CN103606145B - GCV model based on interframe shape constraining segmentation aortic valve ultrasonic image sequence method - Google Patents

Abstract

The invention discloses a kind of GCV model based on interframe shape constraining segmentation aortic valve ultrasonic image sequence method, it comprises the steps: obtain one group of ultrasonoscopy and carry out Wiener filtering；Calculate the gradient vector flow of each image frame by frame and join CV model as energy constraint, obtaining GCV model；By defining initial constraint shapes, add in GCV model as energy constraint item, then minimize energy functional, obtain the segmentation result of the first two field picture；The aortic valve segmentation result of adjacent previous frame image is carried out rolling ball method filtering, and acquired results joins in GCV model as energy constraint item, is calculated the segmentation result of present frame.The present invention be directed to what echocardiographic short axis images carried out operating, not only greatly reduce the workload of doctor, and solve the problem seriously overflowed in aortic valve Ultrasound Image Segmentation in prior art, its segmentation result is very close to manual segmentation result, it is possible to simply and be efficiently partitioned into aortic valve.

Description

GCV model based on interframe shape constraining segmentation aortic valve ultrasonic image sequence Method

Technical field

The present invention relates to Medical Ultrasonic Image Segmentation method, a kind of GCV mould based on interframe shape constraining Type segmentation aortic valve ultrasonic image sequence method.

Background technology

In China, aortic valve class disease be a kind of most common be also the most complicated, dangerous cardiovascular disease, seriously endanger Do harm to the healthy of the mankind.Aortic valve is positioned at left ventricular outflow tract end and aorta intersection, and it acts like one " one-way cock ", anti-Hemostatic Oral Liquid refluxes, it is ensured that cardiac pumping function is the best.Due to the safe noinvasive of B ultrasonic, simple cheap, can The feature repeated, is widely used shape and the position of Echocardiographic Observation aortic valve at present in clinical diagnosis Put.In its ultrasonoscopy, the accurately segmentation of aortic valve is possible not only to assist doctor's clinical diagnosis, is also that image guiding is non-simultaneously Get involved the basis of valve class operation.But, owing to its ultrasonoscopy has low contrast, has that a large amount of speckle is ultrasonic and aorta Lobe constantly carries out the feature of opening and closing campaign, at present in clinical diagnosis, mainly by the doctor couple having a large amount of clinical experience Aortic valve ultrasonoscopy carries out manual segmentation one by one, and this not only considerably increases the workload of doctor, and for For clinical experience compares the doctor of shortcoming, carry out manual segmentation and be also one and compare the work being not easily done.

For solving all inconvenience that manual segmentation exists, many scholars have been had to propose multiple ultrasonic The method automatically or semi-automatically split of image.As at home, 2005, Shang Yefeng et al. proposed based on region shape priori The Geodesic Main method of moving contours segmentation ultrasonoscopy cardiac valve；Abroad, 2006, Sebastien Martin et al. Propose a kind of based on tricuspid semi-automatic method in active contour model segmentation ultrasonoscopy.But these methods existing, The long axial images being mainly both in ultrasoundcardiogram, use these methods that aortic valve ultrasonoscopy is split Time, due to ultrasonoscopy edge blurry and there is much noise, there is a large amount of spilling, so that segmentation result is inadequate in weak edge Accurately, credible.The problem existed based on existing method, researcher attempts the segmentation of the aortic valve to ultrasonic short axis images to be carried out point Cut, but the most little about the research of ultrasonic short axis images dividing method.

Summary of the invention

It is an object of the invention to provide a kind of GCV model based on interframe shape constraining segmentation aortic valve ultrasonoscopy sequence Row method, to solve cannot existing completely when carrying out aortic valve Ultrasound Image Segmentation for ultrasound long axis image of prior art existence The problem of whole and serious spilling.

