CN103177421B - Noise reduction processing method of ultrasound medical image - Google Patents

Abstract

The invention discloses a noise reduction processing method of an ultrasound medical image. The noise reduction processing method of the ultrasound medical image is characterized by comprising the following steps of calculating an edge intensity image E of a gray-scale ultrasound image I; calculating a mask image P in a smooth organization structure area of the gray-scale ultrasound image I; calculating an edge intensity threshold curve Kcurve of the gray-scale ultrasound image I along a depth direction of an ultrasonic probe scanning line; extracting a corresponding edge intensity threshold from the edge intensity threshold curve Kcurve and utilizing a diffusion coefficient function g to obtain a diffusion coefficient for any ith row image Ei of the edge intensity image E; and utilizing a diffusion equation (formula as follows, wherein all portions of the formula respectively stand for an image the gray-scale ultrasound image I evolving to time t to be formed into, an evolution time step length, a divergence operator and a gradient of the gray-scale ultrasound image I) to perform filtering processing on the image according to the generated diffusion coefficient. The noise reduction processing method of the ultrasound medical image has the advantages of enabling the ultrasound medical image to be capable of utilizing different edge intensity thresholds along the depth direction of probe scanning and achieving the effect of speckle noise suppression optimization.

Description

The noise reduction process method of ultrasonic medical image

Technical field

The present invention relates to technical field of medical image processing, especially relate to one kind and threshold value is spread to ash by automatic detection The noise reduction process method that rank ultrasonoscopy is carried out.

Background technology

It is usually present substantial amounts of speckle noise in medical science grayscale ultrasound image (B-mode), can bring to ultrasonograph quality Significantly decline, and mask the pathological changes of some vital tissues, this gives the diagnosis of doctor and identifies that some specific diseases are brought Larger difficulty, and have a risk failed to pinpoint a disease in diagnosis with mistaken diagnosis.

Generally to suppress the HFS of signal using the method for filtering, also can be by useful side while filtering off noise Edge information removes, and such as neighborhood averaging, median filtering method all can be excessively flat by the edge and details that have clinical meaning in image Sliding；Thus, some are rapidly developed based on the filtering algorithm that edge retains, such as anisotropic filtering (anisotropic Diffusion), the Medical Image Denoising technology such as BM3D and nonlocal means.This is carried in medical image post-processing stages High medical image quality provides a huge space.

Anisotropic filter is to be grown up based on the diffusion equation in physicss.Perona and Malik is in 1990 It is introduced into and retains (edge-preserving) smoothing filter as a kind of border in the middle of image procossing, its achievement is delivered In IEEE Transactions on Pattern Analysis and Machine Intelligence, the name in vol.12 In article for " Scale space and edge detection using anisotropic diffusion ".They Diffusion equation is as follows

I_t=div (g (I) I).

Wherein I representative image, I_tFor the partial derivative to the time t that evolves for the original image, g () is 0 related to gradient To 1 scalar, referred to as diffusion coefficient, div () represents divergence operator (divergence operator).

The evolution time, t can also regard the yardstick in metric space as, and that is, t is bigger, and image is more smooth.

According to explicit discretisation scheme, image I can be expressed as in the time t that evolves

I^t=I^t-1+ δ div (g (I) I),

Wherein δ is evolution time step, and in explicit aspect, it takes a positive number that can not be more than 0.25 to ensure numerical value Stability.When using implicit expression or half implicit expression discrete solution, δ can be not limited and take larger value.Implicit expression and half implicit expression from Scattered scheme refers to relevant documentation.

Perona and Malik proposes following two models for diffusion coefficient.

Wherein K is the threshold value of gradient modulus value.According to above-mentioned two Diffusion Coefficient Model, gradient modulus value | | I | | and diffusion Inversely, and when | | I | | is much larger than K, g's coefficient g becomes for 0；On the contrary, when | | I | | is during very little, g becomes for 1.Cause This, is in the big image-region of gradient modulus value, that is, the region that boundary information is strong, diffusion stopping；In the little region of gradient modulus value, Generally interior of articles or background, diffusion increases.It is achieved thereby that while retaining boundary information, other regions of image obtain To smooth effect.

