Content of the invention
Cause for overcoming present in prior art liver area noise in ultrasonoscopy to be connected with other regions greatly
Difficulty extract complete problem, the invention provides a kind of method for extracting liver area in ultrasonoscopy.The technology of the present invention
Scheme is:In a kind of extraction ultrasonoscopy, the method for liver area, comprises the steps:
Step 1:Pretreatment, specifically, processes the even situation of uneven illumination in image, makes liver area in image
Brightness reaches unanimity;
Step 2:Using FCM_I algorithm segmentation figure pictures, specifically, using with the addition of neighborhood relevance information and prior shape
The FCM_I algorithms of information reduce the noise in image and complete image segmentation;
Step 3:The segmentation of foreground and background is obtained, specifically, according to FCM_I algorithm classifications result and half-tone information, is obtained
Obtain foreground area and the background area of image;
Step 4:The liver area in ultrasonoscopy is obtained, specifically, according to distribution and the shape of organ in liver, is obtained
The image of complete liver area.
In some embodiments, also include step 5:Further optimized using active contour algorithm and extract liver area
Edge.
In some embodiments, in the step 1, uneven illumination is processed even that method is:
Formula (1):I (x, y)=I (x, y)-min (NI (x, y)),
Wherein, I (x, y) is a pixel of image, NI (x, y)It is the neighborhood of I (x, y), by the pretreated figure of formula (1)
As dimmed, but the brightness of entirety reaches unanimity.
In some embodiments, in the step 2, the method for segmentation figure picture is:Using FCM algorithms, by continuous iteration
The method of the minimization of object function is made to obtain the value of the degree of membership of pixel in ultrasound wave liver area image, then according to maximum person in servitude
Category degree principle is divided to pixel, realizes the segmentation of image, and by adding neighborhood relevance information and prior shape information
Reduce the noise in image.
In some embodiments, the object function of the FCM_I algorithms of neighborhood relevance information is with the addition of in the step 2
For:
Formula (2) FCM_I:
Wherein, NjRepresent the neighborhood of pixel j, NrThat represented is neighborhood NjThe number of middle pixel, i are pixel class subscripts, its
In, have C classification, j is pixel subscript, wherein, have N number of pixel, m be regulation fuzzy membership weighted index, dij=|
|xj-vi| | for j-th pixel xjWith ith cluster center viGray scale difference value, dir=| | xr-vi| | for pixel xrPoly- with i-th
Class center viGray scale difference value, uijIt is xjRelative to viFuzzy membership, αijIt is the parameter for adjusting control neighbourhood effect.
In some embodiments, the fuzzy membership is the probability that pixel belongs to certain classification, its value belong to [0,
1], for a pixel, its fuzzy membership sum for belonging to each classification is 1, i.e. formula (3):
In some embodiments, extreme value of the formula (2) under the conditions of formula (3) is solved, specifically, using Lagrange multiplier
Method obtains fuzzy membership u to each variable derivationijAnd cluster centre viIterative formula:
Formula (4):
Formula (5):
By formula (4) and the iteration of formula (5), object function, i.e. formula (2) is made, is gradually tended to minimizing, is obtained each pixel
Degree of membership, so as to realize the segmentation of image.
In some embodiments, in using the FCM_I algorithms that with the addition of neighborhood relevance information, add prior shape
Information is limited, and specifically, the contour area for limiting liver is Rg, restriction background area is Rb, it is further provided with transitional region Ri,
Wherein, Rg=Ix> p1, Rb=Ix< p2, in formula, IxFor image, transition region RiFor IxIn except RgAnd RbRegion.
In some embodiments, using formula (6):
Add prior shape information, wherein, α1、α2And α3It is three different parameters, and α1< α2< α3.
Compared with prior art, the invention has the beneficial effects as follows:Algorithm proposed by the present invention is obtained preferably in an experiment
Extraction effect.The method is improve and is likely to occur liver area prospect using the prior information of liver shape in ultrasonoscopy
The contact of pixel, and the contact between the regional background is reduced, so that the region that extracts is more complete.In addition, algorithm
Have also combined the pretreatment of brightness irregularities background, and the operation of complete foreground area is extracted from multiple clusters, so as to
Obtain than more complete liver area.
Specific embodiment
Below in conjunction with drawings and Examples, the present invention will be described in further detail.It should be appreciated that described herein
Specific embodiment only in order to explain the present invention, is not intended to limit the present invention.
The side of liver area in a kind of extraction ultrasonoscopy that Fig. 1 to Fig. 7 show schematically show according to present disclosure
Method.
As shown in figure 1, in a kind of extraction ultrasonoscopy of present disclosure liver area method, using FCM algorithms, profit
Contact with priori control tactics region.But FCM is only a clustering algorithm, it is therefore desirable to coordinate other pretreatment
The task of extraction liver area can be completed with post processing.The following is liver area extraction scheme proposed by the present invention.