It is an object of the invention to by following technical scheme realization:

A kind of GCV model based on interframe shape constraining segmentation aortic valve ultrasonic image sequence method, it includes following Step:

A) obtaining one group of continuous print aortic valve ultrasonic image sequence, quantity is M, and extracts the fan section of each two field picture Territory, the threshold value of non-sector region is 255；Then each two field picture is carried out Wiener filtering；

B), after carrying out Wiener filtering, the gradient vector flow of each two field picture is calculated；And it is the most fixed on each image One initial evolution curve of justice；

More than each described initial evolution curve method vector direction of calculating and each described gradient vector flow angular separation String value, joins CV model framework using gradient vector flow as energy constraint item, obtains the GCV model of each image；

C) four points of manual definition on the 1st two field picture, then utilize B-spline interpolation to form closed curve, by this Guan Bi Curve is as initial constraint shapes；Then shape matching function is utilized this initial constraint shapes to be joined as energy constraint item In the GCV model of the 1st two field picture, then minimizing the energy functional of this GCV model, the aortic valve obtaining the 1st two field picture is divided Cut result；

D) the aortic valve segmentation result of m two field picture being carried out rolling ball method filtering, the result obtained is as m+1 frame figure The non-initial constraint shapes of picture, then utilizes shape matching function this non-initial constraint shapes to be joined as energy constraint item In the GCV model of m+1 two field picture, then minimize the energy functional of this GCV model, obtain the aorta of m+1 two field picture Lobe segmentation result；Wherein, 1≤m≤M-1.

The method of the present invention, b) calculates the method for gradient vector flow of each two field picture specifically: minimize described in step Energy function $E_{\mathrm{gvf}}\left(g\right)=\∫\∫\mathrm{\μ}(u_x^2+u_y^2+v_x^2+v_y^2)+{|\▿f|}^2{|g-\▿f|}^2\mathrm{dxdy},$ Obtain gradient vector flow g (x, y)=(u (x,y),v(x,y))；Wherein μ is regularisation parameter, and (x y) represents the gradient map of ultrasonoscopy to f.

The method of the present invention, b) described in step, gradient vector flow is joined CV model framework as energy constraint item thus Obtain the formula of GCV model specifically:

E^gcv(φ,c₁,c₂)=E^cv(φ,c₁,c₂)+αcos<n(φ),g>∫_ΩH(φ)dxdy；

Wherein, $\begin{array}{c}{E}^{\mathrm{cv}}(\mathrm{\&phi;},c_1{,c}_2)=\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)|\&dtri;\mathrm{\&phi;}|\mathrm{dxdy}+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}\\ +{\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_1|}^{2}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_2|}^2(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}\end{array},$

The span of α is 0.3～1, and φ represents evolution curve, and n represents the normal vector direction of evolution curve φ, and g represents The gradient vector flow path direction of ultrasonoscopy,Take v=0, λ₁=λ₂=1。

The method of the present invention, d) rolling ball method filtering described in step is with the spherical junctions constitutive element master to m frame in mathematical morphology The segmentation result of arterial valve carries out burn into dilation operation, it may be assumed that R (F, B)=F o B=(F Θ B) B；In formula, F represents dividing of m frame Cutting result, B represents spherical junctions constitutive element, and its radius takes 15～22 pixels.

The method of the present invention, c) described in step by initial constraint shapes or d) described in step using non-initial constraint shapes as energy The formula that bound term joins GCV model is as follows:

$\begin{array}{c}E(\mathrm{\&phi;},c_1,c_2)=E^{\mathrm{gcv}}(\mathrm{\&phi;},c_1,c_2)+\mathrm{\&beta;}{E}_{\mathrm{shape}}\left(\mathrm{\&phi;}\right)\\ =\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)|\&dtri;\mathrm{\&phi;}|\mathrm{dxdy}+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}\\ +{\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_1|}^{2}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_2|}^2(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}\\ +\mathrm{\&alpha;}\cos <n\left(\mathrm{\&phi;}\right),g>{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+\mathrm{\&beta;}{\&Integral;}_{\mathrm{\&Omega;}}{(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B)}^2\mathrm{dxdy}\end{array};$

In formula, E_shape(φ) shape energy constraint item, φ are represented_BRepresent c) at the beginning of the initial constraint shapes of step or the non-of d) step Beginning constraint shapes, the value of β is 0.05～0.2；

Then use Euler-Lagrange equation to minimize the above-mentioned energy functional about φ to obtain:

$\frac{\&PartialD;\mathrm{\&phi;}}{\&PartialD;t}=\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)[\mathrm{\&mu;}\mathrm{div}\left(\frac{\&dtri;\mathrm{\&phi;}}{|\&dtri;\mathrm{\&phi;}|}\right)-{\mathrm{\&lambda;}}_1{(u-c_1)}^2+{\mathrm{\&lambda;}}_2{(u-c_2)}^2+\mathrm{\&alpha;}\frac{\&dtri;\mathrm{\&phi;}\&CenterDot;g}{|\&dtri;\mathrm{\&phi;}|\left|g\right|}+2\mathrm{\&beta;}(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B)]=0,$ It is optimum Solution is the aortic valve segmentation result of the 1st two field picture described in c) step or the aortic valve segmentation of d) m+1 two field picture described in step Result.

The present invention be directed to what echocardiographic short axis images carried out operating, by adding on the basis of CV model framework Enter energy constraint item based on GVF and energy constraint item based on interframe constraint shapes, overcome cutting procedure mesopetalum The impact of film motion, solves the problem that in aortic valve Ultrasound Image Segmentation, weak edge overflows；Compared with CV model, gradient is vowed The addition of amount stream has been greatly reinforced the weak marginal information of aortic valve ultrasonoscopy, can effectively suppress to overflow at fuzzy edge Problem；Interframe shape constraining not only reduces man-machine interactively, and evolution curve can be instructed to approach objective contour；Meanwhile, GVF and the addition of interframe shape constraining item, decrease the restriction to evolution curve initial position, can define at random Initially develop curve.

The inventive method only needs four points of manual definition, it becomes possible to be partitioned into aorta accurately from ultrasoundcardiogram Lobe, not only greatly reduces the workload of doctor, and solves in prior art serious in aortic valve Ultrasound Image Segmentation The problem overflowed, its segmentation result is compared with the manual segmentation result having experience doctor to be completed, and its aliasing error is only 4.83%, it is possible to simply and be efficiently partitioned into aortic valve.

Accompanying drawing explanation

Fig. 1 is the inventive method flow chart.

Fig. 2 (a) is one in acquired ultrasonic image sequence.

Fig. 2 (b) is the image after Fig. 2 (a) carries out Wiener filtering.

Fig. 3 is the aortic area of the first two field picture.

Fig. 4 is the gradient vector flow of Fig. 3.

Fig. 5 (a) is the initial constraint shapes defined on initial frame image.

Fig. 5 (b) is the non-initial constraint shapes defined on non-initial two field picture.

Fig. 6 (a) is the result of manual segmentation aortic valve ultrasonoscopy.

Fig. 6 (b) is the result using the inventive method segmentation aortic valve ultrasonoscopy.

Fig. 6 (c) is the result using CV model segmentation aortic valve ultrasonoscopy.

Detailed description of the invention

The present invention will be further described for below in conjunction with the accompanying drawings with one concrete example.

The present embodiment existsDual-Core CPU E5800@3.20GHz, video card is NVIDIA GeForce GT430NVIDIA GeForce GT430, inside saves as 2.00GB, operating system be Window XP computer in realize, whole Dividing method uses C++ and Matlab language to write.

The flow process of this method is carried out by step as shown in Figure 1:

(1) obtain one group of continuous print aortic valve ultrasonic image sequence, and extract the sector region of each frame, such as Fig. 2 A (), the threshold value of non-sector region is 255；Then the image acquired in each frame being carried out Wiener filtering, filtered result is such as Fig. 2 (b).

After carrying out Wiener filtering process, both can remove the speckle noise in ultrasonoscopy, limit can well have been retained again Edge information.

(2) GCV model is built:

(2.1) gradient vector flow of each two field picture (as shown in Figure 3) aortic area is calculated: by minimizing energy Function

$E_{\mathrm{gvf}}\left(g\right)=\&Integral;\&Integral;\mathrm{\&mu;}(u_x^2+u_y^2+v_x^2+v_y^2)+{|\&dtri;f|}^2{|g-\&dtri;f|}^2\mathrm{dxdy},$

Obtain this image gradient vector flow g (x, y)=(u (and x, y), v (x, y)), its result such as Fig. 4, arrow in Fig. 4 Direction represents gradient vector flow path direction；

Wherein, μ is regularisation parameter, and (x y) represents the gradient map of ultrasonoscopy to f.