Should see because diffusion coefficient g be a scalar so that flow (flux) is parallel with gradient direction, institute With Perona-Malik model from strict physical significance for be a kind of nonlinear isotropic (non-linear Inhomogeneous isotropic) diffusion model.The visible Weickert of specific explanations is 1998 by B.G.Teubner Chapter 1 in " Anisotropic Diffusion in Image Processing " book that Stuttgart publishes.For side Just, for the sake of, still Perona-Malik model is called anisotropy in the present invention.

Based on Perona-Malik model, follow-up derive much different smoothing filters, relatively more famous have Yu and Acton is in IEEE Transactions on Image the Processing, " Speckle that vol.11, no.11 deliver reducing anisotropic diffusion".Yu and Acton employs one kind and is called instantaneous Coefficient of variation (ICOV) is substituting Grad as rim detection index, and divides in advance using in image The one piece of noise Area generation diffusion threshold value cut out.

In the practical application of anisotropic filtering, since it is desired that multiple diffusion is to reach target effect, speed becomes resistance Hinder big restraining factors of its application.Acton in 1998 in IEEE Transactions on Image Processing, Deliver the article of entitled " Multigrid anisotropic diffusion " in vol.7, no.3, disclosed and be based on Multiple dimensioned anisotropic filtering device accelerates and removes to reach to lose the purpose of the pseudomorphism causing because of low frequency signal.

In order to realize anisotropic filtering truly, Weickert is in 1999 in International Journal of Computer Vision, has delivered an entitled " Coherence-enhancing in vol.31, no.2 The article of diffusion filtering ".In the text, Weickert employs diffusion tensor (diffusion tensor), and one Individual 2x2 matrix D, replaces diffusion coefficient g.Diffusion equation therefore can be write as：

I_t=div (D I).

Using diffusion tensor so that diffusion can carry out being not limited solely to gradient direction along any direction θ.Along any θ The diffusion in direction can be analyzed to spread along gradient direction with tangentially doing respectively.Diffusion tensor D can be analyzed to

Wherein v₁=I/ | | I | | is the normal direction of current point,For the tangential direction of current point, WhereinWithIt is respectively v₁Component in x-axis with y-axis, g₁And g₂It is respectively the diffusion along normal direction and tangential direction Coefficient, is defined as follows.

g₁=α

Wherein κ is consistency detection (coherence measure), and α is a very little on the occasion of Weickert suggestion takes 0.001, T is a threshold value more than 0.

Consistency detection κ is one from structure tensor (structure tensor), J, the nonnegative number of derivation.Structure tensor J is defined as：

I_xAnd I_yIt is respectively the Grad in x and y direction, K_ρIt is the Gaussian filter of ρ for a variance.Consistency detection κ is fixed Justice is：

κ=(μ₁-μ₂)²,

Wherein, μ₁And μ₂It is respectively the minimum and maximum eigenvalue of J.

In the directional anisotropic model of Weickert, because α is the value of a very little, diffusion in the normal direction It is considered that being suppressed always, diffusion strength in tangential direction to be determined by consistency detection κ.When κ → 0, ladder Degree is close value in normal direction and tangential direction, and diffusion in tangential direction is also suppressed.As κ ＞ ＞ T, now ladder The component in a direction for the degree is much larger than the component in another direction, that is, is most likely to be the place on border, tangential direction On diffusion coefficient by proximity values 1 with reach strengthen border effect.

Anisotropic filtering is that the speckle noise reducing in ultrasonic B-mode image provides a solution framework, but directly Apply mechanically and can bring the problem that false structure occurs in such as excessively smooth and image, this is primarily due to the bad threshold grasped in diffusion Caused by value, such as threshold value T in threshold k and Weickert model in Perona-Malik model.

Content of the invention

The present invention, according to the imaging characteristicses of ultrasound grayscale image, with anisotropic filtering as framework, proposes a kind of being based on certainly Dynamic detection diffusion threshold value realizes the noise reduction process method of grayscale ultrasound image, during solving current medical ultrasonic image noise reduction process Produce smooth or the technical problem that false structure etc. leads to medical ultrasonic image Quality Down occurs.

