.

The extracting method of the 12. tumour elastic characteristics based on ultrasonic elastograph imaging according to claim 1, is characterized in that, described medical image is B ultrasonic image or CT image or MRI image or X-ray image.

Images