(2.2) (the evolution curve on each two field picture is permissible to define an initial evolution curve on each two field picture at random Identical, it is also possible to different), by calculating on this image initial evolution curve method vector direction with gradient vector flow angular separation The cosine value of θ, adds CV model framework using gradient vector flow as new energy constraint item, sets up for each two field picture GCV model, it may be assumed that

E^gcv(φ,c₁,c₂)=E^cv(φ,c₁,c₂)+αcos<n(φ),g>∫_ΩH(φ)dxdy；

Wherein； $\begin{array}{c}{E}^{\mathrm{cv}}(\mathrm{\&phi;},c_1{,c}_2)=\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)|\&dtri;\mathrm{\&phi;}|\mathrm{dxdy}+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}\\ +{\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_1|}^{2}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_2|}^2(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}\end{array},$

The span of α is 0.3～1, for regulating gradient vector current field energy bound term evolution curve is affected big Little；φ represents evolution curve, and n represents the normal vector direction of evolution curve φ, and g represents the gradient vector flow side of ultrasonoscopy To,For simplified operation, generally take v=0, λ₁=λ₂=1。

(4) four points of the aortic area manual definition on the first two field picture, then utilize B-spline interpolation to form Guan Bi Curve, is initial constraint shapes, as shown in Fig. 5 (a), as energy constraint item, this initial constraint shapes is joined GCV mould In type, it may be assumed that

$\begin{array}{c}E(\mathrm{\&phi;},c_1,c_2)=E^{\mathrm{gcv}}(\mathrm{\&phi;},c_1,c_2)+\mathrm{\&beta;}{E}_{\mathrm{shape}}\left(\mathrm{\&phi;}\right)\\ =\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)|\&dtri;\mathrm{\&phi;}|\mathrm{dxdy}+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}\\ +{\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_1|}^{2}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_2|}^2(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}\\ +\mathrm{\&alpha;}\cos <n\left(\mathrm{\&phi;}\right),g>{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+\mathrm{\&beta;}{\&Integral;}_{\mathrm{\&Omega;}}{(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B)}^2\mathrm{dxdy}\end{array},$

Wherein, E_shape(φ) shape energy constraint item, φ are represented_BRepresenting initial constraint shapes, the span of β is 0.05 ～0.2, for regulating the size that evolution curve is affected by energy constraint item；

Then minimize the above-mentioned energy functional about φ with Euler-Lagrange equation to obtain:

$\frac{\&PartialD;\mathrm{\&phi;}}{\&PartialD;t}=\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)[\mathrm{\&mu;}\mathrm{div}\left(\frac{\&dtri;\mathrm{\&phi;}}{|\&dtri;\mathrm{\&phi;}|}\right)-{\mathrm{\&lambda;}}_1{(u-c_1)}^2+{\mathrm{\&lambda;}}_2{(u-c_2)}^2+\mathrm{\&alpha;}\frac{\&dtri;\mathrm{\&phi;}\&CenterDot;g}{|\&dtri;\mathrm{\&phi;}|\left|g\right|}+2\mathrm{\&beta;}(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B)]=0,$

Its optimal solution is the aortic valve segmentation result of the first two field picture.

(5) for the second two field picture, the aortic valve segmentation result of the first two field picture is carried out the filtered knot of rolling ball method Fruit is as the non-initial constraint shapes of this present frame (the i.e. second frame), then using this non-initial constraint shapes as energy constraint item Join in GCV model, it may be assumed that