The present invention adopts the following technical scheme that realization：A kind of noise reduction process method of ultrasonic medical image, it includes step：

Obtain the grayscale ultrasound image I of a width ultrasonic medical image；

Calculate the edge strength image E of grayscale ultrasound image I；

Calculate the mask image P of the smooth tissue structural region in grayscale ultrasound image I；

Calculate grayscale ultrasound image I along ultrasonic probe scan line depth direction edge strength threshold curve K_curve；

Any i-th row image E to edge intensity image E_i, from edge strength threshold curve K_curveThe corresponding side of middle extraction Edge intensity thresholdAnd obtain diffusion coefficient, wherein edge strength threshold value using diffusion coefficient function gStrong for edge Degree threshold curve K_curveIn i-th element；

According to the diffusion coefficient generating, image is filtered process using diffusion equation, wherein diffusion equation is：I^t= I^t-1+ δ div (g I), I^tEvolve to the image of time t for grayscale ultrasound image I, δ is evolution time step, div () For divergence operator, I is the gradient of grayscale ultrasound image I.

Wherein, the step of described calculating mask image P includes：

Calculate the gradient modulus value image G of input grayscale ultrasound image I；

Gradient modulus value image G is done with binary conversion treatment, obtains a width structure mask image S；

Grayscale ultrasound image I is done with binary conversion treatment, obtains a width background mask image B；

Structure mask image S and background mask image B is carried out with NOR operation, obtains covering of smooth tissue structural region Code image P.

Wherein, the step of described calculating mask image P includes：

Calculate the gradient modulus value image G of input grayscale ultrasound image I；

Gradient modulus value image G is done with binary conversion treatment, obtains a width structure mask image S；

Structure mask image S is divided into the N number of not overlapping subregion of M, calculates to every in structure mask image S One sub-regions RS_m,nIn pixel value add up andIf,Value tissue default more than noise threshold T_S, then Think current sub-region RS_m,nFor structure body region, this subregion RS is set_m,nMiddle all pixels value is 1, otherwise arranges this sub-district Domain RS_m,nMiddle all pixels value is 0；

Grayscale ultrasound image I is done with binary conversion treatment, obtains a width background mask image B；

Grayscale ultrasound image I and background mask image B is respectively divided into the N number of not overlapping subregion of M；

To each of grayscale ultrasound image I subregion RI_m,n, calculate pixel averageIfIt is more than One default background noise threshold value T_B, then corresponding subregion RB in setting background mask figure B_m,nIn all pixels value be 1；

Structure mask image S and background mask image B is carried out with NOR operation, obtains covering of smooth tissue structural region Code image P.

Wherein, the gradient modulus value image of grayscale ultrasound image IWherein G_xAnd G_yFor grayscale ultrasound image I With horizontal Sobel Operator g_x, longitudinal Sobel Operator g_yConvolution.

Wherein, to grayscale ultrasound image I and horizontal Sobel Operator g_x, longitudinal Sobel Operator g_yBefore convolution algorithm, Using a low pass filter, grayscale ultrasound image I is smoothed.

Wherein, described edge strength imageWherein G_xAnd G_yBe respectively grayscale ultrasound image I with laterally Sobel Operator g_xWith longitudinal Sobel Operator g_yConvolution.

Wherein, described edge strength figure E can be calculated by equation below and obtain：

The gradient modulus value for grayscale ultrasound image I for wherein | the I |,²I is pixel I in grayscale ultrasound image I_i,jIn coordinate Second dervative on (i, j).

Wherein, described calculating edge strength threshold curve K_curveStep include：

By the i-th row image E any on edge strength image E_iIn belong to pixel E of smooth tissue structural region_i,jAccordingly Edge intensity value computing is put in effective edge strength list El；

Using in efficient frontier intensity list El, the intermediate value of each element or meansigma methodss are as image E_iCorresponding edge strength Threshold valueAnd it is stored in edge strength threshold curve K_curveIn i-th element；

To K_curveIn non-NULL element carry out first-order linear matching, and fitting result is stored in edge strength threshold curve K_curveIn.

Wherein, any i-th row image E on edge strength image E_iDiffusion coefficient function be g_i, g_iFor an edge strength The monotonic decreasing function of E, maximum is 1, and minima is 0；As image E_iArbitrary j-th pixel E in row_i,jValue be much larger than side Edge intensity thresholdWhen, g_iTend to 0；As image E_iArbitrary j-th pixel E in row_i,jValue be much smaller than edge strength threshold valueWhen, g_iTend to 1.