Step 1:Pretreatment, specifically, processes the even situation of uneven illumination in image, makes liver area in image
Brightness reaches unanimity.Not only noise is big for ultrasonoscopy, and is affected by brightness irregularities, as shown in Fig. 2 background luminance is dark
Region is easy to be mistaken for background.It is even that method is to pass through that the present invention processes uneven illumination:Formula (1):I (x, y)=I (x, y)-
min(NI (x, y)) place buries the even situation of uneven illumination in image, wherein, I (x, y) is a pixel of image, NI (x, y)It is I
The neighborhood of (x, y), the image after pre- place buries are as shown in Figure 3, although liver area is dimmed, but the brightness of entirety tends to
Unanimously, more using the extraction of liver area.
Step 2:Using FCM_I algorithm segmentation figure pictures, specifically, using with the addition of neighborhood relevance information and prior shape
The FCM algorithms of information reduce the noise in image and complete image segmentation.In this embodiment of the present invention, divide in step 2
Actual every the method for image it is:Using FCM algorithms, the method for the minimization of object function is made to obtain ultrasound wave liver by continuous iteration
In dirty district area image, the value of the degree of membership of pixel, then divides to pixel according to maximum membership grade principle, realizes image
Segmentation, and reduce the noise in image by adding neighborhood relevance information and prior shape information.
Traditional, if the object function of image I, FCM is:Formula (7):FCM:Wherein, i is
Pixel class subscript, wherein, has C classification, and j is pixel subscript, wherein, has N number of pixel, and m is to adjust fuzzy membership
Weighted index, dij=| | xj-vi| | for j-th pixel xjWith ith cluster center viGray scale difference value, uijIt is xjRelative to
viFuzzy membership.
Fuzzy membership is the probability that pixel belongs to certain classification, and its value belongs to [0,1], and for a pixel, which belongs to
In each classification fuzzy membership sum be 1, i.e.,
Formula (3):
Extreme value of the solution formula (7) under the conditions of formula (3), specifically, using method of Lagrange multipliers to each variable derivation,
Obtain fuzzy membership uijAnd cluster centre viIterative formula:
Formula (8):
Formula (9):
By formula (8) and the iteration of formula (9), object function, i.e. formula (7) is made, is gradually tended to minimizing, is obtained each pixel
Degree of membership, so as to realize the segmentation of image.
The greatest problem of original FCM algorithms is that object function comprising any spatial information, does not cause affected by noise
Very big.Only using the information of pixel itself, this problem that brings is that anti-noise ability is poor, due to noise usually self-existent,
Distinguish substantially with surrounding pixel, therefore noise is easily independently formed a region by algorithm, so that originally complete region
In be filled with hole.Another problem of original FCM is when the irregular colour in a region is even, and it is divided into several areas easily
Domain, increases the difficulty of later stage process.Then, make an inventive point of the present invention, in this embodiment of the present invention, in FCM
Neighborhood relevance information is with the addition of on algorithm, and the object function of the FCM algorithms for obtaining with the addition of neighborhood relevance information is:
Formula (2) FCM_I:
Wherein, NjRepresent the neighborhood of pixel j, NrThat represented is neighborhood NjThe number of middle pixel, i are pixel class subscripts, its
In, have C classification, j is pixel subscript, wherein, have N number of pixel, m be regulation fuzzy membership weighted index, dij=|
|xj-vi| | for j-th pixel xjWith ith cluster center viGray scale difference value, dir=| | xr-vi| | for pixel xrPoly- with i-th
Class center viGray scale difference value, uijIt is xjRelative to viFuzzy membership, αijIt is the parameter for adjusting control neighbourhood effect.
Fuzzy membership is the probability that pixel belongs to certain classification, and its value belongs to [0,1], and for a pixel, which belongs to
In each classification fuzzy membership sum be 1, i.e.,
Formula (3):
Extreme value of the solution formula (2) under the conditions of formula (3), specifically, using method of Lagrange multipliers to each variable derivation,
Obtain fuzzy membership uijAnd cluster centre viIterative formula:
Formula (3):
Formula (4):
By formula (3) and the iteration of formula (4), object function, i.e. formula (2) is made, is gradually tended to minimizing, is obtained each pixel
Degree of membership, so as to realize the segmentation of image.