$\begin{array}{c}E(\mathrm{\&phi;},c_1,c_2)=E^{\mathrm{gcv}}(\mathrm{\&phi;},c_1,c_2)+\mathrm{\&beta;}{E}_{\mathrm{shape}}\left(\mathrm{\&phi;}\right)\\ =\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)|\&dtri;\mathrm{\&phi;}|\mathrm{dxdy}+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}\\ +{\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_1|}^{2}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}{|u(x,y)-c_2|}^2(1-H\left(\mathrm{\&phi;}\right))\mathrm{dxdy}\\ +\mathrm{\&alpha;}\cos <n\left(\mathrm{\&phi;}\right),g>{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)\mathrm{dxdy}+\mathrm{\&beta;}{\&Integral;}_{\mathrm{\&Omega;}}{(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B)}^2\mathrm{dxdy}\end{array},$

Wherein, E_shape(φ) shape constraining energy term, φ are represented_BRepresent constraint shapes, the span of β be 0.05～ 0.2, size evolution curve affected for adjustable shape bound energy type；

Then minimize the above-mentioned energy functional about φ with Euler-Lagrange equation to obtain:

$\frac{\&PartialD;\mathrm{\&phi;}}{\&PartialD;t}=\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)[\mathrm{\&mu;}\mathrm{div}\left(\frac{\&dtri;\mathrm{\&phi;}}{|\&dtri;\mathrm{\&phi;}|}\right)-{\mathrm{\&lambda;}}_1{(u-c_1)}^2+{\mathrm{\&lambda;}}_2{(u-c_2)}^2+\mathrm{\&alpha;}\frac{\&dtri;\mathrm{\&phi;}\&CenterDot;g}{|\&dtri;\mathrm{\&phi;}|\left|g\right|}+2\mathrm{\&beta;}(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B)]=0,$

Its optimal solution is the aortic valve segmentation result of the second two field picture.

For all images after the first frame, all carry out according to the method for step (5), will the active of previous frame image Arteries and veins lobe segmentation result (as shown in curve I in Fig. 5 (b)) carry out the filtered result of rolling ball method as this present frame non-initial about Harness shape (as shown in curve II in Fig. 5 (b)), then is carried out calculating by the above-mentioned formula be given and develops, obtain this present frame Aortic valve segmentation result.

It is to enter the aortic valve segmentation result of previous frame by spherical junctions constitutive element in mathematical morphology that above-mentioned spin sends out filtering Row burn into dilation operation, it may be assumed that R (F, B)=F o B=(F Θ B) B, in formula, F represents the segmentation result of consecutive frame, and B represents spherical Structural elements, typically its radius take 15～22 pixels.

For feasibility and the accuracy of its segmentation result of checking this method further, the aortic valve of continuous 5 frames is surpassed Acoustic image has carried out manual segmentation respectively and has used the segmentation of CV model, and by the segmentation result of both and employing the inventive method The result carrying out splitting compares, result as shown in Figure 6: Fig. 6 (a) is experienced Ultrasonography doctor's manual segmentation aorta The result of lobe, Fig. 6 (b) is the result of the inventive method segmentation aortic valve, and Fig. 6 (c) is for using CV model segmentation aortic valve Result.It can clearly be seen that use the segmentation result that the inventive method obtains with Ultrasonography doctor's manual segmentation from figure Result is about the same, uses CV model to carry out ultrasonoscopy aortic valve segmentation and then there is a large amount of spilling in weak edge.

From above example result of the test, the present invention solves in prior art in aortic valve Ultrasound Image Segmentation The serious problem overflowed, can greatly reduce the workload of doctor, be partitioned into aortic valve the most efficiently.

Claims (3)

1. GCV model based on an interframe shape constraining segmentation aortic valve ultrasonic image sequence method, it is characterised in that bag Include following steps:

A) obtaining one group of continuous print aortic valve ultrasonic image sequence, quantity is M, and extracts the sector region of each two field picture, The threshold value of non-sector region is 255；Then each two field picture is carried out Wiener filtering；

B), after carrying out Wiener filtering, calculate the gradient vector flow of each two field picture, and on each image, define one the most at random Individual initial evolution curve；

By calculating the cosine value of each described initial evolution curve method vector direction and each described gradient vector flow angular separation, Gradient vector flow is joined CV model framework as energy constraint item, obtains the GCV model of each image；