Wherein, any i-th row image E on edge strength image E_iDiffusion coefficient function be g_iFor：

Or be：

Wherein e is natural Exponents.

Compared with prior art, the present invention has the advantages that：

The present invention uses the smooth tissue structural region in grayscale ultrasound image, that is, neither the area of background again non-organization edge Domain, automatically generates a diffusion threshold curve contour with grayscale ultrasound image (each of diffusion threshold curve element generation The table diffusion threshold value of its corresponding image line) so that image can use along scanning probe depth direction different Diffusion threshold value, reaches the effect of optimization suppression speckle noise, and the fall of the top and the bottom of grayscale ultrasound image after process The consistent feature of effect of making an uproar.

Brief description

Fig. 1 is the schematic flow sheet that the present invention realizes grayscale ultrasound image noise-reduction method.

Specific embodiment

The present invention, according to the imaging characteristicses of ultrasound grayscale image, with anisotropic filtering as framework, proposes a kind of being based on and sweeps Retouch the dynamic diffusion threshold value of depth to realize grayscale ultrasound image noise-reduction method.

As shown in figure 1, in a preferred embodiment of the invention, the noise reduction process method realizing grayscale ultrasound image includes Step：

Step S101, obtain the grayscale ultrasound image of a width ultrasonic medical image from image acquisition equipment or image memory I.If obtaining goes ultrasonic medical image not to be gray scale image, first GTG is needed to be processed as grayscale ultrasound image.

Step S102, the edge strength image E of calculating grayscale ultrasound image I.Any point coordinate on grayscale ultrasound image I The edge strength E of (i, j)_i,jRepresent the probability that this point is located at object edge, numerical value is bigger represent that point (i, j) more has can The edge of object can be located at.

In an implementation, similar to Perona-Malik model, gradient modulus value is used as rim detection index. From gradient operator be Sobel (Sobel) operator.Laterally Sobel Operator g_xWith longitudinal Sobel Operator g_yCan represent respectively For

Edge strength imageWherein G_xAnd G_yIt is respectively grayscale ultrasound image I and horizontal Sobel Operator g_xWith longitudinal Sobel Operator g_yConvolution.In a method for optimizing, to grayscale ultrasound image I and horizontal Sobel Operator g_xWith longitudinal Sobel Operator g_yBefore doing convolution, using a low pass filter grayscale ultrasound image I is smoothed with Strengthen the precision of gradient calculation.

In another method for optimizing, ICOV value is used as rim detection index, that is,

Wherein | I | is gradient modulus value,²I_i,j=I_i+1,j+I_i-1,j+I_i,j+1+I_i,j-1-4I_i,jFor pixel I_i,jIn coordinate Second dervative on (i, j).Gradient modulus value | I | can be obtained by foregoing Sobel Operator.

Step S103, the mask image P of the smooth tissue structural region calculating in grayscale ultrasound image I.Smooth tissue is tied Neither the region at background also non-object (detection position of such as human body) edge in structure Regional Representative image.Calculate smooth tissue Configuration mask figure specifically comprises the steps of A1～A4.

Step A1, using gradient operator calculate input grayscale ultrasound image I gradient modulus value image G.In this optimal way Middle selection Sobel Operator.

The gradient modulus value image of grayscale ultrasound image IWherein G_xAnd G_yFor grayscale ultrasound image I and horizontal stroke To Sobel Operator g_x, longitudinal Sobel Operator g_yConvolution.

In another method for optimizing, to grayscale ultrasound image I and horizontal Sobel Operator g_x, longitudinal Sobel Operator g_yBefore convolution algorithm, using a low pass filter, grayscale ultrasound image I is smoothed strengthening gradient calculation Precision.

Step A2, gradient modulus value image G is done with binary conversion treatment, obtain a width structure mask image S.

Wherein S_i,jAnd G_i,jIt is respectively the structure mask image S and gradient modulus value image G value on coordinate (i, j), T_G For Grads threshold.Grads threshold T_GCan be set by hand according to characteristics of image, the statistical information also dependent on present image calculates life Become.Automatically obtained using Da-Jin algorithm (maximum variance between clusters, OTSU) algorithm in the present embodiment.

In structure mask image S, it is worth all point coordinates for 1 and constitutes structure body region, remaining Regional Representative Background area and organ-tissue region.