The FCM_I algorithms that with the addition of neighborhood relevance can be effectively increased anti-noise ability, but the method interpolation is complete
The consistent item of office, not whole region be required for, therefore make another inventive point of the present invention, in this embodiment of the present invention
In, add prior shape information in FCM_I algorithms, that is, add limiting, specifically, contact in region, and the connection outside region
It is not consistent.The restriction of interpolation determined according to different objects, the wheel that be limited to a liver of the present invention in liver is extracted
Wide region Rg, and it is R to limit background areab, another different in view of the liver area in every piece image, in liver area
Transition region R be also add outside domain and background areai, wherein, Rg=Ix> p1, Rb=Ix< p2, in formula, IxFor image, transition region Ri
For IxIn except RgAnd RbRegion.In order to determine liver area, doctor is allowed to sketch out the liver area in several ultrasonoscopys
Come, and be superimposed ballot.The many regions of gained vote are considered as liver area, and it is background to win the vote less or do not have the region that wins the vote, its
His is transitional region.
As preferred, region RgThe contact of interior emphasis prospect, needs to strengthen impact of the foreground area to surrounding, that is, works as vi
For prospect center when, using less α values, increase the proportion of expression formula (2) latter half.On the other hand, region RgInterior minimizing
The contact of background, that is, work as viFor background center when, using larger α values, reduce the proportion of expression formula (2) latter half.Due to
Expression formula iteration tends to minimum, therefore, using α values defined herein, can make RgIn extraction liver area more complete.
Fig. 4 shows the extraction process of liver prior information.Wherein (a) is the liver area of manual extraction, (b) is (a)
The result of binaryzation, is (c) by the result after the figure superposition in (b).Figure in Fig. 4 (c) is Ix.
As preferred, using formula (6):
Add prior shape information, wherein, α1、α2And α3It is three different parameters, and α1< α2< α3.
In step 2 of the present invention, due to the addition of prior information and neighborhood relevance information, compare the liver area of acquisition
Completely.
Step 3:Obtain the substantially segmentation of foreground and background.Specifically, according to classification results and the gray scale of FCM_I algorithms
Information, the foreground and background region for obtaining.
Step 4:Obtain the liver area in ultrasonoscopy.Specifically, in abdominal ultrasound images, liver is typically maximum
Organ, using this priori, according to distribution and the shape of organ in liver, obtain the image of complete liver area.
As preferred, also include step 5:Further optimize the edge for extracting liver area using active contour algorithm.
In order to verify algorithm proposed by the present invention, the method for the present invention is tested on the ultrasound image.Such as Fig. 5 extremely
Shown in Fig. 7, image has larger noise, and occurs in that the situation of brightness irregularities.Algorithm and original FCM and FCM_
S is compared.Clusters number 3, α=1 in FCM_S, it is 5 × 5 that the present invention proposes the pretreatment Size of Neighborhood of algorithm, parameter alpha1
=4, α2=1, α3=0.1.Fig. 5 to Fig. 7 is the example of three kinds of FCM algorithm cluster results, and wherein Fig. 5 is the result of original FCM,
Fig. 6 is the result of FCM_S, and Fig. 7 is the result of FCM_I algorithms, can be seen that from Fig. 5 to Fig. 7 the results contrast of FCM is broken, and
The result of FCM_S then complete some, the result of FCM_I algorithms proposed by the present invention is best.It can be seen that constraint proposed by the present invention
Method can improve the classifying quality of FCM.
In order to more accurately be compared data, by experiment used in ultrasonoscopy carried out artificial segmentation, extract
Right-on hepatic portion is used as reference object.Then liver area is extracted using above-mentioned algorithm, and with correct with reference to knot
Fruit is compared, and obtains and accuracy P=TP/ (TP+FP), recall rate R=TP/AP, and wherein TP is the correct pixel count of classification, and FP is
The pixel count of classification error, AP are the sums of liver area pixel.Accuracy P of FCM_I algorithms and call together as can be seen from Table 1
Rate R of returning increases compared with FCM and FCM_I algorithms.
The Performance comparision of 1 three kinds of FCM of table
In a kind of extraction ultrasonoscopy of present disclosure, the method for liver area is obtained in an experiment
Really.The method is improve and is likely to occur liver area foreground pixel using the prior information of liver shape in ultrasonoscopy
Contact, and the contact between the regional background is reduced, so that the region that extracts is more complete.In addition, algorithm herein in connection with
The pretreatment of brightness irregularities backgrounds, and the operation of complete foreground area is extracted from multiple clusters, so as to obtain
Than more complete liver area.
Described above illustrates and describes the preferred embodiments of the present invention, as previously mentioned, it should be understood that not office of the invention
It is limited to presently disclosed form, is not to be taken as the exclusion to other embodiment, and can be used for various other combinations, modification
And environment, and can be entered by the technology or knowledge of above-mentioned teaching or association area in invention contemplated scope of the present invention
Row is changed.And change that those skilled in the art are carried out and change be without departing from the spirit and scope of the present invention, then all should be in the present invention
In the protection domain of claims.