C) four points of manual definition on the 1st two field picture, then utilize B-spline interpolation to form closed curve, by this closed curve As initial constraint shapes；Then utilize shape matching function that as energy constraint item, this initial constraint shapes is joined the 1st In the GCV model of two field picture, then minimize the energy functional of this GCV model, obtain the aortic valve segmentation knot of the 1st two field picture Really；

D) the aortic valve segmentation result of m two field picture being carried out rolling ball method filtering, the result obtained is as m+1 two field picture Non-initial constraint shapes, then utilizes shape matching function that as energy constraint item, this non-initial constraint shapes is joined m+ In the GCV model of 1 two field picture, then minimizing the energy functional of this GCV model, the aortic valve obtaining m+1 two field picture is divided Cut result；Wherein, 1≤m≤M-1；

Wherein, b) gradient vector flow is joined CV model framework as energy constraint item described in step thus obtain GCV model Formula specifically:

E^gcv(φ,c₁,c₂)=E^cv(φ,c₁,c₂)+αcos<n(φ),g>∫_ΩH(φ)dxdy；

Wherein,The value of α Scope is 0.3～1, and φ represents evolution curve, and n represents the normal vector direction of evolution curve φ, and g represents that the gradient of ultrasonoscopy is vowed Amount flow path direction,Take v=0, λ₁=λ₂=1；

C) by initial constraint shapes or d) non-initial constraint shapes is joined GCV mould as energy constraint item described in step described in step The formula of type is as follows:

$\begin{array}{c}E\left(\mathrm{\&phi;},c_1,c_2\right)=E^{gcv}\left(\mathrm{\&phi;},c_1,c_2\right)+{\mathrm{\&beta;E}}_{shape}\left(\mathrm{\&phi;}\right)\\ =\mathrm{\&mu;}{\&Integral;}_{\mathrm{\&Omega;}}\mathrm{\&delta;}\left(\mathrm{\&phi;}\right)|\&dtri;\mathrm{\&phi;}|dxdy+v{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)dxdy\\ +{\mathrm{\&lambda;}}_1{\&Integral;}_{\mathrm{\&Omega;}}|u\left(x,y\right)-c_1{|}^{2}H\left(\mathrm{\&phi;}\right)dxdy+{\mathrm{\&lambda;}}_2{\&Integral;}_{\mathrm{\&Omega;}}|u\left(x,y\right)-c_2{|}^2\left(1-H\left(\mathrm{\&phi;}\right)\right)dxdy\\ +\mathrm{\&alpha;}\cos <n\left(\mathrm{\&phi;}\right),g>{\&Integral;}_{\mathrm{\&Omega;}}H\left(\mathrm{\&phi;}\right)dxdy+\mathrm{\&beta;}{\&Integral;}_{\mathrm{\&Omega;}}{\left(\mathrm{\&phi;}-{\mathrm{\&phi;}}_B\right)}^{2}dxdy\end{array};$

In formula, E_shape(φ) shape energy constraint item, φ are represented_BRepresent c) step initial constraint shapes or d) step non-initial about Harness shape, the value of β is 0.05～0.2；

Then use Euler-Lagrange equation to minimize the above-mentioned energy functional about φ to obtain:

Its optimal solution is i.e. For the aortic valve segmentation result of the 1st two field picture described in c) step or the aortic valve segmentation result of d) m+1 two field picture described in step.

GCV model based on interframe shape constraining the most according to claim 1 segmentation aortic valve ultrasonic image sequence side Method, is characterized in that, b) calculates the method for gradient vector flow of each two field picture described in step specifically: minimize energy functionObtain gradient vector flow g (x, y)=(u (and x, y), v (x,y))；Wherein μ is regularisation parameter, and (x y) represents the gradient map of ultrasonoscopy to f.

GCV model based on interframe shape constraining the most according to claim 1 segmentation aortic valve ultrasonic image sequence side Method, is characterized in that, d) rolling ball method filtering described in step be with spherical junctions constitutive element in mathematical morphology to the aortic valve of m frame point Cut result and carry out burn into dilation operation, it may be assumed thatIn formula, F represents the segmentation knot of m frame Really, B represents spherical junctions constitutive element, and its radius takes 15～22 pixels.