In order to reduce the noise in structure mask image S, morphology closed operator (morphological can be adopted Close) post processing is carried out to reduce noise to mask image S.

Step A3, grayscale ultrasound image I is done with binary conversion treatment, obtain a width background mask image B.

Wherein B_i,jAnd I_i,jIt is respectively value on coordinate (i, j) for the background mask image B and grayscale ultrasound image I, T_BFor Background noise threshold value.According to the characteristic of grayscale ultrasound image, background noise threshold value T in present implementation_BPreset for a craft Value.With structure mask image S process similar, background mask image B also can using morphology closed operator carry out post processing with Reduce noise.

Step A4, structure mask image S and background mask image B is carried out or non-(NOR) operation, final obtain smooth The mask image P (being also called smooth tissue configuration mask image P) in organizational structure region.

Wherein P_i,jFor value on coordinate (i, j) for the smooth tissue configuration mask image P.Similar, smooth tissue structure is covered Code image P also can carry out post processing using morphology closed operator and reduce noise.

Wherein, the quality in order to reduce operand and improve structure mask image S, is preferable to carry out method at another In, grayscale ultrasound image I piecemeal is processed, specifically includes step B1～B8.Wherein step B1 and B2 and algorithm steps described above Rapid A1 with A2 is identical, and specific descriptions are not repeated.

B1, using gradient operator calculate grayscale ultrasound image I gradient modulus value image G.

B2, gradient modulus value image G is done with binary conversion treatment, obtain a width structure mask image S.

B3, by structure mask image S, vertically decile divides M not overlapping region, along graduation such as parallel directions It is divided into N number of not overlapping region, therefore structure mask image S is divided into the N number of not overlapping subregion of M.

B4, to each of structure mask image S subregion RS_m,n, m and n represents in vertical direction and level respectively Direction sub-zones call number, calculates RS_m,nIn pixel value add up and V_RSm,n, that is, count RS_m,nIn how many pixel generation Table structure.IfValue tissue default more than noise threshold T_SThen it is assumed that current sub-region RS_m,nFor structure Body region, and set RS_m,nMiddle all pixels value is 1, otherwise sets RS_m,nMiddle all pixels value is 0.

B5, generation one width background mask image B.If the size of background mask image B is identical with grayscale ultrasound image I and sets All pixels value is 0.

B6 is similar with above-mentioned steps B3, by grayscale ultrasound image I and background mask image B respectively vertically and water Square to wait point divide the N number of not overlapping region of M.

B7, to each of grayscale ultrasound image I subregion RI_m,n, calculate pixel averageIfGreatly In default background noise threshold value T_BIf, the subregion RB of background mask figure B_m,nIn all pixels value be 1.

B8, structure mask image S and background mask image B is carried out or non-(NOR) operation, finally obtain smooth tissue Configuration mask image P.

Wherein P_i,jFor value on coordinate (i, j) for the smooth tissue configuration mask image P.

Step S104, calculating are along ultrasonic probe scan line depth direction (i.e. image vertical direction) grayscale ultrasound image I's Edge strength threshold curve K_curve(or diffusion threshold curve).Threshold curve K_curveIn i-th elementRepresent arbitrarily I-th row image E_iDiffusion threshold value (or edge strength threshold value).

Calculation procedure is as follows：If a height of H of grayscale ultrasound image I, a width of W；If i is the line index in image vertical direction Number, j is the column index number on image level direction；IfFor edge strength threshold curve K_curveThe edge of correspondence image i row Intensity threshold,

Step C1, take any i-th row image E on edge strength image E_i, calculate the corresponding edge strength threshold of image i row ValueOne efficient frontier intensity list El of setting, if list is sky；To image E_iUpper any one pixel E_i,jIf, phase Value P on coordinate (i, j) for the smooth tissue configuration mask image P answering_i,jEqual to 1, by E_i,jPut into effective edge strength list In El；The number of element in calculations list El, if less than default threshold value T_El, then setIt is worth for sky, otherwise setIntermediate value is the intermediate value of element in list El.

Step C2, to edge intensity threshold curve K_curveIn non-NULL element carry out first-order linear matching, and matching is tied Fruit is stored in K_curveIn.I.e. K_curveIn any one elementEdge strength threshold value corresponding to image i row.

In another method for optimizing, edge strength threshold curveValue be list El in each element meansigma methodss.

Step S105, any i-th row image E to edge intensity image E_i, from edge strength threshold curve K_curveIn carry Take corresponding edge strength threshold valueAnd apply mechanically diffusion coefficient function g and obtain diffusion coefficient.

In a method for optimizing, corresponding diffusion coefficient function g of image i row_iFor

E is natural Exponents.

In another method for optimizing, any row image E on edge strength image E_iCorresponding diffusion coefficient function g_i For

In another method for optimizing, any row image E on edge strength image E_iCorresponding diffusion coefficient function g_i For

In another method for optimizing, any row image E on edge strength image E_iCorresponding diffusion coefficient function g_i For

It can thus be seen that working as image E_iJ-th pixel E_i,jValue be much larger than edge strength threshold valueWhen, diffusion system Number g_iWill be close to 0, thus diffusion is suppressed；As image E_iJ-th pixel E_i,jValue be much smaller than edge strength threshold value When, diffusion coefficient g_iWill be close to 1, thus diffusion is allowed.

If it is desirable, in the evolutionary process of image, can be with repeat step S102～step S105 to obtain renewal Diffusion coefficient.

Step S106, the diffusion coefficient according to generation, are diffused to grayscale ultrasound image I processing according to diffusion equation, Obtain and export finally filtered image I^t.

Wherein, diffusion equation is：

I^t=I^t-1+δ·div(g·▽I).

I^tEvolve to the image of time t for grayscale ultrasound image I, δ is evolution time step, div () is divergence operator, I is the gradient of grayscale ultrasound image I, and the length of grayscale ultrasound image I evolution time is determined by practical application.

To sum up, the image-forming principle according to grayscale ultrasound image, speckle noise feature is in grayscale ultrasound image along probe Scan depths direction and change.The present invention according to this feature, using the smooth tissue structural region in grayscale ultrasound image, that is, Neither background and the region of non-organization edge, automatically generate (the diffusion of a diffusion threshold curve contour with grayscale ultrasound image Each of threshold curve element represents the diffusion threshold value of its corresponding image line) so that image is swept along probe Retouch depth direction and can reach, using different diffusion threshold values, the effect that optimization suppresses speckle noise.It was verified that through this The noise reduction of the top and the bottom of grayscale ultrasound image after invention process is consistent.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all essences in the present invention Any modification, equivalent and improvement made within god and principle etc., should be included within the scope of the present invention.

Claims (9)

1. a kind of noise reduction process method of ultrasonic medical image is it is characterised in that methods described includes step：

Obtain the grayscale ultrasound image I of a width ultrasonic medical image；

Calculate the edge strength image E of grayscale ultrasound image I；

Calculate the mask image P of the smooth tissue structural region in grayscale ultrasound image I；

Calculate grayscale ultrasound image I along ultrasonic probe scan line depth direction edge strength threshold curve K_curve；

Any i-th row image E to edge intensity image E_i, from edge strength threshold curve K_curveThe corresponding edge of middle extraction is strong Degree threshold valueAnd obtain diffusion coefficient, wherein edge strength threshold value using diffusion coefficient function gFor edge strength threshold value Curve K_curveIn i-th element；

According to the diffusion coefficient generating, image is filtered process using diffusion equation, wherein diffusion equation is：I^tEvolve to the image of time t for grayscale ultrasound image I, δ is evolution time step, div () For divergence operator,Gradient for grayscale ultrasound image I；

The step of described calculating mask image P includes：

Calculate the gradient modulus value image G of input grayscale ultrasound image I；

Gradient modulus value image G is done with binary conversion treatment, obtains a width structure mask image S；

Grayscale ultrasound image I is done with binary conversion treatment, obtains a width background mask image B；

Structure mask image S and background mask image B is carried out with NOR operation, obtains the mask figure of smooth tissue structural region As P；

Described calculating grayscale ultrasound image I is along the edge strength threshold curve K of ultrasonic probe scan line depth direction_curveStep Including：

Take any i-th row image E on edge strength image E_i, calculate the corresponding edge strength threshold value of image i rowSetting One efficient frontier intensity list El, if list is sky；To image E_iUpper any one pixel E_i,jIf, corresponding smooth tissue Value P on coordinate (i, j) for the configuration mask image P_i,jEqual to 1, by E_i,jPut in effective edge strength list El；Calculations list The number of element in El, if less than default threshold value T_El, then setIt is worth for sky, otherwise setIntermediate value is list The intermediate value of element in El.

2. according to claim 1 the noise reduction process method of ultrasonic medical image it is characterised in that described calculating mask image The step of P includes：

Calculate the gradient modulus value image G of input grayscale ultrasound image I；

Gradient modulus value image G is done with binary conversion treatment, obtains a width structure mask image S；

Structure mask image S is divided into the N number of not overlapping subregion of M, calculates to each of structure mask image S Subregion RS_m,nIn pixel value add up andIfValue tissue default more than noise threshold T_SThen it is assumed that Current sub-region RS_m,nFor structure body region, this subregion RS is set_m,nMiddle all pixels value is 1, otherwise arranges this subregion RS_m,nMiddle all pixels value is 0；

Grayscale ultrasound image I is done with binary conversion treatment, obtains a width background mask image B；

Grayscale ultrasound image I and background mask image B is respectively divided into the N number of not overlapping subregion of M；

To each of grayscale ultrasound image I subregion RI_m,n, calculate pixel averageIfPre- more than one If background noise threshold value T_B, then corresponding subregion RB in setting background mask figure B_m,nIn all pixels value be 1；

Structure mask image S and background mask image B is carried out with NOR operation, obtains the mask figure of smooth tissue structural region As P.

3. the noise reduction process method of ultrasonic medical image according to claim 1 or claim 2 is it is characterised in that grayscale ultrasound image I Gradient modulus value imageWherein G_xAnd G_yFor grayscale ultrasound image I and horizontal Sobel Operator g_x, longitudinal rope Bel operator g_yConvolution.

4. according to claim 3 the noise reduction process method of ultrasonic medical image it is characterised in that to grayscale ultrasound image I and horizontal Sobel Operator g_x, longitudinal Sobel Operator g_yBefore convolution algorithm, using a low pass filter to gray scale ultrasound Image I is smoothed.

5. according to claim 1 the noise reduction process method of ultrasonic medical image it is characterised in that described edge strength imageWherein G_xAnd G_yIt is respectively grayscale ultrasound image I and horizontal Sobel Operator g_xWith longitudinal Sobel Operator g_y Convolution.

6. according to claim 1 the noise reduction process method of ultrasonic medical image it is characterised in that described edge strength image E can be calculated by equation below and obtain：

$E=\sqrt{\frac{\frac{1}{2}{\left(\frac{|\▿I|}{I}\right)}^2-\frac{1}{16}{\left(\frac{{\▿}^{2}I}{I}\right)}^2}{{\[1+\frac{1}{4}\left(\frac{{\▿}^{2}I}{I}\right)\]}^2}}$

WhereinGradient modulus value for grayscale ultrasound image I,For pixel I in grayscale ultrasound image I_i,jOn coordinate (i, j) Second dervative.

7. according to claim 1 the noise reduction process method of ultrasonic medical image it is characterised in that described calculating edge strength Threshold curve K_curveStep include：

By the i-th row image E any on edge strength image E_iIn belong to pixel E of smooth tissue structural region_i,jCorresponding edge Intensity level is put in effective edge strength list El；

Using in efficient frontier intensity list El, the intermediate value of each element or meansigma methodss are as image E_iCorresponding edge strength threshold valueAnd it is stored in edge strength threshold curve K_curveIn i-th element；

To K_curveIn non-NULL element carry out first-order linear matching, and fitting result is stored in edge strength threshold curve K_curve In.

8. according to claim 1 the noise reduction process method of ultrasonic medical image it is characterised in that on edge strength image E Any i-th row image E_iDiffusion coefficient function be g_i, g_iFor the monotonic decreasing function of an edge strength E, maximum is 1, Little it is worth for 0；

As image E_iJ-th pixel E in row_i,jValue be much larger than edge strength threshold valueWhen, g_iTend to 0；

As image E_iJ-th pixel E in row_i,jValue be much smaller than edge strength threshold valueWhen, g_iTend to 1.

9. according to claim 1 the noise reduction process method of ultrasonic medical image it is characterised in that on edge strength image E Any i-th row image E_iDiffusion coefficient function be g_i：

Or be：

Wherein e is natural Exponents.